From d9c319984511640b6767f2c3c1d93c4f1e24ea2a Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:00:05 -0400 Subject: [PATCH 001/105] Add one shared chronicle-first env helper Chronicle's operational stores migrate by dual-run (chronicle#143, mechanism 3). Every env read was ledger-first with no chronicle-named alternative, and each of the three modules that read configuration had grown its own helper. Add chronicle/env.py: env_value/env_flag expand a name into CHRONICLE_, LEDGER_, POLICYENGINE_LEDGER_ and return the first set value, warning once per process with ChronicleEnvDeprecationWarning when a ledger-era name supplied it. Names outside those three prefixes, such as POLICYENGINE_SUPABASE_URL, are read literally so the helper renames the ledger-era surface only. The warning subclasses FutureWarning, not DeprecationWarning, so operators running the CLI actually see it. The Supabase schema name stays "ledger"; only the env var that overrides it moves to CHRONICLE_SCHEMA. Renaming the schema is a later slice. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 74 ++++++++++------------ chronicle/__init__.py | 1 + chronicle/env.py | 120 ++++++++++++++++++++++++++++++++++++ chronicle/source_package.py | 25 ++------ db/pe_source_inventory.py | 21 +++---- db/supabase_client.py | 23 +++---- 6 files changed, 174 insertions(+), 90 deletions(-) create mode 100644 chronicle/env.py diff --git a/PROGRESS.md b/PROGRESS.md index 24773009..e46b096c 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -1,51 +1,41 @@ -# Lane C5 progress +# Operational rename, slice 1 (chronicle#143, mechanism 3) + +Lane C5's handoff notes previously lived here; its durable record is +`LANE_C5_REPORT.md`. This file now tracks the active lane on this branch. ## State -- Branch: `be-2025-vintages` from `origin/main` at `5c15bfd`. -- Worktree inputs are staged under `.lane-raw/` and must remain uncommitted. -- Lane C5 is complete, validated, independently reviewed, and ready for handoff. -- The requested staged C2 report is absent, but root `LANE_C2_REPORT.md` is byte-identical - to the sibling lane's staged copy (SHA-256 `4590e0dc...50f06e7`) and is the pattern used. +- Branch: `ops-rename-slice1`, cut from `origin/main` at `ff3efd3`. +- Scope: env names, R2 bucket configurability, `ledger.db` -> `chronicle.db`, + and the docs for all three. Code and docs only; no infrastructure changes. +- Out of scope and deliberately untouched: the `ledger` console-script alias, + the Supabase `"ledger"` schema and mirror table names, governance role ids and + concept authorities, hash domains and schema ids, anything under `releases/`. ## Done -- Read the repository Chronicle boundary rules in `AGENTS.md`. -- Read `.lane-raw/SOURCES.md` and confirmed all five named publisher artifacts are present. -- Confirmed the worktree is otherwise clean apart from `.lane-raw/` and the shared `.venv` link. -- Verified all five staged artifact SHA-256 pins exactly. -- Mapped FPB workbook cells: 990 facts across T01/T06/T07/T11/T17/T24, with - 2022–2025 observations and 2026–2031 `source_projection` facts. -- Confirmed PDF boundary evidence: printed page 19 calls 2026 the first projection year; - annex table units appear on printed pages 45, 48, 49, 53, 58, and 65. -- Chosen Eurostat layout: two vintage-specific source-package aliases share new manifest - entries, preserving the prior package YAMLs, raw bytes, and fact outputs unchanged. -- Reproduced the Statbel curator logic: 18 NUTS1 × sex × age-band cells totaling 11,825,551. -- Added the hash-pinned FPB workbook and publication PDF plus the - `fpb-economic-outlook-2026-2031-june-2026` package alias. -- Built 990 line-specific publisher facts (99 per year): 396 observations for - 2022–2025 and 594 `source_projection` facts for 2026–2031. -- Passed FPB `validate-package` and `build-suite`: 990 facts, full cell lineage, - zero acceptance errors, and pinned 2025 cells 320578 / 77771 / 5602 million euro. -- Re-ran the Statbel 2026 curator logic on the 2025 ZIP and added the hash-pinned - raw capture plus its deterministic 18-row curated CSV. -- Passed Statbel 2025 `validate-package` and `build-suite`: 18 facts totaling - 11,825,551, 66 constraints, full lineage, and zero acceptance errors. -- Added the Eurostat `gov_10a_taxag` 2025 and `spr_exp_func` 2024 manifest - entries plus vintage-specific package aliases, without modifying either - prior artifact or prior package specification. -- Passed both new Eurostat package validations and suite builds: 12 tax facts - and 9 ESSPROS facts, full lineage, and zero acceptance errors. -- Extended Belgium and Eurostat regressions for FPB table counts/cells and - assertion boundary, vintage non-overlap, prior-output digests, Statbel pins, - and the declared 0.25% Statbel/FPB population comparison tolerance. -- Passed 43 focused tests and the full merged-bundle regression: 157,177 facts, - 148 packages, zero aggregate-key duplicates, and expected goldens throughout. -- Recorded pins, counts, boundary evidence, curator commands, validation tails, - and consumer fact families in `LANE_C5_REPORT.md`. -- Passed independent `ledger-source-fidelity` and `ledger-boundary` reviews with - no required corrections. +- Read `AGENTS.md`, `docs/storage-architecture.md`, + `docs/agent-source-package-harness.md`, and the mechanism-3 migration spec in + the first comment of PolicyEngine/chronicle#143. +- Enumerated every ledger-named env read in tracked Python: the four real + variables (`LEDGER_SOURCE_ARTIFACT_CACHE_DIR`, `LEDGER_SOURCE_ARTIFACT_FETCH`, + `LEDGER_PE_US_DATA_ROOT`, `LEDGER_PE_UK_DATA_ROOT`) plus + `POLICYENGINE_LEDGER_SCHEMA`. `LEDGER_MIRROR_TABLES`, + `LEDGER_MIRROR_PRIMARY_KEYS`, and `LEDGER_DB_SCHEMA_VERSION` are module + constants, not env reads, and name out-of-scope surfaces. +- Added `chronicle/env.py`: one shared `env_value`/`env_flag`/`env_names` + helper reading `CHRONICLE_` first, then `LEDGER_` and + `POLICYENGINE_LEDGER_` with a once-per-process + `ChronicleEnvDeprecationWarning` naming the preferred variable. +- Replaced all three ad-hoc helpers (`db/supabase_client._env`, + `chronicle/source_package._env_value`/`_truthy_env`, + `db/pe_source_inventory._env_value`) with the shared helper. ## Next -- None; ready for handoff. No push was performed. +- Make R2 bucket names configurable with unchanged `ledger-*` defaults. +- Emit `chronicle.db` for new suite outputs; keep reading `ledger.db`. +- Fix the backwards fallback statement in the docs and sweep every env-name, + bucket-name, and db-filename mention in README/AGENTS/docs. +- Add the hermetic dual-read tests; run pytest, ruff check, ruff format --check. +- Push and open the PR. Do not merge. diff --git a/chronicle/__init__.py b/chronicle/__init__.py index 04d50943..e3ddfb06 100644 --- a/chronicle/__init__.py +++ b/chronicle/__init__.py @@ -13,6 +13,7 @@ "consumer_contract", "core", "database", + "env", "facts", "harness", "jurisdictions", diff --git a/chronicle/env.py b/chronicle/env.py new file mode 100644 index 00000000..80667f4d --- /dev/null +++ b/chronicle/env.py @@ -0,0 +1,120 @@ +"""Environment configuration for the Chronicle rename window. + +Chronicle's operational stores migrate by dual-run (PolicyEngine/chronicle#143, +mechanism 3): every configuration variable gets a ``CHRONICLE_``-prefixed name +that is read first, while the ledger-era ``LEDGER_`` and +``POLICYENGINE_LEDGER_`` names keep working behind a deprecation warning. That +window lets downstream publish flows migrate on their own schedule instead of +breaking the moment Chronicle ships a rename. + +Names that carry none of those three prefixes are read literally: this helper +renames the ledger-era surface, not every PolicyEngine variable. +""" + +from __future__ import annotations + +import os +from typing import TypeVar +import warnings + +__all__ = [ + "CHRONICLE_ENV_PREFIX", + "ChronicleEnvDeprecationWarning", + "LEGACY_ENV_PREFIXES", + "env_flag", + "env_names", + "env_value", + "reset_env_deprecation_state", +] + +CHRONICLE_ENV_PREFIX = "CHRONICLE_" + +# Ordered most specific first so prefix stripping is unambiguous. +LEGACY_ENV_PREFIXES = ("POLICYENGINE_LEDGER_", "LEDGER_") + +TRUTHY_ENV_VALUES = frozenset({"1", "true", "yes", "on"}) + + +class ChronicleEnvDeprecationWarning(FutureWarning): + """A ledger-era environment variable supplied a Chronicle setting. + + Subclasses :class:`FutureWarning` rather than :class:`DeprecationWarning` + so the notice reaches operators running the CLI, who are the people who + have to move the variable. ``DeprecationWarning`` is silenced by default + outside ``__main__``. + """ + + +_Default = TypeVar("_Default") + +_WARNED_LEGACY_NAMES: set[str] = set() + + +def _env_suffix(name: str) -> str | None: + """Return the rename-window suffix of ``name``, or None if it has none.""" + for prefix in (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES): + if name.startswith(prefix) and len(name) > len(prefix): + return name[len(prefix) :] + return None + + +def env_names(name: str) -> tuple[str, ...]: + """Return the lookup order for ``name``. + + The chronicle-preferred name comes first, then the ledger-era names that + remain accepted during the migration window. A name outside the rename + window is returned unchanged, as its own single-element lookup order. + """ + suffix = _env_suffix(name) + if suffix is None: + return (name,) + return ( + f"{CHRONICLE_ENV_PREFIX}{suffix}", + *(f"{prefix}{suffix}" for prefix in LEGACY_ENV_PREFIXES), + ) + + +def _warn_legacy(found: str, preferred: str) -> None: + """Warn once per process that a ledger-era variable supplied a value.""" + if found in _WARNED_LEGACY_NAMES: + return + _WARNED_LEGACY_NAMES.add(found) + warnings.warn( + f"{found} is a ledger-era Chronicle environment variable; " + f"set {preferred} instead. The old name is still honored during the " + "Chronicle rename window and will be removed once consumers migrate.", + ChronicleEnvDeprecationWarning, + stacklevel=3, + ) + + +def reset_env_deprecation_state() -> None: + """Forget which legacy names have already warned. Test-support hook.""" + _WARNED_LEGACY_NAMES.clear() + + +def env_value(*names: str, default: _Default = None) -> str | _Default: + """Read the first set value across ``names``, chronicle-preferred first. + + Each name is expanded through :func:`env_names`, so a caller can pass the + chronicle name and still pick up a value set under a ledger-era name. + Empty values are treated as unset, matching the helpers this replaces. + """ + for name in names: + candidates = env_names(name) + preferred = candidates[0] + for candidate in candidates: + value = os.environ.get(candidate) + if value: + if candidate != preferred: + _warn_legacy(candidate, preferred) + return value + return default + + +def env_flag(*names: str) -> bool: + """Return whether the first set value across ``names`` reads as true.""" + value = env_value(*names) + if value is None: + return False + return value.strip().lower() in TRUTHY_ENV_VALUES diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 5f6a5709..26cc8639 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -3,7 +3,6 @@ from __future__ import annotations import hashlib -import os from dataclasses import dataclass, replace from importlib.resources import files from io import BytesIO @@ -32,6 +31,7 @@ AggregateFact, build_label, ) +from chronicle.env import env_flag, env_value from chronicle.epoch import SCHEMA_IDS, schema_id from chronicle.sources.cells import ( SourceArtifactMetadata, @@ -392,8 +392,8 @@ "usda_snap/fy2025_monthly_state_caseloads" ), } -SOURCE_ARTIFACT_CACHE_ENV = "LEDGER_SOURCE_ARTIFACT_CACHE_DIR" -SOURCE_ARTIFACT_FETCH_ENV = "LEDGER_SOURCE_ARTIFACT_FETCH" +SOURCE_ARTIFACT_CACHE_ENV = "CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR" +SOURCE_ARTIFACT_FETCH_ENV = "CHRONICLE_SOURCE_ARTIFACT_FETCH" DEFAULT_SOURCE_ARTIFACT_CACHE_DIR = ( Path.home() / ".cache" / "policyengine-chronicle" / "source-artifacts" ) @@ -2257,7 +2257,7 @@ def _read_source_artifact_content( if cache_path.exists(): return cache_path.read_bytes() - if not _truthy_env(SOURCE_ARTIFACT_FETCH_ENV): + if not env_flag(SOURCE_ARTIFACT_FETCH_ENV): raise FileNotFoundError( f"Source artifact {spec['filename']} is not packaged and was not " f"found in {cache_path}. Set {SOURCE_ARTIFACT_FETCH_ENV}=1 to fetch " @@ -2279,7 +2279,7 @@ def _read_source_artifact_content( def _source_artifact_cache_path(spec: dict[str, Any]) -> Path: cache_root = Path( - _env_value( + env_value( SOURCE_ARTIFACT_CACHE_ENV, default=DEFAULT_SOURCE_ARTIFACT_CACHE_DIR, ) @@ -2316,21 +2316,6 @@ def _validate_source_artifact_sha( ) -def _env_value(*names: str, default: str | Path) -> str | Path: - for name in names: - value = os.environ.get(name) - if value: - return value - return default - - -def _truthy_env(*names: str) -> bool: - return any( - os.environ.get(name, "").strip().lower() in {"1", "true", "yes", "on"} - for name in names - ) - - def _single_archive_member(archive: ZipFile, *, suffixes: tuple[str, ...]) -> str: members = [ name diff --git a/db/pe_source_inventory.py b/db/pe_source_inventory.py index a52fff23..f64ed870 100644 --- a/db/pe_source_inventory.py +++ b/db/pe_source_inventory.py @@ -2,14 +2,15 @@ from __future__ import annotations -import os from pathlib import Path +from chronicle.env import env_value + from .schema import Jurisdiction from .source_files import SourceArtifactSpec, make_slug, make_url_slug -PE_US_DATA_ROOT_ENV = "LEDGER_PE_US_DATA_ROOT" -PE_UK_DATA_ROOT_ENV = "LEDGER_PE_UK_DATA_ROOT" +PE_US_DATA_ROOT_ENV = "CHRONICLE_PE_US_DATA_ROOT" +PE_UK_DATA_ROOT_ENV = "CHRONICLE_PE_UK_DATA_ROOT" SOURCE_SUFFIXES = { ".csv", @@ -272,21 +273,13 @@ ] -def _env_value(*names: str) -> str | None: - for name in names: - value = os.environ.get(name) - if value: - return value - return None - - def _resolve_required_root( root: Path | None, *, flag: str, env_var: str, ) -> Path: - value = root if root is not None else _env_value(env_var) + value = root if root is not None else env_value(env_var) if value is None: raise ValueError(f"{flag} or {env_var} is required.") path = Path(value).expanduser() @@ -612,8 +605,8 @@ def pe_source_specs( ) -> list[SourceArtifactSpec]: """Return source files used by the PE-US and PE-UK calibration pipelines.""" specs: list[SourceArtifactSpec] = [] - us_configured = pe_us_root is not None or _env_value(PE_US_DATA_ROOT_ENV) - uk_configured = pe_uk_root is not None or _env_value(PE_UK_DATA_ROOT_ENV) + us_configured = pe_us_root is not None or env_value(PE_US_DATA_ROOT_ENV) + uk_configured = pe_uk_root is not None or env_value(PE_UK_DATA_ROOT_ENV) if include_us and (us_configured or not include_uk or not uk_configured): specs.extend(pe_us_source_specs(pe_us_root)) if include_uk and (uk_configured or not include_us or not us_configured): diff --git a/db/supabase_client.py b/db/supabase_client.py index e14bb5bb..50ce4f30 100644 --- a/db/supabase_client.py +++ b/db/supabase_client.py @@ -8,7 +8,6 @@ from __future__ import annotations -import os from dataclasses import dataclass from functools import lru_cache from typing import Any, Dict, List, Optional @@ -16,18 +15,14 @@ from supabase import create_client, Client +from chronicle.env import env_value -def _env(*names: str) -> str | None: - """Read PolicyEngine-owned storage config.""" - for name in names: - value = os.environ.get(name) - if value: - return value - return None - - -LEDGER_SCHEMA = _env("POLICYENGINE_LEDGER_SCHEMA") or "ledger" -TARGETS_SCHEMA = _env("POLICYENGINE_TARGETS_SCHEMA") or "targets" +# The hosted Postgres schema is still named "ledger"; only the environment +# variable that overrides it has moved to the chronicle prefix. Renaming the +# schema itself is a later slice of PolicyEngine/chronicle#143, coordinated +# with the CI writers that already target the ledger schema. +LEDGER_SCHEMA = env_value("CHRONICLE_SCHEMA") or "ledger" +TARGETS_SCHEMA = env_value("POLICYENGINE_TARGETS_SCHEMA") or "targets" @dataclass @@ -49,14 +44,14 @@ def from_env(cls) -> "SupabaseConfig": Raises: ValueError: If required environment variables are missing """ - url = _env("POLICYENGINE_SUPABASE_URL") + url = env_value("POLICYENGINE_SUPABASE_URL") if not url: raise ValueError( "POLICYENGINE_SUPABASE_URL not set. " "Set this to your Supabase project URL." ) - secret_key = _env( + secret_key = env_value( "POLICYENGINE_SUPABASE_SERVICE_KEY", "POLICYENGINE_SUPABASE_SECRET_KEY", ) From e2e0af0e43b60de70c062b38e829a0874ce5ef4b Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:06:52 -0400 Subject: [PATCH 002/105] Make R2 buckets configurable and emit chronicle.db Bucket names: add CHRONICLE_R2_RAW_BUCKET / CHRONICLE_R2_DERIVED_BUCKET, resolved in the function body rather than bound as keyword defaults so the setting reaches long-lived processes. Defaults are unchanged at ledger-raw and ledger-derived; only the follow-up cutover PR flips them. Plumbed through fetch-artifact, publish-raw, publish-derived and bootstrap-r2, whose --r2-bucket/--raw-bucket/--derived-bucket now default to the resolver. Because the bucket can now vary, two manifest write paths could restate recorded storage.r2 blocks. Archived witness records pin raw R2 URLs by hash, so both now preserve history: publish-raw reports recorded_r2_bucket_is_preserved_history instead of uploading, and fetch-artifact keeps an already-recorded r2 block rather than overwriting it with a different bucket. The consumer-fact boundary guard matched the literal strings ledger-derived: and r2://ledger-derived/, so a renamed bucket would have made it silently stop firing. It now matches on shape (bucket ends in -derived, or key starts with derived/) and parses the r2:// URI, which also closes the uri-only hole where a fact carrying no bucket or key slipped past. Database artifact: new suite outputs write chronicle.db. infer_build_id reads chronicle.db then ledger.db (both branches were byte-identical, so the existing fallback was a no-op), and _derived_artifact_kind classifies both names (its set literal held one element twice). The sidecar resource list moves with the write site because _resource_descriptor stats every listed path. Fix db/cli.py, which imported the private env reader that the shared helper replaced. That ImportError broke every db CLI subcommand, including the three the CI job runs, while pytest stayed green. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 104 ++++++++++++++++++++++++++++----- chronicle/consumer_contract.py | 32 ++++++---- chronicle/database.py | 9 +++ chronicle/env.py | 38 ++++++++---- chronicle/harness.py | 53 +++++++++++------ chronicle/suite.py | 10 +++- db/cli.py | 5 +- tests/test_chronicle_mirror.py | 8 +-- tests/test_chronicle_suite.py | 10 ++-- 9 files changed, 204 insertions(+), 65 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index fa56d620..6d1e9781 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -18,14 +18,41 @@ import httpx import yaml +from chronicle.database import ( + CHRONICLE_DB_FILENAME, + CHRONICLE_DB_FILENAMES, + LEGACY_CHRONICLE_DB_FILENAME, +) +from chronicle.env import env_value from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain + +R2_RAW_BUCKET_ENV = "CHRONICLE_R2_RAW_BUCKET" +R2_DERIVED_BUCKET_ENV = "CHRONICLE_R2_DERIVED_BUCKET" + +# The bucket defaults stay at their ledger-era names. Archived witness records +# pin raw R2 URLs by hash, so ledger-raw and ledger-derived are preserved +# read-only forever and no recorded manifest URI is ever rewritten. The env +# vars exist so the cutover in docs/storage-architecture.md can be rehearsed, +# and so flipping to chronicle-raw/chronicle-derived is a default change rather +# than a code change (PolicyEngine/chronicle#143, mechanism 3). DEFAULT_R2_RAW_BUCKET = "ledger-raw" DEFAULT_R2_DERIVED_BUCKET = "ledger-derived" DEFAULT_R2_PREFIX = "raw" DEFAULT_R2_DERIVED_PREFIX = "derived" + +def default_r2_raw_bucket() -> str: + """Resolve the raw bucket: ``$CHRONICLE_R2_RAW_BUCKET`` or the default.""" + return env_value(R2_RAW_BUCKET_ENV, default=DEFAULT_R2_RAW_BUCKET) + + +def default_r2_derived_bucket() -> str: + """Resolve the derived bucket: ``$CHRONICLE_R2_DERIVED_BUCKET`` or default.""" + return env_value(R2_DERIVED_BUCKET_ENV, default=DEFAULT_R2_DERIVED_BUCKET) + + # New UK and New Zealand uploads are namespaced by country. US objects predate # the country segment and deliberately keep their legacy ``raw/{source_id}`` # and ``derived/{source_id}`` shapes. Publisher directories are the stable @@ -392,11 +419,12 @@ def fetch_source_artifact( table: str | None = None, filename: str | None = None, upload_r2: bool = False, - r2_bucket: str = DEFAULT_R2_RAW_BUCKET, + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: """Fetch/register a source artifact and optionally upload it to R2.""" + r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, @@ -482,12 +510,13 @@ def publish_derived_artifacts( package_id: str, year: int, build_id: str | None = None, - r2_bucket: str = DEFAULT_R2_DERIVED_BUCKET, + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", build_artifacts_output: str | Path | None = None, ) -> DerivedArtifactPublishReport: """Upload a deterministic build output directory to the derived R2 bucket.""" + r2_bucket = r2_bucket or default_r2_derived_bucket() input_path = Path(input_dir) if not input_path.exists(): return DerivedArtifactPublishReport( @@ -617,11 +646,12 @@ def publish_source_artifacts( manifest_filename: str = "manifest.yaml", source_id: str | None = None, package_id: str | None = None, - r2_bucket: str = DEFAULT_R2_RAW_BUCKET, + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> RawArtifactPublishReport: """Upload manifest-declared raw source artifacts and record R2 locations.""" + r2_bucket = r2_bucket or default_r2_raw_bucket() root_path = Path(root) if not root_path.exists(): return RawArtifactPublishReport( @@ -791,12 +821,15 @@ def inventory_source_artifacts( def bootstrap_r2_buckets( *, - raw_bucket: str = DEFAULT_R2_RAW_BUCKET, - derived_bucket: str = DEFAULT_R2_DERIVED_BUCKET, + raw_bucket: str | None = None, + derived_bucket: str | None = None, wrangler_command: str = "npx wrangler", ) -> R2BootstrapReport: """Create the R2 buckets Chronicle expects, if Wrangler is authenticated.""" - buckets = (raw_bucket, derived_bucket) + buckets = ( + raw_bucket or default_r2_raw_bucket(), + derived_bucket or default_r2_derived_bucket(), + ) commands: list[ArtifactCommandResult] = [] errors: list[str] = [] @@ -1010,9 +1043,9 @@ def infer_build_id(input_dir: str | Path) -> str | None: if build_id: return str(build_id) - db_path = input_path / "ledger.db" + db_path = input_path / CHRONICLE_DB_FILENAME if not db_path.exists(): - db_path = input_path / "ledger.db" + db_path = input_path / LEGACY_CHRONICLE_DB_FILENAME if db_path.exists(): with sqlite3.connect(db_path) as connection: row = connection.execute( @@ -1043,6 +1076,17 @@ def _filename_from_url(source_url: str) -> str: return Path(unquote(parsed.path)).name +def _recorded_r2(spec: Any) -> dict[str, Any]: + """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" + if not isinstance(spec, dict): + return {} + storage = spec.get("storage") + if not isinstance(storage, dict): + return {} + recorded = storage.get("r2") + return recorded if isinstance(recorded, dict) else {} + + def _upsert_manifest( manifest_path: Path, *, @@ -1076,8 +1120,18 @@ def _upsert_manifest( "size_bytes": size_bytes, "fetched_at": fetched_at, } - if r2_location is not None: - file_entry["storage"] = {"r2": r2_location.to_dict()} + recorded_r2 = _recorded_r2(payload["files"].get(year)) + new_r2 = r2_location.to_dict() if r2_location is not None else None + if recorded_r2 and ( + new_r2 is None or recorded_r2.get("bucket") != new_r2.get("bucket") + ): + # A recorded storage.r2 block is historical truth: archived witness + # records pin raw R2 URLs by hash. Re-fetching under a renamed bucket + # copies bytes; it does not restate where the bytes were first + # published (PolicyEngine/chronicle#143, mechanism 3). + file_entry["storage"] = {"r2": recorded_r2} + elif new_r2 is not None: + file_entry["storage"] = {"r2": new_r2} payload["files"][year] = file_entry manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False), @@ -1169,8 +1223,32 @@ def _publish_raw_manifest_entry( package_path=manifest_path, ), ) - storage = spec.get("storage") if isinstance(spec.get("storage"), dict) else {} - recorded_r2 = storage.get("r2") if isinstance(storage.get("r2"), dict) else {} + recorded_r2 = _recorded_r2(spec) + recorded_bucket = recorded_r2.get("bucket") + if recorded_bucket and recorded_bucket != location.bucket: + # The recorded bucket is preserved history. Publishing the same bytes + # into a renamed bucket is a backfill copy, not a restatement, so the + # manifest must not be rewritten to point at the new bucket. + errors.append( + "recorded_r2_bucket_is_preserved_history:" + f"recorded={recorded_bucket}:requested={location.bucket}" + ) + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(artifact_path), + sha256=sha256_actual, + size_bytes=size_bytes, + r2_location=None, + upload=None, + errors=tuple(errors), + ), + None, + ) recorded_key = recorded_r2.get("key") if recorded_key and recorded_key != location.key: errors.append( @@ -1275,7 +1353,7 @@ def _inventory_entry( def _derived_artifact_kind(artifact_name: str) -> str: - if artifact_name in {"ledger.db", "ledger.db"}: + if artifact_name in CHRONICLE_DB_FILENAMES: return "sqlite_database" if artifact_name.endswith(".jsonl"): return "jsonl" diff --git a/chronicle/consumer_contract.py b/chronicle/consumer_contract.py index a1c78325..f9c93f3a 100644 --- a/chronicle/consumer_contract.py +++ b/chronicle/consumer_contract.py @@ -468,6 +468,24 @@ def validate_consumer_fact_contract( ) +def _r2_uri_parts(uri: str) -> tuple[str, str]: + """Split an ``r2://bucket/key`` URI into its bucket and key.""" + if not uri.startswith("r2://"): + return "", "" + bucket, _, key = uri[len("r2://") :].partition("/") + return bucket, key + + +def _points_at_derived(bucket: str, key: str) -> bool: + """Whether an R2 bucket/key pair addresses derived build output. + + Matched on shape rather than on the ledger-era bucket names, so the guard + keeps firing once the buckets are renamed (PolicyEngine/chronicle#143, + mechanism 3). + """ + return bucket.endswith("-derived") or key.startswith("derived/") + + def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: """Return a boundary error if a fact is a downstream target derivation.""" source = fact.source @@ -489,20 +507,14 @@ def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: "itself. Target construction, aging, and reconciliation belong in " "Microcosm." ) - if source_file.startswith("ledger-derived:"): + source_file_bucket, bucket_separator, _ = source_file.partition(":") + if bucket_separator and source_file_bucket.endswith("-derived"): return ( "Chronicle consumer facts must cite raw publisher artifacts. Derived " "target-construction artifacts belong in Microcosm." ) - if ( - raw_r2_bucket.endswith("-derived") - or raw_r2_key.startswith("derived/") - or raw_r2_uri.startswith( - ( - "r2://ledger-derived/", - "r2://ledger-raw/derived/", - ) - ) + if _points_at_derived(raw_r2_bucket, raw_r2_key) or _points_at_derived( + *_r2_uri_parts(raw_r2_uri) ): return ( "Chronicle consumer facts must point at raw source artifacts, not " diff --git a/chronicle/database.py b/chronicle/database.py index 3f18ef78..5297db16 100644 --- a/chronicle/database.py +++ b/chronicle/database.py @@ -42,6 +42,15 @@ LEDGER_DB_SCHEMA_VERSION = schema_id("relational", Epoch.LEDGER) +# New suite outputs write chronicle.db. Existing builds wrote ledger.db and are +# still read and published unchanged, so the legacy name stays accepted for +# inference and artifact classification (PolicyEngine/chronicle#143, +# mechanism 3). The relational schema id above is a frozen machine surface that +# migrates with the epoch lane, not with this rename. +CHRONICLE_DB_FILENAME = "chronicle.db" +LEGACY_CHRONICLE_DB_FILENAME = "ledger.db" +CHRONICLE_DB_FILENAMES = (CHRONICLE_DB_FILENAME, LEGACY_CHRONICLE_DB_FILENAME) + @dataclass(frozen=True) class ChronicleDbBuildReport: diff --git a/chronicle/env.py b/chronicle/env.py index 80667f4d..3a2f7af6 100644 --- a/chronicle/env.py +++ b/chronicle/env.py @@ -75,7 +75,11 @@ def env_names(name: str) -> tuple[str, ...]: def _warn_legacy(found: str, preferred: str) -> None: - """Warn once per process that a ledger-era variable supplied a value.""" + """Warn once per process that a ledger-era variable supplied a value. + + ``stacklevel=4`` walks out through :func:`_first_set` and its public + wrapper so the notice points at the code that asked for the setting. + """ if found in _WARNED_LEGACY_NAMES: return _WARNED_LEGACY_NAMES.add(found) @@ -84,7 +88,7 @@ def _warn_legacy(found: str, preferred: str) -> None: f"set {preferred} instead. The old name is still honored during the " "Chronicle rename window and will be removed once consumers migrate.", ChronicleEnvDeprecationWarning, - stacklevel=3, + stacklevel=4, ) @@ -93,12 +97,11 @@ def reset_env_deprecation_state() -> None: _WARNED_LEGACY_NAMES.clear() -def env_value(*names: str, default: _Default = None) -> str | _Default: - """Read the first set value across ``names``, chronicle-preferred first. +def _first_set(names: tuple[str, ...]) -> str | None: + """Return the first set value across ``names``, warning on a legacy hit. - Each name is expanded through :func:`env_names`, so a caller can pass the - chronicle name and still pick up a value set under a ledger-era name. - Empty values are treated as unset, matching the helpers this replaces. + Both public readers call this at the same stack depth so the deprecation + warning is always attributed to their caller, not to this module. """ for name in names: candidates = env_names(name) @@ -109,12 +112,27 @@ def env_value(*names: str, default: _Default = None) -> str | _Default: if candidate != preferred: _warn_legacy(candidate, preferred) return value - return default + return None + + +def env_value(*names: str, default: _Default = None) -> str | _Default: + """Read the first set value across ``names``, chronicle-preferred first. + + Each name is expanded through :func:`env_names`, so a caller can pass the + chronicle name and still pick up a value set under a ledger-era name. + Empty values are treated as unset, matching the helpers this replaces. + """ + value = _first_set(names) + return default if value is None else value def env_flag(*names: str) -> bool: - """Return whether the first set value across ``names`` reads as true.""" - value = env_value(*names) + """Return whether the first set value across ``names`` reads as true. + + The chronicle-preferred name wins even when it reads false, so an operator + who has migrated can turn a flag off without unsetting the legacy name. + """ + value = _first_set(names) if value is None: return False return value.strip().lower() in TRUTHY_ENV_VALUES diff --git a/chronicle/harness.py b/chronicle/harness.py index 298a1380..54c14139 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -8,6 +8,8 @@ from pathlib import Path from chronicle.artifacts import ( + DEFAULT_R2_DERIVED_BUCKET, + DEFAULT_R2_RAW_BUCKET, ArtifactFetchReport, ArtifactInventoryReport, DerivedArtifactPublishReport, @@ -335,7 +337,7 @@ def fetch_artifact_file( table: str | None = None, filename: str | None = None, upload_r2: bool = False, - r2_bucket: str = "ledger-raw", + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: @@ -372,7 +374,7 @@ def publish_raw_artifact_files( manifest_filename: str = "manifest.yaml", source_id: str | None = None, package_id: str | None = None, - r2_bucket: str = "ledger-raw", + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> RawArtifactPublishReport: @@ -390,8 +392,8 @@ def publish_raw_artifact_files( def bootstrap_r2_storage( *, - raw_bucket: str = "ledger-raw", - derived_bucket: str = "ledger-derived", + raw_bucket: str | None = None, + derived_bucket: str | None = None, wrangler_command: str = "npx wrangler", ) -> R2BootstrapReport: """Create Chronicle R2 buckets when Wrangler is authenticated.""" @@ -409,7 +411,7 @@ def publish_derived_artifact_files( package_id: str, year: int, build_id: str | None = None, - r2_bucket: str = "ledger-derived", + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", build_artifacts_output: str | Path | None = None, @@ -888,8 +890,11 @@ def main(argv: list[str] | None = None) -> int: ) artifact_parser.add_argument( "--r2-bucket", - default="ledger-raw", - help="R2 bucket for raw artifacts when --upload-r2 is set.", + default=None, + help=( + "R2 bucket for raw artifacts when --upload-r2 is set. Defaults to " + f"$CHRONICLE_R2_RAW_BUCKET, else {DEFAULT_R2_RAW_BUCKET}." + ), ) artifact_parser.add_argument( "--r2-prefix", @@ -923,7 +928,7 @@ def main(argv: list[str] | None = None) -> int: raw_publish_parser = subparsers.add_parser( "publish-raw", - help="Upload manifest-declared raw source artifacts to ledger-raw R2", + help="Upload manifest-declared raw source artifacts to the raw R2 bucket", ) raw_publish_parser.add_argument( "--root", @@ -946,8 +951,11 @@ def main(argv: list[str] | None = None) -> int: ) raw_publish_parser.add_argument( "--r2-bucket", - default="ledger-raw", - help="R2 bucket for immutable raw artifacts.", + default=None, + help=( + "R2 bucket for immutable raw artifacts. Defaults to " + f"$CHRONICLE_R2_RAW_BUCKET, else {DEFAULT_R2_RAW_BUCKET}." + ), ) raw_publish_parser.add_argument( "--r2-prefix", @@ -969,13 +977,19 @@ def main(argv: list[str] | None = None) -> int: ) r2_parser.add_argument( "--raw-bucket", - default="ledger-raw", - help="R2 bucket name for immutable raw source artifacts.", + default=None, + help=( + "R2 bucket name for immutable raw source artifacts. Defaults to " + f"$CHRONICLE_R2_RAW_BUCKET, else {DEFAULT_R2_RAW_BUCKET}." + ), ) r2_parser.add_argument( "--derived-bucket", - default="ledger-derived", - help="R2 bucket name for derived Chronicle build artifacts.", + default=None, + help=( + "R2 bucket name for derived Chronicle build artifacts. Defaults to " + f"$CHRONICLE_R2_DERIVED_BUCKET, else {DEFAULT_R2_DERIVED_BUCKET}." + ), ) r2_parser.add_argument( "--wrangler-command", @@ -985,7 +999,7 @@ def main(argv: list[str] | None = None) -> int: derived_publish_parser = subparsers.add_parser( "publish-derived", - help="Upload deterministic Chronicle build outputs to ledger-derived R2", + help="Upload deterministic Chronicle build outputs to the derived R2 bucket", ) derived_publish_parser.add_argument( "--dir", @@ -1014,13 +1028,16 @@ def main(argv: list[str] | None = None) -> int: help=( "Build ID under an accepted epoch prefix, ledger.build.v1: or " "chronicle.build.v2:; any other form is refused. Defaults to " - "the ID inferred from reports or ledger.db." + "the ID inferred from reports, chronicle.db, or a legacy ledger.db." ), ) derived_publish_parser.add_argument( "--r2-bucket", - default="ledger-derived", - help="R2 bucket for derived build artifacts.", + default=None, + help=( + "R2 bucket for derived build artifacts. Defaults to " + f"$CHRONICLE_R2_DERIVED_BUCKET, else {DEFAULT_R2_DERIVED_BUCKET}." + ), ) derived_publish_parser.add_argument( "--r2-prefix", diff --git a/chronicle/suite.py b/chronicle/suite.py index 157b7420..0705c862 100644 --- a/chronicle/suite.py +++ b/chronicle/suite.py @@ -22,7 +22,11 @@ build_fact_key, validate_facts, ) -from chronicle.database import ChronicleDbBuildReport, build_chronicle_db +from chronicle.database import ( + CHRONICLE_DB_FILENAME, + ChronicleDbBuildReport, + build_chronicle_db, +) from chronicle.epoch import canonicalize_key from chronicle.sources.cells import ( SourceCell, @@ -372,7 +376,7 @@ def build_source_suite( concept_report.to_dict(), ) - db_path = output_path / "ledger.db" + db_path = output_path / CHRONICLE_DB_FILENAME db_report = build_chronicle_db( facts, db_path, @@ -1675,7 +1679,7 @@ def _write_package_sidecars(output_path: Path, *, source: str, year: int) -> Non output_path / "source_regions.jsonl", output_path / "facts.jsonl", output_path / "consumer_facts.jsonl", - output_path / "ledger.db", + output_path / CHRONICLE_DB_FILENAME, output_path / "reports" / "source_rows.json", output_path / "reports" / "source_cells.json", output_path / "reports" / "source_regions.json", diff --git a/db/cli.py b/db/cli.py index 52257240..66538149 100644 --- a/db/cli.py +++ b/db/cli.py @@ -276,14 +276,15 @@ def cmd_query(args): def _pe_source_root_env_default(jurisdiction: str) -> str | None: + from chronicle.env import env_value + from .pe_source_inventory import ( PE_UK_DATA_ROOT_ENV, PE_US_DATA_ROOT_ENV, - _env_value, ) env_var = PE_US_DATA_ROOT_ENV if jurisdiction == "us" else PE_UK_DATA_ROOT_ENV - return _env_value(env_var) + return env_value(env_var) def main(): diff --git a/tests/test_chronicle_mirror.py b/tests/test_chronicle_mirror.py index 787429c1..c06a33bd 100644 --- a/tests/test_chronicle_mirror.py +++ b/tests/test_chronicle_mirror.py @@ -18,7 +18,7 @@ def test_export_chronicle_db_tables_writes_jsonl_and_manifest(tmp_path): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" output_dir = tmp_path / "mirror" build_chronicle_db( build_soi_table_1_1_facts(2023), @@ -53,7 +53,7 @@ def test_export_chronicle_db_tables_writes_jsonl_and_manifest(tmp_path): def test_export_chronicle_db_tables_orders_rows_deterministically(tmp_path): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" first_output_dir = tmp_path / "mirror-first" second_output_dir = tmp_path / "mirror-second" build_chronicle_db( @@ -72,7 +72,7 @@ def test_export_chronicle_db_tables_orders_rows_deterministically(tmp_path): def test_export_db_tables_cli_emits_manifest_summary(tmp_path, capsys): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" output_dir = tmp_path / "mirror" build_chronicle_db( build_soi_table_1_1_facts(2023), @@ -99,7 +99,7 @@ def test_export_db_tables_cli_emits_manifest_summary(tmp_path, capsys): def test_load_supabase_mirror_dry_run_counts_exported_rows(tmp_path): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" output_dir = tmp_path / "mirror" build_chronicle_db( build_soi_table_1_1_facts(2023), diff --git a/tests/test_chronicle_suite.py b/tests/test_chronicle_suite.py index 6911d6e2..c08f8abe 100644 --- a/tests/test_chronicle_suite.py +++ b/tests/test_chronicle_suite.py @@ -96,7 +96,7 @@ def test_build_source_suite_writes_artifacts_and_reports(tmp_path): assert (output_dir / "source_regions.jsonl").exists() assert (output_dir / "facts.jsonl").exists() assert (output_dir / "consumer_facts.jsonl").exists() - assert (output_dir / "ledger.db").exists() + assert (output_dir / "chronicle.db").exists() assert (output_dir / "datapackage.json").exists() assert (output_dir / "ro-crate-metadata.json").exists() assert (output_dir / "reports" / "source_regions.json").exists() @@ -124,7 +124,7 @@ def test_build_source_suite_writes_artifacts_and_reports(tmp_path): "source_regions.jsonl", "facts.jsonl", "consumer_facts.jsonl", - "ledger.db", + "chronicle.db", "reports/build_summary.json", "reports/source_regions.json", "reports/selectors.json", @@ -150,7 +150,7 @@ def test_build_source_suite_writes_artifacts_and_reports(tmp_path): "concept_alignment_validation_skipped" ] - with sqlite3.connect(output_dir / "ledger.db") as connection: + with sqlite3.connect(output_dir / "chronicle.db") as connection: facts_count = connection.execute( "SELECT COUNT(*) FROM aggregate_facts" ).fetchone()[0] @@ -184,7 +184,7 @@ def test_build_source_suite_supports_soi_table_1_4(tmp_path): "concept_alignment_validation_skipped" ) assert (output_dir / "source_regions.jsonl").exists() - assert (output_dir / "ledger.db").exists() + assert (output_dir / "chronicle.db").exists() def test_agent_acceptance_accepts_aggregate_income_range_source_rows(): @@ -739,7 +739,7 @@ def test_build_suite_cli_emits_json_summary(tmp_path, capsys): assert payload["outputs"]["source_regions"] == str( output_dir / "source_regions.jsonl" ) - assert payload["outputs"]["database"] == str(output_dir / "ledger.db") + assert payload["outputs"]["database"] == str(output_dir / "chronicle.db") assert payload["outputs"]["consumer_facts"] == str( output_dir / "consumer_facts.jsonl" ) From de03d3752f26275200a6b01295cb08465752db16 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:25:23 -0400 Subject: [PATCH 003/105] Add hermetic tests for the rename window tests/test_chronicle_env.py covers the shared helper end to end: the lookup ladder, CHRONICLE_* winning over both ledger-era spellings without a warning, each legacy spelling still working with a once-per-process warning attributed to the caller, empty values counting as unset, and a migrated operator being able to turn a flag off without unsetting the stale legacy name. It then exercises the real call sites -- source-package artifact cache and fetch flag, the db CLI's --pe-us-root default, the Supabase schema override, and R2 bucket resolution -- so a helper regression cannot pass by only testing the helper. An autouse fixture strips every rename-window variable from the ambient environment, so the file is hermetic under any shell. The artifacts tests cover the chronicle.db write and ledger.db read fallback, both database names classifying as sqlite_database, publish-derived following the configured bucket, and the two manifest-preservation paths: publish-raw refusing to restate a recorded bucket and fetch-artifact keeping the recorded r2 block while still uploading the backfill copy. Co-Authored-By: Claude Fable 5.1 --- tests/test_chronicle_artifacts.py | 169 +++++++++++++++ tests/test_chronicle_env.py | 334 ++++++++++++++++++++++++++++++ 2 files changed, 503 insertions(+) create mode 100644 tests/test_chronicle_env.py diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 7104ec03..a96d6c25 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -4,6 +4,7 @@ import hashlib import json +import sqlite3 import pytest import yaml @@ -658,3 +659,171 @@ def test_top_level_cli_dispatches_publish_derived(tmp_path, capsys, monkeypatch) assert exc.value.code == 0 assert payload["valid"] + + +def _sqlite_build(path, build_id): + """Write a minimal build database carrying one ledger_builds row.""" + with sqlite3.connect(path) as connection: + connection.execute("CREATE TABLE ledger_builds (build_id TEXT PRIMARY KEY)") + connection.execute("INSERT INTO ledger_builds VALUES (?)", (build_id,)) + + +@pytest.mark.parametrize("db_name", ["chronicle.db", "ledger.db"]) +def test_infer_build_id_reads_new_and_legacy_database_names(tmp_path, db_name): + suite = tmp_path / "suite" + suite.mkdir() + _sqlite_build(suite / db_name, "ledger.build.v1:from-db") + + assert infer_build_id(suite) == "ledger.build.v1:from-db" + + +def test_infer_build_id_prefers_the_chronicle_database(tmp_path): + suite = tmp_path / "suite" + suite.mkdir() + _sqlite_build(suite / "chronicle.db", "ledger.build.v1:chronicle") + _sqlite_build(suite / "ledger.db", "ledger.build.v1:legacy") + + assert infer_build_id(suite) == "ledger.build.v1:chronicle" + + +@pytest.mark.parametrize("db_name", ["chronicle.db", "ledger.db"]) +def test_publish_derived_classifies_both_database_names(tmp_path, db_name): + suite = tmp_path / "suite" + reports = suite / "reports" + reports.mkdir(parents=True) + build_id = "ledger.build.v1:kind" + (reports / "database.json").write_text(json.dumps({"build_id": build_id})) + (suite / db_name).write_bytes(b"db") + wrangler = tmp_path / "wrangler" + wrangler.write_text("#!/bin/sh\necho ok\n") + wrangler.chmod(0o755) + + report = publish_derived_artifacts( + suite, + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + wrangler_command=str(wrangler), + ) + rows = {row["artifact_name"]: row for row in build_artifact_rows(report)} + + assert rows[db_name]["artifact_kind"] == "sqlite_database" + + +def test_publish_derived_uses_the_configured_bucket(tmp_path, monkeypatch): + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-derived") + suite = tmp_path / "suite" + reports = suite / "reports" + reports.mkdir(parents=True) + (reports / "database.json").write_text( + json.dumps({"build_id": "ledger.build.v1:bucket"}) + ) + (suite / "facts.jsonl").write_text("{}\n") + log = tmp_path / "wrangler.log" + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + + report = publish_derived_artifacts( + suite, + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + wrangler_command=str(wrangler), + ) + + assert report.valid + assert report.entries[0].r2_location.bucket == "chronicle-derived" + assert "chronicle-derived/derived/irs_soi/" in log.read_text() + + +def test_publish_raw_refuses_to_restate_a_recorded_bucket(tmp_path, monkeypatch): + """A recorded storage.r2 bucket is preserved history, not a publish target. + + Archived witness records pin raw R2 URLs by hash, so backfilling the same + bytes into a renamed bucket must not rewrite the manifest. + """ + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-1-1" + source = tmp_path / "soi.xlsx" + source.write_bytes(b"official SOI workbook") + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + output_dir=output_dir, + ) + manifest_path = output_dir / "manifest.yaml" + manifest = yaml.safe_load(manifest_path.read_text()) + artifact = manifest["files"][2023] + recorded_key = ( + f"raw/irs_soi/soi-table-1-1/2023/{artifact['sha256']}/{artifact['filename']}" + ) + artifact["storage"] = { + "r2": { + "provider": "r2", + "bucket": "ledger-raw", + "key": recorded_key, + "uri": f"r2://ledger-raw/{recorded_key}", + } + } + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + log = tmp_path / "wrangler.log" + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + unchanged = yaml.safe_load(manifest_path.read_text()) + + assert not report.valid + assert ( + report.entries[0] + .errors[0] + .startswith("recorded_r2_bucket_is_preserved_history:") + ) + assert not log.exists() + assert unchanged["files"][2023]["storage"]["r2"]["bucket"] == "ledger-raw" + + +def test_fetch_artifact_keeps_an_already_recorded_bucket(tmp_path, monkeypatch): + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-1-1" + source = tmp_path / "soi.xlsx" + source.write_bytes(b"official SOI workbook") + log = tmp_path / "wrangler.log" + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + output_dir=output_dir, + upload_r2=True, + wrangler_command=str(wrangler), + ) + manifest_path = output_dir / "manifest.yaml" + first = yaml.safe_load(manifest_path.read_text()) + assert first["files"][2023]["storage"]["r2"]["bucket"] == "ledger-raw" + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + report = fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + output_dir=output_dir, + upload_r2=True, + wrangler_command=str(wrangler), + ) + second = yaml.safe_load(manifest_path.read_text()) + + # The backfill copy really is uploaded to the new bucket, but the manifest + # keeps recording where the bytes were first published. + assert report.r2_location.bucket == "chronicle-raw" + assert "chronicle-raw" in log.read_text() + assert ( + second["files"][2023]["storage"]["r2"] == first["files"][2023]["storage"]["r2"] + ) diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py new file mode 100644 index 00000000..4ffe0206 --- /dev/null +++ b/tests/test_chronicle_env.py @@ -0,0 +1,334 @@ +"""Tests for the chronicle-first environment read window. + +Chronicle's operational stores migrate by dual-run (PolicyEngine/chronicle#143, +mechanism 3): ``CHRONICLE_*`` names win, ledger-era names keep working behind a +deprecation warning. Every test here is hermetic — the fixture strips every +variable in the rename window from the ambient environment first. +""" + +from __future__ import annotations + +import importlib +import os +from pathlib import Path + +import pytest + +from chronicle.artifacts import ( + DEFAULT_R2_DERIVED_BUCKET, + DEFAULT_R2_RAW_BUCKET, + default_r2_derived_bucket, + default_r2_raw_bucket, +) +from chronicle.env import ( + CHRONICLE_ENV_PREFIX, + ChronicleEnvDeprecationWarning, + LEGACY_ENV_PREFIXES, + env_flag, + env_names, + env_value, + reset_env_deprecation_state, +) +from chronicle.harness import main as harness_main +from chronicle.source_package import ( + SOURCE_ARTIFACT_CACHE_ENV, + SOURCE_ARTIFACT_FETCH_ENV, +) + +RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) + + +@pytest.fixture(autouse=True) +def isolated_rename_window_env(monkeypatch): + """Run each test with no rename-window variable inherited from the shell.""" + for name in list(os.environ): + if name.startswith(RENAME_WINDOW_PREFIXES): + monkeypatch.delenv(name, raising=False) + reset_env_deprecation_state() + yield + reset_env_deprecation_state() + + +def _fake_wrangler(tmp_path, log): + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + return wrangler + + +# --------------------------------------------------------------------------- +# Lookup order +# --------------------------------------------------------------------------- + + +def test_env_names_puts_chronicle_first_then_ledger_era_names(): + assert env_names("CHRONICLE_SOURCE_ARTIFACT_FETCH") == ( + "CHRONICLE_SOURCE_ARTIFACT_FETCH", + "POLICYENGINE_LEDGER_SOURCE_ARTIFACT_FETCH", + "LEDGER_SOURCE_ARTIFACT_FETCH", + ) + + +def test_env_names_expands_a_ledger_era_name_to_the_same_ladder(): + assert env_names("LEDGER_PE_US_DATA_ROOT") == env_names("CHRONICLE_PE_US_DATA_ROOT") + + +def test_env_names_leaves_variables_outside_the_rename_window_alone(): + assert env_names("POLICYENGINE_SUPABASE_URL") == ("POLICYENGINE_SUPABASE_URL",) + assert env_names("POLICYENGINE_TARGETS_SCHEMA") == ("POLICYENGINE_TARGETS_SCHEMA",) + + +def test_bare_prefix_is_not_treated_as_a_renamed_variable(): + assert env_names("LEDGER_") == ("LEDGER_",) + + +# --------------------------------------------------------------------------- +# Precedence and the deprecation warning +# --------------------------------------------------------------------------- + + +def test_chronicle_name_wins_over_both_ledger_era_names(monkeypatch, recwarn): + monkeypatch.setenv("CHRONICLE_PE_US_DATA_ROOT", "/chronicle") + monkeypatch.setenv("LEDGER_PE_US_DATA_ROOT", "/ledger") + monkeypatch.setenv("POLICYENGINE_LEDGER_PE_US_DATA_ROOT", "/policyengine-ledger") + + assert env_value("CHRONICLE_PE_US_DATA_ROOT") == "/chronicle" + assert not [ + warning + for warning in recwarn.list + if issubclass(warning.category, ChronicleEnvDeprecationWarning) + ] + + +def test_ledger_name_alone_still_works_and_warns(monkeypatch): + monkeypatch.setenv("LEDGER_PE_US_DATA_ROOT", "/ledger") + + with pytest.warns(ChronicleEnvDeprecationWarning) as warnings_raised: + assert env_value("CHRONICLE_PE_US_DATA_ROOT") == "/ledger" + + message = str(warnings_raised[0].message) + assert "LEDGER_PE_US_DATA_ROOT" in message + assert "CHRONICLE_PE_US_DATA_ROOT" in message + + +def test_policyengine_ledger_name_alone_still_works_and_warns(monkeypatch): + monkeypatch.setenv("POLICYENGINE_LEDGER_SCHEMA", "ledger") + + with pytest.warns(ChronicleEnvDeprecationWarning) as warnings_raised: + assert env_value("CHRONICLE_SCHEMA") == "ledger" + + message = str(warnings_raised[0].message) + assert "POLICYENGINE_LEDGER_SCHEMA" in message + assert "CHRONICLE_SCHEMA" in message + + +def test_deprecation_warning_is_raised_once_per_process(monkeypatch, recwarn): + monkeypatch.setenv("LEDGER_PE_UK_DATA_ROOT", "/ledger") + + for _ in range(3): + assert env_value("CHRONICLE_PE_UK_DATA_ROOT") == "/ledger" + + deprecations = [ + warning + for warning in recwarn.list + if issubclass(warning.category, ChronicleEnvDeprecationWarning) + ] + assert len(deprecations) == 1 + + +def test_deprecation_warning_is_attributed_to_the_calling_module(monkeypatch): + monkeypatch.setenv("LEDGER_SOURCE_ARTIFACT_FETCH", "1") + + with pytest.warns(ChronicleEnvDeprecationWarning) as warnings_raised: + assert env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + # env_flag and env_value must report at the same depth, or operators get a + # notice pointing at Chronicle's own source instead of their call site. + assert Path(warnings_raised[0].filename).name == "test_chronicle_env.py" + + +def test_unset_variables_fall_back_to_the_default(): + assert env_value("CHRONICLE_PE_US_DATA_ROOT") is None + assert env_value("CHRONICLE_PE_US_DATA_ROOT", default="/fallback") == "/fallback" + + +def test_empty_values_count_as_unset(monkeypatch): + monkeypatch.setenv("CHRONICLE_PE_US_DATA_ROOT", "") + monkeypatch.setenv("LEDGER_PE_US_DATA_ROOT", "/ledger") + + with pytest.warns(ChronicleEnvDeprecationWarning): + assert env_value("CHRONICLE_PE_US_DATA_ROOT") == "/ledger" + + +# --------------------------------------------------------------------------- +# Flags +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize("value", ["1", "true", "TRUE", "yes", "on", " on "]) +def test_env_flag_accepts_truthy_spellings(monkeypatch, value): + monkeypatch.setenv("CHRONICLE_SOURCE_ARTIFACT_FETCH", value) + assert env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + +@pytest.mark.parametrize("value", ["0", "false", "no", "off", "maybe"]) +def test_env_flag_rejects_other_values(monkeypatch, value): + monkeypatch.setenv("CHRONICLE_SOURCE_ARTIFACT_FETCH", value) + assert not env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + +def test_env_flag_lets_the_chronicle_name_turn_a_legacy_flag_off(monkeypatch): + monkeypatch.setenv("CHRONICLE_SOURCE_ARTIFACT_FETCH", "0") + monkeypatch.setenv("LEDGER_SOURCE_ARTIFACT_FETCH", "1") + + # An operator who has migrated must be able to turn the flag off without + # first hunting down the stale ledger-era variable. + assert not env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + +# --------------------------------------------------------------------------- +# Real call sites +# --------------------------------------------------------------------------- + + +def test_source_artifact_env_constants_are_chronicle_named(): + assert SOURCE_ARTIFACT_CACHE_ENV == "CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR" + assert SOURCE_ARTIFACT_FETCH_ENV == "CHRONICLE_SOURCE_ARTIFACT_FETCH" + + +@pytest.mark.parametrize( + "name", + ["CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR", "LEDGER_SOURCE_ARTIFACT_CACHE_DIR"], +) +def test_source_artifact_cache_dir_honors_both_names(monkeypatch, tmp_path, name): + from chronicle.source_package import _source_artifact_cache_path + + monkeypatch.setenv(name, str(tmp_path)) + + cache_path = _source_artifact_cache_path( + {"filename": "table.xlsx", "sha256": "abc123"} + ) + + assert cache_path == tmp_path / "abc123" / "table.xlsx" + + +@pytest.mark.parametrize( + "name", ["CHRONICLE_PE_US_DATA_ROOT", "LEDGER_PE_US_DATA_ROOT"] +) +def test_pe_source_root_cli_default_honors_both_names(monkeypatch, name): + from db.cli import _pe_source_root_env_default + + monkeypatch.setenv(name, "/pe-us") + + assert _pe_source_root_env_default("us") == "/pe-us" + + +def test_pe_source_inventory_env_constants_are_chronicle_named(): + from db.pe_source_inventory import PE_UK_DATA_ROOT_ENV, PE_US_DATA_ROOT_ENV + + assert PE_US_DATA_ROOT_ENV == "CHRONICLE_PE_US_DATA_ROOT" + assert PE_UK_DATA_ROOT_ENV == "CHRONICLE_PE_UK_DATA_ROOT" + + +def test_db_cli_parser_builds_with_the_env_backed_defaults(monkeypatch, capsys): + """The db CLI builds its parser before dispatching any subcommand. + + Its --pe-us-root/--pe-uk-root defaults call into the env helper, so an + import error there breaks `chronicle init`, `load` and `stats` alike while + the rest of the test suite stays green. + """ + import db.cli + + monkeypatch.setenv("CHRONICLE_PE_US_DATA_ROOT", "/pe-us") + monkeypatch.setattr("sys.argv", ["chronicle", "--help"]) + + with pytest.raises(SystemExit) as exit_info: + db.cli.main() + + assert exit_info.value.code == 0 + assert "Manage Chronicle target input data" in capsys.readouterr().out + + +@pytest.mark.parametrize("name", ["CHRONICLE_SCHEMA", "POLICYENGINE_LEDGER_SCHEMA"]) +def test_supabase_schema_honors_both_names(monkeypatch, name): + import db.supabase_client + + monkeypatch.setenv(name, "chronicle_probe") + try: + reloaded = importlib.reload(db.supabase_client) + assert reloaded.LEDGER_SCHEMA == "chronicle_probe" + finally: + monkeypatch.delenv(name, raising=False) + importlib.reload(db.supabase_client) + + +def test_supabase_schema_default_is_unchanged(): + import db.supabase_client + + # The hosted schema name itself is out of this slice; only the variable + # that overrides it moved. + assert db.supabase_client.LEDGER_SCHEMA == "ledger" + + +# --------------------------------------------------------------------------- +# R2 bucket configuration +# --------------------------------------------------------------------------- + + +def test_r2_bucket_defaults_are_still_the_ledger_era_names(): + assert DEFAULT_R2_RAW_BUCKET == "ledger-raw" + assert DEFAULT_R2_DERIVED_BUCKET == "ledger-derived" + assert default_r2_raw_bucket() == "ledger-raw" + assert default_r2_derived_bucket() == "ledger-derived" + + +def test_r2_buckets_follow_the_chronicle_env_vars(monkeypatch): + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-derived") + + assert default_r2_raw_bucket() == "chronicle-raw" + assert default_r2_derived_bucket() == "chronicle-derived" + + +def test_r2_buckets_honor_ledger_era_names_with_a_warning(monkeypatch): + monkeypatch.setenv("LEDGER_R2_RAW_BUCKET", "legacy-raw") + + with pytest.warns(ChronicleEnvDeprecationWarning): + assert default_r2_raw_bucket() == "legacy-raw" + + +def test_bootstrap_r2_cli_creates_the_configured_buckets(monkeypatch, tmp_path): + log = tmp_path / "wrangler.log" + wrangler = _fake_wrangler(tmp_path, log) + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-derived") + + exit_code = harness_main(["bootstrap-r2", "--wrangler-command", str(wrangler)]) + + commands = log.read_text() + assert exit_code == 0 + assert "r2 bucket create chronicle-raw" in commands + assert "r2 bucket create chronicle-derived" in commands + + +def test_bootstrap_r2_cli_flags_still_override_the_environment(monkeypatch, tmp_path): + log = tmp_path / "wrangler.log" + wrangler = _fake_wrangler(tmp_path, log) + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + + harness_main( + [ + "bootstrap-r2", + "--raw-bucket", + "explicit-raw", + "--derived-bucket", + "explicit-derived", + "--wrangler-command", + str(wrangler), + ] + ) + + commands = log.read_text() + assert "r2 bucket create explicit-raw" in commands + assert "r2 bucket create explicit-derived" in commands + assert "chronicle-raw" not in commands From 08d950beb5376802298b2e704a56a6b31b07ded6 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:31:20 -0400 Subject: [PATCH 004/105] Document the rename window and the bucket cutover docs/storage-architecture.md stated the fallback direction backwards: it claimed the CHRONICLE_-prefixed variables were the old names kept as migration fallbacks, when they are the names to migrate to. Replaced with an "Environment Variable Rename Window" section that gives the actual lookup order, the once-per-process warning, and a table of all seven variables. Two behaviors get stated outright because both invert a naive fallback: the CHRONICLE_ name wins even when it reads false, so a migrated operator can turn a flag off without hunting down a stale legacy export; and an empty value counts as unset. The section also records what is deliberately NOT aliased -- POLICYENGINE_SUPABASE_* and POLICYENGINE_TARGETS_SCHEMA carry no ledger prefix and are read literally -- and that CHRONICLE_SCHEMA renames the override variable, not the schema value, which moves in a later slice. Added a "Bucket Cutover" section covering the six steps: create the buckets on the account wrangler.toml pins, enumerate, backfill-copy, verify, flip the defaults in a follow-up, and leave the ledger-era buckets read-only forever. The enumeration and verification are given as commands rather than prose. Every raw key ends {sha256}/{filename}, so the key is its own checksum witness and verification needs no manifest lookup. Counted rather than assumed: 186 distinct ledger-raw objects across 154 tracked manifest files, and the doc says to recount rather than trust that number, since source packages land continuously. The derived bucket needs no backfill -- derived artifacts are reproducible and already keyed by {build_id}, so a rebuild republishes them wherever configured. README and the harness doc follow: bucket references become archive roles with the default named, ledger.db becomes chronicle.db with the legacy name noted as still readable, and bootstrap-r2's example drops the hardcoded bucket flags now that they default to the resolver. Left alone: LEDGER_EXPLORER_DATA_DIRS in the README. It configures the explorer app, which is not in this repository, so renaming it here would document a name no shipped code honors -- the same defect this commit fixes in the other direction. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 42 ++++-- README.md | 31 +++-- docs/agent-source-package-harness.md | 27 ++-- docs/storage-architecture.md | 191 +++++++++++++++++++++++++-- 4 files changed, 249 insertions(+), 42 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index e46b096c..da5d13e5 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -11,6 +11,8 @@ Lane C5's handoff notes previously lived here; its durable record is - Out of scope and deliberately untouched: the `ledger` console-script alias, the Supabase `"ledger"` schema and mirror table names, governance role ids and concept authorities, hash domains and schema ids, anything under `releases/`. +- This PR does not touch the source-data boundary. No package spec, parser, + selector, manifest, or fact value changes. ## Done @@ -24,18 +26,40 @@ Lane C5's handoff notes previously lived here; its durable record is `LEDGER_MIRROR_PRIMARY_KEYS`, and `LEDGER_DB_SCHEMA_VERSION` are module constants, not env reads, and name out-of-scope surfaces. - Added `chronicle/env.py`: one shared `env_value`/`env_flag`/`env_names` - helper reading `CHRONICLE_` first, then `LEDGER_` and - `POLICYENGINE_LEDGER_` with a once-per-process - `ChronicleEnvDeprecationWarning` naming the preferred variable. + helper reading `CHRONICLE_` first, then `POLICYENGINE_LEDGER_` and + `LEDGER_` with a once-per-process `ChronicleEnvDeprecationWarning` naming + the preferred variable. - Replaced all three ad-hoc helpers (`db/supabase_client._env`, `chronicle/source_package._env_value`/`_truthy_env`, `db/pe_source_inventory._env_value`) with the shared helper. +- Made the R2 bucket names configurable via `CHRONICLE_R2_RAW_BUCKET` and + `CHRONICLE_R2_DERIVED_BUCKET`, plumbed through fetch-artifact, publish-raw, + publish-derived and bootstrap-r2. Defaults unchanged at `ledger-raw` and + `ledger-derived`. Both manifest write paths now preserve a recorded + `storage.r2` block instead of restating it under a renamed bucket. +- Emitted `chronicle.db` for new suite outputs, with `ledger.db` still accepted + on read and on derived-artifact kind inference. +- Added `tests/test_chronicle_env.py` plus artifact tests: 75 hermetic tests + covering the lookup ladder, precedence, the once-per-process warning, and every + real call site. +- Swept the docs. `docs/storage-architecture.md` gained an "Environment Variable + Rename Window" section (the old text stated the fallback direction backwards) + and a "Bucket Cutover" section; `docs/agent-source-package-harness.md` and + `README.md` follow. Verified 186 distinct `ledger-raw` objects across 154 + tracked manifest files, every key content-addressed by sha256. + +## Verification + +- `uv run pytest -q`: green. +- `uv run ruff check .`: clean. +- `uv run ruff format --check .`: clean for every file this branch touches. 13 + files are unformatted on `main` already and are byte-identical here; CI runs + `ruff check` only, so they are pre-existing and out of scope. +- CI's db CLI gate (`chronicle init` / `load all` / `stats`): passes. ## Next -- Make R2 bucket names configurable with unchanged `ledger-*` defaults. -- Emit `chronicle.db` for new suite outputs; keep reading `ledger.db`. -- Fix the backwards fallback statement in the docs and sweep every env-name, - bucket-name, and db-filename mention in README/AGENTS/docs. -- Add the hermetic dual-read tests; run pytest, ruff check, ruff format --check. -- Push and open the PR. Do not merge. +- Push and open the PR against `main`. Do not merge. +- Follow-up PR, after Max creates and backfills the new buckets: flip + `DEFAULT_R2_RAW_BUCKET` / `DEFAULT_R2_DERIVED_BUCKET` to `chronicle-raw` / + `chronicle-derived`. diff --git a/README.md b/README.md index f75fe697..e0ade235 100644 --- a/README.md +++ b/README.md @@ -65,10 +65,12 @@ contract that aligns it to another period (see | Microcosm Target Contracts | Selection, measurement bindings, and active subset | Period alignment, support-aware activation, solver inputs, diagnostics | The storage split is documented in -[`docs/storage-architecture.md`](docs/storage-architecture.md): `ledger-raw` -stores immutable source bytes, `ledger-derived` stores reproducible build +[`docs/storage-architecture.md`](docs/storage-architecture.md): a raw R2 archive +stores immutable source bytes, a derived R2 archive stores reproducible build artifacts, and Supabase/Postgres hosts the queryable relational Chronicle registry -mirrored from accepted builds. +mirrored from accepted builds. The bucket names are configuration +(`$CHRONICLE_R2_RAW_BUCKET` and `$CHRONICLE_R2_DERIVED_BUCKET`), still defaulting +to the ledger-era `ledger-raw` and `ledger-derived`. ## Repository Model @@ -247,7 +249,7 @@ This writes: source_regions.jsonl facts.jsonl consumer_facts.jsonl - ledger.db + chronicle.db reports/ source_rows.json source_cells.json @@ -349,8 +351,9 @@ needed, even when your Cloudflare user belongs to several accounts: # One-time per machine (opens a browser consent page): bunx wrangler login -# One-time per account (already done for the PolicyEngine account): -uv run chronicle bootstrap-r2 --raw-bucket ledger-raw --derived-bucket ledger-derived +# One-time per account (already done for the PolicyEngine account). The bucket +# flags default to $CHRONICLE_R2_RAW_BUCKET / $CHRONICLE_R2_DERIVED_BUCKET: +uv run chronicle bootstrap-r2 # Fetch/register a source artifact, write db/data/.../manifest.yaml, and upload # the exact bytes to R2 when Wrangler is authenticated: @@ -367,8 +370,8 @@ uv run chronicle fetch-artifact \ # Audit local manifests and checksums: uv run chronicle inventory-artifacts --root db/data -# Upload all existing manifest-declared local artifacts to ledger-raw and write -# storage.r2 metadata back into the manifests: +# Upload all existing manifest-declared local artifacts to the raw archive and +# write storage.r2 metadata back into the manifests: uv run chronicle publish-raw --root db/data ``` @@ -406,10 +409,10 @@ To prepare the deterministic SQLite artifact for a hosted Supabase/Postgres mirror, export each relational table to JSONL plus a manifest: ```bash -uv run chronicle export-db-tables --db /tmp/chronicle-suite/ledger.db --out /tmp/chronicle-mirror --replace +uv run chronicle export-db-tables --db /tmp/chronicle-suite/chronicle.db --out /tmp/chronicle-mirror --replace ``` -To publish the deterministic build outputs to the `ledger-derived` R2 bucket: +To publish the deterministic build outputs to the derived R2 archive: ```bash uv run chronicle publish-derived \ @@ -437,6 +440,14 @@ uv run chronicle load-supabase-mirror \ Use `--dry-run` first to validate JSONL row counts and file coverage without writing to Supabase. +Chronicle settings are read chronicle-first: `CHRONICLE_X` wins, and the +ledger-era `POLICYENGINE_LEDGER_X` and `LEDGER_X` spellings still work behind a +one-time deprecation warning naming the variable to move to. +[`docs/storage-architecture.md`](docs/storage-architecture.md#environment-variable-rename-window) +lists every variable in that window, and +[Bucket Cutover](docs/storage-architecture.md#bucket-cutover) covers the R2 +bucket rename. + Chronicle facts keep source concepts and canonical concepts separately. For example, the SOI Table 1.1 adjusted gross income column is preserved as `irs_soi.adjusted_gross_income`, while the canonical concept is diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 5f300356..77872ae8 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -14,8 +14,9 @@ lineage, provenance, constraints, and a passing `build-suite` report. The first gate for a new package is source-artifact acquisition. Agents should register raw source files with `uv run chronicle fetch-artifact` before authoring selectors. This writes the local artifact, captures checksum and retrieval -metadata in `manifest.yaml`, and can upload the exact bytes to the private -`ledger-raw` R2 bucket when Wrangler is authenticated. Agents can audit the local +metadata in `manifest.yaml`, and can upload the exact bytes to the private raw +R2 bucket (`ledger-raw` today; overridable with `CHRONICLE_R2_RAW_BUCKET`) when +Wrangler is authenticated. Agents can audit the local artifact registry with `uv run chronicle inventory-artifacts --root db/data`. For already-downloaded manifest artifacts, agents should run `uv run chronicle publish-raw --root db/data` to upload checksum-verified bytes to @@ -23,12 +24,15 @@ R2 and write `storage.r2` metadata back into each manifest entry. Builds do not require production raw bytes to be committed to Git. Source packages first read packaged fixture bytes, then -`LEDGER_SOURCE_ARTIFACT_CACHE_DIR` (defaulting to +`CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` (defaulting to `~/.cache/policyengine-chronicle/source-artifacts`). If a manifest artifact is -missing locally, set `LEDGER_SOURCE_ARTIFACT_FETCH=1` to fetch it from the +missing locally, set `CHRONICLE_SOURCE_ARTIFACT_FETCH=1` to fetch it from the manifest `source_url`, verify the declared SHA-256, and write it to that cache. -The old `CHRONICLE_`-prefixed environment variables remain accepted only as -migration fallbacks. +The ledger-era spellings `LEDGER_SOURCE_ARTIFACT_CACHE_DIR` and +`LEDGER_SOURCE_ARTIFACT_FETCH` are still honored during the rename window and +emit a one-time deprecation warning naming the `CHRONICLE_` variable to set +instead; see "Environment Variable Rename Window" in +[`docs/storage-architecture.md`](storage-architecture.md#environment-variable-rename-window). For broad PE source migration, generate the agent queue from the manifest before assigning work: @@ -645,16 +649,19 @@ uv run chronicle build-suite packages/irs_soi/table_1_1 \ --require-axiom-validation ``` -The SQLite `ledger.db` is the source of hosted mirrors. To prepare tables for +The SQLite `chronicle.db` is the source of hosted mirrors. To prepare tables for Supabase/Postgres bulk loading, export the DB artifact rather than inserting cells through the Supabase client: ```bash -uv run chronicle export-db-tables --db /tmp/chronicle-suite/ledger.db --out /tmp/chronicle-mirror --replace +uv run chronicle export-db-tables --db /tmp/chronicle-suite/chronicle.db --out /tmp/chronicle-mirror --replace ``` -Accepted build-suite outputs can be published to the private `ledger-derived` R2 -bucket after validation: +Builds produced before the rename wrote `ledger.db`. That name is still read and +published unchanged, so point `--db` at whichever file the build emitted. + +Accepted build-suite outputs can be published to the private derived R2 bucket +after validation: ```bash uv run chronicle publish-derived \ diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index cf898933..e381353f 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -9,17 +9,24 @@ class of Chronicle data belongs. Chronicle uses three storage layers with different jobs. -`ledger-raw` is the immutable source-byte archive. It stores exact publisher +The raw archive is the immutable source-byte store. It holds exact publisher artifacts as fetched: workbooks, CSVs, PDFs, ZIPs, HTML snapshots, and similar government-statistics release files. Raw objects are content-addressed by checksum and should never be overwritten in place. -`ledger-derived` is the reproducible artifact archive. It stores build outputs +The derived archive is the reproducible artifact store. It holds build outputs that Chronicle can regenerate from raw bytes, package specs, parser code, and build configuration. Examples include parsed-cell or parsed-row Parquet/JSONL files, -source record outputs, `ledger.db`, mirror JSONL exports, QA reports, Data +source record outputs, `chronicle.db`, mirror JSONL exports, QA reports, Data Package metadata, and RO-Crate metadata. +Both bucket names are configuration, not constants. The raw archive is +`$CHRONICLE_R2_RAW_BUCKET` and the derived archive is +`$CHRONICLE_R2_DERIVED_BUCKET`; the shipped defaults are still the ledger-era +`ledger-raw` and `ledger-derived`. [Bucket Cutover](#bucket-cutover) records how +those defaults move to `chronicle-raw` and `chronicle-derived` and why the +ledger-era buckets are preserved read-only rather than retired. + Supabase/Postgres is the queryable relational registry for accepted Chronicle builds. It stores rows that applications, agents, and downstream systems need to search and join: source artifacts, source rows/cells, source records, @@ -32,8 +39,8 @@ Hosted tables mirror accepted build outputs and provide a shared query surface. ## Ownership Matrix -| Data class | Git/local package | `ledger-raw` R2 | `ledger-derived` R2 | SQLite `ledger.db` | Supabase/Postgres | -|------------|-------------------|---------------|-------------------|------------------|-------------------| +| Data class | Git/local package | Raw R2 | Derived R2 | SQLite `chronicle.db` | Supabase/Postgres | +|------------|-------------------|--------|------------|-----------------------|-------------------| | Source package specs | Authoritative YAML and parser code | No | Optional packaged snapshot | No | Metadata only | | Raw publisher files | Tiny fixtures only | Authoritative bytes | No | Metadata only | Metadata plus R2 pointer | | Source manifests | Authoritative checked metadata | No | Optional snapshot | Metadata loaded into tables | Queryable artifact registry | @@ -76,7 +83,7 @@ Examples: ```text derived/uk/ons/ons-mye-2024-uk/2024/{build_id}/source_cells.jsonl -derived/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{build_id}/ledger.db +derived/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{build_id}/chronicle.db ``` Legacy US derived keys likewise remain `derived/{source_id}/...`. @@ -114,20 +121,21 @@ The intended flow is: 1. Register raw source artifacts with `uv run chronicle fetch-artifact`, which writes local bytes, records checksums in `manifest.yaml`, and can upload the - exact bytes to `ledger-raw`. Existing manifest-declared artifacts can be + exact bytes to the raw archive. Existing manifest-declared artifacts can be checksum-validated, uploaded, and linked with `uv run chronicle publish-raw`. Production package specs may omit raw bytes from Git as long as the manifest keeps `source_url` and SHA-256 metadata; builds can fill - `LEDGER_SOURCE_ARTIFACT_CACHE_DIR` by setting - `LEDGER_SOURCE_ARTIFACT_FETCH=1`. The old `CHRONICLE_`-prefixed environment - variables remain accepted only as migration fallbacks. + `CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` by setting + `CHRONICLE_SOURCE_ARTIFACT_FETCH=1`. Ledger-era spellings of both still work; + see [Environment Variable Rename Window](#environment-variable-rename-window). 2. Validate and build a source package with `uv run chronicle validate-package` and `uv run chronicle build-suite`. 3. Produce local deterministic outputs: parsed rows/cells, source records, - aggregate facts, `ledger.db`, QA reports, Data Package metadata, and RO-Crate - metadata. + aggregate facts, `chronicle.db`, QA reports, Data Package metadata, and + RO-Crate metadata. Builds before this rename wrote `ledger.db`; every reader + still accepts that name. 4. Export relational mirror files with `uv run chronicle export-db-tables`. -5. Publish derived build outputs to `ledger-derived`: +5. Publish derived build outputs to the derived archive: ```bash uv run chronicle publish-derived \ @@ -150,6 +158,163 @@ The Supabase project must have the checked migration applied and the `chronicle` schema exposed in PostgREST/Data API settings before the REST loader can write to it. Use `--dry-run` to verify local JSONL files without writing. +## Environment Variable Rename Window + +Every Chronicle setting is read chronicle-first by one shared helper, +`chronicle/env.py`. For a setting `X`, the lookup order is: + +1. `CHRONICLE_X` +2. `POLICYENGINE_LEDGER_X` +3. `LEDGER_X` + +The first name that holds a non-empty value wins. When that name is a ledger-era +one, the process emits a single `ChronicleEnvDeprecationWarning` naming the +`CHRONICLE_`-prefixed variable to set instead. The warning fires once per legacy +name per process, and it subclasses `FutureWarning` rather than +`DeprecationWarning` so it actually reaches operators running the CLI. + +Two consequences are worth stating outright, because both are the reverse of +what a naive fallback would do: + +- The `CHRONICLE_` name wins even when its value reads false. An operator who + has migrated can set `CHRONICLE_SOURCE_ARTIFACT_FETCH=0` and have the flag + turn off, without first hunting down a stale `LEDGER_SOURCE_ARTIFACT_FETCH=1` + somewhere in their profile. +- An empty value counts as unset, so exporting an empty `CHRONICLE_` name does + not mask a set legacy name. + +| Chronicle name | Ledger-era names still accepted | Meaning | +|----------------|--------------------------------|---------| +| `CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` | `LEDGER_SOURCE_ARTIFACT_CACHE_DIR` | Where fetched raw bytes are cached; defaults to `~/.cache/policyengine-chronicle/source-artifacts` | +| `CHRONICLE_SOURCE_ARTIFACT_FETCH` | `LEDGER_SOURCE_ARTIFACT_FETCH` | Fetch a missing manifest artifact from its `source_url` during a build | +| `CHRONICLE_PE_US_DATA_ROOT` | `LEDGER_PE_US_DATA_ROOT` | Local checkout root for PE US source inventory | +| `CHRONICLE_PE_UK_DATA_ROOT` | `LEDGER_PE_UK_DATA_ROOT` | Local checkout root for PE UK source inventory | +| `CHRONICLE_SCHEMA` | `POLICYENGINE_LEDGER_SCHEMA` | Postgres schema the Supabase client reads and writes | +| `CHRONICLE_R2_RAW_BUCKET` | `LEDGER_R2_RAW_BUCKET` | Raw R2 archive bucket; defaults to `ledger-raw` | +| `CHRONICLE_R2_DERIVED_BUCKET` | `LEDGER_R2_DERIVED_BUCKET` | Derived R2 archive bucket; defaults to `ledger-derived` | + +The two R2 rows are new in this window rather than renamed: those buckets were +hardcoded before, so the ledger-era spellings are accepted for consistency, not +because anything ever set them. + +Variables carrying none of the three prefixes are read literally. This helper +renames the ledger-era surface, not every PolicyEngine variable, so +`POLICYENGINE_SUPABASE_URL`, `POLICYENGINE_SUPABASE_SERVICE_KEY` and +`POLICYENGINE_TARGETS_SCHEMA` keep their names and gain no aliases. + +The hosted schema *value* is a separate migration. `CHRONICLE_SCHEMA` renames +the variable that overrides the schema; the schema still defaults to `ledger`, +and the mirror table names are unchanged. Those move in a later slice +coordinated with the CI writers. + +## Bucket Cutover + +Chronicle's operational stores migrate by dual-run +(PolicyEngine/chronicle#143, mechanism 3): stand up the chronicle-named home, +backfill it, repoint writers, retire the old home. The R2 buckets take one +exception to the last step. Archived witness records pin raw R2 URLs by hash, so +`ledger-raw` and `ledger-derived` are preserved read-only forever rather than +deleted, and manifests keep the `storage.r2` URIs they already recorded as +historical truth. A backfill copies bytes into the new bucket; it never rewrites +where those bytes were first published. `publish-raw` and `fetch-artifact` +enforce that: both refuse to restate a recorded `storage.r2` block under a +different bucket. + +The cutover therefore has one irreversible-looking step that is in fact additive +(creating and filling the new buckets), one cheap reversible step (flipping the +defaults, which is a one-line change in `chronicle/artifacts.py`), and no +deletion step at all. + +### 1. Create the new buckets + +Bucket creation needs a Cloudflare login carrying R2 permissions, so it is an +operator step rather than something CI can do. `wrangler.toml` already pins the +PolicyEngine account (`account_id = "20d90f557651969925eece96e58e24dc"`), so no +`CLOUDFLARE_ACCOUNT_ID` is needed even for a user who belongs to several +accounts: + +```bash +bunx wrangler login +uv run chronicle bootstrap-r2 --raw-bucket chronicle-raw --derived-bucket chronicle-derived +``` + +`bootstrap-r2` verifies authentication with `wrangler whoami` before creating +anything, and creating a bucket that already exists is not an error. + +### 2. Enumerate what has to be copied + +Tracked manifests are the authoritative registry of raw objects. Every one of +them points at `ledger-raw` today: + +```bash +git ls-files '*manifest*.yaml' '*manifest*.yml' \ + | xargs grep -ho 'r2://ledger-raw/[^"'"'"' ]*' | sort -u > /tmp/chronicle-raw-objects.txt +wc -l < /tmp/chronicle-raw-objects.txt +``` + +That is 186 distinct objects at `ff3efd3`, spread over 154 manifest files. Recount +rather than trusting the number: source packages land continuously, and each new +package adds objects. + +The derived bucket needs no backfill. Derived artifacts are reproducible by +definition and are already keyed by `{build_id}`, so a rebuild republishes them +into whichever bucket is configured. + +### 3. Backfill-copy the raw objects + +Keys are content-addressed and identical across buckets, so the copy is a +straight get/put per object: + +```bash +mkdir -p /tmp/chronicle-r2-backfill +while read -r uri; do + key=${uri#r2://ledger-raw/} + dest=/tmp/chronicle-r2-backfill/$key + mkdir -p "$(dirname "$dest")" + bunx wrangler r2 object get "ledger-raw/$key" --file "$dest" --remote + bunx wrangler r2 object put "chronicle-raw/$key" --file "$dest" --remote +done < /tmp/chronicle-raw-objects.txt +``` + +### 4. Verify the copy against the keys themselves + +Every raw key ends `.../{sha256}/{filename}`, so the key is its own checksum +witness and verification needs no manifest lookup: + +```bash +while read -r uri; do + key=${uri#r2://ledger-raw/} + expected=$(printf '%s\n' "$key" | awk -F/ '{print $(NF-1)}') + actual=$(shasum -a 256 "/tmp/chronicle-r2-backfill/$key" | cut -d' ' -f1) + [ "$expected" = "$actual" ] || echo "MISMATCH $key" +done < /tmp/chronicle-raw-objects.txt +``` + +Silence means every downloaded object hashes to the checksum its key claims. +That covers the read from `ledger-raw`; to cover the write to `chronicle-raw`, +re-download each key from the new bucket into a second directory and rerun the +same loop against it. + +### 5. Flip the defaults, in a follow-up PR + +Once the new buckets are filled and verified, change `DEFAULT_R2_RAW_BUCKET` and +`DEFAULT_R2_DERIVED_BUCKET` in `chronicle/artifacts.py` to `chronicle-raw` and +`chronicle-derived`. Until then, operators can opt in per-shell: + +```bash +export CHRONICLE_R2_RAW_BUCKET=chronicle-raw +export CHRONICLE_R2_DERIVED_BUCKET=chronicle-derived +``` + +New raw publications land in the new bucket from that point. Manifests written +before the flip keep pointing at `ledger-raw`, which is why the old bucket stays +readable. + +### 6. Set the ledger-era buckets read-only + +`ledger-raw` and `ledger-derived` keep serving archived witness records after the +flip. They should accept no further writes and should never be deleted. + ## Non-Goals Supabase should not store large raw binary artifacts. It should point to R2. From d032ec58d2e6e6d42305655e4ce8c12ac19f25c2 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:34:42 -0400 Subject: [PATCH 005/105] Pin the boundary guard against the bucket rename The consumer-fact boundary guard now matches derived artifacts on shape, but nothing tested the cases that motivated the change. Both of these pass every check the old literal matching applied, so they would have slipped through silently once the buckets are renamed: - a fact carrying no bucket and no key, only a `r2://chronicle-derived/derived/source/fact.json` URI. The old code compared the URI against two hardcoded prefixes, `r2://ledger-derived/` and `r2://ledger-raw/derived/`, and matched neither. - a `source_file` prefixed `chronicle-derived:`. The old code matched the literal prefix `ledger-derived:`. Co-Authored-By: Claude Fable 5.1 --- tests/test_chronicle_consumer_contract.py | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+) diff --git a/tests/test_chronicle_consumer_contract.py b/tests/test_chronicle_consumer_contract.py index cda97b66..30ac88d5 100644 --- a/tests/test_chronicle_consumer_contract.py +++ b/tests/test_chronicle_consumer_contract.py @@ -886,6 +886,29 @@ def test_export_consumer_facts_cli_rejects_contract_invalid_facts(tmp_path, caps }, "irs_soi.ty2024.table.us.taxable_interest_amount.ledger_derived", ), + # The guard used to match two hardcoded URI prefixes, so a URI naming + # any derived bucket other than `ledger-derived` did not match. Once the + # buckets are renamed (PolicyEngine/chronicle#143, mechanism 3) that is + # every derived URI, so the guard has to match on shape. + ( + { + "source_name": "irs_soi", + "source_file": "publisher.xlsx", + "raw_r2_bucket": None, + "raw_r2_key": None, + "raw_r2_uri": "r2://chronicle-derived/derived/source/fact.json", + }, + "publisher.raw.fact", + ), + ( + { + "source_name": "irs_soi", + "source_file": "chronicle-derived:taxable_interest.json", + "raw_r2_bucket": "ledger-raw", + "raw_r2_uri": "r2://ledger-raw/raw/source/publisher.xlsx", + }, + "publisher.raw.fact", + ), ], ) def test_consumer_contract_rejects_downstream_derived_target_facts( From b8d717ddcf12e2b847d3e3b0d600e3dabbb38343 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:29:41 -0400 Subject: [PATCH 006/105] Isolate the rename window for every test, not one module MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The env-isolation fixture lived in tests/test_chronicle_env.py, so only that module ran with CHRONICLE_/POLICYENGINE_LEDGER_/LEDGER_ variables stripped. Tests elsewhere assert the defaults those variables override — the raw and derived bucket names, the Supabase schema — and failed when an operator's shell had them set: CHRONICLE_R2_RAW_BUCKET=zzz CHRONICLE_SCHEMA=zzz uv run pytest -q FAILED tests/test_chronicle_artifacts.py::test_publish_source_artifacts_uploads_manifest_entries The fixture moves to tests/conftest.py as a suite-wide autouse fixture, and tests/test_chronicle_env.py gains an assertion that no rename-window variable reaches a test. db.supabase_client resolves LEDGER_SCHEMA and TARGETS_SCHEMA at import, which happens during collection — before any fixture runs — so the namespace test re-imports the module under the cleared environment instead of asserting on the constant it bound at collection time. Co-Authored-By: Claude Fable 5.1 --- tests/conftest.py | 33 +++++++++++++++++++++++++++++++ tests/test_chronicle_env.py | 33 +++++++++++++++++-------------- tests/test_chronicle_namespace.py | 24 +++++++++++++++------- 3 files changed, 68 insertions(+), 22 deletions(-) create mode 100644 tests/conftest.py diff --git a/tests/conftest.py b/tests/conftest.py new file mode 100644 index 00000000..378fa0b3 --- /dev/null +++ b/tests/conftest.py @@ -0,0 +1,33 @@ +"""Shared fixtures for the Chronicle test suite. + +Chronicle is mid-rename (PolicyEngine/chronicle#143, mechanism 3), so its +settings answer to three prefixes at once: ``CHRONICLE_``, and the ledger-era +``POLICYENGINE_LEDGER_`` and ``LEDGER_``. Any of them can be set in an +operator's shell, and many tests assert the defaults those variables override. +Isolation therefore belongs to the whole suite, not to one module. +""" + +from __future__ import annotations + +import os + +import pytest + +from chronicle.env import ( + CHRONICLE_ENV_PREFIX, + LEGACY_ENV_PREFIXES, + reset_env_deprecation_state, +) + +RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) + + +@pytest.fixture(autouse=True) +def isolated_rename_window_env(monkeypatch): + """Run every test with no rename-window variable inherited from the shell.""" + for name in list(os.environ): + if name.startswith(RENAME_WINDOW_PREFIXES): + monkeypatch.delenv(name, raising=False) + reset_env_deprecation_state() + yield + reset_env_deprecation_state() diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py index 4ffe0206..65029d14 100644 --- a/tests/test_chronicle_env.py +++ b/tests/test_chronicle_env.py @@ -2,7 +2,8 @@ Chronicle's operational stores migrate by dual-run (PolicyEngine/chronicle#143, mechanism 3): ``CHRONICLE_*`` names win, ledger-era names keep working behind a -deprecation warning. Every test here is hermetic — the fixture strips every +deprecation warning. Every test here is hermetic — the suite-wide +``isolated_rename_window_env`` fixture in ``tests/conftest.py`` strips every variable in the rename window from the ambient environment first. """ @@ -27,7 +28,6 @@ env_flag, env_names, env_value, - reset_env_deprecation_state, ) from chronicle.harness import main as harness_main from chronicle.source_package import ( @@ -35,19 +35,6 @@ SOURCE_ARTIFACT_FETCH_ENV, ) -RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) - - -@pytest.fixture(autouse=True) -def isolated_rename_window_env(monkeypatch): - """Run each test with no rename-window variable inherited from the shell.""" - for name in list(os.environ): - if name.startswith(RENAME_WINDOW_PREFIXES): - monkeypatch.delenv(name, raising=False) - reset_env_deprecation_state() - yield - reset_env_deprecation_state() - def _fake_wrangler(tmp_path, log): wrangler = tmp_path / "wrangler" @@ -61,6 +48,22 @@ def _fake_wrangler(tmp_path, log): # --------------------------------------------------------------------------- +def test_every_test_runs_with_the_rename_window_cleared(): + """Isolation is suite-wide (tests/conftest.py), not module-scoped. + + Modules well outside this one assert the defaults these variables override + — the raw and derived bucket names, the Supabase schema — so an operator's + shell must not reach any test. + """ + leaked = sorted( + name + for name in os.environ + if name.startswith((CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES)) + ) + + assert leaked == [] + + def test_env_names_puts_chronicle_first_then_ledger_era_names(): assert env_names("CHRONICLE_SOURCE_ARTIFACT_FETCH") == ( "CHRONICLE_SOURCE_ARTIFACT_FETCH", diff --git a/tests/test_chronicle_namespace.py b/tests/test_chronicle_namespace.py index 9b0ffd41..e8890948 100644 --- a/tests/test_chronicle_namespace.py +++ b/tests/test_chronicle_namespace.py @@ -1,5 +1,7 @@ """Tests for the Chronicle namespace.""" +import importlib + from chronicle.client import get_supabase_client from chronicle.normalization import convert_units from chronicle.targets import ( @@ -8,11 +10,7 @@ query_targets, ) from db.schema import Target as DbTarget -from db.supabase_client import ( - LEDGER_SCHEMA, - TARGETS_SCHEMA, - query_targets as db_query_targets, -) +from db.supabase_client import query_targets as db_query_targets def test_chronicle_targets_reexport_schema_objects(): @@ -29,8 +27,20 @@ def test_chronicle_client_reexports_supabase_client(): def test_chronicle_supabase_schema_boundaries_are_defaulted(): - assert LEDGER_SCHEMA == "ledger" - assert TARGETS_SCHEMA == "targets" + """The schema names are import-time constants, so re-read them here. + + ``db.supabase_client`` resolves them from the environment when it is first + imported, which happens at collection — before the suite-wide + ``isolated_rename_window_env`` fixture clears an operator's + ``CHRONICLE_SCHEMA``. Reloading under the cleared environment is what makes + this a test of the defaults rather than of the shell. + """ + import db.supabase_client + + supabase_client = importlib.reload(db.supabase_client) + + assert supabase_client.LEDGER_SCHEMA == "ledger" + assert supabase_client.TARGETS_SCHEMA == "targets" def test_chronicle_normalization_exports_helpers(): From 5cb357a1cee078c77c72a2e9c7c55b37cbd48bb2 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:29:53 -0400 Subject: [PATCH 007/105] Refuse to attach a recorded R2 URI to different bytes MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit A raw R2 key is content-addressed, so a recorded storage.r2 block is a claim about specific bytes. fetch-artifact preserved that block whenever one existed and the fetch did not re-upload into the same bucket, then rewrote the entry's sha256 and size_bytes from the newly fetched bytes. When a publisher re-publishes under the same URL and vintage — the IRS did exactly this to the 2022 IRA tables today (PolicyEngine/chronicle#225) — the manifest ended up describing the new bytes while pointing at the old object's key and URI. Verified against this branch's parent: a second fetch of different bytes leaves sha256=109dcf49… in an entry whose storage.r2 key is addressed by c63744a4…, both when the fetch only registers the bytes and when the bucket default has moved. Identity now decides, not the bucket: - the recorded key's last two segments (sha256, filename) are compared with the fetched bytes. Identical: the recorded block is preserved exactly, whichever bucket is configured now. Different: SourceArtifactRevisionError, raised before the cached artifact or its manifest entry is touched, naming recorded and fetched sha256/size_bytes and the ADR rule that the same vintage with new bytes is a new release revision. - --record-revision opts in: the fetched bytes get their own content-addressed key under the configured bucket, never the old key, and the superseded block moves to storage.previous_r2 with its sha256, size_bytes and fetched_at so the earlier bytes stay addressable. - publish-raw applies the same check before treating a recorded block as history, so a local file the recorded object does not hold is refused rather than uploaded. storage.previous_r2 is a sibling of storage.r2: every reader (inventory-artifacts, publish-raw, source_package._artifact_content and the suite's raw-R2-link acceptance check) reads storage.r2 alone, and publish-raw already spreads the existing storage block when it rewrites, so a revision survives publication untouched. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 330 ++++++++++++++++++++++++------ chronicle/harness.py | 57 ++++-- tests/test_chronicle_artifacts.py | 287 ++++++++++++++++++++++++++ 3 files changed, 599 insertions(+), 75 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 6d1e9781..8bf9c7cd 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -53,6 +53,18 @@ def default_r2_derived_bucket() -> str: return env_value(R2_DERIVED_BUCKET_ENV, default=DEFAULT_R2_DERIVED_BUCKET) +class SourceArtifactRevisionError(RuntimeError): + """Fetched bytes are not the bytes the recorded R2 object holds. + + Raw R2 keys are content-addressed, so a recorded ``storage.r2`` block is a + claim about specific bytes. When a publisher re-publishes under the same + URL and vintage, keeping that block would attach its provenance to bytes it + never described. Chronicle refuses instead: same vintage plus new bytes is a + new release revision (docs/adr-chronicle-fact-identity-v2.md), registered + with ``fetch-artifact --record-revision``. + """ + + # New UK and New Zealand uploads are namespaced by country. US objects predate # the country segment and deliberately keep their legacy ``raw/{source_id}`` # and ``derived/{source_id}`` shapes. Publisher directories are the stable @@ -419,11 +431,19 @@ def fetch_source_artifact( table: str | None = None, filename: str | None = None, upload_r2: bool = False, + record_revision: bool = False, r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: - """Fetch/register a source artifact and optionally upload it to R2.""" + """Fetch/register a source artifact and optionally upload it to R2. + + ``record_revision`` opts into registering a publisher revision: the fetched + bytes get their own content-addressed key under the configured bucket and + the superseded object moves to ``storage.previous_r2``. Without it, bytes + that disagree with the recorded object raise + :class:`SourceArtifactRevisionError` before anything is overwritten. + """ r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) resolved_r2_prefix = resolve_r2_prefix( @@ -438,14 +458,26 @@ def fetch_source_artifact( if not artifact_filename: raise ValueError("Could not infer artifact filename; pass --filename.") - output.mkdir(parents=True, exist_ok=True) - local_path = output / artifact_filename - local_path.write_bytes(content) - sha256 = hashlib.sha256(content).hexdigest() size_bytes = len(content) manifest_path = output / "manifest.yaml" + # Guard before the cached artifact is touched. A rejected fetch must leave + # the recorded bytes and their manifest entry exactly as they were. + _assert_recorded_object_holds_these_bytes( + manifest_path, + year=year, + filename=artifact_filename, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=r2_bucket, + record_revision=record_revision, + ) + + output.mkdir(parents=True, exist_ok=True) + local_path = output / artifact_filename + local_path.write_bytes(content) + r2_location = ArtifactStorageLocation( provider="r2", bucket=r2_bucket, @@ -484,6 +516,7 @@ def fetch_source_artifact( size_bytes=size_bytes, fetched_at=fetched_at, r2_location=(r2_location if upload_r2 and r2_upload and r2_upload.ok else None), + record_revision=record_revision, ) return ArtifactFetchReport( @@ -1076,17 +1109,181 @@ def _filename_from_url(source_url: str) -> str: return Path(unquote(parsed.path)).name -def _recorded_r2(spec: Any) -> dict[str, Any]: - """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" +def _read_manifest(manifest_path: Path) -> dict[str, Any]: + """Return a manifest's parsed payload, or an empty mapping.""" + if not manifest_path.exists(): + return {} + payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} + return payload if isinstance(payload, dict) else {} + + +def _manifest_file_spec(payload: dict[str, Any], year: Any) -> dict[str, Any]: + """Return one manifest ``files`` entry, or an empty mapping.""" + files = payload.get("files") + if not isinstance(files, dict): + return {} + spec = files.get(year) + return spec if isinstance(spec, dict) else {} + + +def _recorded_storage(spec: Any) -> dict[str, Any]: + """Return a manifest file spec's recorded ``storage`` block, if any.""" if not isinstance(spec, dict): return {} storage = spec.get("storage") - if not isinstance(storage, dict): - return {} - recorded = storage.get("r2") + return storage if isinstance(storage, dict) else {} + + +def _recorded_r2(spec: Any) -> dict[str, Any]: + """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" + recorded = _recorded_storage(spec).get("r2") return recorded if isinstance(recorded, dict) else {} +def _r2_key_identity(key: Any) -> tuple[str, str]: + """Return the ``(sha256, filename)`` a raw R2 key is addressed by. + + Raw keys are ``{prefix}/{source_id}/{package_id}/{year}/{sha256}/{filename}`` + (see :func:`build_r2_key`), so the last two segments say which bytes the + object holds. A key in any other shape yields empty strings and therefore + never matches fetched bytes. + """ + if not isinstance(key, str): + return ("", "") + parts = [part for part in key.split("/") if part] + if len(parts) < 2: + return ("", "") + return (parts[-2], parts[-1]) + + +def _r2_holds_these_bytes( + recorded_r2: dict[str, Any], + *, + sha256: str, + filename: str, +) -> bool: + """Whether a recorded ``storage.r2`` block addresses exactly these bytes.""" + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + return bool(recorded_sha256) and (recorded_sha256, recorded_filename) == ( + sha256, + Path(filename).name, + ) + + +def _revision_error_message( + *, + manifest_path: Path, + year: Any, + filename: str, + recorded_spec: dict[str, Any], + recorded_r2: dict[str, Any], + sha256: str, + size_bytes: int, + r2_bucket: str, +) -> str: + """Explain a refused fetch: recorded identity, fetched identity, next step.""" + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + declared_sha256 = recorded_spec.get("sha256") + recorded_size = ( + recorded_spec.get("size_bytes") if declared_sha256 == recorded_sha256 else None + ) + message = ( + f"{manifest_path} entry {year!r} already records the R2 object " + f"{recorded_r2.get('uri') or recorded_r2.get('key')}, which holds " + f"sha256={recorded_sha256 or 'unknown'} " + f"filename={recorded_filename or 'unknown'} " + f"size_bytes={recorded_size if recorded_size is not None else 'unknown'}. " + f"The fetched bytes are sha256={sha256} filename={Path(filename).name} " + f"size_bytes={size_bytes}. Chronicle will not attach a recorded, " + "content-addressed R2 URI to bytes it does not describe." + ) + if declared_sha256 and declared_sha256 != recorded_sha256: + message += ( + f" (The entry also declares sha256={declared_sha256}, which its own " + "R2 key contradicts: an earlier fetch rewrote the hash without " + "moving the object.)" + ) + return message + ( + " The same vintage with new bytes is a new release revision " + "(docs/adr-chronicle-fact-identity-v2.md). Re-run with " + "--record-revision to store the fetched bytes under their own " + f"content-addressed key in {r2_bucket} and keep the superseded object " + "in storage.previous_r2." + ) + + +def _assert_recorded_object_holds_these_bytes( + manifest_path: Path, + *, + year: Any, + filename: str, + sha256: str, + size_bytes: int, + r2_bucket: str, + record_revision: bool, +) -> None: + """Refuse a publisher revision that has not been opted into.""" + if record_revision: + return + recorded_spec = _manifest_file_spec(_read_manifest(manifest_path), year) + recorded_r2 = _recorded_r2(recorded_spec) + if not recorded_r2: + return + if _r2_holds_these_bytes(recorded_r2, sha256=sha256, filename=filename): + return + raise SourceArtifactRevisionError( + _revision_error_message( + manifest_path=manifest_path, + year=year, + filename=filename, + recorded_spec=recorded_spec, + recorded_r2=recorded_r2, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=r2_bucket, + ) + ) + + +def _superseding_storage( + recorded_spec: dict[str, Any], + *, + new_r2: dict[str, Any] | None, + superseded_at: str, +) -> dict[str, Any]: + """Return a storage block in which the recorded object becomes history. + + ``storage.r2`` only ever names the object that holds the entry's current + bytes. The superseded block is appended, oldest first, to + ``storage.previous_r2`` so the earlier bytes stay addressable by the URI + archived witness records already pin. + """ + storage = dict(_recorded_storage(recorded_spec)) + previous = storage.get("previous_r2") + entries = list(previous) if isinstance(previous, list) else [] + recorded_r2 = _recorded_r2(recorded_spec) + if recorded_r2: + entry = dict(recorded_r2) + recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2.get("key")) + if recorded_sha256: + entry["sha256"] = recorded_sha256 + if recorded_spec.get("sha256") == recorded_sha256: + # Only carry metadata the superseded key agrees with: a manifest + # can arrive here already describing the new bytes. + for field in ("size_bytes", "fetched_at", "source_url"): + value = recorded_spec.get(field) + if value is not None: + entry[field] = value + entry["superseded_at"] = superseded_at + entries.append(entry) + storage["previous_r2"] = entries + if new_r2 is None: + storage.pop("r2", None) + else: + storage["r2"] = new_r2 + return storage + + def _upsert_manifest( manifest_path: Path, *, @@ -1102,11 +1299,9 @@ def _upsert_manifest( size_bytes: int, fetched_at: str, r2_location: ArtifactStorageLocation | None, + record_revision: bool = False, ) -> None: - if manifest_path.exists(): - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - else: - payload = {} + payload = _read_manifest(manifest_path) payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) payload.setdefault("dataset", dataset) @@ -1120,18 +1315,44 @@ def _upsert_manifest( "size_bytes": size_bytes, "fetched_at": fetched_at, } - recorded_r2 = _recorded_r2(payload["files"].get(year)) + recorded_spec = _manifest_file_spec(payload, year) + recorded_storage = _recorded_storage(recorded_spec) + recorded_r2 = _recorded_r2(recorded_spec) new_r2 = r2_location.to_dict() if r2_location is not None else None - if recorded_r2 and ( - new_r2 is None or recorded_r2.get("bucket") != new_r2.get("bucket") + if recorded_r2 and _r2_holds_these_bytes( + recorded_r2, sha256=sha256, filename=filename ): - # A recorded storage.r2 block is historical truth: archived witness - # records pin raw R2 URLs by hash. Re-fetching under a renamed bucket - # copies bytes; it does not restate where the bytes were first - # published (PolicyEngine/chronicle#143, mechanism 3). - file_entry["storage"] = {"r2": recorded_r2} + # A recorded storage.r2 block for these exact bytes is historical + # truth: archived witness records pin raw R2 URLs by hash. Re-fetching + # under a renamed bucket copies bytes; it does not restate where the + # bytes were first published (PolicyEngine/chronicle#143, mechanism 3). + file_entry["storage"] = {**recorded_storage, "r2": recorded_r2} + elif recorded_r2: + # Different bytes under the same vintage. The guard in + # fetch_source_artifact refuses this without --record-revision; repeat + # the check here so no caller can reach a false-provenance write. + if not record_revision: + raise SourceArtifactRevisionError( + _revision_error_message( + manifest_path=manifest_path, + year=year, + filename=filename, + recorded_spec=recorded_spec, + recorded_r2=recorded_r2, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), + ) + ) + file_entry["storage"] = _superseding_storage( + recorded_spec, + new_r2=new_r2, + superseded_at=fetched_at, + ) elif new_r2 is not None: - file_entry["storage"] = {"r2": new_r2} + file_entry["storage"] = {**recorded_storage, "r2": new_r2} + elif recorded_storage: + file_entry["storage"] = dict(recorded_storage) payload["files"][year] = file_entry manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False), @@ -1192,7 +1413,10 @@ def _publish_raw_manifest_entry( if sha256_expected and sha256_actual != sha256_expected: errors.append("checksum_mismatch") - if errors: + def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: + """Report the entry unpublished, with nothing uploaded or rewritten.""" + if reason is not None: + errors.append(reason) return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), @@ -1210,6 +1434,27 @@ def _publish_raw_manifest_entry( None, ) + if errors: + return refuse() + + recorded_r2 = _recorded_r2(spec) + if recorded_r2 and not _r2_holds_these_bytes( + recorded_r2, sha256=sha256_actual or "", filename=filename + ): + # The recorded object is addressed by different bytes, so it is not + # this file's history. Uploading anyway would either publish under a + # key that misdescribes its content or restate a URI that belongs to + # the superseded bytes. Registering a publisher revision is + # `fetch-artifact --record-revision`, not a publish-time rewrite. + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + return refuse( + "recorded_r2_identity_mismatch:" + f"recorded_sha256={recorded_sha256 or 'unknown'}:" + f"recorded_filename={recorded_filename or 'unknown'}:" + f"local_sha256={sha256_actual}:" + f"local_filename={Path(filename).name}" + ) + location = ArtifactStorageLocation( provider="r2", bucket=r2_bucket, @@ -1223,54 +1468,21 @@ def _publish_raw_manifest_entry( package_path=manifest_path, ), ) - recorded_r2 = _recorded_r2(spec) recorded_bucket = recorded_r2.get("bucket") if recorded_bucket and recorded_bucket != location.bucket: # The recorded bucket is preserved history. Publishing the same bytes # into a renamed bucket is a backfill copy, not a restatement, so the # manifest must not be rewritten to point at the new bucket. - errors.append( + return refuse( "recorded_r2_bucket_is_preserved_history:" f"recorded={recorded_bucket}:requested={location.bucket}" ) - return ( - RawArtifactPublishEntry( - manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, - year=str(year), - filename=filename, - local_path=str(artifact_path), - sha256=sha256_actual, - size_bytes=size_bytes, - r2_location=None, - upload=None, - errors=tuple(errors), - ), - None, - ) recorded_key = recorded_r2.get("key") if recorded_key and recorded_key != location.key: - errors.append( + return refuse( "recorded_r2_key_disagrees_with_country_prefix:" f"recorded={recorded_key}:expected={location.key}" ) - return ( - RawArtifactPublishEntry( - manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, - year=str(year), - filename=filename, - local_path=str(artifact_path), - sha256=sha256_actual, - size_bytes=size_bytes, - r2_location=None, - upload=None, - errors=tuple(errors), - ), - None, - ) upload = _upload_r2_object( location, artifact_path, diff --git a/chronicle/harness.py b/chronicle/harness.py index 54c14139..dc9df8ba 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -5,6 +5,7 @@ import argparse import json import shlex +import sys from pathlib import Path from chronicle.artifacts import ( @@ -15,6 +16,7 @@ DerivedArtifactPublishReport, R2BootstrapReport, RawArtifactPublishReport, + SourceArtifactRevisionError, bootstrap_r2_buckets, fetch_source_artifact, inventory_source_artifacts, @@ -337,11 +339,17 @@ def fetch_artifact_file( table: str | None = None, filename: str | None = None, upload_r2: bool = False, + record_revision: bool = False, r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: - """Fetch/register a raw source artifact and optionally upload it to R2.""" + """Fetch/register a raw source artifact and optionally upload it to R2. + + Raises :class:`SourceArtifactRevisionError` when the fetched bytes are not + the bytes the manifest's recorded R2 object holds, unless + ``record_revision`` opts into registering the publisher revision. + """ return fetch_source_artifact( source_url, source_id=source_id, @@ -353,6 +361,7 @@ def fetch_artifact_file( table=table, filename=filename, upload_r2=upload_r2, + record_revision=record_revision, r2_bucket=r2_bucket, r2_prefix=r2_prefix, wrangler_command=wrangler_command, @@ -888,6 +897,17 @@ def main(argv: list[str] | None = None) -> int: action="store_true", help="Upload the artifact to R2 after local checksum capture.", ) + artifact_parser.add_argument( + "--record-revision", + action="store_true", + help=( + "Register a publisher revision: the fetched bytes get their own " + "content-addressed key under the configured bucket and the " + "superseded object moves to storage.previous_r2. Without this " + "flag, bytes that disagree with the recorded R2 object are " + "refused." + ), + ) artifact_parser.add_argument( "--r2-bucket", default=None, @@ -1322,21 +1342,26 @@ def main(argv: list[str] | None = None) -> int: print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if args.command == "fetch-artifact": - report = fetch_artifact_file( - args.url, - source_id=args.source_id, - package_id=args.package_id, - year=args.year, - output_dir=args.out_dir, - dataset=args.dataset, - source_page=args.source_page, - table=args.table, - filename=args.filename, - upload_r2=args.upload_r2, - r2_bucket=args.r2_bucket, - r2_prefix=args.r2_prefix, - wrangler_command=args.wrangler_command, - ) + try: + report = fetch_artifact_file( + args.url, + source_id=args.source_id, + package_id=args.package_id, + year=args.year, + output_dir=args.out_dir, + dataset=args.dataset, + source_page=args.source_page, + table=args.table, + filename=args.filename, + upload_r2=args.upload_r2, + record_revision=args.record_revision, + r2_bucket=args.r2_bucket, + r2_prefix=args.r2_prefix, + wrangler_command=args.wrangler_command, + ) + except SourceArtifactRevisionError as error: + print(f"error: {error}", file=sys.stderr) + return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 if args.command == "inventory-artifacts": diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index a96d6c25..7346584c 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -11,6 +11,7 @@ from chronicle.cli import main as cli_main from chronicle.artifacts import ( + SourceArtifactRevisionError, build_artifact_key, build_artifact_rows, build_derived_r2_key, @@ -827,3 +828,289 @@ def test_fetch_artifact_keeps_an_already_recorded_bucket(tmp_path, monkeypatch): assert ( second["files"][2023]["storage"]["r2"] == first["files"][2023]["storage"]["r2"] ) + + +# --------------------------------------------------------------------------- +# Publisher revisions +# +# A raw R2 key is content-addressed, so a recorded storage.r2 block is a claim +# about specific bytes. On 2026-09-02 the IRS re-published 22in05ira.xlsx and +# 22in06ira.xlsx under their existing URLs (PolicyEngine/chronicle#225): a +# repeated fetch must never pair those new bytes with the old object's URI. +# --------------------------------------------------------------------------- + +REPUBLISHED_URL = "https://www.irs.gov/pub/irs-soi/22in05ira.xlsx" +REPUBLISHED_FILENAME = "22in05ira.xlsx" +FIRST_PUBLICATION = b"IRA table 5, first publication" +SECOND_PUBLICATION = b"IRA table 5, silently re-published with revised rows" + + +def _wrangler_stub(tmp_path, log): + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + return wrangler + + +def _serve(monkeypatch, content): + """Serve ``content`` from the publisher URL, without touching the network.""" + + def _fake_read_artifact(source_url): + assert source_url == REPUBLISHED_URL + return content, REPUBLISHED_FILENAME + + monkeypatch.setattr("chronicle.artifacts._read_artifact", _fake_read_artifact) + + +def _fetch_republished(output_dir, wrangler, *, upload_r2=True, **kwargs): + return fetch_source_artifact( + REPUBLISHED_URL, + source_id="irs_soi", + package_id="soi-table-5", + year=2022, + output_dir=output_dir, + upload_r2=upload_r2, + wrangler_command=str(wrangler), + **kwargs, + ) + + +def test_repeated_fetch_of_identical_bytes_preserves_the_recorded_block( + tmp_path, monkeypatch +): + """Same bytes: the recorded block survives whatever bucket is configured.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + _fetch_republished(output_dir, wrangler) + first = yaml.safe_load(manifest_path.read_text())["files"][2022] + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + report = _fetch_republished(output_dir, wrangler) + second = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert report.valid + # The backfill copy really goes to the renamed bucket, but the manifest + # keeps recording where these bytes were first published. + assert report.r2_location.bucket == "chronicle-raw" + assert "chronicle-raw" in log.read_text() + assert second["storage"] == first["storage"] + assert second["storage"]["r2"]["bucket"] == "ledger-raw" + assert "previous_r2" not in second["storage"] + assert second["sha256"] == first["sha256"] + + +@pytest.mark.parametrize( + ("upload_r2", "configured_bucket"), + [ + pytest.param(True, None, id="reuploaded"), + # The two routes that reached a manifest in the wild: a fetch that only + # registers the bytes, and a fetch once the bucket default has moved. + # Both preserved the recorded block while rewriting sha256/size_bytes. + pytest.param(False, None, id="registered-without-upload"), + pytest.param(True, "chronicle-raw", id="after-the-bucket-rename"), + ], +) +def test_repeated_fetch_of_different_bytes_is_refused( + tmp_path, monkeypatch, upload_r2, configured_bucket +): + """A publisher revision must not inherit the recorded object's provenance.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + artifact_path = output_dir / REPUBLISHED_FILENAME + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + manifest_before = manifest_path.read_bytes() + uploads_before = log.read_text() + + if configured_bucket: + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", configured_bucket) + _serve(monkeypatch, SECOND_PUBLICATION) + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_republished(output_dir, wrangler, upload_r2=upload_r2) + + message = str(raised.value) + assert first_report.sha256 in message + assert hashlib.sha256(SECOND_PUBLICATION).hexdigest() in message + assert f"size_bytes={len(FIRST_PUBLICATION)}" in message + assert f"size_bytes={len(SECOND_PUBLICATION)}" in message + assert "release revision" in message + assert "--record-revision" in message + # Nothing was overwritten, copied or uploaded on the way to the refusal. + assert manifest_path.read_bytes() == manifest_before + assert artifact_path.read_bytes() == FIRST_PUBLICATION + assert log.read_text() == uploads_before + + +def test_record_revision_writes_a_new_key_and_keeps_the_previous_object( + tmp_path, monkeypatch +): + """The opt-in records the new bytes' own key under the configured bucket.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + superseded = yaml.safe_load(manifest_path.read_text())["files"][2022] + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + _serve(monkeypatch, SECOND_PUBLICATION) + report = _fetch_republished(output_dir, wrangler, record_revision=True) + revised = yaml.safe_load(manifest_path.read_text())["files"][2022] + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + + assert report.valid + # storage.r2 names the object that holds the entry's current bytes... + assert revised["sha256"] == revised_sha256 + assert revised["size_bytes"] == len(SECOND_PUBLICATION) + assert revised["storage"]["r2"]["bucket"] == "chronicle-raw" + assert revised["storage"]["r2"]["key"] == ( + f"raw/irs_soi/soi-table-5/2022/{revised_sha256}/{REPUBLISHED_FILENAME}" + ) + assert revised["storage"]["r2"]["uri"] == ( + f"r2://chronicle-raw/{revised['storage']['r2']['key']}" + ) + # ...and never the superseded key, which stays addressable as history. + previous = revised["storage"]["previous_r2"] + assert [entry["uri"] for entry in previous] == [superseded["storage"]["r2"]["uri"]] + assert previous[0]["bucket"] == "ledger-raw" + assert previous[0]["sha256"] == first_report.sha256 + assert previous[0]["size_bytes"] == len(FIRST_PUBLICATION) + assert previous[0]["fetched_at"] == superseded["fetched_at"] + assert previous[0]["superseded_at"] == revised["fetched_at"] + assert (output_dir / REPUBLISHED_FILENAME).read_bytes() == SECOND_PUBLICATION + assert f"chronicle-raw/{revised['storage']['r2']['key']}" in log.read_text() + + +def test_a_revised_manifest_still_reads_as_one_r2_linked_artifact( + tmp_path, monkeypatch +): + """storage.previous_r2 is a sibling key, so every storage.r2 reader is intact.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + _serve(monkeypatch, FIRST_PUBLICATION) + _fetch_republished(output_dir, wrangler) + _serve(monkeypatch, SECOND_PUBLICATION) + _fetch_republished(output_dir, wrangler, record_revision=True) + + inventory = inventory_source_artifacts(output_dir) + + assert inventory.valid + assert inventory.counts["r2_link_count"] == 1 + assert inventory.counts["checksum_mismatch_count"] == 0 + assert inventory.entries[0].r2["bucket"] == "ledger-raw" + assert inventory.entries[0].sha256_actual == ( + hashlib.sha256(SECOND_PUBLICATION).hexdigest() + ) + + +def test_publish_raw_refuses_a_file_the_recorded_object_does_not_hold( + tmp_path, monkeypatch +): + """Recorded sha256 != local sha256 is a revision, not a backfill.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + + # Reproduce the state a pre-fix fetch left behind: new bytes on disk, the + # entry's own hash rewritten, the recorded key still addressing the old + # bytes. + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + (output_dir / REPUBLISHED_FILENAME).write_bytes(SECOND_PUBLICATION) + manifest = yaml.safe_load(manifest_path.read_text()) + manifest["files"][2022]["sha256"] = revised_sha256 + manifest["files"][2022]["size_bytes"] = len(SECOND_PUBLICATION) + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + manifest_before = manifest_path.read_bytes() + uploads_before = log.read_text() + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + + assert not report.valid + assert report.entries[0].upload is None + assert report.entries[0].r2_location is None + assert report.entries[0].errors == ( + "recorded_r2_identity_mismatch:" + f"recorded_sha256={first_report.sha256}:" + f"recorded_filename={REPUBLISHED_FILENAME}:" + f"local_sha256={revised_sha256}:" + f"local_filename={REPUBLISHED_FILENAME}", + ) + assert log.read_text() == uploads_before + assert manifest_path.read_bytes() == manifest_before + + +def test_publish_raw_uploads_a_registered_revision_and_keeps_its_history( + tmp_path, monkeypatch +): + """Once the revision is registered, publishing it is ordinary work.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + _serve(monkeypatch, SECOND_PUBLICATION) + _fetch_republished(output_dir, wrangler, record_revision=True) + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + published = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert report.valid + assert published["storage"]["r2"]["key"].endswith( + f"/{revised_sha256}/{REPUBLISHED_FILENAME}" + ) + assert [entry["sha256"] for entry in published["storage"]["previous_r2"]] == [ + first_report.sha256 + ] + + +def test_fetch_artifact_cli_refuses_a_revision_then_records_it_on_request( + tmp_path, monkeypatch, capsys +): + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + argv = [ + "fetch-artifact", + "--url", + REPUBLISHED_URL, + "--source-id", + "irs_soi", + "--package-id", + "soi-table-5", + "--year", + "2022", + "--out-dir", + str(output_dir), + "--upload-r2", + "--wrangler-command", + str(wrangler), + ] + _serve(monkeypatch, FIRST_PUBLICATION) + assert harness_main(argv) == 0 + capsys.readouterr() + + _serve(monkeypatch, SECOND_PUBLICATION) + refused = harness_main(argv) + refusal = capsys.readouterr() + + assert refused == 1 + assert "--record-revision" in refusal.err + assert refusal.out == "" + + assert harness_main([*argv, "--record-revision"]) == 0 + recorded = json.loads(capsys.readouterr().out) + + assert recorded["sha256"] == hashlib.sha256(SECOND_PUBLICATION).hexdigest() + assert recorded["r2_location"]["key"].endswith( + f"/{recorded['sha256']}/{REPUBLISHED_FILENAME}" + ) From 58706dd61a57b8839c64fd83afacd2dff161bb39 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:30:52 -0400 Subject: [PATCH 008/105] Document publisher revisions alongside the bucket rename The rename-window text said a recorded storage.r2 block is preserved because the bucket may have moved. That is the weaker half of the rule: the block is preserved because it addresses the bytes in hand. Adds a Publisher Revisions section covering the identity check, the refusal, --record-revision, the storage.previous_r2 shape and why every existing storage.r2 reader is unaffected, and points the bucket-cutover, publish-flow and harness passages at it. Co-Authored-By: Claude Fable 5.1 --- README.md | 7 +++ docs/agent-source-package-harness.md | 9 ++++ docs/storage-architecture.md | 73 ++++++++++++++++++++++++++-- 3 files changed, 86 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index e0ade235..6b6e170c 100644 --- a/README.md +++ b/README.md @@ -367,6 +367,13 @@ uv run chronicle fetch-artifact \ --table "Publication 1304 Table 1.2" \ --upload-r2 +# Re-fetching is safe: identical bytes keep the recorded storage.r2 block, and +# bytes that disagree with it are refused. When a publisher has re-published +# under the same URL and vintage, register the revision explicitly — the new +# bytes get their own content-addressed key and the superseded object is kept +# in storage.previous_r2: +uv run chronicle fetch-artifact ... --record-revision + # Audit local manifests and checksums: uv run chronicle inventory-artifacts --root db/data diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 77872ae8..fafce828 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -22,6 +22,15 @@ For already-downloaded manifest artifacts, agents should run `uv run chronicle publish-raw --root db/data` to upload checksum-verified bytes to R2 and write `storage.r2` metadata back into each manifest entry. +Both commands treat a recorded `storage.r2` block as a claim about specific +bytes, because raw keys are content-addressed. Re-fetching or publishing bytes +the recorded object does not hold is refused; when a publisher has re-published +under the same URL and vintage, register the revision with +`uv run chronicle fetch-artifact ... --record-revision`, which stores the new +bytes under their own key and keeps the superseded object in +`storage.previous_r2`. See +[Publisher Revisions](storage-architecture.md#publisher-revisions). + Builds do not require production raw bytes to be committed to Git. Source packages first read packaged fixture bytes, then `CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` (defaulting to diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index e381353f..6f4cfb00 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -91,6 +91,68 @@ Legacy US derived keys likewise remain `derived/{source_id}/...`. Derived artifacts are reproducible and may be replaced by a new build, but a specific `{build_id}` path should be immutable once published. +## Publisher Revisions + +A raw key embeds the sha256 of the bytes it holds, so a manifest's recorded +`storage.r2` block is a claim about specific bytes, not a pointer to a file +name. Publishers do not always honor that: on 2026-09-02 the IRS re-published +`22in05ira.xlsx` and `22in06ira.xlsx` under their existing URLs +(PolicyEngine/chronicle#225). + +`fetch-artifact` therefore compares the recorded key's `{sha256}/{filename}` +tail with the bytes it just fetched, before it writes anything: + +- **Identical** — the recorded block is preserved exactly, whichever bucket is + configured now. Re-fetching after the bucket rename copies bytes; it does not + restate where they were first published. +- **Different** — the fetch is refused. Nothing is overwritten: not the cached + artifact, not the manifest entry, and no object is uploaded. The error names + the recorded and the fetched `sha256`/`size_bytes`. Per + `docs/adr-chronicle-fact-identity-v2.md`, the same vintage with new bytes is a + new release revision, so registering it is a decision an operator makes, not a + silent rewrite. + +`--record-revision` makes that decision explicit. The fetched bytes get their +own content-addressed key under the configured bucket — never the recorded key — +and the superseded block moves to `storage.previous_r2`: + +```yaml +files: + 2022: + filename: 22in05ira.xlsx + sha256: + size_bytes: + fetched_at: "2026-09-02T17:04:11+00:00" + storage: + r2: + provider: r2 + bucket: chronicle-raw + key: raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + uri: r2://chronicle-raw/raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + previous_r2: + - provider: r2 + bucket: ledger-raw + key: raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + uri: r2://ledger-raw/raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + sha256: + size_bytes: + fetched_at: "2026-06-11T14:22:05+00:00" + superseded_at: "2026-09-02T17:04:11+00:00" +``` + +`storage.r2` only ever names the object that holds the entry's current bytes, +and `previous_r2` lists superseded objects oldest first, so the bytes an +archived witness record pinned stay addressable at the URI it pinned. Every +reader — `inventory-artifacts`, `publish-raw`, source-package artifact loading, +and the suite's raw-R2-link acceptance check — reads `storage.r2` alone, so a +revised entry reads exactly like an unrevised one; `publish-raw` preserves the +rest of the `storage` block when it writes back. + +`publish-raw` applies the same identity check before treating a recorded block +as history. A local file the recorded object does not hold is reported as +`recorded_r2_identity_mismatch` and nothing is uploaded: registering a revision +is a fetch-time decision, not a publish-time rewrite. + ## Relational Registry Contract The hosted `chronicle` schema should be the lookup surface for Chronicle, not the place @@ -121,7 +183,10 @@ The intended flow is: 1. Register raw source artifacts with `uv run chronicle fetch-artifact`, which writes local bytes, records checksums in `manifest.yaml`, and can upload the - exact bytes to the raw archive. Existing manifest-declared artifacts can be + exact bytes to the raw archive. Re-fetching an entry whose bytes the + publisher has changed is refused unless the revision is registered with + `--record-revision`; see [Publisher Revisions](#publisher-revisions). + Existing manifest-declared artifacts can be checksum-validated, uploaded, and linked with `uv run chronicle publish-raw`. Production package specs may omit raw bytes from Git as long as the manifest keeps `source_url` and SHA-256 metadata; builds can fill @@ -217,8 +282,10 @@ exception to the last step. Archived witness records pin raw R2 URLs by hash, so deleted, and manifests keep the `storage.r2` URIs they already recorded as historical truth. A backfill copies bytes into the new bucket; it never rewrites where those bytes were first published. `publish-raw` and `fetch-artifact` -enforce that: both refuse to restate a recorded `storage.r2` block under a -different bucket. +enforce that: a recorded block that addresses the bytes in hand is preserved +whichever bucket is configured, and `publish-raw` refuses to restate it under a +different one. Bytes that the recorded object does not hold are not that +object's history at all; see [Publisher Revisions](#publisher-revisions). The cutover therefore has one irreversible-looking step that is in fact additive (creating and filling the new buckets), one cheap reversible step (flipping the From 0e7e095bcde1cfc12d4f09730f73957665d5b3a1 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:32:09 -0400 Subject: [PATCH 009/105] Record the gate round-1 fixes in PROGRESS.md Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 40 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 40 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index da5d13e5..27b1babe 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -48,6 +48,42 @@ Lane C5's handoff notes previously lived here; its durable record is `README.md` follow. Verified 186 distinct `ledger-raw` objects across 154 tracked manifest files, every key content-addressed by sha256. +## Review fixes (gate round 1) + +The Fable+Sol gate requested changes; both findings are applied on this branch. + +- **[high] `fetch-artifact` could attach a recorded R2 URI to new bytes.** The + preserve rule keyed on the bucket, so a repeated fetch that did not re-upload + into the same bucket kept the recorded `storage.r2` block while rewriting the + entry's `sha256`/`size_bytes`. Reproduced against this branch's parent: the + entry ends up declaring `109dcf49…` with a key addressed by `c63744a4…`, both + when the fetch only registers the bytes and when the bucket default has moved. + The rule now keys on identity — the recorded key's `{sha256}/{filename}` tail + against the fetched bytes. Identical preserves the block exactly; different + raises `SourceArtifactRevisionError` before the cached artifact or its + manifest entry is touched, naming recorded and fetched `sha256`/`size_bytes` + and the ADR rule that same vintage plus new bytes is a new release revision. + `--record-revision` opts in: the new bytes get their own content-addressed key + under the configured bucket, never the old key, and the superseded block moves + to `storage.previous_r2`. `publish-raw` applies the same check before treating + a recorded block as history (`recorded_r2_identity_mismatch`, nothing + uploaded). +- **[low] Env isolation was scoped to one module.** The autouse fixture moved to + `tests/conftest.py` and now clears all three prefixes for every test. + `db.supabase_client` resolves `LEDGER_SCHEMA` at import — during collection, + before any fixture — so `tests/test_chronicle_namespace.py` re-imports it + under the cleared environment instead of asserting the constant it bound at + collection time. + +`storage.previous_r2` is a sibling key, chosen because every reader +(`inventory-artifacts`, `publish-raw`, `source_package._artifact_content`, the +suite's raw-R2-link acceptance check) reads `storage.r2` alone, and +`publish-raw` already spreads the rest of the `storage` block when it writes +back, so a revision survives publication untouched. All 180 tracked manifest +entries that carry a `storage.r2` block are content-addressed and agree with +their declared `sha256` and `filename`, so the identity check never fires on +tracked data. + ## Verification - `uv run pytest -q`: green. @@ -56,6 +92,10 @@ Lane C5's handoff notes previously lived here; its durable record is files are unformatted on `main` already and are byte-identical here; CI runs `ruff check` only, so they are pre-existing and out of scope. - CI's db CLI gate (`chronicle init` / `load all` / `stats`): passes. +- `CHRONICLE_R2_RAW_BUCKET=zzz CHRONICLE_SCHEMA=zzz uv run pytest -q`: green. + Before the shared fixture it failed five tests — four bucket-default + assertions in `tests/test_chronicle_artifacts.py` and the collection-time + schema constant in `tests/test_chronicle_namespace.py`. ## Next From 83b8ba08b92fb8f9ff42bdc497d584318f268042 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:33:18 -0400 Subject: [PATCH 010/105] Read the recorded object's identity from its URI too A storage.r2 block written by Chronicle always carries both key and uri, but a hand-written one need not. The identity read now falls back to the uri, whose last two segments are the same {sha256}/{filename}, so a block that records only where the object lives is still recognized as history for the bytes it holds instead of being refused as a revision. Also covers registering a revision without an upload: the entry is left with no storage.r2 at all rather than a pointer to bytes R2 does not hold, the superseded object stays in storage.previous_r2, and publish-raw completes the registration. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 22 +++++----- tests/test_chronicle_artifacts.py | 69 +++++++++++++++++++++++++++++++ 2 files changed, 81 insertions(+), 10 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 8bf9c7cd..9421ab08 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1140,17 +1140,19 @@ def _recorded_r2(spec: Any) -> dict[str, Any]: return recorded if isinstance(recorded, dict) else {} -def _r2_key_identity(key: Any) -> tuple[str, str]: - """Return the ``(sha256, filename)`` a raw R2 key is addressed by. +def _r2_key_identity(recorded_r2: dict[str, Any]) -> tuple[str, str]: + """Return the ``(sha256, filename)`` a recorded R2 object is addressed by. Raw keys are ``{prefix}/{source_id}/{package_id}/{year}/{sha256}/{filename}`` (see :func:`build_r2_key`), so the last two segments say which bytes the - object holds. A key in any other shape yields empty strings and therefore - never matches fetched bytes. + object holds; the URI ends in the same two segments and stands in for a + block that records only that. A locator in any other shape yields empty + strings and therefore never matches fetched bytes. """ - if not isinstance(key, str): + locator = recorded_r2.get("key") or recorded_r2.get("uri") + if not isinstance(locator, str): return ("", "") - parts = [part for part in key.split("/") if part] + parts = [part for part in locator.split("/") if part] if len(parts) < 2: return ("", "") return (parts[-2], parts[-1]) @@ -1163,7 +1165,7 @@ def _r2_holds_these_bytes( filename: str, ) -> bool: """Whether a recorded ``storage.r2`` block addresses exactly these bytes.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) return bool(recorded_sha256) and (recorded_sha256, recorded_filename) == ( sha256, Path(filename).name, @@ -1182,7 +1184,7 @@ def _revision_error_message( r2_bucket: str, ) -> str: """Explain a refused fetch: recorded identity, fetched identity, next step.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) declared_sha256 = recorded_spec.get("sha256") recorded_size = ( recorded_spec.get("size_bytes") if declared_sha256 == recorded_sha256 else None @@ -1264,7 +1266,7 @@ def _superseding_storage( recorded_r2 = _recorded_r2(recorded_spec) if recorded_r2: entry = dict(recorded_r2) - recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2) if recorded_sha256: entry["sha256"] = recorded_sha256 if recorded_spec.get("sha256") == recorded_sha256: @@ -1446,7 +1448,7 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: # key that misdescribes its content or restate a URI that belongs to # the superseded bytes. Registering a publisher revision is # `fetch-artifact --record-revision`, not a publish-time rewrite. - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) return refuse( "recorded_r2_identity_mismatch:" f"recorded_sha256={recorded_sha256 or 'unknown'}:" diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 7346584c..329ac1cf 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1114,3 +1114,72 @@ def test_fetch_artifact_cli_refuses_a_revision_then_records_it_on_request( assert recorded["r2_location"]["key"].endswith( f"/{recorded['sha256']}/{REPUBLISHED_FILENAME}" ) + + +def test_record_revision_without_an_upload_records_no_current_object( + tmp_path, monkeypatch +): + """An offline revision keeps history without claiming the new bytes exist. + + Registering a revision without ``--upload-r2`` leaves the entry with no + ``storage.r2`` at all rather than a pointer to bytes R2 does not hold. The + superseded object stays addressable, and a later publish-raw completes the + registration. + """ + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + + _serve(monkeypatch, SECOND_PUBLICATION) + _fetch_republished(output_dir, wrangler, upload_r2=False, record_revision=True) + registered = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert "r2" not in registered["storage"] + assert [entry["sha256"] for entry in registered["storage"]["previous_r2"]] == [ + first_report.sha256 + ] + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + published = yaml.safe_load(manifest_path.read_text())["files"][2022] + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + + assert report.valid + assert published["storage"]["r2"]["key"].endswith( + f"/{revised_sha256}/{REPUBLISHED_FILENAME}" + ) + assert [entry["sha256"] for entry in published["storage"]["previous_r2"]] == [ + first_report.sha256 + ] + + +def test_a_recorded_block_that_only_carries_a_uri_is_still_recognized( + tmp_path, monkeypatch +): + """Identity reads the URI when a hand-written block records no key.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + _fetch_republished(output_dir, wrangler) + manifest = yaml.safe_load(manifest_path.read_text()) + recorded = manifest["files"][2022]["storage"]["r2"] + manifest["files"][2022]["storage"]["r2"] = { + "provider": recorded["provider"], + "bucket": recorded["bucket"], + "uri": recorded["uri"], + } + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + uri_only = manifest["files"][2022]["storage"]["r2"] + + report = _fetch_republished(output_dir, wrangler) + preserved = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert report.valid + assert preserved["storage"]["r2"] == uri_only + + _serve(monkeypatch, SECOND_PUBLICATION) + with pytest.raises(SourceArtifactRevisionError): + _fetch_republished(output_dir, wrangler) From 45b7d25effbffbab2548a7cb071b9be894429e31 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:35:35 -0400 Subject: [PATCH 011/105] Pin the preserved block's field order too Dict equality would pass on a block that was rebuilt with the same values in a different order, which is not what "preserved" means for a manifest that gets dumped back to YAML. Co-Authored-By: Claude Fable 5.1 --- tests/test_chronicle_artifacts.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 329ac1cf..93b4d429 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -897,6 +897,8 @@ def test_repeated_fetch_of_identical_bytes_preserves_the_recorded_block( assert report.r2_location.bucket == "chronicle-raw" assert "chronicle-raw" in log.read_text() assert second["storage"] == first["storage"] + # Field order too, so the block is byte-for-byte identical once dumped. + assert list(second["storage"]["r2"].items()) == list(first["storage"]["r2"].items()) assert second["storage"]["r2"]["bucket"] == "ledger-raw" assert "previous_r2" not in second["storage"] assert second["sha256"] == first["sha256"] From 8cf288638c9a8bc3817fd9f84c0df4caa53db2cf Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:37:48 -0400 Subject: [PATCH 012/105] Say what the reproduction actually ran The hashes named the fix lane's two-body fixture, not IRS bytes. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 27b1babe..38e669e2 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -55,9 +55,11 @@ The Fable+Sol gate requested changes; both findings are applied on this branch. - **[high] `fetch-artifact` could attach a recorded R2 URI to new bytes.** The preserve rule keyed on the bucket, so a repeated fetch that did not re-upload into the same bucket kept the recorded `storage.r2` block while rewriting the - entry's `sha256`/`size_bytes`. Reproduced against this branch's parent: the - entry ends up declaring `109dcf49…` with a key addressed by `c63744a4…`, both - when the fetch only registers the bytes and when the bucket default has moved. + entry's `sha256`/`size_bytes`. Reproduced against this branch's parent by + serving two different bodies from one URL: the entry ends up declaring the + fetched bytes' `sha256` under a key addressed by the superseded bytes' one, + both when the fetch only registers the bytes and when the bucket default has + moved. The rule now keys on identity — the recorded key's `{sha256}/{filename}` tail against the fetched bytes. Identical preserves the block exactly; different raises `SourceArtifactRevisionError` before the cached artifact or its From c074341d359f0753ca3020912b2329266c319c7b Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:09:03 -0400 Subject: [PATCH 013/105] Record the round-2 gate findings and the corpus scan The seven findings, in the order the fixes depend on each other, plus what a scan of every tracked manifest says about how strict the locator check can be: 187 entries, all content-addressed, no contradictory field. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 33 +++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 38e669e2..e51ea2ca 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -86,6 +86,39 @@ entries that carry a `storage.r2` block are content-addressed and agree with their declared `sha256` and `filename`, so the identity check never fires on tracked data. +## Review fixes (gate round 2) + +The second Fable+Sol gate requested changes again. Seven findings, each fixed +with a regression test on this branch. Plan, in dependency order: + +1. **[high] `CHRONICLE_SCHEMA` does not reach the Supabase mirror writer.** + `chronicle/harness.py` and `chronicle/mirror.py` default the schema to the + literal `"ledger"`; only `db.supabase_client` reads the renamed variable. + Resolve through the shared helper whenever no explicit `--schema` is given. +2. **[high] The derived-fact boundary check is not rename-safe.** + `chronicle/consumer_contract.py` matches the `.ledger_derived` suffix only. +3. **[high] `fetch-artifact` cannot address a package's non-default manifest.** + Seven tracked packages keep a `manifest_*_source_package.yaml`; three + directories keep two. A fetch into one of them writes a third manifest and + never sees the recorded block. +4. **[high] Revision protection vanishes when the entry has no `storage.r2`.** +5. **[medium] Recorded-R2 locator fields must be cross-checked**, not read as + key-or-URI, before a block is preserved or published. +6. **[medium] `_read_manifest` must reject a malformed document**, not treat a + non-mapping YAML payload as an absent manifest. +7. **[low] Schema resolution must be lazy** so no legacy variable is read at + collection, before the autouse isolation fixture runs. + +## State (round 2) + +- Read both gate rounds on PolicyEngine/chronicle#226 and the code each finding + names. +- Scanned all 154 tracked manifest files (187 `files` entries, every one + carrying `storage.r2`): every recorded block supplies provider, bucket, key + and uri; every key is content-addressed; every declared `sha256`/`filename` + agrees with its key tail; no `uri` contradicts its `key`. Strict locator + validation therefore refuses nothing that is tracked today. + ## Verification - `uv run pytest -q`: green. From 5e41b75226faa4f6b5fc591adf039694d7a3c0ab Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:09:55 -0400 Subject: [PATCH 014/105] Reject the chronicle spelling of the derived-row marker The consumer boundary matched the `.ledger_derived` suffix literally, so a producer that renamed its derived rows with everything else would have walked a downstream target fact straight through the guard. Match the whole final dot-segment against both spellings instead. Without the fix the two new cases fail: the chronicle-spelled record id validates clean. Co-Authored-By: Claude Fable 5.1 --- chronicle/consumer_contract.py | 16 ++++++- tests/test_chronicle_consumer_contract.py | 52 +++++++++++++++++++++++ 2 files changed, 67 insertions(+), 1 deletion(-) diff --git a/chronicle/consumer_contract.py b/chronicle/consumer_contract.py index f9c93f3a..17315a13 100644 --- a/chronicle/consumer_contract.py +++ b/chronicle/consumer_contract.py @@ -476,6 +476,20 @@ def _r2_uri_parts(uri: str) -> tuple[str, str]: return bucket, key +# The marker a downstream target row carries in its source_record_id. It moves +# with the rename window: producers write `ledger_derived` today and +# `chronicle_derived` once they migrate (PolicyEngine/chronicle#143, mechanism +# 3), so the boundary has to reject both spellings identically or the guard +# stops firing the moment a producer renames. +DERIVED_SOURCE_RECORD_SUFFIXES = frozenset({"ledger_derived", "chronicle_derived"}) + + +def _is_derived_source_record_id(source_record_id: str) -> bool: + """Whether a source_record_id marks a downstream derived target row.""" + _, separator, suffix = source_record_id.rpartition(".") + return bool(separator) and suffix in DERIVED_SOURCE_RECORD_SUFFIXES + + def _points_at_derived(bucket: str, key: str) -> bool: """Whether an R2 bucket/key pair addresses derived build output. @@ -520,7 +534,7 @@ def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: "Chronicle consumer facts must point at raw source artifacts, not " "derived build artifacts." ) - if source_record_id.endswith(".ledger_derived"): + if _is_derived_source_record_id(source_record_id): return ( "Chronicle source_record_id must identify a publisher-backed row, not " "a downstream derived target row." diff --git a/tests/test_chronicle_consumer_contract.py b/tests/test_chronicle_consumer_contract.py index 30ac88d5..9cc730c1 100644 --- a/tests/test_chronicle_consumer_contract.py +++ b/tests/test_chronicle_consumer_contract.py @@ -909,6 +909,18 @@ def test_export_consumer_facts_cli_rejects_contract_invalid_facts(tmp_path, caps }, "publisher.raw.fact", ), + # The derived-row marker renames with everything else, so the guard has + # to reject the chronicle spelling the same way it rejects the ledger + # one (PolicyEngine/chronicle#143, mechanism 3). + ( + { + "source_name": "irs_soi", + "source_file": "publisher.xlsx", + "raw_r2_bucket": "ledger-raw", + "raw_r2_uri": "r2://ledger-raw/raw/source/publisher.xlsx", + }, + "irs_soi.ty2024.table.us.taxable_interest_amount.chronicle_derived", + ), ], ) def test_consumer_contract_rejects_downstream_derived_target_facts( @@ -928,6 +940,46 @@ def test_consumer_contract_rejects_downstream_derived_target_facts( assert "derived_fact_provenance" in {error.code for error in report.errors} +def test_derived_record_marker_is_rejected_in_either_spelling(): + """Both rename-window spellings produce the identical boundary error.""" + fact = _soi_agi_fact() + base = "irs_soi.ty2024.table.us.taxable_interest_amount" + + reports = { + suffix: validate_consumer_fact_contract( + [replace(fact, source_record_id=f"{base}.{suffix}")] + ) + for suffix in ("ledger_derived", "chronicle_derived") + } + + ledger_errors = [ + (error.code, error.message) for error in reports["ledger_derived"].errors + ] + chronicle_errors = [ + (error.code, error.message) for error in reports["chronicle_derived"].errors + ] + assert ledger_errors == chronicle_errors + assert "derived_fact_provenance" in {code for code, _ in ledger_errors} + + +@pytest.mark.parametrize( + "source_record_id", + [ + # A publisher-backed row that merely contains the marker as a word, or + # carries it without the separating dot, is not a derived target row. + "irs_soi.ty2024.table.us.chronicle_derived_totals", + "irs_soi.ty2024.table.us.ledger_derived_totals", + "chronicle_derived", + ], +) +def test_derived_record_marker_matches_the_whole_final_segment(source_record_id): + fact = replace(_soi_agi_fact(), source_record_id=source_record_id) + + report = validate_consumer_fact_contract([fact]) + + assert report.valid + + def test_export_consumer_facts_cli_preserves_decimal_values(tmp_path, capsys): input_path = tmp_path / "facts.jsonl" output_path = tmp_path / "consumer_facts.jsonl" From fa4a82f110f28a2a041c489ee652e244c5b51558 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:15:03 -0400 Subject: [PATCH 015/105] Let CHRONICLE_SCHEMA configure the mirror writer, at call time The hosted mirror loader is the primary writer into the schema, and both it and its CLI defaulted to the literal "ledger": setting CHRONICLE_SCHEMA to rehearse a cutover moved the read-side client and left the writer pointed at production. The default now resolves through the shared helper -- one home for the CHRONICLE_SCHEMA -> POLICYENGINE_LEDGER_SCHEMA -> LEDGER_SCHEMA -> "ledger" ladder -- whenever no explicit --schema is supplied. Defaults are unchanged. Resolution is a function rather than a module constant, so db.supabase_client no longer binds the schema while being imported. That import happens at collection, before any fixture, which is why the namespace test needed a reload dance to assert the defaults; it now just calls the resolver. A pytest_configure hook clears the rename window before collection too, so no module can read (or warn from) an operator's shell on the way in. Six tests fail against the previous code: the loader and its CLI ignore the variable, and the namespace assertions see the collection-time constant. Co-Authored-By: Claude Fable 5.1 --- chronicle/env.py | 23 ++++++ chronicle/harness.py | 16 +++- chronicle/mirror.py | 13 +++- db/supabase_client.py | 43 ++++++++--- tests/conftest.py | 17 ++++ tests/test_chronicle_env.py | 45 ++++++++--- tests/test_chronicle_mirror.py | 124 ++++++++++++++++++++++++++++++ tests/test_chronicle_namespace.py | 34 ++++---- 8 files changed, 276 insertions(+), 39 deletions(-) diff --git a/chronicle/env.py b/chronicle/env.py index 3a2f7af6..0d4bbbad 100644 --- a/chronicle/env.py +++ b/chronicle/env.py @@ -19,8 +19,11 @@ __all__ = [ "CHRONICLE_ENV_PREFIX", + "CHRONICLE_SCHEMA_ENV", "ChronicleEnvDeprecationWarning", + "DEFAULT_CHRONICLE_SCHEMA", "LEGACY_ENV_PREFIXES", + "default_chronicle_schema", "env_flag", "env_names", "env_value", @@ -34,6 +37,14 @@ TRUTHY_ENV_VALUES = frozenset({"1", "true", "yes", "on"}) +CHRONICLE_SCHEMA_ENV = "CHRONICLE_SCHEMA" + +# The hosted Postgres schema is still named "ledger". Renaming the schema value +# is a later slice of PolicyEngine/chronicle#143, coordinated with the CI +# writers that already target it; only the variable that overrides the name has +# moved to the chronicle prefix. +DEFAULT_CHRONICLE_SCHEMA = "ledger" + class ChronicleEnvDeprecationWarning(FutureWarning): """A ledger-era environment variable supplied a Chronicle setting. @@ -126,6 +137,18 @@ def env_value(*names: str, default: _Default = None) -> str | _Default: return default if value is None else value +def default_chronicle_schema() -> str: + """Resolve the Chronicle schema: ``$CHRONICLE_SCHEMA``, else the default. + + Every reader of the setting goes through this function so the lookup ladder + and the default have one home. It resolves at call time rather than at + import: a module-level constant binds whatever the shell held when the + module was first imported, which for a library means an arbitrary moment + the caller cannot control, and for the test suite means collection. + """ + return env_value(CHRONICLE_SCHEMA_ENV, default=DEFAULT_CHRONICLE_SCHEMA) + + def env_flag(*names: str) -> bool: """Return whether the first set value across ``names`` reads as true. diff --git a/chronicle/harness.py b/chronicle/harness.py index dc9df8ba..a5646553 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -37,6 +37,7 @@ ) from chronicle.core import AggregateFact, ValidationReport, validate_facts from chronicle.database import ChronicleDbBuildReport, build_chronicle_db +from chronicle.env import DEFAULT_CHRONICLE_SCHEMA from chronicle.mirror import ( ChronicleMirrorExportReport, SupabaseMirrorLoadReport, @@ -452,12 +453,16 @@ def export_chronicle_db_table_files( def load_supabase_mirror_files( input_dir: str | Path, *, - schema: str = "ledger", + schema: str | None = None, batch_size: int = 500, dry_run: bool = False, build_artifacts_path: str | Path | None = None, ) -> SupabaseMirrorLoadReport: - """Load exported Chronicle JSONL mirror files into Supabase/Postgres.""" + """Load exported Chronicle JSONL mirror files into Supabase/Postgres. + + ``schema`` of None resolves to ``$CHRONICLE_SCHEMA``, else the default + schema, so the hosted mirror writer answers to the renamed variable. + """ table_paths = ( {"build_artifacts": Path(build_artifacts_path)} if build_artifacts_path is not None @@ -1112,8 +1117,11 @@ def main(argv: list[str] | None = None) -> int: ) mirror_load_parser.add_argument( "--schema", - default="ledger", - help="Supabase/Postgres schema to load into.", + default=None, + help=( + "Supabase/Postgres schema to load into. Defaults to " + f"$CHRONICLE_SCHEMA, else {DEFAULT_CHRONICLE_SCHEMA}." + ), ) mirror_load_parser.add_argument( "--batch-size", diff --git a/chronicle/mirror.py b/chronicle/mirror.py index 7a3ccaf3..f9c95dc2 100644 --- a/chronicle/mirror.py +++ b/chronicle/mirror.py @@ -11,6 +11,8 @@ from pathlib import Path from typing import Any +from chronicle.env import default_chronicle_schema + LEDGER_MIRROR_TABLES = ( "ledger_builds", "build_artifacts", @@ -200,15 +202,22 @@ def export_chronicle_db_tables( def load_supabase_mirror( input_dir: str | Path, *, - schema: str = "ledger", + schema: str | None = None, batch_size: int = 500, dry_run: bool = False, table_paths: dict[str, str | Path] | None = None, client: Any | None = None, ) -> SupabaseMirrorLoadReport: - """Load exported Chronicle JSONL mirror files into Supabase/Postgres.""" + """Load exported Chronicle JSONL mirror files into Supabase/Postgres. + + ``schema`` defaults to :func:`chronicle.env.default_chronicle_schema`, so + the writer that owns the hosted mirror answers to ``CHRONICLE_SCHEMA`` (and + the ledger-era names behind it) exactly like every other reader of the + setting. The resolved name is reported back in the load report. + """ if batch_size < 1: raise ValueError("batch_size must be at least 1.") + schema = schema or default_chronicle_schema() input_path = Path(input_dir) tables: list[SupabaseTableLoad] = [] errors: list[str] = [] diff --git a/db/supabase_client.py b/db/supabase_client.py index 50ce4f30..190d8b7e 100644 --- a/db/supabase_client.py +++ b/db/supabase_client.py @@ -15,14 +15,33 @@ from supabase import create_client, Client -from chronicle.env import env_value +from chronicle.env import default_chronicle_schema, env_value -# The hosted Postgres schema is still named "ledger"; only the environment -# variable that overrides it has moved to the chronicle prefix. Renaming the -# schema itself is a later slice of PolicyEngine/chronicle#143, coordinated -# with the CI writers that already target the ledger schema. -LEDGER_SCHEMA = env_value("CHRONICLE_SCHEMA") or "ledger" -TARGETS_SCHEMA = env_value("POLICYENGINE_TARGETS_SCHEMA") or "targets" +TARGETS_SCHEMA_ENV = "POLICYENGINE_TARGETS_SCHEMA" +DEFAULT_TARGETS_SCHEMA = "targets" + + +def chronicle_schema() -> str: + """Resolve the hosted Chronicle schema for a query. + + Read at call time, not bound at import: an import-time constant fixes the + schema at whatever the environment held when this module was first + imported, which the caller does not control (in the test suite that moment + is collection, before any fixture has isolated the environment). The + hosted schema is still named "ledger" -- only the variable that overrides + it has moved to the chronicle prefix, and renaming the schema value is a + later slice of PolicyEngine/chronicle#143. + """ + return default_chronicle_schema() + + +def targets_schema() -> str: + """Resolve the hosted targets schema. Read at call time, as above. + + ``POLICYENGINE_TARGETS_SCHEMA`` names a surface outside the ledger rename + window, so it is read literally. + """ + return env_value(TARGETS_SCHEMA_ENV, default=DEFAULT_TARGETS_SCHEMA) @dataclass @@ -109,7 +128,7 @@ def query_sources( List of source records """ client = get_supabase_client() - query = _table(client, LEDGER_SCHEMA, "sources").select("*") + query = _table(client, chronicle_schema(), "sources").select("*") if jurisdiction: query = query.eq("jurisdiction", jurisdiction) @@ -138,7 +157,9 @@ def query_strata( List of strata records with nested constraints """ client = get_supabase_client() - query = _table(client, TARGETS_SCHEMA, "strata").select("*, stratum_constraints(*)") + query = _table(client, targets_schema(), "strata").select( + "*, stratum_constraints(*)" + ) if jurisdiction: query = query.eq("jurisdiction", jurisdiction) @@ -167,7 +188,7 @@ def query_targets( """ client = get_supabase_client() # Nested join: strata with their stratum_constraints - query = _table(client, TARGETS_SCHEMA, "targets").select( + query = _table(client, targets_schema(), "targets").select( "*, strata(*, stratum_constraints(*)), sources(*)" ) @@ -214,7 +235,7 @@ def insert_targets_batch( for i in range(0, len(targets), chunk_size): chunk = targets[i : i + chunk_size] - _table(client, TARGETS_SCHEMA, "targets").insert(chunk).execute() + _table(client, targets_schema(), "targets").insert(chunk).execute() total += len(chunk) return total diff --git a/tests/conftest.py b/tests/conftest.py index 378fa0b3..3d315441 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -22,6 +22,23 @@ RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) +def pytest_configure(config): + """Strip the rename window before collection imports a single module. + + The autouse fixture below runs per test, which is too late for anything a + module does while being imported. Collection happens after this hook, so + clearing here means no module can read an operator's ``LEDGER_SCHEMA`` + (warning as it goes, or freezing it into a constant) before a fixture has + had the chance to isolate it. Modules should resolve settings at call time + rather than at import; this hook makes that property testable instead of + depending on which shell ran pytest. + """ + for name in list(os.environ): + if name.startswith(RENAME_WINDOW_PREFIXES): + del os.environ[name] + reset_env_deprecation_state() + + @pytest.fixture(autouse=True) def isolated_rename_window_env(monkeypatch): """Run every test with no rename-window variable inherited from the shell.""" diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py index 65029d14..202d86f5 100644 --- a/tests/test_chronicle_env.py +++ b/tests/test_chronicle_env.py @@ -24,7 +24,9 @@ from chronicle.env import ( CHRONICLE_ENV_PREFIX, ChronicleEnvDeprecationWarning, + DEFAULT_CHRONICLE_SCHEMA, LEGACY_ENV_PREFIXES, + default_chronicle_schema, env_flag, env_names, env_value, @@ -252,17 +254,18 @@ def test_db_cli_parser_builds_with_the_env_backed_defaults(monkeypatch, capsys): assert "Manage Chronicle target input data" in capsys.readouterr().out -@pytest.mark.parametrize("name", ["CHRONICLE_SCHEMA", "POLICYENGINE_LEDGER_SCHEMA"]) -def test_supabase_schema_honors_both_names(monkeypatch, name): +@pytest.mark.parametrize( + "name", + ["CHRONICLE_SCHEMA", "POLICYENGINE_LEDGER_SCHEMA", "LEDGER_SCHEMA"], +) +def test_supabase_schema_honors_every_name_in_the_window(monkeypatch, name): + """Set after import and still honored: the schema is read at call time.""" import db.supabase_client monkeypatch.setenv(name, "chronicle_probe") - try: - reloaded = importlib.reload(db.supabase_client) - assert reloaded.LEDGER_SCHEMA == "chronicle_probe" - finally: - monkeypatch.delenv(name, raising=False) - importlib.reload(db.supabase_client) + + assert db.supabase_client.chronicle_schema() == "chronicle_probe" + assert default_chronicle_schema() == "chronicle_probe" def test_supabase_schema_default_is_unchanged(): @@ -270,7 +273,31 @@ def test_supabase_schema_default_is_unchanged(): # The hosted schema name itself is out of this slice; only the variable # that overrides it moved. - assert db.supabase_client.LEDGER_SCHEMA == "ledger" + assert DEFAULT_CHRONICLE_SCHEMA == "ledger" + assert default_chronicle_schema() == "ledger" + assert db.supabase_client.chronicle_schema() == "ledger" + assert db.supabase_client.targets_schema() == "targets" + + +def test_supabase_schema_is_not_bound_at_import(monkeypatch): + """No module-level constant may freeze the schema at import time. + + A reload under a set variable is the pre-fix behavior this guards against: + it proves nothing about a module that resolved the value once, at + collection, and answers with the stale constant forever after. + """ + import db.supabase_client + + assert not [ + name + for name, value in vars(db.supabase_client).items() + if name.isupper() and value == "ledger" + ] + + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + unreloaded = importlib.import_module("db.supabase_client") + + assert unreloaded.chronicle_schema() == "chronicle_probe" # --------------------------------------------------------------------------- diff --git a/tests/test_chronicle_mirror.py b/tests/test_chronicle_mirror.py index c06a33bd..6b2e694d 100644 --- a/tests/test_chronicle_mirror.py +++ b/tests/test_chronicle_mirror.py @@ -4,6 +4,9 @@ import json +import pytest + +from chronicle.env import ChronicleEnvDeprecationWarning from chronicle.harness import main as harness_main from chronicle.mirror import ( LEDGER_MIRROR_TABLES, @@ -189,6 +192,127 @@ def test_load_supabase_mirror_cli_dry_run(tmp_path, capsys): assert payload["table_count"] == len(LEDGER_MIRROR_TABLES) +# --------------------------------------------------------------------------- +# Schema configuration +# +# The mirror loader is the primary writer into the hosted schema, so it is the +# call site CHRONICLE_SCHEMA has to reach (PolicyEngine/chronicle#143, +# mechanism 3). It defaulted to the literal "ledger" while only the read-side +# client honored the renamed variable, which would have sent a rehearsal load +# into production the moment an operator set it. +# --------------------------------------------------------------------------- + + +def _empty_mirror(tmp_path): + mirror_dir = tmp_path / "mirror" + mirror_dir.mkdir() + for table in LEDGER_MIRROR_TABLES: + (mirror_dir / f"{table}.jsonl").write_text("") + return mirror_dir + + +def _one_build_artifact(tmp_path): + path = tmp_path / "build_artifacts.jsonl" + path.write_text( + json.dumps( + { + "build_artifact_key": "ledger.build_artifact.v1:test", + "build_id": "ledger.build.v1:test", + "artifact_kind": "json", + "artifact_name": "reports/build_summary.json", + "sha256": "abc", + "size_bytes": 3, + "r2_bucket": "ledger-derived", + "r2_key": "derived/test", + "r2_uri": "r2://ledger-derived/derived/test", + }, + sort_keys=True, + ) + + "\n" + ) + return path + + +def _load_into_fake_client(tmp_path, **kwargs): + client = _FakeSupabaseClient() + report = load_supabase_mirror( + _empty_mirror(tmp_path), + table_paths={"build_artifacts": _one_build_artifact(tmp_path)}, + client=client, + **kwargs, + ) + return report, client + + +def test_load_supabase_mirror_defaults_to_the_ledger_schema(tmp_path): + report, client = _load_into_fake_client(tmp_path) + + assert report.schema == "ledger" + assert [upsert[0] for upsert in client.upserts] == ["ledger"] + + +def test_load_supabase_mirror_writes_to_the_chronicle_schema(tmp_path, monkeypatch): + """The renamed variable configures the writer, not just the reader.""" + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + + report, client = _load_into_fake_client(tmp_path) + + assert report.schema == "chronicle_probe" + assert [upsert[0] for upsert in client.upserts] == ["chronicle_probe"] + + +@pytest.mark.parametrize("name", ["POLICYENGINE_LEDGER_SCHEMA", "LEDGER_SCHEMA"]) +def test_load_supabase_mirror_honors_a_ledger_era_schema_name( + tmp_path, monkeypatch, name +): + monkeypatch.setenv(name, "legacy_probe") + + with pytest.warns(ChronicleEnvDeprecationWarning): + report, client = _load_into_fake_client(tmp_path) + + assert report.schema == "legacy_probe" + assert [upsert[0] for upsert in client.upserts] == ["legacy_probe"] + + +def test_an_explicit_schema_still_wins_over_the_environment(tmp_path, monkeypatch): + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + + report, client = _load_into_fake_client(tmp_path, schema="explicit_probe") + + assert report.schema == "explicit_probe" + assert [upsert[0] for upsert in client.upserts] == ["explicit_probe"] + + +def test_load_supabase_mirror_cli_writes_to_the_configured_schema( + tmp_path, monkeypatch, capsys +): + """The CLI resolves the same way when no --schema is supplied.""" + client = _FakeSupabaseClient() + monkeypatch.setattr("chronicle.mirror._get_supabase_client", lambda: client) + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + argv = [ + "load-supabase-mirror", + "--dir", + str(_empty_mirror(tmp_path)), + "--build-artifacts", + str(_one_build_artifact(tmp_path)), + ] + + exit_code = harness_main(argv) + payload = json.loads(capsys.readouterr().out) + + assert exit_code == 0 + assert payload["schema"] == "chronicle_probe" + assert [upsert[0] for upsert in client.upserts] == ["chronicle_probe"] + + assert harness_main([*argv, "--schema", "explicit_probe"]) == 0 + assert json.loads(capsys.readouterr().out)["schema"] == "explicit_probe" + assert [upsert[0] for upsert in client.upserts] == [ + "chronicle_probe", + "explicit_probe", + ] + + class _FakeSupabaseClient: def __init__(self): self.upserts = [] diff --git a/tests/test_chronicle_namespace.py b/tests/test_chronicle_namespace.py index e8890948..5dad695e 100644 --- a/tests/test_chronicle_namespace.py +++ b/tests/test_chronicle_namespace.py @@ -1,7 +1,5 @@ """Tests for the Chronicle namespace.""" -import importlib - from chronicle.client import get_supabase_client from chronicle.normalization import convert_units from chronicle.targets import ( @@ -27,20 +25,30 @@ def test_chronicle_client_reexports_supabase_client(): def test_chronicle_supabase_schema_boundaries_are_defaulted(): - """The schema names are import-time constants, so re-read them here. - - ``db.supabase_client`` resolves them from the environment when it is first - imported, which happens at collection — before the suite-wide - ``isolated_rename_window_env`` fixture clears an operator's - ``CHRONICLE_SCHEMA``. Reloading under the cleared environment is what makes - this a test of the defaults rather than of the shell. + """The schema names resolve per call, so this reads the cleared window. + + ``db.supabase_client`` is imported at collection, before any fixture runs. + Resolving the schema there — as an import-time constant — would bind an + operator's ``CHRONICLE_SCHEMA`` (or a ledger-era name, warning as it went) + into the module for the whole session, and no fixture could take it back. + Reading at call time is what makes this a test of the defaults rather than + of the shell. """ - import db.supabase_client + from db import supabase_client + + assert supabase_client.chronicle_schema() == "ledger" + assert supabase_client.targets_schema() == "targets" + + +def test_chronicle_supabase_schema_follows_the_environment(monkeypatch): + """The renamed variable reaches the client after it has been imported.""" + from db import supabase_client - supabase_client = importlib.reload(db.supabase_client) + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + monkeypatch.setenv("POLICYENGINE_TARGETS_SCHEMA", "targets_probe") - assert supabase_client.LEDGER_SCHEMA == "ledger" - assert supabase_client.TARGETS_SCHEMA == "targets" + assert supabase_client.chronicle_schema() == "chronicle_probe" + assert supabase_client.targets_schema() == "targets_probe" def test_chronicle_normalization_exports_helpers(): From e68fdd0ff1d99f3f8708858b22f226988490035d Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:26:41 -0400 Subject: [PATCH 016/105] Address the manifest a fetch is actually revising, and read it strictly Four defects in the state fetch-artifact reads before it writes: - It always targeted manifest.yaml. Seven tracked packages keep a manifest_*_source_package.yaml instead, and three publisher directories keep two of them, so the IRA revision workflow the docs cite would have written a third manifest beside the real ones and never seen the recorded block. A --manifest filename now selects it; the name has to stay inside the package. - Revision protection vanished when the entry had no storage.r2 -- a fetch that only registered bytes, or one whose upload failed, which is the state #225 landed in. A manifest entry identifies its bytes by its declared sha256 whether or not it has been published, and a fetch of different bytes over either identity is refused unless --record-revision opts in. - Recorded R2 validation read the key or the uri, whichever came first. A block whose key and uri named different objects was preserved verbatim, so the entry kept publishing a URI for bytes it no longer described. Every supplied locator field is now cross-checked against every other and against the content-addressed key shape; a contradiction is an error at fetch time and a refusal at publish time, never a silent preserve. - A manifest that parsed as anything but a mapping was treated as absent, so the fetch would replace it with a single entry. It is now refused before the publisher is read at all, and inventory-artifacts and publish-raw report it rather than crashing on it. The refusals share a SourceArtifactManifestError base, so the CLI reports all of them as an exit-1 message with nothing written. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 525 +++++++++++++++++++++++------- chronicle/harness.py | 29 +- tests/test_chronicle_artifacts.py | 416 +++++++++++++++++++++++ 3 files changed, 846 insertions(+), 124 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 9421ab08..1b3f202a 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -9,6 +9,7 @@ from datetime import UTC, datetime from pathlib import Path import posixpath +import re import shlex import sqlite3 import subprocess @@ -42,6 +43,27 @@ DEFAULT_R2_PREFIX = "raw" DEFAULT_R2_DERIVED_PREFIX = "derived" +# Most packages keep one manifest.yaml. Publisher directories that feed several +# source packages keep one manifest each -- db/data/irs_soi/ira_contributions +# holds manifest_traditional_source_package.yaml beside the Roth one -- so the +# name is an input, not a constant, wherever a caller addresses a package. +DEFAULT_MANIFEST_FILENAME = "manifest.yaml" + + +def _manifest_path(output: Path, manifest_filename: str) -> Path: + """Return the named manifest inside ``output``. + + The name is a filename, not a path: it selects among the manifests a + package directory keeps, and must not reach outside it. + """ + name = manifest_filename.strip() + if not name or name in (".", "..") or name != Path(name).name: + raise ValueError( + "Manifest must name a file inside the package directory, not " + f"{manifest_filename!r}." + ) + return output / name + def default_r2_raw_bucket() -> str: """Resolve the raw bucket: ``$CHRONICLE_R2_RAW_BUCKET`` or the default.""" @@ -53,15 +75,44 @@ def default_r2_derived_bucket() -> str: return env_value(R2_DERIVED_BUCKET_ENV, default=DEFAULT_R2_DERIVED_BUCKET) -class SourceArtifactRevisionError(RuntimeError): - """Fetched bytes are not the bytes the recorded R2 object holds. +class SourceArtifactManifestError(RuntimeError): + """A manifest refuses the write a fetch is about to make. + + Every subclass is raised before anything is downloaded, cached, uploaded or + rewritten, so a refusal leaves the package exactly as it was. + """ + + +class SourceArtifactRevisionError(SourceArtifactManifestError): + """Fetched bytes are not the bytes the manifest entry identifies. + + A manifest entry identifies specific bytes: by its declared ``sha256``, and + -- once published -- by a content-addressed R2 key that repeats them. When + a publisher re-publishes under the same URL and vintage, rewriting that + entry would attach its provenance, and any recorded URI, to bytes it never + described. Chronicle refuses instead: same vintage plus new bytes is a new + release revision (docs/adr-chronicle-fact-identity-v2.md), registered with + ``fetch-artifact --record-revision``. + """ + + +class MalformedManifestError(SourceArtifactManifestError): + """A manifest document, or a block inside one, is not a mapping. + + Reading such a file as an absent manifest would let a fetch replace it with + a single entry, dropping whatever the unreadable document recorded. + """ + - Raw R2 keys are content-addressed, so a recorded ``storage.r2`` block is a - claim about specific bytes. When a publisher re-publishes under the same - URL and vintage, keeping that block would attach its provenance to bytes it - never described. Chronicle refuses instead: same vintage plus new bytes is a - new release revision (docs/adr-chronicle-fact-identity-v2.md), registered - with ``fetch-artifact --record-revision``. +class RecordedR2LocatorError(SourceArtifactManifestError): + """A recorded ``storage.r2`` block does not locate exactly one object. + + ``provider``, ``bucket``, ``key`` and ``uri`` all describe the same object, + so any that are supplied have to agree, and the key has to carry the + ``{sha256}/{filename}`` tail that says which bytes it holds. A block whose + fields contradict each other has no single answer to "which bytes does this + entry claim R2 holds", and preserving or publishing under it would ship + whichever field the reader happened to consult. """ @@ -114,6 +165,60 @@ def to_dict(self) -> dict[str, str]: } +# A raw key ends in {sha256}/{filename} (see build_r2_key), so the segment +# before the filename is what says which bytes the object holds. +_SHA256_KEY_SEGMENT = re.compile(r"[0-9a-f]{64}") + + +@dataclass(frozen=True) +class RecordedR2Object: + """The R2 object a manifest entry's ``storage.r2`` block claims exists. + + Built only by :func:`_validated_recorded_r2`, so every instance names one + object whose locator fields agree with each other. + """ + + provider: str + bucket: str + key: str + sha256: str + filename: str + + @property + def uri(self) -> str: + """Return the storage URI the recorded fields spell out.""" + return f"{self.provider}://{self.bucket}/{self.key}" + + +@dataclass(frozen=True) +class RecordedIdentity: + """The bytes a manifest entry says its vintage currently holds. + + From the recorded object's content-addressed key once the entry has been + published, and from the entry's own declared ``sha256``/``filename`` before + that. ``r2`` is None in the second case: protection does not wait for an + upload to have happened. + """ + + sha256: str + filename: str + size_bytes: int | None + declared_sha256: str | None + r2: RecordedR2Object | None + + def holds(self, *, sha256: str, filename: str) -> bool: + """Whether this identity is exactly the given bytes under that name. + + The filename participates only when the entry records one: a published + key always carries it, an entry that declares bytes and no name does + not, and inventing a mismatch there would refuse a re-fetch of the very + bytes the entry describes. + """ + if self.sha256 != sha256: + return False + return not self.filename or self.filename == Path(filename).name + + @dataclass(frozen=True) class ArtifactCommandResult: """Result from a storage command.""" @@ -430,6 +535,7 @@ def fetch_source_artifact( source_page: str | None = None, table: str | None = None, filename: str | None = None, + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -438,20 +544,36 @@ def fetch_source_artifact( ) -> ArtifactFetchReport: """Fetch/register a source artifact and optionally upload it to R2. + ``manifest_filename`` names the manifest inside ``output_dir`` the entry + belongs to. Packages that split one publisher directory across several + source packages keep one manifest each, so a fetch that always wrote + ``manifest.yaml`` would write a fresh manifest beside the real ones and + never see the entry it is revising. + ``record_revision`` opts into registering a publisher revision: the fetched bytes get their own content-addressed key under the configured bucket and the superseded object moves to ``storage.previous_r2``. Without it, bytes - that disagree with the recorded object raise + that disagree with the entry's recorded identity raise :class:`SourceArtifactRevisionError` before anything is overwritten. """ r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) + manifest_path = _manifest_path(output, manifest_filename) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, source_id=source_id, package_path=output, ) + # Read and validate the entry being written before anything is fetched: a + # manifest Chronicle cannot read, or a recorded block that names two + # different objects, is a refusal that need not touch the publisher. + recorded_identity = _recorded_identity( + _manifest_file_spec(_read_manifest(manifest_path), year), + manifest_path=manifest_path, + year=year, + ) + fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() content, inferred_filename = _read_artifact(source_url) artifact_filename = filename or inferred_filename @@ -460,12 +582,12 @@ def fetch_source_artifact( sha256 = hashlib.sha256(content).hexdigest() size_bytes = len(content) - manifest_path = output / "manifest.yaml" # Guard before the cached artifact is touched. A rejected fetch must leave # the recorded bytes and their manifest entry exactly as they were. - _assert_recorded_object_holds_these_bytes( - manifest_path, + _assert_recorded_identity_holds_these_bytes( + recorded_identity, + manifest_path=manifest_path, year=year, filename=artifact_filename, sha256=sha256, @@ -676,7 +798,7 @@ def publish_derived_artifacts( def publish_source_artifacts( root: str | Path, *, - manifest_filename: str = "manifest.yaml", + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, source_id: str | None = None, package_id: str | None = None, r2_bucket: str | None = None, @@ -697,8 +819,8 @@ def publish_source_artifacts( errors: list[str] = [] for manifest_path in sorted(root_path.rglob(manifest_filename)): try: - manifest = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - except (OSError, yaml.YAMLError) as exc: + manifest = _read_manifest(manifest_path) + except (OSError, MalformedManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue @@ -801,7 +923,7 @@ def write_build_artifacts_jsonl( def inventory_source_artifacts( root: str | Path, *, - manifest_filename: str = "manifest.yaml", + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, ) -> ArtifactInventoryReport: """Inventory manifest-declared source artifacts under a root directory.""" root_path = Path(root) @@ -824,9 +946,8 @@ def inventory_source_artifacts( manifests = sorted(root_path.rglob(manifest_filename)) for manifest_path in manifests: try: - manifest = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - files = manifest.get("files") or {} - except (OSError, yaml.YAMLError) as exc: + files = _read_manifest(manifest_path).get("files") or {} + except (OSError, MalformedManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue if not isinstance(files, dict): @@ -1110,11 +1231,31 @@ def _filename_from_url(source_url: str) -> str: def _read_manifest(manifest_path: Path) -> dict[str, Any]: - """Return a manifest's parsed payload, or an empty mapping.""" + """Return a manifest's parsed payload, refusing a document it cannot read. + + An absent or empty manifest reads as an empty mapping: ``fetch-artifact`` + writes the first entry into a package that has none. A document that parses + as anything else -- a list, a scalar, a truncated or half-merged file -- is + not an absent manifest, and treating it as one would let the fetch replace + it with a single entry and drop everything it recorded. + """ if not manifest_path.exists(): return {} - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - return payload if isinstance(payload, dict) else {} + try: + payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) + except yaml.YAMLError as exc: + raise MalformedManifestError(f"{manifest_path} is not valid YAML: {exc}") from ( + exc + ) + if payload is None: + return {} + if not isinstance(payload, dict): + raise MalformedManifestError( + f"{manifest_path} must be a YAML mapping; it parses as a " + f"{type(payload).__name__}. Chronicle will not overwrite a manifest " + "it cannot read." + ) + return payload def _manifest_file_spec(payload: dict[str, Any], year: Any) -> dict[str, Any]: @@ -1135,40 +1276,182 @@ def _recorded_storage(spec: Any) -> dict[str, Any]: def _recorded_r2(spec: Any) -> dict[str, Any]: - """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" + """Return the recorded ``storage.r2`` block verbatim, if any. + + Raw access, for callers that carry the block forward as history. Callers + that reason about which object it names go through + :func:`_validated_recorded_r2` instead. + """ recorded = _recorded_storage(spec).get("r2") return recorded if isinstance(recorded, dict) else {} -def _r2_key_identity(recorded_r2: dict[str, Any]) -> tuple[str, str]: - """Return the ``(sha256, filename)`` a recorded R2 object is addressed by. +def _split_r2_uri(uri: str) -> tuple[str, str, str] | None: + """Split ``provider://bucket/key`` into its three parts, or None.""" + provider, separator, remainder = uri.partition("://") + if not separator or not provider: + return None + bucket, separator, key = remainder.partition("/") + if not separator or not bucket or not key: + return None + return (provider, bucket, key) - Raw keys are ``{prefix}/{source_id}/{package_id}/{year}/{sha256}/{filename}`` - (see :func:`build_r2_key`), so the last two segments say which bytes the - object holds; the URI ends in the same two segments and stands in for a - block that records only that. A locator in any other shape yields empty - strings and therefore never matches fetched bytes. + +def _validated_recorded_r2( + spec: Any, + *, + manifest_path: Path, + year: Any, +) -> RecordedR2Object | None: + """Return the object a recorded ``storage.r2`` block names, or None. + + Every locator field the block supplies is cross-checked against every + other: ``key`` against the URI's path, ``bucket`` against its authority, + ``provider`` against its scheme, and the resulting key against the + canonical content-addressed shape :func:`build_r2_key` writes. Reading one + field and trusting the rest is what lets a block that says two different + things survive a preserve or a publish. """ - locator = recorded_r2.get("key") or recorded_r2.get("uri") - if not isinstance(locator, str): - return ("", "") - parts = [part for part in locator.split("/") if part] - if len(parts) < 2: - return ("", "") - return (parts[-2], parts[-1]) + storage = _recorded_storage_block(spec, manifest_path=manifest_path, year=year) + if "r2" not in storage: + return None + block = storage["r2"] + where = f"{manifest_path} entry {year!r} storage.r2" + if not isinstance(block, dict): + raise MalformedManifestError( + f"{where} must be a mapping; it is a {type(block).__name__}." + ) + + supplied: dict[str, str] = {} + for field in ("provider", "bucket", "key", "uri"): + value = block.get(field) + if value is None: + continue + if not isinstance(value, str) or not value.strip(): + raise RecordedR2LocatorError( + f"{where}: {field} must be a non-empty string, not {value!r}." + ) + supplied[field] = value + + provider = supplied.get("provider") + bucket = supplied.get("bucket") + key = supplied.get("key") + uri = supplied.get("uri") + if uri is not None: + parts = _split_r2_uri(uri) + if parts is None: + raise RecordedR2LocatorError( + f"{where}: uri {uri!r} is not provider://bucket/key." + ) + for field, value, from_uri in zip( + ("provider", "bucket", "key"), (provider, bucket, key), parts + ): + if value is not None and value != from_uri: + raise RecordedR2LocatorError( + f"{where}: {field}={value!r} contradicts uri {uri!r}, which " + f"names {from_uri!r}. The block records two different " + "objects, so Chronicle cannot say which bytes it claims." + ) + provider, bucket, key = ( + provider or parts[0], + bucket or parts[1], + key or parts[2], + ) + + missing = [ + field + for field, value in ( + ("provider", provider), + ("bucket", bucket), + ("key", key), + ) + if not value + ] + if missing: + raise RecordedR2LocatorError( + f"{where}: records no {', '.join(missing)}. A recorded block has to " + "locate its object, by key and bucket or by uri." + ) + + segments = key.split("/") + if ( + len(segments) < 2 + or not all(segments) + or not _SHA256_KEY_SEGMENT.fullmatch(segments[-2]) + ): + raise RecordedR2LocatorError( + f"{where}: key {key!r} is not content-addressed. A raw key ends in " + "{sha256}/{filename}, which is what says the object holds the " + "entry's bytes; Chronicle will not guess for a key that does not." + ) + return RecordedR2Object( + provider=provider, + bucket=bucket, + key=key, + sha256=segments[-2], + filename=segments[-1], + ) -def _r2_holds_these_bytes( - recorded_r2: dict[str, Any], +def _recorded_storage_block( + spec: Any, *, - sha256: str, - filename: str, -) -> bool: - """Whether a recorded ``storage.r2`` block addresses exactly these bytes.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) - return bool(recorded_sha256) and (recorded_sha256, recorded_filename) == ( - sha256, - Path(filename).name, + manifest_path: Path, + year: Any, +) -> dict[str, Any]: + """Return the entry's ``storage`` mapping, refusing a malformed one.""" + if not isinstance(spec, dict) or "storage" not in spec: + return {} + storage = spec["storage"] + if not isinstance(storage, dict): + raise MalformedManifestError( + f"{manifest_path} entry {year!r} storage must be a mapping; it is a " + f"{type(storage).__name__}." + ) + return storage + + +def _recorded_identity( + spec: Any, + *, + manifest_path: Path, + year: Any, +) -> RecordedIdentity | None: + """Return what a manifest entry says its vintage holds, if anything. + + A published entry is identified by its recorded object's content-addressed + key. An entry that has not been published yet -- registered without an + upload, or left behind by a failed one -- is identified by its own declared + ``sha256`` and ``filename``. Both are recorded identities, and a fetch of + different bytes over either one is a publisher revision. + """ + recorded_r2 = _validated_recorded_r2(spec, manifest_path=manifest_path, year=year) + declared_sha256 = spec.get("sha256") if isinstance(spec, dict) else None + declared_sha256 = declared_sha256 if isinstance(declared_sha256, str) else None + declared_filename = spec.get("filename") if isinstance(spec, dict) else None + declared_filename = ( + declared_filename if isinstance(declared_filename, str) else None + ) + size_bytes = spec.get("size_bytes") if isinstance(spec, dict) else None + size_bytes = size_bytes if isinstance(size_bytes, int) else None + if recorded_r2 is not None: + return RecordedIdentity( + sha256=recorded_r2.sha256, + filename=recorded_r2.filename, + # Only report a size the recorded key agrees with: an entry can + # arrive here already describing the new bytes. + size_bytes=size_bytes if declared_sha256 == recorded_r2.sha256 else None, + declared_sha256=declared_sha256, + r2=recorded_r2, + ) + if not declared_sha256: + return None + return RecordedIdentity( + sha256=declared_sha256, + filename=Path(declared_filename).name if declared_filename else "", + size_bytes=size_bytes, + declared_sha256=declared_sha256, + r2=None, ) @@ -1177,33 +1460,35 @@ def _revision_error_message( manifest_path: Path, year: Any, filename: str, - recorded_spec: dict[str, Any], - recorded_r2: dict[str, Any], + identity: RecordedIdentity, sha256: str, size_bytes: int, r2_bucket: str, ) -> str: """Explain a refused fetch: recorded identity, fetched identity, next step.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) - declared_sha256 = recorded_spec.get("sha256") - recorded_size = ( - recorded_spec.get("size_bytes") if declared_sha256 == recorded_sha256 else None + records = ( + f"already records the R2 object {identity.r2.uri}, which holds" + if identity.r2 is not None + else "already records" ) message = ( - f"{manifest_path} entry {year!r} already records the R2 object " - f"{recorded_r2.get('uri') or recorded_r2.get('key')}, which holds " - f"sha256={recorded_sha256 or 'unknown'} " - f"filename={recorded_filename or 'unknown'} " - f"size_bytes={recorded_size if recorded_size is not None else 'unknown'}. " + f"{manifest_path} entry {year!r} {records} " + f"sha256={identity.sha256} " + f"filename={identity.filename or 'unknown'} " + f"size_bytes=" + f"{identity.size_bytes if identity.size_bytes is not None else 'unknown'}. " f"The fetched bytes are sha256={sha256} filename={Path(filename).name} " - f"size_bytes={size_bytes}. Chronicle will not attach a recorded, " - "content-addressed R2 URI to bytes it does not describe." + f"size_bytes={size_bytes}. Chronicle will not rewrite a vintage that " + "identifies specific bytes to describe bytes it never identified." ) - if declared_sha256 and declared_sha256 != recorded_sha256: + if identity.r2 is not None and identity.declared_sha256 not in ( + None, + identity.sha256, + ): message += ( - f" (The entry also declares sha256={declared_sha256}, which its own " - "R2 key contradicts: an earlier fetch rewrote the hash without " - "moving the object.)" + f" (The entry also declares sha256={identity.declared_sha256}, which " + "its own R2 key contradicts: an earlier fetch rewrote the hash " + "without moving the object.)" ) return message + ( " The same vintage with new bytes is a new release revision " @@ -1214,9 +1499,10 @@ def _revision_error_message( ) -def _assert_recorded_object_holds_these_bytes( - manifest_path: Path, +def _assert_recorded_identity_holds_these_bytes( + identity: RecordedIdentity | None, *, + manifest_path: Path, year: Any, filename: str, sha256: str, @@ -1225,21 +1511,16 @@ def _assert_recorded_object_holds_these_bytes( record_revision: bool, ) -> None: """Refuse a publisher revision that has not been opted into.""" - if record_revision: - return - recorded_spec = _manifest_file_spec(_read_manifest(manifest_path), year) - recorded_r2 = _recorded_r2(recorded_spec) - if not recorded_r2: + if record_revision or identity is None: return - if _r2_holds_these_bytes(recorded_r2, sha256=sha256, filename=filename): + if identity.holds(sha256=sha256, filename=filename): return raise SourceArtifactRevisionError( _revision_error_message( manifest_path=manifest_path, year=year, filename=filename, - recorded_spec=recorded_spec, - recorded_r2=recorded_r2, + identity=identity, sha256=sha256, size_bytes=size_bytes, r2_bucket=r2_bucket, @@ -1250,6 +1531,7 @@ def _assert_recorded_object_holds_these_bytes( def _superseding_storage( recorded_spec: dict[str, Any], *, + recorded_r2: RecordedR2Object | None, new_r2: dict[str, Any] | None, superseded_at: str, ) -> dict[str, Any]: @@ -1258,18 +1540,16 @@ def _superseding_storage( ``storage.r2`` only ever names the object that holds the entry's current bytes. The superseded block is appended, oldest first, to ``storage.previous_r2`` so the earlier bytes stay addressable by the URI - archived witness records already pin. + archived witness records already pin. An entry that was never published has + no object to supersede, and gets no ``previous_r2`` key. """ storage = dict(_recorded_storage(recorded_spec)) previous = storage.get("previous_r2") entries = list(previous) if isinstance(previous, list) else [] - recorded_r2 = _recorded_r2(recorded_spec) - if recorded_r2: - entry = dict(recorded_r2) - recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2) - if recorded_sha256: - entry["sha256"] = recorded_sha256 - if recorded_spec.get("sha256") == recorded_sha256: + if recorded_r2 is not None: + entry = dict(_recorded_r2(recorded_spec)) + entry["sha256"] = recorded_r2.sha256 + if recorded_spec.get("sha256") == recorded_r2.sha256: # Only carry metadata the superseded key agrees with: a manifest # can arrive here already describing the new bytes. for field in ("size_bytes", "fetched_at", "source_url"): @@ -1278,7 +1558,8 @@ def _superseding_storage( entry[field] = value entry["superseded_at"] = superseded_at entries.append(entry) - storage["previous_r2"] = entries + if entries: + storage["previous_r2"] = entries if new_r2 is None: storage.pop("r2", None) else: @@ -1319,42 +1600,45 @@ def _upsert_manifest( } recorded_spec = _manifest_file_spec(payload, year) recorded_storage = _recorded_storage(recorded_spec) - recorded_r2 = _recorded_r2(recorded_spec) + identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=year) new_r2 = r2_location.to_dict() if r2_location is not None else None - if recorded_r2 and _r2_holds_these_bytes( - recorded_r2, sha256=sha256, filename=filename - ): - # A recorded storage.r2 block for these exact bytes is historical - # truth: archived witness records pin raw R2 URLs by hash. Re-fetching - # under a renamed bucket copies bytes; it does not restate where the - # bytes were first published (PolicyEngine/chronicle#143, mechanism 3). - file_entry["storage"] = {**recorded_storage, "r2": recorded_r2} - elif recorded_r2: + holds = identity is not None and identity.holds(sha256=sha256, filename=filename) + if identity is not None and not holds and not record_revision: # Different bytes under the same vintage. The guard in # fetch_source_artifact refuses this without --record-revision; repeat # the check here so no caller can reach a false-provenance write. - if not record_revision: - raise SourceArtifactRevisionError( - _revision_error_message( - manifest_path=manifest_path, - year=year, - filename=filename, - recorded_spec=recorded_spec, - recorded_r2=recorded_r2, - sha256=sha256, - size_bytes=size_bytes, - r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), - ) + raise SourceArtifactRevisionError( + _revision_error_message( + manifest_path=manifest_path, + year=year, + filename=filename, + identity=identity, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), ) - file_entry["storage"] = _superseding_storage( + ) + if holds and identity.r2 is not None: + # A recorded storage.r2 block for these exact bytes is historical + # truth: archived witness records pin raw R2 URLs by hash. Re-fetching + # under a renamed bucket copies bytes; it does not restate where the + # bytes were first published (PolicyEngine/chronicle#143, mechanism 3). + storage = {**recorded_storage, "r2": _recorded_r2(recorded_spec)} + elif identity is not None and not holds: + storage = _superseding_storage( recorded_spec, + recorded_r2=identity.r2, new_r2=new_r2, superseded_at=fetched_at, ) elif new_r2 is not None: - file_entry["storage"] = {**recorded_storage, "r2": new_r2} - elif recorded_storage: - file_entry["storage"] = dict(recorded_storage) + storage = {**recorded_storage, "r2": new_r2} + else: + storage = dict(recorded_storage) + # An entry that has no storage to record carries no empty block: a + # revision over a never-published entry supersedes nothing. + if storage: + file_entry["storage"] = storage payload["files"][year] = file_entry manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False), @@ -1439,20 +1723,27 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: if errors: return refuse() - recorded_r2 = _recorded_r2(spec) - if recorded_r2 and not _r2_holds_these_bytes( - recorded_r2, sha256=sha256_actual or "", filename=filename + try: + recorded_r2 = _validated_recorded_r2( + spec, manifest_path=manifest_path, year=year + ) + except SourceArtifactManifestError as error: + # A block that does not name one object cannot be treated as history, + # and publishing under it would ship whichever field was read. + return refuse(f"recorded_r2_locator_invalid:{error}") + if recorded_r2 is not None and (recorded_r2.sha256, recorded_r2.filename) != ( + sha256_actual or "", + Path(filename).name, ): # The recorded object is addressed by different bytes, so it is not # this file's history. Uploading anyway would either publish under a # key that misdescribes its content or restate a URI that belongs to # the superseded bytes. Registering a publisher revision is # `fetch-artifact --record-revision`, not a publish-time rewrite. - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) return refuse( "recorded_r2_identity_mismatch:" - f"recorded_sha256={recorded_sha256 or 'unknown'}:" - f"recorded_filename={recorded_filename or 'unknown'}:" + f"recorded_sha256={recorded_r2.sha256}:" + f"recorded_filename={recorded_r2.filename}:" f"local_sha256={sha256_actual}:" f"local_filename={Path(filename).name}" ) @@ -1470,7 +1761,7 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: package_path=manifest_path, ), ) - recorded_bucket = recorded_r2.get("bucket") + recorded_bucket = recorded_r2.bucket if recorded_r2 is not None else None if recorded_bucket and recorded_bucket != location.bucket: # The recorded bucket is preserved history. Publishing the same bytes # into a renamed bucket is a backfill copy, not a restatement, so the @@ -1479,7 +1770,7 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: "recorded_r2_bucket_is_preserved_history:" f"recorded={recorded_bucket}:requested={location.bucket}" ) - recorded_key = recorded_r2.get("key") + recorded_key = recorded_r2.key if recorded_r2 is not None else None if recorded_key and recorded_key != location.key: return refuse( "recorded_r2_key_disagrees_with_country_prefix:" diff --git a/chronicle/harness.py b/chronicle/harness.py index a5646553..906cac55 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -9,6 +9,7 @@ from pathlib import Path from chronicle.artifacts import ( + DEFAULT_MANIFEST_FILENAME, DEFAULT_R2_DERIVED_BUCKET, DEFAULT_R2_RAW_BUCKET, ArtifactFetchReport, @@ -16,7 +17,7 @@ DerivedArtifactPublishReport, R2BootstrapReport, RawArtifactPublishReport, - SourceArtifactRevisionError, + SourceArtifactManifestError, bootstrap_r2_buckets, fetch_source_artifact, inventory_source_artifacts, @@ -339,6 +340,7 @@ def fetch_artifact_file( source_page: str | None = None, table: str | None = None, filename: str | None = None, + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -347,8 +349,9 @@ def fetch_artifact_file( ) -> ArtifactFetchReport: """Fetch/register a raw source artifact and optionally upload it to R2. - Raises :class:`SourceArtifactRevisionError` when the fetched bytes are not - the bytes the manifest's recorded R2 object holds, unless + ``manifest_filename`` selects which of the package directory's manifests + the entry belongs to. Raises :class:`SourceArtifactRevisionError` when the + fetched bytes are not the bytes that manifest's entry identifies, unless ``record_revision`` opts into registering the publisher revision. """ return fetch_source_artifact( @@ -361,6 +364,7 @@ def fetch_artifact_file( source_page=source_page, table=table, filename=filename, + manifest_filename=manifest_filename, upload_r2=upload_r2, record_revision=record_revision, r2_bucket=r2_bucket, @@ -852,7 +856,7 @@ def main(argv: list[str] | None = None) -> int: artifact_parser = subparsers.add_parser( "fetch-artifact", - help="Fetch/register a raw source artifact and update manifest.yaml", + help="Fetch/register a raw source artifact and update its manifest", ) artifact_parser.add_argument( "--url", @@ -873,13 +877,23 @@ def main(argv: list[str] | None = None) -> int: "--year", type=int, required=True, - help="Artifact vintage year to record in manifest.yaml", + help="Artifact vintage year to record in the manifest", ) artifact_parser.add_argument( "--out-dir", type=Path, required=True, - help="Directory where the raw artifact and manifest.yaml should live", + help="Directory where the raw artifact and its manifest should live", + ) + artifact_parser.add_argument( + "--manifest", + default=DEFAULT_MANIFEST_FILENAME, + help=( + "Manifest filename inside --out-dir. A publisher directory that " + "feeds several source packages keeps one manifest each, and the " + "entry being revised lives in exactly one of them. Defaults to " + f"{DEFAULT_MANIFEST_FILENAME}." + ), ) artifact_parser.add_argument( "--dataset", @@ -1361,13 +1375,14 @@ def main(argv: list[str] | None = None) -> int: source_page=args.source_page, table=args.table, filename=args.filename, + manifest_filename=args.manifest, upload_r2=args.upload_r2, record_revision=args.record_revision, r2_bucket=args.r2_bucket, r2_prefix=args.r2_prefix, wrangler_command=args.wrangler_command, ) - except SourceArtifactRevisionError as error: + except SourceArtifactManifestError as error: print(f"error: {error}", file=sys.stderr) return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 93b4d429..4629e06c 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -11,6 +11,8 @@ from chronicle.cli import main as cli_main from chronicle.artifacts import ( + MalformedManifestError, + RecordedR2LocatorError, SourceArtifactRevisionError, build_artifact_key, build_artifact_rows, @@ -1185,3 +1187,417 @@ def test_a_recorded_block_that_only_carries_a_uri_is_still_recognized( _serve(monkeypatch, SECOND_PUBLICATION) with pytest.raises(SourceArtifactRevisionError): _fetch_republished(output_dir, wrangler) + + +# --------------------------------------------------------------------------- +# Manifest addressing, entry identity, and recorded locators +# +# Everything below concerns the state a fetch reads before it writes: which +# manifest it reads, what that manifest's entry says its vintage holds, and +# whether the recorded R2 block names one object or two. +# --------------------------------------------------------------------------- + +TRADITIONAL_MANIFEST = "manifest_traditional_source_package.yaml" +ROTH_MANIFEST = "manifest_roth_source_package.yaml" + + +def _publish(tmp_path, name, content): + """Write bytes a fetch can read as a local publisher path.""" + path = tmp_path / "publisher" / name + path.parent.mkdir(parents=True, exist_ok=True) + path.write_bytes(content) + return path + + +def _fetch_local(output_dir, source_path, *, package_id="soi-table-5", **kwargs): + return fetch_source_artifact( + str(source_path), + source_id="irs_soi", + package_id=package_id, + year=2022, + output_dir=output_dir, + **kwargs, + ) + + +def _entry(manifest_path): + return yaml.safe_load(manifest_path.read_text())["files"][2022] + + +def test_fetch_artifact_writes_the_manifest_it_was_given(tmp_path): + """One publisher directory, two source packages, two manifests. + + db/data/irs_soi/ira_contributions keeps the traditional and Roth IRA + packages side by side. A fetch that always wrote manifest.yaml would write + a third manifest neither package reads. + """ + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + roth = _publish(tmp_path, "22in06ira.xlsx", b"roth IRA table") + + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, + ) + _fetch_local( + package, + roth, + package_id="soi-ira-roth-contributions-2022", + manifest_filename=ROTH_MANIFEST, + ) + + assert sorted(path.name for path in package.glob("manifest*.yaml")) == [ + ROTH_MANIFEST, + TRADITIONAL_MANIFEST, + ] + assert not (package / "manifest.yaml").exists() + assert _entry(package / TRADITIONAL_MANIFEST)["filename"] == "22in05ira.xlsx" + assert _entry(package / ROTH_MANIFEST)["filename"] == "22in06ira.xlsx" + assert _entry(package / TRADITIONAL_MANIFEST)["sha256"] == ( + hashlib.sha256(b"traditional IRA table").hexdigest() + ) + + +def test_a_revision_is_refused_in_the_manifest_that_records_it(tmp_path): + """The IRA revision workflow the docs cite, on a two-manifest package.""" + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, + ) + recorded = (package / TRADITIONAL_MANIFEST).read_bytes() + + # The IRS re-publishes under the same URL and vintage. + traditional.write_bytes(b"traditional IRA table, revised rows") + + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, + ) + + assert TRADITIONAL_MANIFEST in str(raised.value) + assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + assert (package / "22in05ira.xlsx").read_bytes() == b"traditional IRA table" + + # Without the flag the same fetch addresses a manifest that has no entry to + # protect -- which is exactly why the flag exists. + report = _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + ) + + assert report.valid + assert report.manifest_path.endswith("manifest.yaml") + assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + + +@pytest.mark.parametrize( + "manifest_filename", + ["../manifest.yaml", "nested/manifest.yaml", "", " ", ".", ".."], +) +def test_a_manifest_name_must_stay_inside_the_package(tmp_path, manifest_filename): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "table.xlsx", b"table") + + with pytest.raises(ValueError, match="inside the package directory"): + _fetch_local(package, source, manifest_filename=manifest_filename) + + assert not package.exists() + + +def test_fetch_artifact_cli_targets_the_named_manifest(tmp_path, capsys): + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + argv = [ + "fetch-artifact", + "--url", + str(traditional), + "--source-id", + "irs_soi", + "--package-id", + "soi-ira-traditional-contributions-2022", + "--year", + "2022", + "--out-dir", + str(package), + "--manifest", + TRADITIONAL_MANIFEST, + ] + + assert harness_main(argv) == 0 + payload = json.loads(capsys.readouterr().out) + + assert payload["manifest_path"].endswith(TRADITIONAL_MANIFEST) + assert not (package / "manifest.yaml").exists() + + traditional.write_bytes(b"traditional IRA table, revised rows") + + assert harness_main(argv) == 1 + assert TRADITIONAL_MANIFEST in capsys.readouterr().err + + +# --------------------------------------------------------------------------- +# Identity without a recorded R2 object +# --------------------------------------------------------------------------- + + +def _failing_wrangler(tmp_path, log): + wrangler = tmp_path / "failing-wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\nexit 1\n") + wrangler.chmod(0o755) + return wrangler + + +def test_a_registered_entry_is_protected_before_it_is_ever_published(tmp_path): + """No storage.r2 yet is not no identity: the entry declares its bytes.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5, first publication") + first = _fetch_local(package, source, upload_r2=False) + recorded = (package / "manifest.yaml").read_bytes() + + assert "storage" not in _entry(package / "manifest.yaml") + + # Same bytes: an ordinary repeated fetch, not a revision. + assert _fetch_local(package, source, upload_r2=False).sha256 == first.sha256 + + source.write_bytes(b"IRA table 5, silently re-published") + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_local(package, source, upload_r2=False) + + message = str(raised.value) + assert first.sha256 in message + assert hashlib.sha256(b"IRA table 5, silently re-published").hexdigest() in message + assert "size_bytes=30" in message + assert "--record-revision" in message + assert (package / "manifest.yaml").read_bytes() == recorded + assert (package / "22in05ira.xlsx").read_bytes() == ( + b"IRA table 5, first publication" + ) + + +def test_a_failed_upload_does_not_disable_revision_protection(tmp_path): + """The state #225 hit: bytes registered, upload failed, no storage.r2.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _failing_wrangler(tmp_path, log) + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5, first publication") + + report = _fetch_local( + package, source, upload_r2=True, wrangler_command=str(wrangler) + ) + recorded = (package / "manifest.yaml").read_bytes() + + assert report.errors == ("r2_upload_failed",) + assert "storage" not in _entry(package / "manifest.yaml") + + source.write_bytes(b"IRA table 5, silently re-published") + with pytest.raises(SourceArtifactRevisionError): + _fetch_local(package, source, upload_r2=True, wrangler_command=str(wrangler)) + + assert (package / "manifest.yaml").read_bytes() == recorded + + +def test_record_revision_over_an_unpublished_entry_supersedes_nothing(tmp_path): + """There is no object to keep, so the entry gets no previous_r2 key.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5, first publication") + _fetch_local(package, source, upload_r2=False) + + source.write_bytes(b"IRA table 5, silently re-published") + report = _fetch_local(package, source, upload_r2=False, record_revision=True) + revised = _entry(package / "manifest.yaml") + + assert report.valid + assert revised["sha256"] == ( + hashlib.sha256(b"IRA table 5, silently re-published").hexdigest() + ) + assert "storage" not in revised + + +# --------------------------------------------------------------------------- +# Recorded locator cross-checks +# --------------------------------------------------------------------------- + + +def _recorded_package(tmp_path, content=b"IRA table 5, first publication"): + """A package whose entry records a published, content-addressed object.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + source = _publish(tmp_path, "22in05ira.xlsx", content) + report = _fetch_local( + package, source, upload_r2=True, wrangler_command=str(wrangler) + ) + return package, source, report + + +def _rewrite_recorded_r2(package, mutate, manifest="manifest.yaml"): + manifest_path = package / manifest + payload = yaml.safe_load(manifest_path.read_text()) + mutate(payload["files"][2022]["storage"]) + manifest_path.write_text(yaml.safe_dump(payload, sort_keys=False)) + return manifest_path + + +def _other_sha256(): + return hashlib.sha256(b"some other object entirely").hexdigest() + + +def _contradict_key(storage): + key = storage["r2"]["key"] + storage["r2"]["key"] = key.replace(key.split("/")[-2], _other_sha256()) + + +def _contradict_bucket(storage): + storage["r2"]["bucket"] = "some-other-bucket" + + +def _contradict_provider(storage): + storage["r2"]["provider"] = "s3" + + +def _mangle_uri(storage): + storage["r2"]["uri"] = "r2:/ledger-raw-missing-a-slash" + + +def _drop_the_locator(storage): + storage["r2"] = {"provider": "r2", "bucket": "ledger-raw"} + + +def _flatten_the_key(storage): + storage["r2"]["key"] = "raw/irs_soi/22in05ira.xlsx" + storage["r2"]["uri"] = f"r2://ledger-raw/{storage['r2']['key']}" + + +@pytest.mark.parametrize( + ("mutate", "expected"), + [ + pytest.param(_contradict_key, "contradicts uri", id="key-vs-uri"), + pytest.param(_contradict_bucket, "contradicts uri", id="bucket-vs-uri"), + pytest.param(_contradict_provider, "contradicts uri", id="provider-vs-uri"), + pytest.param(_mangle_uri, "is not provider://bucket/key", id="uri-shape"), + pytest.param(_drop_the_locator, "records no key", id="no-locator"), + pytest.param( + _flatten_the_key, "is not content-addressed", id="not-content-addressed" + ), + ], +) +def test_a_recorded_block_that_names_two_objects_is_refused(tmp_path, mutate, expected): + """A contradictory locator is an error, never a silently preserved block. + + The key-vs-uri case is the one that used to pass: identity was read from + the key alone, so a block whose uri named different bytes was carried + forward verbatim, and the manifest kept publishing a URI for an object it + no longer described. + """ + package, source, _ = _recorded_package(tmp_path) + manifest_path = _rewrite_recorded_r2(package, mutate) + recorded = manifest_path.read_bytes() + + # Identical bytes: the fetch would otherwise preserve the recorded block. + with pytest.raises(RecordedR2LocatorError) as raised: + _fetch_local(package, source, upload_r2=False) + + assert expected in str(raised.value) + assert manifest_path.read_bytes() == recorded + + +def test_a_malformed_storage_block_is_not_treated_as_absent(tmp_path): + package, source, _ = _recorded_package(tmp_path) + manifest_path = _rewrite_recorded_r2( + package, lambda storage: storage.update({"r2": ["r2://ledger-raw/raw/key"]}) + ) + recorded = manifest_path.read_bytes() + + with pytest.raises(MalformedManifestError, match="must be a mapping"): + _fetch_local(package, source, upload_r2=False) + + assert manifest_path.read_bytes() == recorded + + +def test_publish_raw_refuses_a_contradictory_recorded_block(tmp_path): + """Nothing is uploaded under a block that does not name one object.""" + package, _, _ = _recorded_package(tmp_path) + log = tmp_path / "publish.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = _rewrite_recorded_r2(package, _contradict_key) + recorded = manifest_path.read_bytes() + + report = publish_source_artifacts(package, wrangler_command=str(wrangler)) + + assert not report.valid + assert report.entries[0].upload is None + assert report.entries[0].errors[0].startswith("recorded_r2_locator_invalid:") + assert "contradicts uri" in report.entries[0].errors[0] + assert not log.exists() + assert manifest_path.read_bytes() == recorded + + +# --------------------------------------------------------------------------- +# Malformed manifests +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + "document", + [ + pytest.param("- one entry\n- another\n", id="list"), + pytest.param("a bare scalar\n", id="scalar"), + pytest.param("files: [\n", id="unparseable"), + ], +) +def test_a_malformed_manifest_is_refused_before_anything_is_fetched(tmp_path, document): + """Not an absent manifest: refusing it protects what it still records. + + The publisher path does not exist, so reaching the fetch at all would raise + FileNotFoundError instead. Getting MalformedManifestError is what says the + manifest was read and refused first. + """ + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text(document) + + with pytest.raises(MalformedManifestError): + _fetch_local(package, tmp_path / "publisher" / "never-read.xlsx") + + assert manifest_path.read_text() == document + assert list(package.iterdir()) == [manifest_path] + + +@pytest.mark.parametrize("document", ["", "\n", "{}\n", "# only a comment\n"]) +def test_an_empty_manifest_still_reads_as_absent(tmp_path, document): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + (package / "manifest.yaml").write_text(document) + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5") + + report = _fetch_local(package, source, upload_r2=False) + + assert report.valid + assert _entry(package / "manifest.yaml")["filename"] == "22in05ira.xlsx" + + +def test_a_malformed_manifest_is_reported_by_inventory_and_publish(tmp_path): + """Neither sweep may crash on, or silently skip, a document it cannot read.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + (package / "manifest.yaml").write_text("- not a mapping\n") + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package) + + assert not inventory.valid + assert inventory.entries == () + assert "must be a YAML mapping" in inventory.errors[0] + assert not published.valid + assert published.entries == () + assert "must be a YAML mapping" in published.errors[0] From ad828b1c51076123c0a02b320561d1a6617878a0 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:38:08 -0400 Subject: [PATCH 017/105] Document identity, manifest selection and locator checks What a manifest entry identifies (declared sha256 from registration, the content-addressed key once published), which manifest a fetch addresses in a package that keeps more than one, what a recorded storage.r2 block has to say for itself, and that load-supabase-mirror takes its schema default from CHRONICLE_SCHEMA like every other reader. Co-Authored-By: Claude Fable 5.1 --- README.md | 8 ++++-- docs/agent-source-package-harness.md | 11 +++++--- docs/storage-architecture.md | 41 ++++++++++++++++++++++++++-- 3 files changed, 51 insertions(+), 9 deletions(-) diff --git a/README.md b/README.md index 6b6e170c..7c526a97 100644 --- a/README.md +++ b/README.md @@ -356,7 +356,9 @@ bunx wrangler login uv run chronicle bootstrap-r2 # Fetch/register a source artifact, write db/data/.../manifest.yaml, and upload -# the exact bytes to R2 when Wrangler is authenticated: +# the exact bytes to R2 when Wrangler is authenticated. Pass --manifest when the +# package directory keeps more than one manifest (ira_contributions keeps a +# traditional and a Roth one): uv run chronicle fetch-artifact \ --url https://www.irs.gov/pub/irs-soi/23in12ms.xls \ --source-id irs_soi \ @@ -368,7 +370,9 @@ uv run chronicle fetch-artifact \ --upload-r2 # Re-fetching is safe: identical bytes keep the recorded storage.r2 block, and -# bytes that disagree with it are refused. When a publisher has re-published +# bytes that disagree with what the entry identifies -- its declared sha256, or +# its recorded content-addressed key once published -- are refused. When a +# publisher has re-published # under the same URL and vintage, register the revision explicitly — the new # bytes get their own content-addressed key and the superseded object is kept # in storage.previous_r2: diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index fafce828..48f09b76 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -16,15 +16,18 @@ register raw source files with `uv run chronicle fetch-artifact` before authorin selectors. This writes the local artifact, captures checksum and retrieval metadata in `manifest.yaml`, and can upload the exact bytes to the private raw R2 bucket (`ledger-raw` today; overridable with `CHRONICLE_R2_RAW_BUCKET`) when -Wrangler is authenticated. Agents can audit the local +Wrangler is authenticated. A publisher directory that feeds several source +packages keeps one manifest each, so pass `--manifest ` to address +the right one. Agents can audit the local artifact registry with `uv run chronicle inventory-artifacts --root db/data`. For already-downloaded manifest artifacts, agents should run `uv run chronicle publish-raw --root db/data` to upload checksum-verified bytes to R2 and write `storage.r2` metadata back into each manifest entry. -Both commands treat a recorded `storage.r2` block as a claim about specific -bytes, because raw keys are content-addressed. Re-fetching or publishing bytes -the recorded object does not hold is refused; when a publisher has re-published +Both commands treat a manifest entry as a claim about specific bytes: by its +declared `sha256` from the moment it is registered, and by the content-addressed +key of its recorded `storage.r2` block once it is published. Re-fetching or +publishing bytes the entry does not identify is refused; when a publisher has re-published under the same URL and vintage, register the revision with `uv run chronicle fetch-artifact ... --record-revision`, which stores the new bytes under their own key and keeps the superseded object in diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 6f4cfb00..3607e82b 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -99,8 +99,12 @@ name. Publishers do not always honor that: on 2026-09-02 the IRS re-published `22in05ira.xlsx` and `22in06ira.xlsx` under their existing URLs (PolicyEngine/chronicle#225). -`fetch-artifact` therefore compares the recorded key's `{sha256}/{filename}` -tail with the bytes it just fetched, before it writes anything: +`fetch-artifact` therefore compares the entry's recorded identity with the +bytes it just fetched, before it writes anything. That identity is the recorded +key's `{sha256}/{filename}` tail once the entry has been published, and the +entry's own declared `sha256` before then — an entry that was registered +without an upload, or whose upload failed, still identifies its bytes, and gets +the same protection: - **Identical** — the recorded block is preserved exactly, whichever bucket is configured now. Re-fetching after the bucket rename copies bytes; it does not @@ -153,6 +157,31 @@ as history. A local file the recorded object does not hold is reported as `recorded_r2_identity_mismatch` and nothing is uploaded: registering a revision is a fetch-time decision, not a publish-time rewrite. +### Which manifest + +Most packages keep one `manifest.yaml`. A publisher directory that feeds +several source packages keeps one manifest each — +`db/data/irs_soi/ira_contributions/` holds +`manifest_traditional_source_package.yaml` beside +`manifest_roth_source_package.yaml` — and the entry being revised lives in +exactly one of them. `fetch-artifact --manifest ` selects it; +defaulting to `manifest.yaml` there would write a third manifest neither +package reads, and the recorded block would never be compared at all. The name +must be a filename inside `--out-dir`, not a path. + +### What a recorded block has to say + +A `storage.r2` block's `provider`, `bucket`, `key` and `uri` all describe one +object, so every field that is present is cross-checked against every other: +the key against the URI's path, the bucket against its authority, the provider +against its scheme, and the resulting key against the content-addressed +`{sha256}/{filename}` shape. A block whose fields disagree does not answer +"which bytes does this entry claim R2 holds", so it is an error rather than +something to preserve or publish under. Likewise a manifest that parses as +anything other than a mapping is refused rather than treated as absent — +reading it as absent would let the next fetch replace the file with a single +entry. + ## Relational Registry Contract The hosted `chronicle` schema should be the lookup surface for Chronicle, not the place @@ -254,7 +283,7 @@ what a naive fallback would do: | `CHRONICLE_SOURCE_ARTIFACT_FETCH` | `LEDGER_SOURCE_ARTIFACT_FETCH` | Fetch a missing manifest artifact from its `source_url` during a build | | `CHRONICLE_PE_US_DATA_ROOT` | `LEDGER_PE_US_DATA_ROOT` | Local checkout root for PE US source inventory | | `CHRONICLE_PE_UK_DATA_ROOT` | `LEDGER_PE_UK_DATA_ROOT` | Local checkout root for PE UK source inventory | -| `CHRONICLE_SCHEMA` | `POLICYENGINE_LEDGER_SCHEMA` | Postgres schema the Supabase client reads and writes | +| `CHRONICLE_SCHEMA` | `POLICYENGINE_LEDGER_SCHEMA`, `LEDGER_SCHEMA` | Postgres schema the Supabase client reads and `load-supabase-mirror` writes; defaults to `ledger` | | `CHRONICLE_R2_RAW_BUCKET` | `LEDGER_R2_RAW_BUCKET` | Raw R2 archive bucket; defaults to `ledger-raw` | | `CHRONICLE_R2_DERIVED_BUCKET` | `LEDGER_R2_DERIVED_BUCKET` | Derived R2 archive bucket; defaults to `ledger-derived` | @@ -272,6 +301,12 @@ the variable that overrides the schema; the schema still defaults to `ledger`, and the mirror table names are unchanged. Those move in a later slice coordinated with the CI writers. +Every reader of the setting resolves it the same way, at call time. That +includes the writer: `load-supabase-mirror` takes its `--schema` default from +`CHRONICLE_SCHEMA`, so setting the variable to rehearse a cutover moves the +mirror load with the client rather than leaving it pointed at `ledger`. An +explicit `--schema` still wins. + ## Bucket Cutover Chronicle's operational stores migrate by dual-run From 7b68988ff00e8212e0e244cfaf5e041b2f3eb27b Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:39:25 -0400 Subject: [PATCH 018/105] Name the strict storage reader for the pair it belongs to _validated_recorded_storage sits next to _validated_recorded_r2 and above its only caller, instead of a _recorded_storage_block that read like a variant of the lenient accessor. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 44 +++++++++++++++++++++--------------------- 1 file changed, 22 insertions(+), 22 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 1b3f202a..1a81b918 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1244,9 +1244,9 @@ def _read_manifest(manifest_path: Path) -> dict[str, Any]: try: payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) except yaml.YAMLError as exc: - raise MalformedManifestError(f"{manifest_path} is not valid YAML: {exc}") from ( - exc - ) + raise MalformedManifestError( + f"{manifest_path} is not valid YAML: {exc}" + ) from exc if payload is None: return {} if not isinstance(payload, dict): @@ -1297,6 +1297,24 @@ def _split_r2_uri(uri: str) -> tuple[str, str, str] | None: return (provider, bucket, key) +def _validated_recorded_storage( + spec: Any, + *, + manifest_path: Path, + year: Any, +) -> dict[str, Any]: + """Return the entry's ``storage`` mapping, refusing a malformed one.""" + if not isinstance(spec, dict) or "storage" not in spec: + return {} + storage = spec["storage"] + if not isinstance(storage, dict): + raise MalformedManifestError( + f"{manifest_path} entry {year!r} storage must be a mapping; it is a " + f"{type(storage).__name__}." + ) + return storage + + def _validated_recorded_r2( spec: Any, *, @@ -1312,7 +1330,7 @@ def _validated_recorded_r2( field and trusting the rest is what lets a block that says two different things survive a preserve or a publish. """ - storage = _recorded_storage_block(spec, manifest_path=manifest_path, year=year) + storage = _validated_recorded_storage(spec, manifest_path=manifest_path, year=year) if "r2" not in storage: return None block = storage["r2"] @@ -1393,24 +1411,6 @@ def _validated_recorded_r2( ) -def _recorded_storage_block( - spec: Any, - *, - manifest_path: Path, - year: Any, -) -> dict[str, Any]: - """Return the entry's ``storage`` mapping, refusing a malformed one.""" - if not isinstance(spec, dict) or "storage" not in spec: - return {} - storage = spec["storage"] - if not isinstance(storage, dict): - raise MalformedManifestError( - f"{manifest_path} entry {year!r} storage must be a mapping; it is a " - f"{type(storage).__name__}." - ) - return storage - - def _recorded_identity( spec: Any, *, From 272283edb3f4ce8fb77df8d90a2e2a16d725e7a5 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:40:10 -0400 Subject: [PATCH 019/105] Record what each round-2 fix does and how it was reproduced Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 50 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index e51ea2ca..79bafcd9 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -118,6 +118,56 @@ with a regression test on this branch. Plan, in dependency order: and uri; every key is content-addressed; every declared `sha256`/`filename` agrees with its key tail; no `uri` contradicts its `key`. Strict locator validation therefore refuses nothing that is tracked today. +- All seven findings are applied, each with a regression test, and each + reproduced against this branch's previous head (`34d1d0f`) first. + +### What each fix does + +1. `chronicle/env.py` gains `default_chronicle_schema()`: one home for the + `CHRONICLE_SCHEMA` -> `POLICYENGINE_LEDGER_SCHEMA` -> `LEDGER_SCHEMA` -> + `"ledger"` ladder. `load_supabase_mirror`, its harness wrapper and the + `--schema` CLI default all resolve through it when no schema is supplied; + an explicit `--schema` still wins. Defaults unchanged. +2. `chronicle/consumer_contract.py` matches the whole final dot-segment of a + `source_record_id` against both `ledger_derived` and `chronicle_derived`. +3. `fetch-artifact --manifest ` selects which of a package's + manifests the entry belongs to (default `manifest.yaml`); the name must be + a filename inside `--out-dir`. +4. Revision protection now compares against the entry's recorded identity -- + the recorded key's `{sha256}/{filename}` once published, the declared + `sha256` before that -- so a registered-but-unpublished entry, or one whose + upload failed, is protected exactly like a published one. +5. `_validated_recorded_r2` cross-checks every supplied locator field against + every other and against the content-addressed key shape. A contradiction is + `RecordedR2LocatorError` at fetch time and `recorded_r2_locator_invalid` at + publish time, never a preserved block. +6. `_read_manifest` refuses a non-mapping or unparseable document + (`MalformedManifestError`) before the publisher is read at all; + `inventory-artifacts` and `publish-raw` report it instead of crashing. +7. `db.supabase_client` resolves both schemas per call rather than at import, + and `tests/conftest.py` strips the rename window in `pytest_configure`, so + no module can read or warn from an operator's shell during collection. + +All four refusals share a `SourceArtifactManifestError` base, so the +`fetch-artifact` CLI reports every one as exit 1 with nothing written. + +### Reproduced against `34d1d0f` (the round-1 head) + +Running the same operations against a checkout of the previous head: + +1. `load_supabase_mirror` default `schema='ledger'`; with + `CHRONICLE_SCHEMA=chronicle_probe` the load still reports `schema='ledger'`. +2. `'.chronicle_derived'.endswith('.ledger_derived')` is False: the boundary + never fired for the chronicle spelling. +3. `fetch_source_artifact()` rejects `manifest_filename` as an unexpected + keyword; a fetch into `ira_contributions/` writes `manifest.yaml`. +4. A fetch of different bytes over a registered (unpublished) entry was + accepted silently: the entry's `sha256` was rewritten with no refusal. +5. A block whose `key` and `uri` named different objects was preserved + verbatim, key sha `c63744a4...` beside uri sha `1e9b3fdb...`. +6. A list-valued `manifest.yaml` was overwritten by the fetch. +7. Importing `db.supabase_client` under `LEDGER_SCHEMA=zzz` bound + `LEDGER_SCHEMA='zzz'` and emitted a `FutureWarning` at collection. ## Verification From 7f628d1186045e8eaa74289f8548ad38513f6ce4 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 14:15:39 -0400 Subject: [PATCH 020/105] Say which fields a recorded block actually needs The message named key and bucket while the check also requires provider, so a hand-written block missing it read as a contradiction between the error and the rule. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 1a81b918..00807ade 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1388,7 +1388,8 @@ def _validated_recorded_r2( if missing: raise RecordedR2LocatorError( f"{where}: records no {', '.join(missing)}. A recorded block has to " - "locate its object, by key and bucket or by uri." + "locate its object: provider, bucket and key, or a uri that " + "supplies them." ) segments = key.split("/") From 0605c470ea3f51e5330f095ba8d403fd0d84e40c Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 14:16:15 -0400 Subject: [PATCH 021/105] Say what --record-revision is an opt-in over The flag's help still described the recorded R2 object alone, which is no longer the only identity a fetch is refused against. Co-Authored-By: Claude Fable 5.1 --- chronicle/harness.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/chronicle/harness.py b/chronicle/harness.py index 906cac55..ee7f6ef9 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -923,8 +923,9 @@ def main(argv: list[str] | None = None) -> int: "Register a publisher revision: the fetched bytes get their own " "content-addressed key under the configured bucket and the " "superseded object moves to storage.previous_r2. Without this " - "flag, bytes that disagree with the recorded R2 object are " - "refused." + "flag, bytes that disagree with what the entry identifies -- its " + "declared sha256, or its recorded content-addressed key once " + "published -- are refused." ), ) artifact_parser.add_argument( From 3abacef682d9a7742b4430f339dcfd0b53a6db39 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 14:16:41 -0400 Subject: [PATCH 022/105] Say when the manifest guards actually run They run before the publisher is read, and again before the manifest is rewritten; the base class claimed the second one also predated the cache write. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 00807ade..58437120 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -78,8 +78,10 @@ def default_r2_derived_bucket() -> str: class SourceArtifactManifestError(RuntimeError): """A manifest refuses the write a fetch is about to make. - Every subclass is raised before anything is downloaded, cached, uploaded or - rewritten, so a refusal leaves the package exactly as it was. + The checks that raise these run before the publisher is read, so an + ordinary refusal costs nothing and leaves the package exactly as it was. + They are repeated immediately before the manifest is rewritten, so no + caller can reach a false-provenance write by another route. """ From f6bce08198b73992fdf3aeec4b8e2c8148108950 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 00:22:35 -0400 Subject: [PATCH 023/105] Drop a blank line ruff format rejects after the rebase onto #228 --- chronicle/artifacts.py | 1 - 1 file changed, 1 deletion(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 58437120..0a746ca7 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -28,7 +28,6 @@ from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain - R2_RAW_BUCKET_ENV = "CHRONICLE_R2_RAW_BUCKET" R2_DERIVED_BUCKET_ENV = "CHRONICLE_R2_DERIVED_BUCKET" From 73a1826bddaa0a48488d329ce2cff135c79b2582 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 00:57:46 -0400 Subject: [PATCH 024/105] Gate round: manifest-name refusals reach the CLI, files blocks are validated before I/O, renames are refused by name, preserved-bucket objects are skipped not red, stray default manifests are refused - `_manifest_path` raises ManifestNameError (a SourceArtifactManifestError that is still a ValueError), so `chronicle fetch-artifact --manifest ../x` prints the ordinary `error:` line and exits 1 instead of a traceback. - A `files` block that is not a mapping is refused by the fetch before the publisher is read, with the same wording inventory-artifacts and publish-raw use; nothing is overwritten or uploaded first. - Identical bytes under a different filename are refused as a rename, naming both filenames and the --filename that keeps the recorded identity; the refusal holds with --record-revision too, since a rename is not a revision. - publish-raw reports an entry whose recorded object (in another bucket) holds exactly the local bytes as `skipped`, with its recorded location and no upload or rewrite; the report stays valid and gains `skipped_count`, so the documented post-flip sweep exits 0. Docs updated (Bucket Cutover, step 5). - Fetching with the default manifest name into a directory that keeps manifest_*.yaml files and no manifest.yaml is refused (AmbiguousManifestError) naming the candidates; the #225 path is closed without the flag, on the API and the CLI. - previous_r2 doc example carries source_url as the code does. --- chronicle/artifacts.py | 144 ++++++++++++++++++++++++--- docs/storage-architecture.md | 12 ++- tests/test_chronicle_artifacts.py | 159 +++++++++++++++++++++++++++--- 3 files changed, 283 insertions(+), 32 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 0a746ca7..7c3032f0 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -57,13 +57,66 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: """ name = manifest_filename.strip() if not name or name in (".", "..") or name != Path(name).name: - raise ValueError( + raise ManifestNameError( "Manifest must name a file inside the package directory, not " f"{manifest_filename!r}." ) return output / name +def _sibling_manifests(output: Path) -> list[str]: + """Return the ``manifest_*.yaml`` files a package directory keeps.""" + if not output.is_dir(): + return [] + return sorted( + path.name + for pattern in ("manifest_*.yaml", "manifest_*.yml") + for path in output.glob(pattern) + if path.is_file() + ) + + +def _refuse_a_stray_default_manifest(output: Path, manifest_path: Path) -> None: + """Refuse to create ``manifest.yaml`` beside a package's named manifests. + + A publisher directory that feeds several source packages keeps one + ``manifest_.yaml`` per package and no ``manifest.yaml``. A fetch + that omits ``--manifest`` there would create a third manifest none of the + packages read, and would bypass the revision guard of the one it should + have addressed (PolicyEngine/chronicle#225). + """ + if manifest_path.name != DEFAULT_MANIFEST_FILENAME or manifest_path.exists(): + return + siblings = _sibling_manifests(output) + if not siblings: + return + raise AmbiguousManifestError( + f"{output} keeps {', '.join(siblings)} and no {DEFAULT_MANIFEST_FILENAME}; " + "pass --manifest to name the manifest this fetch records into rather " + f"than creating {DEFAULT_MANIFEST_FILENAME} beside them." + ) + + +def _manifest_files(payload: dict[str, Any], manifest_path: Path) -> dict[str, Any]: + """Return a manifest's ``files`` block, refusing one that is not a mapping. + + ``inventory-artifacts`` and ``publish-raw`` report the same document as + ``files must be a mapping``; a fetch must refuse it before reading the + publisher, or the write fails only after the local artifact has been + overwritten and any upload has run. + """ + files = payload.get("files") + if files is None: + return {} + if not isinstance(files, dict): + raise MalformedManifestError( + f"{manifest_path} files must be a mapping; it parses as a " + f"{type(files).__name__}. Chronicle will not overwrite a manifest " + "it cannot read." + ) + return files + + def default_r2_raw_bucket() -> str: """Resolve the raw bucket: ``$CHRONICLE_R2_RAW_BUCKET`` or the default.""" return env_value(R2_RAW_BUCKET_ENV, default=DEFAULT_R2_RAW_BUCKET) @@ -97,6 +150,19 @@ class SourceArtifactRevisionError(SourceArtifactManifestError): """ +class ManifestNameError(SourceArtifactManifestError, ValueError): + """A manifest name is not a bare filename inside the package directory. + + Also a :class:`ValueError` for callers that validated the name that way + before the CLI learned to report it as an ordinary manifest refusal. + """ + + +class AmbiguousManifestError(SourceArtifactManifestError): + """The default manifest name would create a manifest beside the ones a + package already keeps (PolicyEngine/chronicle#225).""" + + class MalformedManifestError(SourceArtifactManifestError): """A manifest document, or a block inside one, is not a mapping. @@ -366,16 +432,24 @@ class RawArtifactPublishEntry: r2_location: ArtifactStorageLocation | None upload: ArtifactCommandResult | None errors: tuple[str, ...] = () + skipped: str | None = None + + @property + def uploaded(self) -> bool: + """Whether this run uploaded the artifact.""" + return self.upload is not None and self.upload.ok @property def valid(self) -> bool: - """Whether this raw artifact uploaded and was registered.""" - return not self.errors and self.upload is not None and self.upload.ok + """Whether this raw artifact is published: uploaded now, or already + held by the recorded object in a preserved bucket (``skipped``).""" + return not self.errors and (self.skipped is not None or self.uploaded) def to_dict(self) -> dict[str, Any]: """Return a JSON-serializable entry.""" return { "valid": self.valid, + "skipped": self.skipped, "manifest_path": self.manifest_path, "source_id": self.source_id, "package_id": self.package_id, @@ -410,7 +484,10 @@ def counts(self) -> dict[str, int]: return { "manifest_count": len(manifest_paths), "artifact_count": len(self.entries), - "uploaded_count": sum(1 for entry in self.entries if entry.valid), + "uploaded_count": sum(1 for entry in self.entries if entry.uploaded), + "skipped_count": sum( + 1 for entry in self.entries if entry.skipped is not None + ), "failed_count": sum(1 for entry in self.entries if not entry.valid), "r2_link_count": sum( 1 for entry in self.entries if entry.r2_location is not None @@ -569,8 +646,11 @@ def fetch_source_artifact( # Read and validate the entry being written before anything is fetched: a # manifest Chronicle cannot read, or a recorded block that names two # different objects, is a refusal that need not touch the publisher. + _refuse_a_stray_default_manifest(output, manifest_path) + existing_manifest = _read_manifest(manifest_path) + _manifest_files(existing_manifest, manifest_path) recorded_identity = _recorded_identity( - _manifest_file_spec(_read_manifest(manifest_path), year), + _manifest_file_spec(existing_manifest, year), manifest_path=manifest_path, year=year, ) @@ -1513,9 +1593,21 @@ def _assert_recorded_identity_holds_these_bytes( record_revision: bool, ) -> None: """Refuse a publisher revision that has not been opted into.""" - if record_revision or identity is None: + if identity is None or identity.holds(sha256=sha256, filename=filename): return - if identity.holds(sha256=sha256, filename=filename): + if identity.sha256 == sha256: + # The recorded object holds exactly these bytes under another name. A + # rename is not a publisher revision, so --record-revision does not + # apply, and silently adopting the new name would leave the entry's + # filename disagreeing with the key its own storage block records. + raise SourceArtifactRevisionError( + f"{manifest_path} entry {year!r} already records these exact bytes " + f"(sha256={sha256}) as filename={identity.filename}; this fetch " + f"names them {Path(filename).name}. A rename is not a release " + "revision, so --record-revision does not apply. Re-run with " + f"--filename {identity.filename} to keep the recorded identity." + ) + if record_revision: return raise SourceArtifactRevisionError( _revision_error_message( @@ -1764,13 +1856,37 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: ), ) recorded_bucket = recorded_r2.bucket if recorded_r2 is not None else None - if recorded_bucket and recorded_bucket != location.bucket: - # The recorded bucket is preserved history. Publishing the same bytes - # into a renamed bucket is a backfill copy, not a restatement, so the - # manifest must not be rewritten to point at the new bucket. - return refuse( - "recorded_r2_bucket_is_preserved_history:" - f"recorded={recorded_bucket}:requested={location.bucket}" + if recorded_r2 is not None and recorded_bucket != location.bucket: + # The recorded bucket is preserved history and, per the identity check + # above, its object holds exactly these bytes: the artifact is already + # published. Restating it under the configured bucket would rewrite + # where the bytes were first published (a backfill copy is not a + # restatement), so the entry is reported as skipped with nothing + # uploaded or rewritten. After the bucket-default flip every entry + # published before it takes this path, and the sweep stays green. + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(artifact_path), + sha256=sha256_actual, + size_bytes=size_bytes, + r2_location=ArtifactStorageLocation( + provider="r2", + bucket=recorded_r2.bucket, + key=recorded_r2.key, + ), + upload=None, + errors=(), + skipped=( + "recorded_r2_bucket_is_preserved_history:" + f"recorded={recorded_bucket}:requested={location.bucket}" + ), + ), + None, ) recorded_key = recorded_r2.key if recorded_r2 is not None else None if recorded_key and recorded_key != location.key: diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 3607e82b..b91d06fa 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -141,6 +141,7 @@ files: sha256: size_bytes: fetched_at: "2026-06-11T14:22:05+00:00" + source_url: https://www.irs.gov/pub/irs-soi/22in05ira.xlsx superseded_at: "2026-09-02T17:04:11+00:00" ``` @@ -318,8 +319,10 @@ deleted, and manifests keep the `storage.r2` URIs they already recorded as historical truth. A backfill copies bytes into the new bucket; it never rewrites where those bytes were first published. `publish-raw` and `fetch-artifact` enforce that: a recorded block that addresses the bytes in hand is preserved -whichever bucket is configured, and `publish-raw` refuses to restate it under a -different one. Bytes that the recorded object does not hold are not that +whichever bucket is configured, and `publish-raw` reports such an entry as +`skipped` (already published under the recorded bucket) rather than restating +it under a different one, so a sweep over a fully published tree stays green +after the flip. Bytes that the recorded object does not hold are not that object's history at all; see [Publisher Revisions](#publisher-revisions). The cutover therefore has one irreversible-looking step that is in fact additive @@ -410,7 +413,10 @@ export CHRONICLE_R2_DERIVED_BUCKET=chronicle-derived New raw publications land in the new bucket from that point. Manifests written before the flip keep pointing at `ledger-raw`, which is why the old bucket stays -readable. +readable. A `publish-raw --root db/data` sweep after the flip reports every +already-published entry as `skipped` with its recorded `ledger-raw` location +(`skipped_count` in the report) and exits 0; only bytes that no recorded object +holds are uploaded, into `chronicle-raw`. ### 6. Set the ledger-era buckets read-only diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 4629e06c..224e73bd 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -11,6 +11,7 @@ from chronicle.cli import main as cli_main from chronicle.artifacts import ( + AmbiguousManifestError, MalformedManifestError, RecordedR2LocatorError, SourceArtifactRevisionError, @@ -246,6 +247,7 @@ def test_publish_source_artifacts_uploads_manifest_entries(tmp_path): "failed_count": 0, "manifest_count": 1, "r2_link_count": 1, + "skipped_count": 0, "uploaded_count": 1, } assert storage["bucket"] == "ledger-raw" @@ -740,11 +742,15 @@ def test_publish_derived_uses_the_configured_bucket(tmp_path, monkeypatch): assert "chronicle-derived/derived/irs_soi/" in log.read_text() -def test_publish_raw_refuses_to_restate_a_recorded_bucket(tmp_path, monkeypatch): +def test_publish_raw_skips_an_object_already_held_by_a_preserved_bucket( + tmp_path, monkeypatch +): """A recorded storage.r2 bucket is preserved history, not a publish target. Archived witness records pin raw R2 URLs by hash, so backfilling the same - bytes into a renamed bucket must not rewrite the manifest. + bytes into a renamed bucket must not rewrite the manifest. The entry is + already published, so the sweep reports it skipped and stays green: after + the bucket-default flip every entry published before it takes this path. """ monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-1-1" @@ -777,17 +783,26 @@ def test_publish_raw_refuses_to_restate_a_recorded_bucket(tmp_path, monkeypatch) wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") wrangler.chmod(0o755) + before = manifest_path.read_bytes() report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) - unchanged = yaml.safe_load(manifest_path.read_text()) + entry = report.entries[0] - assert not report.valid - assert ( - report.entries[0] - .errors[0] - .startswith("recorded_r2_bucket_is_preserved_history:") + assert report.valid + assert entry.errors == () + assert entry.upload is None + assert entry.skipped == ( + "recorded_r2_bucket_is_preserved_history:" + "recorded=ledger-raw:requested=chronicle-raw" ) + assert entry.r2_location is not None + assert entry.r2_location.bucket == "ledger-raw" + assert entry.r2_location.key == recorded_key + assert entry.to_dict()["skipped"] == entry.skipped + assert report.counts["skipped_count"] == 1 + assert report.counts["uploaded_count"] == 0 + assert report.counts["failed_count"] == 0 assert not log.exists() - assert unchanged["files"][2023]["storage"]["r2"]["bucket"] == "ledger-raw" + assert manifest_path.read_bytes() == before def test_fetch_artifact_keeps_an_already_recorded_bucket(tmp_path, monkeypatch): @@ -1287,17 +1302,76 @@ def test_a_revision_is_refused_in_the_manifest_that_records_it(tmp_path): assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded assert (package / "22in05ira.xlsx").read_bytes() == b"traditional IRA table" - # Without the flag the same fetch addresses a manifest that has no entry to - # protect -- which is exactly why the flag exists. - report = _fetch_local( + # Without the flag the same fetch would address a manifest.yaml that no + # package reads and that protects nothing: the #225 path. It is refused, + # naming the manifests the directory keeps, and nothing is written. + with pytest.raises(AmbiguousManifestError) as stray: + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + ) + + assert TRADITIONAL_MANIFEST in str(stray.value) + assert "--manifest" in str(stray.value) + assert not (package / "manifest.yaml").exists() + assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + assert (package / "22in05ira.xlsx").read_bytes() == b"traditional IRA table" + + +def test_fetch_artifact_cli_refuses_a_stray_default_manifest(tmp_path, capsys): + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + _fetch_local( package, traditional, package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, ) + argv = [ + "fetch-artifact", + "--url", + str(traditional), + "--source-id", + "irs_soi", + "--package-id", + "soi-ira-traditional-contributions-2022", + "--year", + "2022", + "--out-dir", + str(package), + ] - assert report.valid - assert report.manifest_path.endswith("manifest.yaml") - assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + assert harness_main(argv) == 1 + + err = capsys.readouterr().err + assert err.startswith("error: ") + assert TRADITIONAL_MANIFEST in err + assert not (package / "manifest.yaml").exists() + + +def test_a_same_bytes_rename_is_refused_by_name_not_as_a_revision(tmp_path): + """Identical bytes under another filename are neither a revision nor a + re-fetch: the entry's filename must keep agreeing with its recorded key.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5") + _fetch_local(package, source, filename="table-5.xlsx", upload_r2=False) + recorded = (package / "manifest.yaml").read_bytes() + + for record_revision in (False, True): + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_local( + package, source, upload_r2=False, record_revision=record_revision + ) + message = str(raised.value) + assert "rename is not a release revision" in message + assert "filename=table-5.xlsx" in message + assert "names them 22in05ira.xlsx" in message + assert "--filename table-5.xlsx" in message + + assert (package / "manifest.yaml").read_bytes() == recorded + assert not (package / "22in05ira.xlsx").exists() + assert (package / "table-5.xlsx").read_bytes() == b"IRA table 5" @pytest.mark.parametrize( @@ -1314,6 +1388,36 @@ def test_a_manifest_name_must_stay_inside_the_package(tmp_path, manifest_filenam assert not package.exists() +def test_fetch_artifact_cli_reports_a_manifest_name_outside_the_package( + tmp_path, capsys +): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "table.xlsx", b"table") + argv = [ + "fetch-artifact", + "--url", + str(source), + "--source-id", + "irs_soi", + "--package-id", + "soi-table-5", + "--year", + "2022", + "--out-dir", + str(package), + "--manifest", + "../manifest.yaml", + ] + + assert harness_main(argv) == 1 + + err = capsys.readouterr().err + assert err.startswith("error: ") + assert "inside the package directory" in err + assert not package.exists() + assert not (tmp_path / "db" / "data" / "irs_soi" / "manifest.yaml").exists() + + def test_fetch_artifact_cli_targets_the_named_manifest(tmp_path, capsys): package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") @@ -1573,6 +1677,31 @@ def test_a_malformed_manifest_is_refused_before_anything_is_fetched(tmp_path, do assert list(package.iterdir()) == [manifest_path] +@pytest.mark.parametrize( + "document", + [ + "files:\n- not a mapping\n", + "files: 3\n", + "source_id: irs_soi\nfiles: text\n", + ], +) +def test_a_non_mapping_files_block_is_refused_before_anything_is_fetched( + tmp_path, document +): + """The same document inventory-artifacts and publish-raw report as + 'files must be a mapping'; a fetch must not overwrite the artifact first.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text(document) + + with pytest.raises(MalformedManifestError, match="files must be a mapping"): + _fetch_local(package, tmp_path / "publisher" / "never-read.xlsx") + + assert manifest_path.read_text() == document + assert list(package.iterdir()) == [manifest_path] + + @pytest.mark.parametrize("document", ["", "\n", "{}\n", "# only a comment\n"]) def test_an_empty_manifest_still_reads_as_absent(tmp_path, document): package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" From d911e1f9db40b5946d7d300b7845daffe22fbac5 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 01:53:56 -0400 Subject: [PATCH 025/105] Record into a fresh mapping when a manifest's files block is an explicit null A bare `files:` line parses as None. `_manifest_files` reads it as no entries (like an absent block) and `_upsert_manifest` now replaces the null with a mapping before recording, instead of `setdefault` keeping the null and the item assignment failing after the artifact was written and any upload ran. Covered by two new `files:` cases in test_an_empty_manifest_still_reads_as_absent. --- chronicle/artifacts.py | 7 ++++++- tests/test_chronicle_artifacts.py | 14 +++++++++++++- 2 files changed, 19 insertions(+), 2 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 7c3032f0..e67f151d 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -107,6 +107,8 @@ def _manifest_files(payload: dict[str, Any], manifest_path: Path) -> dict[str, A """ files = payload.get("files") if files is None: + # A bare ``files:`` line parses as None: no entries, like an absent + # block. The writer normalizes it to a mapping before recording into it. return {} if not isinstance(files, dict): raise MalformedManifestError( @@ -1684,7 +1686,10 @@ def _upsert_manifest( payload.setdefault("dataset", dataset) payload.setdefault("source_page", source_page) payload.setdefault("table", table) - payload.setdefault("files", {}) + if payload.get("files") is None: + # setdefault keeps an explicit null (a bare ``files:`` line); the + # entry below needs a mapping to record into. + payload["files"] = {} file_entry: dict[str, Any] = { "filename": filename, "source_url": source_url, diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 224e73bd..f41f3938 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1702,8 +1702,20 @@ def test_a_non_mapping_files_block_is_refused_before_anything_is_fetched( assert list(package.iterdir()) == [manifest_path] -@pytest.mark.parametrize("document", ["", "\n", "{}\n", "# only a comment\n"]) +@pytest.mark.parametrize( + "document", + [ + "", + "\n", + "{}\n", + "# only a comment\n", + "files:\n", + "source_id: irs_soi\nfiles:\n", + ], +) def test_an_empty_manifest_still_reads_as_absent(tmp_path, document): + """Including a bare ``files:`` line, which parses as an explicit null: the + fetch records into a fresh mapping rather than failing after the write.""" package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" package.mkdir(parents=True) (package / "manifest.yaml").write_text(document) From 2e8aae01770ad6ec8e5c9e54d33c43533ae76345 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:00:58 -0400 Subject: [PATCH 026/105] Start Sol gate round progress log --- PROGRESS.md | 36 ++++++++++++++++++++++++++++++++++++ 1 file changed, 36 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 79bafcd9..0cf8920b 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -188,3 +188,39 @@ Running the same operations against a checkout of the previous head: - Follow-up PR, after Max creates and backfills the new buckets: flip `DEFAULT_R2_RAW_BUCKET` / `DEFAULT_R2_DERIVED_BUCKET` to `chronicle-raw` / `chronicle-derived`. + +## Review fixes (Sol gate round 3) + +### State + +- Detached HEAD: `fb1bc1df`, the PR #226 head supplied for the ten-finding Sol + gate round. +- Scope: ten operational-rename findings in artifact fetch/publish/inventory, + Supabase compatibility aliases, and the README cutover procedure. No tracked + `db/data/**` manifest will be changed. +- `CLAUDE.md`, one of the requested initial reads, is absent from both this + worktree and `/Users/maxghenis/PolicyEngine/chronicle`; `AGENTS.md` and the + remaining requested guidance/code/tests are present. +- PR #227 is available read-only at + `/Users/maxghenis/PolicyEngine/_worktrees/chronicle-227-fix`. Applicable + non-microdata hunks from `daafac0` and `c0d9d74` will be ported with the same + function names and shapes. + +### Done + +- Re-established the lane state in this committed progress log before making + gate-round code or test changes. +- Read `AGENTS.md`, the Bucket Cutover and Publisher Revisions contracts in + `docs/storage-architecture.md`, the README cutover instructions, and the + named implementation/test surfaces. Confirmed Chronicle must preserve + publisher bytes and provenance, refuse unsafe fetches before I/O, and leave + schema/bucket value cutovers explicit. + +### Next + +- Add and run focused failing regression tests for findings 1-8, recording the + exact commands and observed failures for `out.md`. +- Port the matching #227 validation, year-key selection, in-place entry update, + and package-manifest discovery hunks; implement the remaining artifact fixes. +- Reproduce and fix findings 9-10, then run the required lint, format, full-test, + and tracked-USDA sweep verification. From 7c06b1a3573eca7a098da776531088e1c4207043 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:10:00 -0400 Subject: [PATCH 027/105] Refuse unsafe fetch manifest rewrites --- PROGRESS.md | 17 ++- chronicle/artifacts.py | 134 ++++++++++++++-- tests/test_chronicle_artifacts.py | 246 ++++++++++++++++++++++++++++++ 3 files changed, 380 insertions(+), 17 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 0cf8920b..0e76c20f 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -215,12 +215,21 @@ Running the same operations against a checkout of the previous head: named implementation/test surfaces. Confirmed Chronicle must preserve publisher bytes and provenance, refuse unsafe fetches before I/O, and leave schema/bucket value cutovers explicit. +- Reproduced findings 1, 4, 5, and 8 with 11 failing cases: mismatched manifest + identity reached the publisher read, quoted year keys bypassed revision + protection, duplicate year spellings and malformed entries reached I/O, and + both refetch and revision discarded entry metadata. +- Ported the non-microdata parts of #227's `_assert_manifest_identifies`, + `_select_vintage_entry`, `_FETCH_OWNED_FIELDS`, and in-place + `_upsert_manifest` flow. Fetch now validates manifest identity and entry + shape before I/O, resolves either year-key spelling while refusing both, + preserves the recorded key spelling, and carries forward every field it does + not own. The 11 focused cases now pass. ### Next -- Add and run focused failing regression tests for findings 1-8, recording the - exact commands and observed failures for `out.md`. -- Port the matching #227 validation, year-key selection, in-place entry update, - and package-manifest discovery hunks; implement the remaining artifact fixes. +- Reproduce and fix root discovery/type validation (findings 2 and 7), then R2 + canonical-location/provider validation (findings 3 and 6). +- Port #227's `package_manifest_paths` helper shape for the default root sweeps. - Reproduce and fix findings 9-10, then run the required lint, format, full-test, and tracked-USDA sweep verification. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index e67f151d..5aaa2a1a 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -119,6 +119,30 @@ def _manifest_files(payload: dict[str, Any], manifest_path: Path) -> dict[str, A return files +def _assert_manifest_identifies( + existing_manifest: dict[str, Any], + manifest_path: Path, + *, + source_id: str, + package_id: str, +) -> None: + """Refuse to fetch into a manifest that identifies another package. + + The R2 key and registration identity are built from the fetch arguments. + Recording them in a manifest that declares different identifiers would + leave one entry making two incompatible provenance claims. + """ + for field, value in (("source_id", source_id), ("package_id", package_id)): + declared = existing_manifest.get(field) + declared = declared.strip() if isinstance(declared, str) else declared + if declared not in (None, "") and str(declared) != value: + raise SourceArtifactManifestError( + f"{manifest_path} declares {field}={declared!r}; refusing to " + f"fetch {field}={value!r} into it. Fetch into the package the " + "manifest identifies, or into that package's own directory." + ) + + def default_r2_raw_bucket() -> str: """Resolve the raw bucket: ``$CHRONICLE_R2_RAW_BUCKET`` or the default.""" return env_value(R2_RAW_BUCKET_ENV, default=DEFAULT_R2_RAW_BUCKET) @@ -651,11 +675,22 @@ def fetch_source_artifact( _refuse_a_stray_default_manifest(output, manifest_path) existing_manifest = _read_manifest(manifest_path) _manifest_files(existing_manifest, manifest_path) - recorded_identity = _recorded_identity( - _manifest_file_spec(existing_manifest, year), + _assert_manifest_identifies( + existing_manifest, + manifest_path, + source_id=source_id, + package_id=package_id, + ) + vintage_key, _existing_value, selected_spec, _index = _select_vintage_entry( + existing_manifest, manifest_path=manifest_path, year=year, ) + recorded_identity = _recorded_identity( + selected_spec, + manifest_path=manifest_path, + year=vintage_key, + ) fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() content, inferred_filename = _read_artifact(source_url) @@ -671,7 +706,7 @@ def fetch_source_artifact( _assert_recorded_identity_holds_these_bytes( recorded_identity, manifest_path=manifest_path, - year=year, + year=vintage_key, filename=artifact_filename, sha256=sha256, size_bytes=size_bytes, @@ -1341,13 +1376,58 @@ def _read_manifest(manifest_path: Path) -> dict[str, Any]: return payload -def _manifest_file_spec(payload: dict[str, Any], year: Any) -> dict[str, Any]: - """Return one manifest ``files`` entry, or an empty mapping.""" - files = payload.get("files") +def _select_vintage_entry( + payload: dict[str, Any], + *, + manifest_path: Path, + year: Any, +) -> tuple[Any, Any, dict[str, Any], int | None]: + """Locate the entry a fetch revises: ``(key, files[key], entry, index)``. + + Integer and quoted-integer keys are two spellings of one vintage. Preserve + the spelling already present, refuse a manifest that contains both, and + reject any present non-mapping entry before publisher I/O. The final tuple + slot matches the stacked #227 selector; table manifests never use a list + index. + """ + files = payload.get("files") if isinstance(payload, dict) else None + if files is None: + return year, None, {}, None if not isinstance(files, dict): - return {} - spec = files.get(year) - return spec if isinstance(spec, dict) else {} + raise MalformedManifestError( + f"{manifest_path} files must be a mapping; it is a " + f"{type(files).__name__}. Chronicle will not write into a manifest " + "it cannot read." + ) + + forms: tuple[Any, ...] + if isinstance(year, bool): + forms = (year,) + elif isinstance(year, int): + forms = (year, str(year)) + else: + text = str(year) + if text.isdecimal() and (text == "0" or not text.startswith("0")): + forms = (year, int(text)) + else: + forms = (year,) + present = [form for form in forms if form in files] + if len(present) > 1: + raise MalformedManifestError( + f"{manifest_path}: Vintage {year!r} is recorded under both keys " + f"{present!r}; one vintage has one key. Merge the entries by hand " + "first. Chronicle will not choose which entry is the record." + ) + if not present: + return year, None, {}, None + key = present[0] + existing = files[key] + if not isinstance(existing, dict): + raise MalformedManifestError( + f"{manifest_path} entry {key!r} must be a mapping; it is a " + f"{type(existing).__name__}." + ) + return key, existing, existing, None def _recorded_storage(spec: Any) -> dict[str, Any]: @@ -1663,6 +1743,20 @@ def _superseding_storage( return storage +#: Entry fields a fetch owns. Every other field already recorded on the entry +#: is carried forward during a refetch or explicit publisher revision. +_FETCH_OWNED_FIELDS: frozenset[str] = frozenset( + { + "filename", + "source_url", + "sha256", + "size_bytes", + "fetched_at", + "storage", + } +) + + def _upsert_manifest( manifest_path: Path, *, @@ -1681,6 +1775,13 @@ def _upsert_manifest( record_revision: bool = False, ) -> None: payload = _read_manifest(manifest_path) + _manifest_files(payload, manifest_path) + _assert_manifest_identifies( + payload, + manifest_path, + source_id=source_id, + package_id=package_id, + ) payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) payload.setdefault("dataset", dataset) @@ -1690,6 +1791,11 @@ def _upsert_manifest( # setdefault keeps an explicit null (a bare ``files:`` line); the # entry below needs a mapping to record into. payload["files"] = {} + key, _existing_value, recorded_spec, _index = _select_vintage_entry( + payload, + manifest_path=manifest_path, + year=year, + ) file_entry: dict[str, Any] = { "filename": filename, "source_url": source_url, @@ -1697,9 +1803,11 @@ def _upsert_manifest( "size_bytes": size_bytes, "fetched_at": fetched_at, } - recorded_spec = _manifest_file_spec(payload, year) + for field, value in recorded_spec.items(): + if field not in _FETCH_OWNED_FIELDS and field not in file_entry: + file_entry[field] = value recorded_storage = _recorded_storage(recorded_spec) - identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=year) + identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=key) new_r2 = r2_location.to_dict() if r2_location is not None else None holds = identity is not None and identity.holds(sha256=sha256, filename=filename) if identity is not None and not holds and not record_revision: @@ -1709,7 +1817,7 @@ def _upsert_manifest( raise SourceArtifactRevisionError( _revision_error_message( manifest_path=manifest_path, - year=year, + year=key, filename=filename, identity=identity, sha256=sha256, @@ -1738,7 +1846,7 @@ def _upsert_manifest( # revision over a never-published entry supersedes nothing. if storage: file_entry["storage"] = storage - payload["files"][year] = file_entry + payload["files"][key] = file_entry manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False), encoding="utf-8", diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index f41f3938..7099b79e 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -14,6 +14,7 @@ AmbiguousManifestError, MalformedManifestError, RecordedR2LocatorError, + SourceArtifactManifestError, SourceArtifactRevisionError, build_artifact_key, build_artifact_rows, @@ -1742,3 +1743,248 @@ def test_a_malformed_manifest_is_reported_by_inventory_and_publish(tmp_path): assert not published.valid assert published.entries == () assert "must be a YAML mapping" in published.errors[0] + + +# --------------------------------------------------------------------------- +# Sol gate round 3: fetch preflight and in-place manifest updates +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + ( + "mismatched_field", + "declared_source_id", + "declared_package_id", + "source_id", + "package_id", + ), + [ + pytest.param( + "source_id", + "other_source", + "requested-package", + "requested_source", + "requested-package", + id="source-id", + ), + pytest.param( + "package_id", + "usda_snap", + "usda-snap-fy69-to-current", + "usda_snap", + "usda-snap-fy2025-monthly-state-caseloads", + id="package-id", + ), + ], +) +def test_fetch_refuses_a_selected_manifest_for_another_package_before_io( + tmp_path, + monkeypatch, + mismatched_field, + declared_source_id, + declared_package_id, + source_id, + package_id, +): + package = tmp_path / "db" / "data" / "usda_snap" / "fy69_to_current" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": declared_source_id, + "package_id": declared_package_id, + "files": {}, + }, + sort_keys=False, + ) + ) + before = manifest_path.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("a mismatched manifest must be refused before I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(SourceArtifactManifestError) as raised: + fetch_source_artifact( + "https://example.test/snap-zip-fy69tocurrent-6.zip", + source_id=source_id, + package_id=package_id, + year=2025, + output_dir=package, + ) + + message = str(raised.value) + declared = { + "source_id": declared_source_id, + "package_id": declared_package_id, + }[mismatched_field] + requested = {"source_id": source_id, "package_id": package_id}[mismatched_field] + assert f"{mismatched_field}={declared!r}" in message + assert f"{mismatched_field}={requested!r}" in message + assert manifest_path.read_bytes() == before + assert list(package.iterdir()) == [manifest_path] + + +def test_fetch_uses_a_quoted_year_key_for_revision_protection(tmp_path): + package = tmp_path / "db" / "data" / "irs_soi" / "table" + package.mkdir(parents=True) + artifact_path = package / "table.xlsx" + original = b"original publisher bytes" + revised = b"silently revised publisher bytes" + artifact_path.write_bytes(original) + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "irs_soi", + "package_id": "soi-table", + "files": { + "2024": { + "filename": artifact_path.name, + "source_url": "https://example.test/table.xlsx", + "sha256": hashlib.sha256(original).hexdigest(), + "size_bytes": len(original), + } + }, + }, + sort_keys=False, + ) + ) + source = _publish(tmp_path, artifact_path.name, revised) + before = manifest_path.read_bytes() + + with pytest.raises(SourceArtifactRevisionError): + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + ) + + assert manifest_path.read_bytes() == before + assert artifact_path.read_bytes() == original + + +def test_fetch_refuses_both_spellings_of_one_year_before_io(tmp_path, monkeypatch): + package = tmp_path / "db" / "data" / "irs_soi" / "table" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "irs_soi", + "package_id": "soi-table", + "files": { + 2024: {"filename": "numeric.xlsx"}, + "2024": {"filename": "quoted.xlsx"}, + }, + }, + sort_keys=False, + ) + ) + before = manifest_path.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("ambiguous year keys must be refused before I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(MalformedManifestError, match="both keys"): + fetch_source_artifact( + "https://example.test/table.xlsx", + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + ) + + assert manifest_path.read_bytes() == before + assert list(package.iterdir()) == [manifest_path] + + +@pytest.mark.parametrize( + "file_spec", + [ + pytest.param([], id="list"), + pytest.param("not a mapping", id="string"), + pytest.param(0, id="zero"), + pytest.param(False, id="false"), + pytest.param(None, id="null"), + ], +) +def test_fetch_refuses_a_non_mapping_year_entry_before_io( + tmp_path, monkeypatch, file_spec +): + package = tmp_path / "db" / "data" / "irs_soi" / "table" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "irs_soi", + "package_id": "soi-table", + "files": {2024: file_spec}, + }, + sort_keys=False, + ) + ) + before = manifest_path.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("a malformed year entry must be refused before I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(MalformedManifestError, match="entry 2024.*mapping"): + fetch_source_artifact( + "https://example.test/table.xlsx", + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + ) + + assert manifest_path.read_bytes() == before + assert list(package.iterdir()) == [manifest_path] + + +@pytest.mark.parametrize("revision", [False, True], ids=["refetch", "revision"]) +def test_fetch_carries_forward_fields_it_does_not_own(tmp_path, revision): + package = tmp_path / "db" / "data" / "irs_soi" / "table" + source = _publish(tmp_path, "table.xlsx", b"original publisher bytes") + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + ) + manifest_path = package / "manifest.yaml" + manifest = yaml.safe_load(manifest_path.read_text()) + metadata = { + "source_table": "Publisher table 7", + "notes": "Keep this review note.", + "source_urls": ["https://example.test/landing-page"], + "archive_member": "table.csv", + "year": 2024, + } + manifest["files"][2024].update(metadata) + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + if revision: + source.write_bytes(b"publisher revision") + + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + record_revision=revision, + ) + + updated = yaml.safe_load(manifest_path.read_text())["files"][2024] + for field, value in metadata.items(): + assert updated.get(field) == value From 19f0cf29461b427e17c55130f01d71cb6a2f83f6 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:12:30 -0400 Subject: [PATCH 028/105] Sweep every package manifest safely --- PROGRESS.md | 13 ++- chronicle/artifacts.py | 76 +++++++++++++---- tests/test_chronicle_artifacts.py | 131 ++++++++++++++++++++++++++++++ 3 files changed, 201 insertions(+), 19 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 0e76c20f..392ea44a 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -225,11 +225,18 @@ Running the same operations against a checkout of the previous head: shape before I/O, resolves either year-key spelling while refusing both, preserves the recorded key spelling, and carries forward every field it does not own. The 11 focused cases now pass. +- Reproduced finding 2 with both default sweeps reporting only one of four + supported manifest names, and finding 7 with all four falsy non-mapping + `files` values producing `(inventory.valid, publish.valid) == (True, True)`. +- Ported #227's `is_manifest_filename` / `package_manifest_paths` shapes and + made default root sweeps discover `manifest.yaml`, `manifest.yml`, and both + `manifest_` extensions. Both sweeps now share `_manifest_files`, so + only `None` means absent and every other non-mapping value is reported. All + 106 artifact tests pass. ### Next -- Reproduce and fix root discovery/type validation (findings 2 and 7), then R2 - canonical-location/provider validation (findings 3 and 6). -- Port #227's `package_manifest_paths` helper shape for the default root sweeps. +- Reproduce and fix R2 canonical-location/provider validation (findings 3 and + 6). - Reproduce and fix findings 9-10, then run the required lint, format, full-test, and tracked-USDA sweep verification. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 5aaa2a1a..6f9f06f4 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -48,6 +48,59 @@ # name is an input, not a constant, wherever a caller addresses a package. DEFAULT_MANIFEST_FILENAME = "manifest.yaml" +# The names a package directory's manifests may carry. Match case-insensitively +# because Chronicle is also used on case-insensitive filesystems. +_MANIFEST_FILENAME_RE = re.compile( + r"^manifest(?:_[^/\\]+)?\.ya?ml$", + re.IGNORECASE, +) + + +def is_manifest_filename(value: Any) -> bool: + """Whether ``value`` is a package-manifest filename.""" + if not isinstance(value, str) or not value or value != value.strip(): + return False + return value == Path(value).name and bool(_MANIFEST_FILENAME_RE.fullmatch(value)) + + +def package_manifest_paths(package_dir: Path) -> list[Path]: + """Return every manifest file a package directory keeps, sorted by name.""" + directory = Path(package_dir) + if not directory.is_dir(): + return [] + return sorted( + path + for path in directory.iterdir() + if path.is_file() and is_manifest_filename(path.name) + ) + + +def _root_manifest_paths(root: Path, manifest_filename: str) -> list[Path]: + """Return the manifests a root sweep addresses. + + The default is package discovery, not one literal filename: both YAML + extensions and every ``manifest_`` sibling participate. A caller + that supplies another filename keeps the historical exact-name override. + """ + if manifest_filename != DEFAULT_MANIFEST_FILENAME: + return sorted(path for path in root.rglob(manifest_filename) if path.is_file()) + package_dirs = { + path.parent + for pattern in ( + "manifest.yaml", + "manifest.yml", + "manifest_*.yaml", + "manifest_*.yml", + ) + for path in root.rglob(pattern) + if path.is_file() + } + return sorted( + manifest_path + for package_dir in package_dirs + for manifest_path in package_manifest_paths(package_dir) + ) + def _manifest_path(output: Path, manifest_filename: str) -> Path: """Return the named manifest inside ``output``. @@ -66,13 +119,10 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: def _sibling_manifests(output: Path) -> list[str]: """Return the ``manifest_*.yaml`` files a package directory keeps.""" - if not output.is_dir(): - return [] return sorted( path.name - for pattern in ("manifest_*.yaml", "manifest_*.yml") - for path in output.glob(pattern) - if path.is_file() + for path in package_manifest_paths(output) + if path.stem.startswith("manifest_") ) @@ -935,26 +985,22 @@ def publish_source_artifacts( entries: list[RawArtifactPublishEntry] = [] errors: list[str] = [] - for manifest_path in sorted(root_path.rglob(manifest_filename)): + for manifest_path in _root_manifest_paths(root_path, manifest_filename): try: manifest = _read_manifest(manifest_path) + files = _manifest_files(manifest, manifest_path) except (OSError, MalformedManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue manifest_source_id = source_id or manifest.get("source_id") manifest_package_id = package_id or manifest.get("package_id") - files = manifest.get("files") or {} if not manifest_source_id: errors.append(f"Manifest missing source_id: {manifest_path}") continue if not manifest_package_id: errors.append(f"Manifest missing package_id: {manifest_path}") continue - if not isinstance(files, dict): - errors.append(f"Manifest files must be a mapping: {manifest_path}") - continue - try: resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, @@ -1061,16 +1107,14 @@ def inventory_source_artifacts( errors=(f"Root does not exist: {root_path}",), ) - manifests = sorted(root_path.rglob(manifest_filename)) + manifests = _root_manifest_paths(root_path, manifest_filename) for manifest_path in manifests: try: - files = _read_manifest(manifest_path).get("files") or {} + manifest = _read_manifest(manifest_path) + files = _manifest_files(manifest, manifest_path) except (OSError, MalformedManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue - if not isinstance(files, dict): - errors.append(f"Manifest files must be a mapping: {manifest_path}") - continue for year, spec in files.items(): entries.append(_inventory_entry(manifest_path, year, spec)) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 7099b79e..9303b9b1 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1988,3 +1988,134 @@ def test_fetch_carries_forward_fields_it_does_not_own(tmp_path, revision): updated = yaml.safe_load(manifest_path.read_text())["files"][2024] for field, value in metadata.items(): assert updated.get(field) == value + + +# --------------------------------------------------------------------------- +# Sol gate round 3: whole-tree manifest discovery and files-block shape +# --------------------------------------------------------------------------- + + +def _write_sweep_manifests(root): + manifest_names = ( + "manifest.yaml", + "manifest.yml", + "manifest_named.yaml", + "manifest_named.yml", + ) + for index, manifest_name in enumerate(manifest_names): + package = root / f"package-{index}" + package.mkdir(parents=True) + content = f"publisher artifact {index}".encode() + filename = f"artifact-{index}.csv" + (package / filename).write_bytes(content) + (package / manifest_name).write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": f"package-{index}", + "files": { + 2024: { + "filename": filename, + "sha256": hashlib.sha256(content).hexdigest(), + "size_bytes": len(content), + } + }, + }, + sort_keys=False, + ) + ) + decoy = root / "decoy" / "manifest-not-a-package.yaml" + decoy.parent.mkdir() + decoy.write_text("this: is not a package manifest\n") + + +def test_inventory_default_sweep_discovers_every_package_manifest(tmp_path): + root = tmp_path / "data" + _write_sweep_manifests(root) + + report = inventory_source_artifacts(root) + + assert report.valid + assert report.counts["manifest_count"] == 4 + assert report.counts["artifact_count"] == 4 + assert {entry.manifest_path.rsplit("/", 1)[-1] for entry in report.entries} == { + "manifest.yaml", + "manifest.yml", + "manifest_named.yaml", + "manifest_named.yml", + } + + +def test_publish_default_sweep_discovers_every_package_manifest(tmp_path): + root = tmp_path / "data" + _write_sweep_manifests(root) + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + + report = publish_source_artifacts(root, wrangler_command=str(wrangler)) + + assert report.valid + assert report.counts["manifest_count"] == 4 + assert report.counts["artifact_count"] == 4 + assert report.counts["uploaded_count"] == 4 + assert len(log.read_text().splitlines()) == 4 + + +@pytest.mark.parametrize( + "files", + [ + pytest.param([], id="empty-list"), + pytest.param("", id="empty-string"), + pytest.param(0, id="zero"), + pytest.param(False, id="false"), + ], +) +def test_sweeps_reject_falsy_non_mapping_files_blocks(tmp_path, files): + package = tmp_path / "data" / "package" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": "package", + "files": files, + }, + sort_keys=False, + ) + ) + before = manifest_path.read_bytes() + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package, wrangler_command=str(wrangler)) + + assert (inventory.valid, published.valid) == (False, False) + assert "files must be a mapping" in inventory.errors[0] + assert "files must be a mapping" in published.errors[0] + assert inventory.entries == () + assert published.entries == () + assert not log.exists() + assert manifest_path.read_bytes() == before + + +def test_sweeps_treat_a_null_files_block_as_absent(tmp_path): + package = tmp_path / "data" / "package" + package.mkdir(parents=True) + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": "package", + "files": None, + }, + sort_keys=False, + ) + ) + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package) + + assert inventory.valid + assert published.valid From 404281979b922fca6eff7eab3c9e803e8c3df07e Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:14:07 -0400 Subject: [PATCH 029/105] Validate R2 routes before cutover skips --- PROGRESS.md | 10 +++- chronicle/artifacts.py | 21 +++++-- tests/test_chronicle_artifacts.py | 96 +++++++++++++++++++++++++++++++ 3 files changed, 119 insertions(+), 8 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 392ea44a..0772c3cb 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -233,10 +233,16 @@ Running the same operations against a checkout of the previous head: `manifest_` extensions. Both sweeps now share `_manifest_files`, so only `None` means absent and every other non-mapping value is reported. All 106 artifact tests pass. +- Reproduced finding 3 as a green preserved-bucket skip for a key routed to the + wrong package/year, and finding 6 as both a fetch reaching I/O and a green + publish skip for a self-consistent `s3://` block under `storage.r2`. +- Raw publish now compares the recorded key with the canonical + source/package/year key before any bucket-change skip. Recorded R2 validation + also requires `provider='r2'` (and therefore an `r2://` effective URI). The + focused canonical-key/provider tests, including the existing URI-only and + stale-country cases, pass without uploads or manifest rewrites on refusal. ### Next -- Reproduce and fix R2 canonical-location/provider validation (findings 3 and - 6). - Reproduce and fix findings 9-10, then run the required lint, format, full-test, and tracked-USDA sweep verification. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 6f9f06f4..5a2a341e 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1598,6 +1598,12 @@ def _validated_recorded_r2( "locate its object: provider, bucket and key, or a uri that " "supplies them." ) + if provider != "r2": + raise RecordedR2LocatorError( + f"{where}: provider={provider!r} does not identify R2. A block " + "under storage.r2 must use provider='r2' and an r2:// URI, not " + f"{provider}://." + ) segments = key.split("/") if ( @@ -2012,6 +2018,15 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: package_path=manifest_path, ), ) + recorded_key = recorded_r2.key if recorded_r2 is not None else None + if recorded_key and recorded_key != location.key: + # Validate the full source/package/year route before the preserved- + # bucket shortcut below. A bucket rename does not make a misrouted + # object valid history. + return refuse( + "recorded_r2_key_disagrees_with_country_prefix:" + f"recorded={recorded_key}:expected={location.key}" + ) recorded_bucket = recorded_r2.bucket if recorded_r2 is not None else None if recorded_r2 is not None and recorded_bucket != location.bucket: # The recorded bucket is preserved history and, per the identity check @@ -2045,12 +2060,6 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: ), None, ) - recorded_key = recorded_r2.key if recorded_r2 is not None else None - if recorded_key and recorded_key != location.key: - return refuse( - "recorded_r2_key_disagrees_with_country_prefix:" - f"recorded={recorded_key}:expected={location.key}" - ) upload = _upload_r2_object( location, artifact_path, diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 9303b9b1..b1738035 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -2119,3 +2119,99 @@ def test_sweeps_treat_a_null_files_block_as_absent(tmp_path): assert inventory.valid assert published.valid + + +# --------------------------------------------------------------------------- +# Sol gate round 3: canonical R2 locators before bucket-cutover skips +# --------------------------------------------------------------------------- + + +def test_publish_checks_the_canonical_key_before_a_preserved_bucket_skip( + tmp_path, monkeypatch +): + package = tmp_path / "db" / "data" / "irs_soi" / "table" + source = _publish(tmp_path, "table.xlsx", b"publisher table") + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + ) + manifest_path = package / "manifest.yaml" + manifest = yaml.safe_load(manifest_path.read_text()) + spec = manifest["files"][2024] + wrong_key = f"raw/irs_soi/other-package/2023/{spec['sha256']}/{spec['filename']}" + spec["storage"] = { + "r2": { + "provider": "r2", + "bucket": "ledger-raw", + "key": wrong_key, + "uri": f"r2://ledger-raw/{wrong_key}", + } + } + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_bytes() + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + + report = publish_source_artifacts(package, wrangler_command=str(wrangler)) + + assert not report.valid + assert report.entries[0].upload is None + assert report.entries[0].skipped is None + assert ( + report.entries[0] + .errors[0] + .startswith("recorded_r2_key_disagrees_with_country_prefix:") + ) + assert not log.exists() + assert manifest_path.read_bytes() == before + + +def _make_recorded_locator_use_s3(package): + manifest_path = package / "manifest.yaml" + manifest = yaml.safe_load(manifest_path.read_text()) + r2 = manifest["files"][2022]["storage"]["r2"] + r2["provider"] = "s3" + r2["uri"] = f"s3://{r2['bucket']}/{r2['key']}" + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + return manifest_path + + +def test_fetch_refuses_a_self_consistent_non_r2_locator_before_io( + tmp_path, monkeypatch +): + package, source, _report = _recorded_package(tmp_path) + manifest_path = _make_recorded_locator_use_s3(package) + before = manifest_path.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("a non-R2 storage.r2 locator must be refused before I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(RecordedR2LocatorError, match="provider.*r2"): + _fetch_local(package, source, upload_r2=False) + + assert manifest_path.read_bytes() == before + + +def test_publish_refuses_a_self_consistent_non_r2_locator(tmp_path, monkeypatch): + package, _source, _report = _recorded_package(tmp_path) + manifest_path = _make_recorded_locator_use_s3(package) + before = manifest_path.read_bytes() + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + log = tmp_path / "publish.log" + wrangler = _wrangler_stub(tmp_path, log) + + report = publish_source_artifacts(package, wrangler_command=str(wrangler)) + + assert not report.valid + assert report.entries[0].upload is None + assert report.entries[0].skipped is None + assert report.entries[0].errors[0].startswith("recorded_r2_locator_invalid:") + assert "provider" in report.entries[0].errors[0] + assert not log.exists() + assert manifest_path.read_bytes() == before From c218dcc521988f7c4587a15b226f1f17fdf8f5fc Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:15:26 -0400 Subject: [PATCH 030/105] Restore Supabase schema compatibility aliases --- PROGRESS.md | 13 +++++++++++-- db/supabase_client.py | 17 ++++++++++++++++- tests/test_chronicle_env.py | 19 ++++++++++--------- tests/test_chronicle_namespace.py | 13 +++++++++++++ 4 files changed, 50 insertions(+), 12 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 0772c3cb..7a1023b5 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -241,8 +241,17 @@ Running the same operations against a checkout of the previous head: also requires `provider='r2'` (and therefore an `r2://` effective URI). The focused canonical-key/provider tests, including the existing URI-only and stale-country cases, pass without uploads or manifest rewrites on refusal. +- Reproduced finding 9 with the shipped import raising `ImportError` for + `LEDGER_SCHEMA`, then restored `LEDGER_SCHEMA` and `TARGETS_SCHEMA` as + deprecated aliases of the stable defaults. Runtime queries remain on the + lazy functions and still honor post-import environment changes; 55 focused + namespace/env/client cases pass (1 skipped for absent real credentials). +- Reproduced both parts of finding 10: the README did not state that an + unqualified mirror load writes to `ledger`, and it named the absent + `supabase/migrations/20260504_chronicle_bronze.sql` file. ### Next -- Reproduce and fix findings 9-10, then run the required lint, format, full-test, - and tracked-USDA sweep verification. +- Fix the README cutover instructions and commit their two regression tests. +- Run the required lint, format, full-test, and tracked-USDA sweep verification, + then write the final `out.md` report. diff --git a/db/supabase_client.py b/db/supabase_client.py index 190d8b7e..24f89f18 100644 --- a/db/supabase_client.py +++ b/db/supabase_client.py @@ -4,6 +4,11 @@ Provides connection to PolicyEngine Supabase database for: - Source metadata and dataset registries - Target inputs + +``LEDGER_SCHEMA`` and ``TARGETS_SCHEMA`` remain as deprecated compatibility +aliases for their default schema names. Runtime code should call +``chronicle_schema()`` and ``targets_schema()`` so environment overrides are +resolved at the time of use. """ from __future__ import annotations @@ -15,11 +20,21 @@ from supabase import create_client, Client -from chronicle.env import default_chronicle_schema, env_value +from chronicle.env import ( + DEFAULT_CHRONICLE_SCHEMA, + default_chronicle_schema, + env_value, +) TARGETS_SCHEMA_ENV = "POLICYENGINE_TARGETS_SCHEMA" DEFAULT_TARGETS_SCHEMA = "targets" +# Deprecated import compatibility. These are deliberately aliases for the +# defaults, not environment-backed runtime values; query paths below remain on +# the lazy resolver functions. +LEDGER_SCHEMA = DEFAULT_CHRONICLE_SCHEMA +TARGETS_SCHEMA = DEFAULT_TARGETS_SCHEMA + def chronicle_schema() -> str: """Resolve the hosted Chronicle schema for a query. diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py index 202d86f5..28bce190 100644 --- a/tests/test_chronicle_env.py +++ b/tests/test_chronicle_env.py @@ -280,24 +280,25 @@ def test_supabase_schema_default_is_unchanged(): def test_supabase_schema_is_not_bound_at_import(monkeypatch): - """No module-level constant may freeze the schema at import time. + """Compatibility constants do not freeze the runtime schema resolver. - A reload under a set variable is the pre-fix behavior this guards against: - it proves nothing about a module that resolved the value once, at - collection, and answers with the stale constant forever after. + The deprecated names expose only stable defaults for existing importers. + Query code calls the functions, which still honor an environment change + made after module import. """ import db.supabase_client - assert not [ - name - for name, value in vars(db.supabase_client).items() - if name.isupper() and value == "ledger" - ] + assert db.supabase_client.LEDGER_SCHEMA == "ledger" + assert db.supabase_client.TARGETS_SCHEMA == "targets" monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + monkeypatch.setenv("POLICYENGINE_TARGETS_SCHEMA", "targets_probe") unreloaded = importlib.import_module("db.supabase_client") assert unreloaded.chronicle_schema() == "chronicle_probe" + assert unreloaded.targets_schema() == "targets_probe" + assert unreloaded.LEDGER_SCHEMA == "ledger" + assert unreloaded.TARGETS_SCHEMA == "targets" # --------------------------------------------------------------------------- diff --git a/tests/test_chronicle_namespace.py b/tests/test_chronicle_namespace.py index 5dad695e..a8ce2b63 100644 --- a/tests/test_chronicle_namespace.py +++ b/tests/test_chronicle_namespace.py @@ -40,6 +40,19 @@ def test_chronicle_supabase_schema_boundaries_are_defaulted(): assert supabase_client.targets_schema() == "targets" +def test_chronicle_supabase_schema_compatibility_aliases_are_defaulted(): + """The shipped module keeps its pre-rename import surface.""" + from chronicle.env import DEFAULT_CHRONICLE_SCHEMA + from db.supabase_client import ( + DEFAULT_TARGETS_SCHEMA, + LEDGER_SCHEMA, + TARGETS_SCHEMA, + ) + + assert LEDGER_SCHEMA == DEFAULT_CHRONICLE_SCHEMA == "ledger" + assert TARGETS_SCHEMA == DEFAULT_TARGETS_SCHEMA == "targets" + + def test_chronicle_supabase_schema_follows_the_environment(monkeypatch): """The renamed variable reaches the client after it has been imported.""" from db import supabase_client From 701712c864bf19376ec78eab59f07314db758249 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:16:40 -0400 Subject: [PATCH 031/105] Make the Supabase cutover instructions truthful --- PROGRESS.md | 6 +++++- README.md | 17 ++++++++++------ docs/agent-source-package-harness.md | 17 +++++++++------- docs/storage-architecture.md | 15 ++++++++------ tests/test_chronicle_mirror.py | 30 +++++++++++++++++++++++++++- 5 files changed, 64 insertions(+), 21 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 7a1023b5..30da7be4 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -249,9 +249,13 @@ Running the same operations against a checkout of the previous head: - Reproduced both parts of finding 10: the README did not state that an unqualified mirror load writes to `ledger`, and it named the absent `supabase/migrations/20260504_chronicle_bronze.sql` file. +- README now instructs operators to create and apply the deployment migration, + states the `ledger` runtime default, and gives both supported ways to target + `chronicle`. The storage architecture and source-package harness use the same + truthful procedure; no documentation names the absent SQL file. Both README + regression tests pass. ### Next -- Fix the README cutover instructions and commit their two regression tests. - Run the required lint, format, full-test, and tracked-USDA sweep verification, then write the final `out.md` report. diff --git a/README.md b/README.md index 7c526a97..56dc70e2 100644 --- a/README.md +++ b/README.md @@ -434,13 +434,14 @@ uv run chronicle publish-derived \ --build-artifacts-out /tmp/chronicle-build-artifacts.jsonl ``` -The Supabase schema for this mirror lives at -`supabase/migrations/20260504_chronicle_bronze.sql`. Raw government spreadsheets are -mirrored as artifact metadata plus one row per parsed cell, not one tidy table -per sheet. Chronicle does not host raw survey microdata tables. +Before loading, create and apply a Supabase/Postgres migration that creates the +mirror tables in the schema selected for the load, then expose that schema +through the Supabase Data API. Raw government spreadsheets are mirrored as +artifact metadata plus one row per parsed cell, not one tidy table per sheet. +Chronicle does not host raw survey microdata tables. -After the migration is applied and the `chronicle` schema is exposed through the -Supabase Data API, accepted mirror exports can be upserted with: +After that deployment migration is applied, accepted mirror exports can be +upserted with: ```bash uv run chronicle load-supabase-mirror \ @@ -448,6 +449,10 @@ uv run chronicle load-supabase-mirror \ --build-artifacts /tmp/chronicle-build-artifacts.jsonl ``` +With neither `CHRONICLE_SCHEMA` nor `--schema`, this command writes to `ledger`. +To load a migrated `chronicle` schema instead, set `CHRONICLE_SCHEMA=chronicle` +or pass `--schema chronicle`. + Use `--dry-run` first to validate JSONL row counts and file coverage without writing to Supabase. diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 48f09b76..3a96a314 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -684,11 +684,11 @@ uv run chronicle publish-derived \ --build-artifacts-out /tmp/chronicle-build-artifacts.jsonl ``` -The SQL schema is checked in at -`supabase/migrations/20260504_chronicle_bronze.sql`. Spreadsheet publications are -stored as immutable artifact metadata and one parsed-cell row per workbook cell. -Agents should not try to normalize irregular government worksheets into tidy -sheet tables before selector specs interpret them. +Before loading, create and apply a Supabase/Postgres migration that creates the +mirror tables in the selected schema. Spreadsheet publications are stored as +immutable artifact metadata and one parsed-cell row per workbook cell. Agents +should not try to normalize irregular government worksheets into tidy sheet +tables before selector specs interpret them. After the DB export and derived publish, agents can validate and load the hosted mirror: @@ -704,8 +704,11 @@ uv run chronicle load-supabase-mirror \ ``` The live load requires `POLICYENGINE_SUPABASE_URL` and -`POLICYENGINE_SUPABASE_SERVICE_KEY`, the Chronicle mirror migration applied, and the -`chronicle` schema exposed by the Supabase Data API. +`POLICYENGINE_SUPABASE_SERVICE_KEY`, the deployment migration applied, and the +selected schema exposed by the Supabase Data API. With neither +`CHRONICLE_SCHEMA` nor `--schema`, the selected schema is `ledger`; set +`CHRONICLE_SCHEMA=chronicle` or pass `--schema chronicle` to load a migrated +`chronicle` schema. ## Declarative Authoring Contract diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index b91d06fa..f6b53fb8 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -203,9 +203,10 @@ The registry should expose: authority, legal vintage, and evidence; - build metadata, validation status, and derived artifact R2 bucket/key/URI. -The current Supabase migration mirrors the core relational tables and includes -R2 location fields for raw source artifacts and derived build artifacts, so the -registry can serve as the shared index over both R2 buckets. +A deployment migration for the selected Supabase schema must mirror the core +relational tables and include R2 location fields for raw source artifacts and +derived build artifacts, so the registry can serve as the shared index over +both R2 buckets. ## Build And Publish Flow @@ -249,9 +250,11 @@ The intended flow is: --build-artifacts /tmp/chronicle-build-artifacts.jsonl ``` -The Supabase project must have the checked migration applied and the `chronicle` -schema exposed in PostgREST/Data API settings before the REST loader can write -to it. Use `--dry-run` to verify local JSONL files without writing. +The Supabase project must have a deployment migration for the selected schema +applied and that schema exposed in PostgREST/Data API settings before the REST +loader can write to it. The load defaults to `ledger`; set +`CHRONICLE_SCHEMA=chronicle` or pass `--schema chronicle` to target a migrated +`chronicle` schema. Use `--dry-run` to verify local JSONL files without writing. ## Environment Variable Rename Window diff --git a/tests/test_chronicle_mirror.py b/tests/test_chronicle_mirror.py index 6b2e694d..5b0b31a0 100644 --- a/tests/test_chronicle_mirror.py +++ b/tests/test_chronicle_mirror.py @@ -3,10 +3,12 @@ from __future__ import annotations import json +from pathlib import Path +import re import pytest -from chronicle.env import ChronicleEnvDeprecationWarning +from chronicle.env import ChronicleEnvDeprecationWarning, DEFAULT_CHRONICLE_SCHEMA from chronicle.harness import main as harness_main from chronicle.mirror import ( LEDGER_MIRROR_TABLES, @@ -313,6 +315,32 @@ def test_load_supabase_mirror_cli_writes_to_the_configured_schema( ] +def _readme_supabase_cutover_section(): + readme = (Path(__file__).parents[1] / "README.md").read_text() + return readme[ + readme.index("To prepare the deterministic SQLite artifact") : readme.index( + "Chronicle settings are read" + ) + ] + + +def test_readme_supabase_cutover_documents_the_runtime_schema_default(): + section = _readme_supabase_cutover_section() + + assert f"writes to `{DEFAULT_CHRONICLE_SCHEMA}`" in section + assert "`CHRONICLE_SCHEMA=chronicle`" in section + assert "`--schema chronicle`" in section + + +def test_readme_supabase_cutover_only_names_checked_in_migrations(): + section = _readme_supabase_cutover_section() + repository = Path(__file__).parents[1] + migration_paths = re.findall(r"`([^`\n]+[.]sql)`", section) + missing = [path for path in migration_paths if not (repository / path).is_file()] + + assert missing == [] + + class _FakeSupabaseClient: def __init__(self): self.upserts = [] From 2881a8c81f1f1d63e8e1b5f3c6799591db5222a0 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:36:12 -0400 Subject: [PATCH 032/105] Complete strict artifact preflight coverage --- PROGRESS.md | 24 ++++- README.md | 6 +- chronicle/artifacts.py | 144 +++++++++++++++++++-------- docs/agent-source-package-harness.md | 4 +- docs/storage-architecture.md | 9 +- tests/test_chronicle_artifacts.py | 126 +++++++++++++++++++++-- tests/test_chronicle_mirror.py | 2 + 7 files changed, 255 insertions(+), 60 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 30da7be4..19135e4c 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -254,8 +254,28 @@ Running the same operations against a checkout of the previous head: `chronicle`. The storage architecture and source-package harness use the same truthful procedure; no documentation names the absent SQL file. Both README regression tests pass. +- Adversarial review tightened the same contracts before final verification. + Mixed-case package manifests were reproduced as omissions from both root + sweeps and from the stray-default guard; default discovery now filters every + recursive filename through the case-insensitive #227 helper. Missing + `provider` and missing `uri` locators were each reproduced reaching publisher + I/O; `storage.r2` now requires explicit `provider: r2` and an `r2://` URI. + The artifact file's 112 tests pass. +- Clarified that *all* schema environment overrides, including deprecated + spellings, precede the `ledger` default. The README test now also requires + the create/apply-migration instruction, so deleting the guidance cannot pass + vacuously. All 14 mirror tests pass. +- Completed the remaining non-microdata preflight port from #227. Reproduced + four invalid explicit/inferred artifact filenames reaching `_read_artifact`, + including `filename=manifest.yaml`, which could overwrite the selected + manifest, and reproduced an undiscoverable `custom.yaml` manifest reaching + publisher I/O. Ported `is_bare_filename`, `bare_filename`, + `_infer_artifact_filename`, the manifest-like artifact refusal from + `daafac0`, and c0d9d74's discoverable `_manifest_path` restriction. All 117 + artifact tests pass after the pre-I/O fix. ### Next -- Run the required lint, format, full-test, and tracked-USDA sweep verification, - then write the final `out.md` report. +- Commit the adversarial-review corrections, run the required lint, format, + full-test, and tracked-USDA sweep verification, then write the final `out.md` + report. diff --git a/README.md b/README.md index 56dc70e2..3574847a 100644 --- a/README.md +++ b/README.md @@ -449,9 +449,9 @@ uv run chronicle load-supabase-mirror \ --build-artifacts /tmp/chronicle-build-artifacts.jsonl ``` -With neither `CHRONICLE_SCHEMA` nor `--schema`, this command writes to `ledger`. -To load a migrated `chronicle` schema instead, set `CHRONICLE_SCHEMA=chronicle` -or pass `--schema chronicle`. +With no schema environment override and no `--schema`, this command +writes to `ledger`. To load a migrated `chronicle` schema instead, set +`CHRONICLE_SCHEMA=chronicle` or pass `--schema chronicle`. Use `--dry-run` first to validate JSONL row counts and file coverage without writing to Supabase. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 5a2a341e..30cb34ef 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -56,11 +56,37 @@ ) +def is_bare_filename(value: Any) -> bool: + """Whether ``value`` names a file inside a directory, with no path.""" + if value is None: + return False + text = str(value).strip() + if not text or text != str(value) or text in (".", ".."): + return False + if "/" in text or "\\" in text or "\x00" in text: + return False + return Path(text).name == text + + +def bare_filename(value: Any, *, what: str = "filename") -> str: + """Return ``value`` as a bare filename, refusing any other spelling. + + ``./table.csv``, ``nested/table.csv`` and an absolute path all resolve + outside the one-name manifest contract once joined under a package, so + fetch validates the spelling before reading publisher bytes. + """ + if not is_bare_filename(value): + raise ArtifactFilenameError( + f"{what} must be a bare filename inside the package directory, not " + f"{value!r}; it may not carry a directory, '.', '..', a trailing " + "slash, surrounding whitespace, or an absolute path." + ) + return str(value) + + def is_manifest_filename(value: Any) -> bool: """Whether ``value`` is a package-manifest filename.""" - if not isinstance(value, str) or not value or value != value.strip(): - return False - return value == Path(value).name and bool(_MANIFEST_FILENAME_RE.fullmatch(value)) + return is_bare_filename(value) and bool(_MANIFEST_FILENAME_RE.fullmatch(str(value))) def package_manifest_paths(package_dir: Path) -> list[Path]: @@ -84,21 +110,10 @@ def _root_manifest_paths(root: Path, manifest_filename: str) -> list[Path]: """ if manifest_filename != DEFAULT_MANIFEST_FILENAME: return sorted(path for path in root.rglob(manifest_filename) if path.is_file()) - package_dirs = { - path.parent - for pattern in ( - "manifest.yaml", - "manifest.yml", - "manifest_*.yaml", - "manifest_*.yml", - ) - for path in root.rglob(pattern) - if path.is_file() - } return sorted( - manifest_path - for package_dir in package_dirs - for manifest_path in package_manifest_paths(package_dir) + path + for path in root.rglob("*") + if path.is_file() and is_manifest_filename(path.name) ) @@ -114,6 +129,13 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: "Manifest must name a file inside the package directory, not " f"{manifest_filename!r}." ) + if not is_manifest_filename(name): + raise ManifestNameError( + f"Manifest must be named {DEFAULT_MANIFEST_FILENAME} or " + f"manifest_.yaml, not {manifest_filename!r}: the sweeps " + "address a package's manifests by those names, and a manifest " + "under any other name is invisible to them." + ) return output / name @@ -122,7 +144,7 @@ def _sibling_manifests(output: Path) -> list[str]: return sorted( path.name for path in package_manifest_paths(output) - if path.stem.startswith("manifest_") + if path.stem.lower().startswith("manifest_") ) @@ -234,6 +256,10 @@ class ManifestNameError(SourceArtifactManifestError, ValueError): """ +class ArtifactFilenameError(SourceArtifactManifestError, ValueError): + """An artifact filename is not a bare, non-manifest package filename.""" + + class AmbiguousManifestError(SourceArtifactManifestError): """The default manifest name would create a manifest beside the ones a package already keeps (PolicyEngine/chronicle#225).""" @@ -250,12 +276,13 @@ class MalformedManifestError(SourceArtifactManifestError): class RecordedR2LocatorError(SourceArtifactManifestError): """A recorded ``storage.r2`` block does not locate exactly one object. - ``provider``, ``bucket``, ``key`` and ``uri`` all describe the same object, - so any that are supplied have to agree, and the key has to carry the - ``{sha256}/{filename}`` tail that says which bytes it holds. A block whose - fields contradict each other has no single answer to "which bytes does this - entry claim R2 holds", and preserving or publishing under it would ship - whichever field the reader happened to consult. + The provider and URI must explicitly identify R2. ``provider``, ``bucket``, + ``key`` and ``uri`` all describe the same object, so any additional fields + have to agree, and the key has to carry the ``{sha256}/{filename}`` tail + that says which bytes it holds. A block whose fields contradict each other + has no single answer to "which bytes does this entry claim R2 holds", and + preserving or publishing under it would ship whichever field the reader + happened to consult. """ @@ -713,6 +740,19 @@ def fetch_source_artifact( r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) manifest_path = _manifest_path(output, manifest_filename) + what = ( + "--filename" if filename is not None else "The filename inferred from the URL" + ) + artifact_filename = bare_filename( + filename if filename is not None else _infer_artifact_filename(source_url), + what=what, + ) + if is_manifest_filename(artifact_filename): + raise ArtifactFilenameError( + f"{what} {artifact_filename!r} is a manifest name. An artifact may " + "not be named like a manifest, which it would overwrite; pass " + "--filename with the publisher's name for the bytes." + ) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, @@ -743,10 +783,7 @@ def fetch_source_artifact( ) fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() - content, inferred_filename = _read_artifact(source_url) - artifact_filename = filename or inferred_filename - if not artifact_filename: - raise ValueError("Could not infer artifact filename; pass --filename.") + content, _inferred_filename = _read_artifact(source_url) sha256 = hashlib.sha256(content).hexdigest() size_bytes = len(content) @@ -1392,6 +1429,23 @@ def _filename_from_url(source_url: str) -> str: return Path(unquote(parsed.path)).name +def _infer_artifact_filename(source_url: str) -> str: + """Return the filename :func:`_read_artifact` would report, without I/O. + + The name is a pure function of the URL: the last path segment for http(s) + and ``file://`` URLs, the basename for a bare path. Resolving it before + the read lets every filename guard run before the publisher is touched. + """ + parsed = urlparse(source_url) + if parsed.scheme in ("http", "https"): + return _filename_from_url(source_url) + if parsed.scheme == "file": + return Path(unquote(parsed.path)).name + if not parsed.scheme: + return Path(source_url).name + raise ValueError(f"Unsupported source URL scheme: {parsed.scheme}") + + def _read_manifest(manifest_path: Path) -> dict[str, Any]: """Return a manifest's parsed payload, refusing a document it cannot read. @@ -1530,12 +1584,13 @@ def _validated_recorded_r2( ) -> RecordedR2Object | None: """Return the object a recorded ``storage.r2`` block names, or None. - Every locator field the block supplies is cross-checked against every - other: ``key`` against the URI's path, ``bucket`` against its authority, - ``provider`` against its scheme, and the resulting key against the - canonical content-addressed shape :func:`build_r2_key` writes. Reading one - field and trusting the rest is what lets a block that says two different - things survive a preserve or a publish. + The block must explicitly record provider ``r2`` and an ``r2://`` URI. + Every additional locator field it supplies is cross-checked against that + URI: ``key`` against its path, ``bucket`` against its authority, and the + resulting key against the canonical content-addressed shape + :func:`build_r2_key` writes. Reading one field and trusting the rest is + what lets a block that says two different things survive a preserve or a + publish. """ storage = _validated_recorded_storage(spec, manifest_path=manifest_path, year=year) if "r2" not in storage: @@ -1562,6 +1617,20 @@ def _validated_recorded_r2( bucket = supplied.get("bucket") key = supplied.get("key") uri = supplied.get("uri") + missing_required = [ + field for field in ("provider", "uri") if not supplied.get(field) + ] + if missing_required: + raise RecordedR2LocatorError( + f"{where}: records no {', '.join(missing_required)}. A block under " + "storage.r2 must explicitly record provider='r2' and an r2:// URI." + ) + if provider != "r2": + raise RecordedR2LocatorError( + f"{where}: provider={provider!r} does not identify R2. A block " + "under storage.r2 must use provider='r2' and an r2:// URI, not " + f"{provider}://." + ) if uri is not None: parts = _split_r2_uri(uri) if parts is None: @@ -1598,13 +1667,6 @@ def _validated_recorded_r2( "locate its object: provider, bucket and key, or a uri that " "supplies them." ) - if provider != "r2": - raise RecordedR2LocatorError( - f"{where}: provider={provider!r} does not identify R2. A block " - "under storage.r2 must use provider='r2' and an r2:// URI, not " - f"{provider}://." - ) - segments = key.split("/") if ( len(segments) < 2 diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 3a96a314..a8da1801 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -705,8 +705,8 @@ uv run chronicle load-supabase-mirror \ The live load requires `POLICYENGINE_SUPABASE_URL` and `POLICYENGINE_SUPABASE_SERVICE_KEY`, the deployment migration applied, and the -selected schema exposed by the Supabase Data API. With neither -`CHRONICLE_SCHEMA` nor `--schema`, the selected schema is `ledger`; set +selected schema exposed by the Supabase Data API. With no schema environment +override and no `--schema`, the selected schema is `ledger`; set `CHRONICLE_SCHEMA=chronicle` or pass `--schema chronicle` to load a migrated `chronicle` schema. diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index f6b53fb8..4e1f47ee 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -172,10 +172,11 @@ must be a filename inside `--out-dir`, not a path. ### What a recorded block has to say -A `storage.r2` block's `provider`, `bucket`, `key` and `uri` all describe one -object, so every field that is present is cross-checked against every other: -the key against the URI's path, the bucket against its authority, the provider -against its scheme, and the resulting key against the content-addressed +A `storage.r2` block must explicitly say `provider: r2` and carry an `r2://` +URI. Its `provider`, `bucket`, `key` and `uri` all describe one object, so every +additional field that is present is cross-checked against the URI: the key +against its path, the bucket against its authority, the provider against its +scheme, and the resulting key against the content-addressed `{sha256}/{filename}` shape. A block whose fields disagree does not answer "which bytes does this entry claim R2 holds", so it is an error rather than something to preserve or publish under. Likewise a manifest that parses as diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index b1738035..f6cb8410 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -13,6 +13,7 @@ from chronicle.artifacts import ( AmbiguousManifestError, MalformedManifestError, + ManifestNameError, RecordedR2LocatorError, SourceArtifactManifestError, SourceArtifactRevisionError, @@ -350,6 +351,9 @@ def test_publish_source_artifacts_refuses_stale_country_key(tmp_path): ), } } + artifact["storage"]["r2"]["uri"] = ( + f"r2://ledger-raw/{artifact['storage']['r2']['key']}" + ) manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) log = tmp_path / "wrangler.log" wrangler = tmp_path / "wrangler" @@ -1351,6 +1355,26 @@ def test_fetch_artifact_cli_refuses_a_stray_default_manifest(tmp_path, capsys): assert not (package / "manifest.yaml").exists() +def test_fetch_refuses_a_stray_default_beside_a_case_variant_named_manifest( + tmp_path, monkeypatch +): + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + package.mkdir(parents=True) + named_manifest = package / "MANIFEST_TRADITIONAL.YML" + named_manifest.write_text("source_id: irs_soi\nfiles: {}\n") + source = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + + def unexpected_read(_source_url): + raise AssertionError("a case-variant named manifest did not block I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(AmbiguousManifestError, match="MANIFEST_TRADITIONAL.YML"): + _fetch_local(package, source) + + assert not (package / "manifest.yaml").exists() + + def test_a_same_bytes_rename_is_refused_by_name_not_as_a_revision(tmp_path): """Identical bytes under another filename are neither a revision nor a re-fetch: the entry's filename must keep agreeing with its recorded key.""" @@ -1389,6 +1413,66 @@ def test_a_manifest_name_must_stay_inside_the_package(tmp_path, manifest_filenam assert not package.exists() +@pytest.mark.parametrize( + ("source_url", "filename", "message"), + [ + pytest.param("publisher.csv", "manifest.yaml", "manifest name", id="default"), + pytest.param( + "publisher.csv", "MANIFEST_NAMED.YML", "manifest name", id="named" + ), + pytest.param( + "publisher.csv", "nested/publisher.csv", "bare filename", id="nested" + ), + pytest.param( + "https://publisher.test/manifest.yaml", + None, + "manifest name", + id="inferred", + ), + ], +) +def test_artifact_filename_is_refused_before_publisher_io( + tmp_path, monkeypatch, source_url, filename, message +): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text("source_id: irs_soi\npackage_id: soi-table\nfiles: {}\n") + before = manifest_path.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("an invalid artifact filename reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(SourceArtifactManifestError, match=message): + fetch_source_artifact( + source_url, + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + filename=filename, + ) + + assert manifest_path.read_bytes() == before + + +def test_manifest_name_must_be_discoverable_before_publisher_io(tmp_path, monkeypatch): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table" + source = _publish(tmp_path, "table.csv", b"publisher table") + + def unexpected_read(_source_url): + raise AssertionError("an undiscoverable manifest name reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(ManifestNameError, match="invisible"): + _fetch_local(package, source, manifest_filename="custom.yaml") + + assert not package.exists() + + def test_fetch_artifact_cli_reports_a_manifest_name_outside_the_package( tmp_path, capsys ): @@ -1587,9 +1671,11 @@ def _flatten_the_key(storage): [ pytest.param(_contradict_key, "contradicts uri", id="key-vs-uri"), pytest.param(_contradict_bucket, "contradicts uri", id="bucket-vs-uri"), - pytest.param(_contradict_provider, "contradicts uri", id="provider-vs-uri"), + pytest.param( + _contradict_provider, "does not identify R2", id="provider-vs-uri" + ), pytest.param(_mangle_uri, "is not provider://bucket/key", id="uri-shape"), - pytest.param(_drop_the_locator, "records no key", id="no-locator"), + pytest.param(_drop_the_locator, "records no uri", id="no-locator"), pytest.param( _flatten_the_key, "is not content-addressed", id="not-content-addressed" ), @@ -2001,6 +2087,7 @@ def _write_sweep_manifests(root): "manifest.yml", "manifest_named.yaml", "manifest_named.yml", + "Manifest_Mixed.YAML", ) for index, manifest_name in enumerate(manifest_names): package = root / f"package-{index}" @@ -2036,13 +2123,14 @@ def test_inventory_default_sweep_discovers_every_package_manifest(tmp_path): report = inventory_source_artifacts(root) assert report.valid - assert report.counts["manifest_count"] == 4 - assert report.counts["artifact_count"] == 4 + assert report.counts["manifest_count"] == 5 + assert report.counts["artifact_count"] == 5 assert {entry.manifest_path.rsplit("/", 1)[-1] for entry in report.entries} == { "manifest.yaml", "manifest.yml", "manifest_named.yaml", "manifest_named.yml", + "Manifest_Mixed.YAML", } @@ -2055,10 +2143,10 @@ def test_publish_default_sweep_discovers_every_package_manifest(tmp_path): report = publish_source_artifacts(root, wrangler_command=str(wrangler)) assert report.valid - assert report.counts["manifest_count"] == 4 - assert report.counts["artifact_count"] == 4 - assert report.counts["uploaded_count"] == 4 - assert len(log.read_text().splitlines()) == 4 + assert report.counts["manifest_count"] == 5 + assert report.counts["artifact_count"] == 5 + assert report.counts["uploaded_count"] == 5 + assert len(log.read_text().splitlines()) == 5 @pytest.mark.parametrize( @@ -2198,6 +2286,28 @@ def unexpected_read(_source_url): assert manifest_path.read_bytes() == before +@pytest.mark.parametrize("missing_field", ["provider", "uri"]) +def test_fetch_refuses_an_incomplete_r2_locator_before_io( + tmp_path, monkeypatch, missing_field +): + package, source, _report = _recorded_package(tmp_path) + manifest_path = package / "manifest.yaml" + manifest = yaml.safe_load(manifest_path.read_text()) + manifest["files"][2022]["storage"]["r2"].pop(missing_field) + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("an incomplete storage.r2 locator reached I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(RecordedR2LocatorError, match=missing_field): + _fetch_local(package, source, upload_r2=False) + + assert manifest_path.read_bytes() == before + + def test_publish_refuses_a_self_consistent_non_r2_locator(tmp_path, monkeypatch): package, _source, _report = _recorded_package(tmp_path) manifest_path = _make_recorded_locator_use_s3(package) diff --git a/tests/test_chronicle_mirror.py b/tests/test_chronicle_mirror.py index 5b0b31a0..b9026cae 100644 --- a/tests/test_chronicle_mirror.py +++ b/tests/test_chronicle_mirror.py @@ -328,6 +328,7 @@ def test_readme_supabase_cutover_documents_the_runtime_schema_default(): section = _readme_supabase_cutover_section() assert f"writes to `{DEFAULT_CHRONICLE_SCHEMA}`" in section + assert "With no schema environment override and no `--schema`" in section assert "`CHRONICLE_SCHEMA=chronicle`" in section assert "`--schema chronicle`" in section @@ -338,6 +339,7 @@ def test_readme_supabase_cutover_only_names_checked_in_migrations(): migration_paths = re.findall(r"`([^`\n]+[.]sql)`", section) missing = [path for path in migration_paths if not (repository / path).is_file()] + assert "create and apply a Supabase/Postgres migration" in section assert missing == [] From 0c042601ecc89d76a9b1dd16b44c91186c31595d Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 09:38:21 -0400 Subject: [PATCH 033/105] Record Sol gate final verification --- PROGRESS.md | 27 +++++++++++++++++++++------ 1 file changed, 21 insertions(+), 6 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 19135e4c..19d36e99 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -270,12 +270,27 @@ Running the same operations against a checkout of the previous head: including `filename=manifest.yaml`, which could overwrite the selected manifest, and reproduced an undiscoverable `custom.yaml` manifest reaching publisher I/O. Ported `is_bare_filename`, `bare_filename`, - `_infer_artifact_filename`, the manifest-like artifact refusal from - `daafac0`, and c0d9d74's discoverable `_manifest_path` restriction. All 117 - artifact tests pass after the pre-I/O fix. + `_infer_artifact_filename` from `daafac0`, and the manifest-like artifact + refusal plus discoverable `_manifest_path` restriction from `c0d9d74`. All + 117 artifact tests pass after the pre-I/O fix. ### Next -- Commit the adversarial-review corrections, run the required lint, format, - full-test, and tracked-USDA sweep verification, then write the final `out.md` - report. +- None in this lane. All ten Sol findings, the non-microdata #227 port + completeness pass, and adversarial follow-ups are committed and verified. + The final evidence and per-finding commit map are in the runner's external + `-o out.md` report, not a repository-root file. + +### Final verification + +- `uv run ruff check .`: exit 0 (`All checks passed!`). +- `uv run ruff format --check` on all six changed Python files: exit 0 + (`6 files already formatted`). +- Full `uv run pytest -q -p no:cacheprovider`: direct exit 0, 995 passed, + 1 skipped, 18 warnings in 1539.05 seconds. +- A fresh `/tmp` copy of tracked USDA `fy69_to_current` reports two manifests + and two artifacts in inventory. Publish includes both: FY2024 is one safe + preserved-bucket skip and the known misrouted FY2025 package/year key is one + explicit failure, with zero uploads. No tracked manifest was changed. +- The external `-o out.md` report records every failing-first command and + observation, regression test, fix commit, and exact #227 port provenance. From 8688cbde47f36120dbaccf9bd8103d238846a716 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:05:28 -0400 Subject: [PATCH 034/105] Start eight-finding Sol gate progress log --- PROGRESS.md | 42 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 42 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 19d36e99..8326493e 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -294,3 +294,45 @@ Running the same operations against a checkout of the previous head: explicit failure, with zero uploads. No tracked manifest was changed. - The external `-o out.md` report records every failing-first command and observation, regression test, fix commit, and exact #227 port provenance. + +## Review fixes (Sol gate round 2, eight findings) + +### State + +- Detached HEAD began at `c36f3fc8`, the supplied head of PR #226; base is + `main` at `9da02431`. The worktree was clean at intake. +- Scope is the eight supplied findings: cutover compatibility, shared-file + revision ownership, artifact/manifest path safety, duplicate YAML keys, + malformed revision history, and import-time Supabase alias compatibility. +- No tracked `db/data/**` manifest will be modified. The whole-tree cutover + regression will operate on a temporary copy and use a non-writing uploader. +- PR #227 is available read-only at + `/Users/maxghenis/PolicyEngine/_worktrees/chronicle-227-fix` at `5557cb9e`. + Applicable non-microdata hunks will be ported with the same function names + and shapes so its later rebase stays straightforward. + +### Done + +- Read the existing `PROGRESS.md`, the Bucket Cutover and Publisher Revisions + sections of `docs/storage-architecture.md`, and all six requested code/test + files: `chronicle/artifacts.py`, `chronicle/harness.py`, + `chronicle/source_package.py`, `db/supabase_client.py`, + `tests/test_chronicle_artifacts.py`, and `tests/test_chronicle_env.py`. +- Read the GitNexus debugging/refactoring workflow guidance. No GitNexus MCP + graph tools are exposed in this session, so dependency tracing will use + repository search, focused tests, and the stacked PR's committed diffs. +- Confirmed the baseline gaps in the named code: `_read_manifest` still uses + `yaml.safe_load`; `_root_manifest_paths` passes a non-default value to + `rglob`; manifest-declared filenames reach direct path joins and reads; + fetch updates only its selected manifest; `_superseding_storage` converts a + non-list `previous_r2` to an empty history; and the Supabase compatibility + constants are hard-coded defaults. + +### Next + +1. Audit and port the exact non-microdata #227 hunks, preserving names/shapes. +2. Add focused failing regressions before each fix and capture every red + command/observation for the external `-o out.md` report. +3. Commit each coherent red-test and implementation step, updating this log. +4. Run focused tests, lint/format checks on changed files, and the full suite + with the directly captured exit code and counts. From 7cabf11527d8e9b3d0dd114aee6473784de82bb9 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:07:24 -0400 Subject: [PATCH 035/105] Reproduce duplicate manifest key loss --- tests/test_chronicle_artifacts.py | 66 +++++++++++++++++++++++++++++++ 1 file changed, 66 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index f6cb8410..a599dfed 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1831,6 +1831,72 @@ def test_a_malformed_manifest_is_reported_by_inventory_and_publish(tmp_path): assert "must be a YAML mapping" in published.errors[0] +@pytest.mark.parametrize( + "duplicate_document", + [ + pytest.param( + "source_id: hidden_source\n" + "source_id: irs_soi\n" + "package_id: soi-table-5\n" + "files: {}\n", + id="source-id", + ), + pytest.param( + "source_id: irs_soi\n" + "package_id: hidden-package\n" + "package_id: soi-table-5\n" + "files: {}\n", + id="package-id", + ), + pytest.param( + "source_id: irs_soi\n" + "package_id: soi-table-5\n" + "files:\n" + " 2022:\n" + " filename: hidden.xlsx\n" + f" sha256: {hashlib.sha256(b'hidden bytes').hexdigest()}\n" + "files: {}\n", + id="files", + ), + pytest.param( + "source_id: irs_soi\n" + "package_id: soi-table-5\n" + "files:\n" + " 2022:\n" + " filename: hidden.xlsx\n" + f" sha256: {hashlib.sha256(b'hidden bytes').hexdigest()}\n" + " 2022: {}\n", + id="vintage", + ), + ], +) +def test_fetch_refuses_duplicate_manifest_keys_before_publisher_io( + tmp_path, monkeypatch, duplicate_document +): + """A lossy YAML parse must never decide which identity gets rewritten.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text(duplicate_document) + before = manifest_path.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("duplicate manifest keys reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(MalformedManifestError, match="duplicate key"): + fetch_source_artifact( + "https://example.test/table.xlsx", + source_id="irs_soi", + package_id="soi-table-5", + year=2022, + output_dir=package, + ) + + assert manifest_path.read_bytes() == before + + # --------------------------------------------------------------------------- # Sol gate round 3: fetch preflight and in-place manifest updates # --------------------------------------------------------------------------- From a430d2d9708793fc918667c21cf28a2009248c51 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:09:04 -0400 Subject: [PATCH 036/105] Read artifact manifests with strict YAML loader --- chronicle/artifacts.py | 3 +- chronicle/registration.py | 67 +++++++++++++++++++++++++++++++++++++ chronicle/source_package.py | 3 +- 3 files changed, 71 insertions(+), 2 deletions(-) create mode 100644 chronicle/registration.py diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 30cb34ef..ae60614e 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -26,6 +26,7 @@ ) from chronicle.env import env_value from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain +from chronicle.registration import load_manifest_document R2_RAW_BUCKET_ENV = "CHRONICLE_R2_RAW_BUCKET" @@ -1458,7 +1459,7 @@ def _read_manifest(manifest_path: Path) -> dict[str, Any]: if not manifest_path.exists(): return {} try: - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) + payload = load_manifest_document(manifest_path.read_text(encoding="utf-8")) except yaml.YAMLError as exc: raise MalformedManifestError( f"{manifest_path} is not valid YAML: {exc}" diff --git a/chronicle/registration.py b/chronicle/registration.py new file mode 100644 index 00000000..b3452e5b --- /dev/null +++ b/chronicle/registration.py @@ -0,0 +1,67 @@ +"""Shared source-artifact registration primitives. + +This module holds manifest parsing and filename identity rules used at every +artifact boundary. PR #227 extends the same surface with access-specific +registration; keeping the common functions here lets that stacked work rebase +without inventing parallel helpers. +""" + +from __future__ import annotations + +from typing import Any + +import yaml + + +class StrictManifestLoader(yaml.SafeLoader): + """A YAML loader that refuses a mapping with duplicate keys. + + PyYAML keeps the last of two equal keys, so ``files:`` recorded twice, or + a vintage recorded as ``2023`` and again as ``2_023`` (the same integer), + would read as one entry and the shadowed entry would be dropped by the + next write. A manifest is the record the byte boundary is decided from, + so a document the loader cannot represent faithfully is malformed. + """ + + def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: + if not isinstance(node, yaml.MappingNode): + raise yaml.constructor.ConstructorError( + None, + None, + f"expected a mapping node, but found {node.id}", + node.start_mark, + ) + self.flatten_mapping(node) + mapping: dict[Any, Any] = {} + for key_node, value_node in node.value: + key = self.construct_object(key_node, deep=deep) + try: + hash(key) + except TypeError as exc: + raise yaml.constructor.ConstructorError( + "while constructing a mapping", + node.start_mark, + f"found unhashable key ({exc})", + key_node.start_mark, + ) from exc + if key in mapping: + raise yaml.constructor.ConstructorError( + "while constructing a mapping", + node.start_mark, + f"found duplicate key {key!r}", + key_node.start_mark, + ) + mapping[key] = self.construct_object(value_node, deep=deep) + return mapping + + +def load_manifest_document(text: str) -> Any: + """Parse a manifest document, refusing duplicate keys. + + Raises :class:`yaml.YAMLError` (a ``ConstructorError`` naming the + duplicate key) for a document YAML would otherwise silently collapse. + """ + return yaml.load(text, Loader=StrictManifestLoader) # noqa: S506 + + +__all__ = ["StrictManifestLoader", "load_manifest_document"] diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 26cc8639..60e4ff87 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -33,6 +33,7 @@ ) from chronicle.env import env_flag, env_value from chronicle.epoch import SCHEMA_IDS, schema_id +from chronicle.registration import load_manifest_document from chronicle.sources.cells import ( SourceArtifactMetadata, SourceCell, @@ -877,7 +878,7 @@ def _artifact_content( self.manifest, ) with manifest_path.open("r", encoding="utf-8") as file: - manifest = yaml.safe_load(file) + manifest = load_manifest_document(file.read()) spec = _year_mapping(manifest["files"], self.artifact_year or year) artifact_path = files(self.resource_package).joinpath( self.resource_directory, From a49e1f430e78547ee7dddc30ae1f2b58a030f45c Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:10:03 -0400 Subject: [PATCH 037/105] Record strict manifest loader progress --- PROGRESS.md | 17 +++++++++++++---- 1 file changed, 13 insertions(+), 4 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 8326493e..50814a20 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -327,12 +327,21 @@ Running the same operations against a checkout of the previous head: fetch updates only its selected manifest; `_superseding_storage` converts a non-list `previous_r2` to an empty history; and the Supabase compatibility constants are hard-coded defaults. +- **Finding 6 reproduced and fixed.** Red command: + `UV_CACHE_DIR=/tmp/chronicle-uv-cache uv run pytest -q -p + no:cacheprovider tests/test_chronicle_artifacts.py::test_fetch_refuses_duplicate_manifest_keys_before_publisher_io` + exited 1 with four failures; every duplicate (`source_id`, `package_id`, + `files`, vintage) reached the publisher-read sentinel. Test-only commit: + `56f9ce1`. Ported #227 commit `7f9bfe6`'s `StrictManifestLoader` and + `load_manifest_document` verbatim into `chronicle/registration.py`, switched + artifact manifest reads to it, and shared it with source-package artifact + manifest reads. Fix commit: `77e6fda`. The same focused command now exits 0 + with 4 passed. ### Next -1. Audit and port the exact non-microdata #227 hunks, preserving names/shapes. -2. Add focused failing regressions before each fix and capture every red +1. Add focused failing regressions before each remaining fix and capture every red command/observation for the external `-o out.md` report. -3. Commit each coherent red-test and implementation step, updating this log. -4. Run focused tests, lint/format checks on changed files, and the full suite +2. Commit each coherent red-test and implementation step, updating this log. +3. Run focused tests, lint/format checks on changed files, and the full suite with the directly captured exit code and counts. From 82d05b4239b79d91568dc753b5c3f861006c01e7 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:15:28 -0400 Subject: [PATCH 038/105] Reproduce artifact and manifest path escapes --- tests/test_chronicle_artifacts.py | 214 ++++++++++++++++++++++++++++++ 1 file changed, 214 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index a599dfed..2fbe51b8 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1534,6 +1534,129 @@ def test_fetch_artifact_cli_targets_the_named_manifest(tmp_path, capsys): assert TRADITIONAL_MANIFEST in capsys.readouterr().err +@pytest.mark.parametrize( + ("existing_name", "requested_name"), + [ + pytest.param("manifest.yml", "manifest.yaml", id="yml-default"), + pytest.param("Manifest.yaml", "manifest.yaml", id="case-variant-default"), + pytest.param( + "manifest_monthly_source_package.yaml", + "manifest_monthy_source_package.yaml", + id="mistyped-named-manifest", + ), + ], +) +def test_fetch_refuses_to_create_any_manifest_beside_an_existing_registry( + tmp_path, monkeypatch, existing_name, requested_name +): + package = tmp_path / "db" / "data" / "usda_snap" / "fy69_to_current" + package.mkdir(parents=True) + existing = package / existing_name + existing.write_text( + "source_id: usda_snap\npackage_id: usda-snap-fy69-to-current\nfiles: {}\n" + ) + before = existing.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("ambiguous manifest creation reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(AmbiguousManifestError, match=existing_name): + fetch_source_artifact( + "https://example.test/snap.zip", + source_id="usda_snap", + package_id="usda-snap-fy69-to-current", + year=2024, + output_dir=package, + manifest_filename=requested_name, + ) + + assert existing.read_bytes() == before + assert not (package / requested_name).exists() + + +def test_fetch_refuses_a_symlinked_manifest_before_publisher_io(tmp_path, monkeypatch): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table" + package.mkdir(parents=True) + outside_manifest = tmp_path / "outside-manifest.yaml" + outside_manifest.write_text( + "source_id: irs_soi\npackage_id: soi-table\nfiles: {}\n" + ) + manifest_path = package / "manifest.yaml" + manifest_path.symlink_to(outside_manifest) + before = outside_manifest.read_bytes() + + def unexpected_read(_source_url): + raise AssertionError("a symlinked manifest reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(MalformedManifestError, match="symlink"): + fetch_source_artifact( + "https://example.test/table.xlsx", + source_id="irs_soi", + package_id="soi-table", + year=2024, + output_dir=package, + ) + + assert manifest_path.is_symlink() + assert outside_manifest.read_bytes() == before + + +@pytest.mark.parametrize( + "manifest_filename", + [ + pytest.param("../manifest.yaml", id="parent"), + pytest.param("manifest_*.yaml", id="star-glob"), + pytest.param("manifest_?.yml", id="question-glob"), + pytest.param("manifest_[ab].yaml", id="character-class-glob"), + ], +) +def test_sweep_manifest_selector_must_be_a_literal_supported_filename( + tmp_path, manifest_filename +): + root = tmp_path / "requested-root" + package = root / "package" + package.mkdir(parents=True) + content = b"publisher table" + (package / "table.csv").write_bytes(content) + (package / "manifest_a.yaml").write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": "package", + "files": { + 2024: { + "filename": "table.csv", + "sha256": hashlib.sha256(content).hexdigest(), + } + }, + }, + sort_keys=False, + ) + ) + outside_manifest = root.parent / "manifest.yaml" + outside_manifest.write_text("files: {}\n") + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + before = { + path: path.read_bytes() + for path in (package / "manifest_a.yaml", outside_manifest) + } + + with pytest.raises(ManifestNameError, match="Manifest"): + publish_source_artifacts( + root, + manifest_filename=manifest_filename, + wrangler_command=str(wrangler), + ) + + assert {path: path.read_bytes() for path in before} == before + assert not log.exists() + + # --------------------------------------------------------------------------- # Identity without a recorded R2 object # --------------------------------------------------------------------------- @@ -2275,6 +2398,97 @@ def test_sweeps_treat_a_null_files_block_as_absent(tmp_path): assert published.valid +# --------------------------------------------------------------------------- +# Manifest-declared artifact paths +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize("path_kind", ["absolute", "parent"]) +def test_sweeps_refuse_non_bare_artifact_filenames_without_reading_them( + tmp_path, path_kind +): + package = tmp_path / "data" / "package" + package.mkdir(parents=True) + outside = tmp_path / "data" / "outside.csv" + outside.write_bytes(b"outside publisher bytes") + filename = str(outside) if path_kind == "absolute" else "../outside.csv" + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": "package", + "files": { + 2024: { + "filename": filename, + "sha256": hashlib.sha256(outside.read_bytes()).hexdigest(), + } + }, + }, + sort_keys=False, + ) + ) + before = manifest_path.read_bytes() + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package, wrangler_command=str(wrangler)) + expected = f"non_canonical_filename:{filename}" + + assert not inventory.valid + assert inventory.entries[0].errors == (expected,) + assert inventory.entries[0].local_path == str(package) + assert not published.valid + assert published.entries[0].errors == (expected,) + assert published.entries[0].upload is None + assert published.entries[0].local_path == str(package) + assert not log.exists() + assert manifest_path.read_bytes() == before + + +def test_sweeps_refuse_a_symlinked_artifact_without_reading_it(tmp_path): + package = tmp_path / "data" / "package" + package.mkdir(parents=True) + outside = tmp_path / "outside.csv" + outside.write_bytes(b"outside publisher bytes") + artifact_path = package / "table.csv" + artifact_path.symlink_to(outside) + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": "package", + "files": { + 2024: { + "filename": artifact_path.name, + "sha256": hashlib.sha256(outside.read_bytes()).hexdigest(), + } + }, + }, + sort_keys=False, + ) + ) + before = manifest_path.read_bytes() + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package, wrangler_command=str(wrangler)) + expected = "artifact_path_is_symlink:table.csv" + + assert not inventory.valid + assert inventory.entries[0].errors == (expected,) + assert not inventory.entries[0].exists + assert not published.valid + assert published.entries[0].errors == (expected,) + assert published.entries[0].upload is None + assert not log.exists() + assert manifest_path.read_bytes() == before + assert artifact_path.is_symlink() + + # --------------------------------------------------------------------------- # Sol gate round 3: canonical R2 locators before bucket-cutover skips # --------------------------------------------------------------------------- From f2432cdb0a404ee3be4b35fdd25e4db82b20119b Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:20:45 -0400 Subject: [PATCH 039/105] Harden artifact and manifest path boundaries --- chronicle/artifacts.py | 155 +++++++++++++++++------------- chronicle/registration.py | 94 +++++++++++++++++- tests/test_chronicle_artifacts.py | 13 ++- 3 files changed, 194 insertions(+), 68 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index ae60614e..eb514c07 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -26,7 +26,13 @@ ) from chronicle.env import env_value from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain -from chronicle.registration import load_manifest_document +from chronicle.registration import ( + is_bare_filename, + is_manifest_filename, + load_manifest_document, + matching_directory_entry, + package_manifest_paths, +) R2_RAW_BUCKET_ENV = "CHRONICLE_R2_RAW_BUCKET" @@ -49,25 +55,6 @@ # name is an input, not a constant, wherever a caller addresses a package. DEFAULT_MANIFEST_FILENAME = "manifest.yaml" -# The names a package directory's manifests may carry. Match case-insensitively -# because Chronicle is also used on case-insensitive filesystems. -_MANIFEST_FILENAME_RE = re.compile( - r"^manifest(?:_[^/\\]+)?\.ya?ml$", - re.IGNORECASE, -) - - -def is_bare_filename(value: Any) -> bool: - """Whether ``value`` names a file inside a directory, with no path.""" - if value is None: - return False - text = str(value).strip() - if not text or text != str(value) or text in (".", ".."): - return False - if "/" in text or "\\" in text or "\x00" in text: - return False - return Path(text).name == text - def bare_filename(value: Any, *, what: str = "filename") -> str: """Return ``value`` as a bare filename, refusing any other spelling. @@ -85,23 +72,6 @@ def bare_filename(value: Any, *, what: str = "filename") -> str: return str(value) -def is_manifest_filename(value: Any) -> bool: - """Whether ``value`` is a package-manifest filename.""" - return is_bare_filename(value) and bool(_MANIFEST_FILENAME_RE.fullmatch(str(value))) - - -def package_manifest_paths(package_dir: Path) -> list[Path]: - """Return every manifest file a package directory keeps, sorted by name.""" - directory = Path(package_dir) - if not directory.is_dir(): - return [] - return sorted( - path - for path in directory.iterdir() - if path.is_file() and is_manifest_filename(path.name) - ) - - def _root_manifest_paths(root: Path, manifest_filename: str) -> list[Path]: """Return the manifests a root sweep addresses. @@ -109,8 +79,13 @@ def _root_manifest_paths(root: Path, manifest_filename: str) -> list[Path]: extensions and every ``manifest_`` sibling participate. A caller that supplies another filename keeps the historical exact-name override. """ - if manifest_filename != DEFAULT_MANIFEST_FILENAME: - return sorted(path for path in root.rglob(manifest_filename) if path.is_file()) + selected_name = _manifest_path(Path(), manifest_filename).name + if selected_name != DEFAULT_MANIFEST_FILENAME: + return sorted( + path + for path in root.rglob("*") + if path.is_file() and path.name == selected_name + ) return sorted( path for path in root.rglob("*") @@ -125,7 +100,12 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: package directory keeps, and must not reach outside it. """ name = manifest_filename.strip() - if not name or name in (".", "..") or name != Path(name).name: + if ( + not name + or name in (".", "..") + or name != Path(name).name + or any(character in name for character in "*?[]") + ): raise ManifestNameError( "Manifest must name a file inside the package directory, not " f"{manifest_filename!r}." @@ -140,33 +120,29 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: return output / name -def _sibling_manifests(output: Path) -> list[str]: - """Return the ``manifest_*.yaml`` files a package directory keeps.""" - return sorted( - path.name - for path in package_manifest_paths(output) - if path.stem.lower().startswith("manifest_") - ) - - def _refuse_a_stray_default_manifest(output: Path, manifest_path: Path) -> None: - """Refuse to create ``manifest.yaml`` beside a package's named manifests. + """Refuse to create any new manifest beside a package's registry. - A publisher directory that feeds several source packages keeps one - ``manifest_.yaml`` per package and no ``manifest.yaml``. A fetch - that omits ``--manifest`` there would create a third manifest none of the - packages read, and would bypass the revision guard of the one it should - have addressed (PolicyEngine/chronicle#225). + Every supported spelling participates: a missing ``manifest.yaml`` beside + ``manifest.yml`` or ``Manifest.yaml`` is just as ambiguous as one beside a + named manifest, and a mistyped named selector must not create a parallel + registry. Operators may create an intentional empty sibling explicitly, + then select that existing file. """ - if manifest_path.name != DEFAULT_MANIFEST_FILENAME or manifest_path.exists(): + paths = package_manifest_paths(output) + if ( + any(path.name == manifest_path.name for path in paths) + or manifest_path.is_symlink() + ): return - siblings = _sibling_manifests(output) + siblings = [path.name for path in paths] if not siblings: return raise AmbiguousManifestError( - f"{output} keeps {', '.join(siblings)} and no {DEFAULT_MANIFEST_FILENAME}; " - "pass --manifest to name the manifest this fetch records into rather " - f"than creating {DEFAULT_MANIFEST_FILENAME} beside them." + f"{output} already keeps {', '.join(siblings)}; refusing to create " + f"{manifest_path.name} beside that registry. Pass --manifest to name " + "an existing manifest, or create an intentional empty sibling " + "explicitly before fetching into it." ) @@ -1456,6 +1432,11 @@ def _read_manifest(manifest_path: Path) -> dict[str, Any]: not an absent manifest, and treating it as one would let the fetch replace it with a single entry and drop everything it recorded. """ + if manifest_path.is_symlink(): + raise MalformedManifestError( + f"{manifest_path} is a symlink; manifest reads and writes require " + "a regular file at its lexical package path." + ) if not manifest_path.exists(): return {} try: @@ -2004,13 +1985,35 @@ def _publish_raw_manifest_entry( spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") - artifact_path = manifest_path.parent / filename + if filename and not is_bare_filename(filename): + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(manifest_path.parent), + sha256=None, + size_bytes=None, + r2_location=None, + upload=None, + errors=(f"non_canonical_filename:{filename}",), + ), + None, + ) + artifact_path = ( + matching_directory_entry(manifest_path.parent, filename) + or manifest_path.parent / filename + ) sha256_expected = spec.get("sha256") sha256_actual = None size_bytes = None if not filename: errors.append("missing_filename") - elif not artifact_path.exists(): + elif artifact_path.is_symlink(): + errors.append(f"artifact_path_is_symlink:{filename}") + elif not artifact_path.is_file(): errors.append("missing_file") else: content = artifact_path.read_bytes() @@ -2172,13 +2175,35 @@ def _inventory_entry( spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") - artifact_path = manifest_path.parent / filename - exists = bool(filename) and artifact_path.exists() + storage = spec.get("storage") if isinstance(spec, dict) else None + r2 = storage.get("r2") if isinstance(storage, dict) else None + if filename and not is_bare_filename(filename): + return ArtifactInventoryEntry( + manifest_path=str(manifest_path), + year=str(year), + filename=filename, + local_path=str(manifest_path.parent), + exists=False, + sha256_expected=spec.get("sha256"), + sha256_actual=None, + size_bytes=None, + source_url=spec.get("source_url"), + r2=r2, + errors=(f"non_canonical_filename:{filename}",), + ) + artifact_path = ( + matching_directory_entry(manifest_path.parent, filename) + or manifest_path.parent / filename + ) + symlink = bool(filename) and artifact_path.is_symlink() + exists = bool(filename) and not symlink and artifact_path.is_file() sha256_expected = spec.get("sha256") sha256_actual = None size_bytes = None if not filename: errors.append("missing_filename") + elif symlink: + errors.append(f"artifact_path_is_symlink:{filename}") elif not exists: errors.append("missing_file") else: @@ -2187,8 +2212,6 @@ def _inventory_entry( size_bytes = len(content) if sha256_expected and sha256_actual != sha256_expected: errors.append("checksum_mismatch") - storage = spec.get("storage") if isinstance(spec, dict) else None - r2 = storage.get("r2") if isinstance(storage, dict) else None return ArtifactInventoryEntry( manifest_path=str(manifest_path), year=str(year), diff --git a/chronicle/registration.py b/chronicle/registration.py index b3452e5b..2315024a 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -8,11 +8,93 @@ from __future__ import annotations +from pathlib import Path +import re from typing import Any +import unicodedata import yaml +class ArtifactFilenameError(ValueError): + """Raised when a filename is not a bare name inside a package directory.""" + + +def is_bare_filename(value: Any) -> bool: + """Whether ``value`` names a file inside a directory, with no path.""" + if value is None: + return False + text = str(value).strip() + if not text or text != str(value) or text in (".", ".."): + return False + if "/" in text or "\\" in text or "\x00" in text: + return False + return Path(text).name == text + + +def bare_filename(value: Any, *, what: str = "filename") -> str: + """Return ``value`` as a bare filename, refusing any other spelling. + + ``./adult.tab``, ``sub/../adult.tab``, ``adult.tab/`` and an absolute path + all resolve to the same file as ``adult.tab`` once joined under the package + directory, so the manifest and every guard use one spelling. + """ + if not is_bare_filename(value): + raise ArtifactFilenameError( + f"{what} must be a bare filename inside the package directory, not " + f"{value!r}; it may not carry a directory, '.', '..', a trailing " + "slash, surrounding whitespace, or an absolute path." + ) + return str(value) + + +def filename_key(value: Any) -> str: + """Return the case-folded, Unicode-normalized comparison key for a name.""" + return unicodedata.normalize("NFC", Path(str(value)).name).casefold() + + +_MANIFEST_FILENAME_RE = re.compile( + r"^manifest(?:_[^/\\]+)?\.ya?ml$", + re.IGNORECASE, +) + + +def is_manifest_filename(value: Any) -> bool: + """Whether ``value`` is a package-manifest filename.""" + return is_bare_filename(value) and bool(_MANIFEST_FILENAME_RE.fullmatch(str(value))) + + +def package_manifest_paths(package_dir: Path) -> list[Path]: + """Return every manifest file a package directory keeps, sorted by name.""" + directory = Path(package_dir) + if not directory.is_dir(): + return [] + return sorted( + path + for path in directory.iterdir() + if path.is_file() and is_manifest_filename(path.name) + ) + + +def matching_directory_entry(directory: Any, filename: Any) -> Any | None: + """Return the actual directory entry matching a bare filename's safe key. + + Scanning real entries makes the identity rule the same on case-sensitive + and case-folding filesystems, including Unicode-normalized aliases. + """ + if not is_bare_filename(filename) or not directory.is_dir(): + return None + wanted = filename_key(filename) + return next( + ( + path + for path in sorted(directory.iterdir(), key=lambda item: item.name) + if filename_key(path.name) == wanted + ), + None, + ) + + class StrictManifestLoader(yaml.SafeLoader): """A YAML loader that refuses a mapping with duplicate keys. @@ -64,4 +146,14 @@ def load_manifest_document(text: str) -> Any: return yaml.load(text, Loader=StrictManifestLoader) # noqa: S506 -__all__ = ["StrictManifestLoader", "load_manifest_document"] +__all__ = [ + "ArtifactFilenameError", + "StrictManifestLoader", + "bare_filename", + "filename_key", + "is_bare_filename", + "is_manifest_filename", + "load_manifest_document", + "matching_directory_entry", + "package_manifest_paths", +] diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 2fbe51b8..89b995be 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1254,6 +1254,17 @@ def test_fetch_artifact_writes_the_manifest_it_was_given(tmp_path): package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") roth = _publish(tmp_path, "22in06ira.xlsx", b"roth IRA table") + package.mkdir(parents=True) + for name, package_id in ( + (TRADITIONAL_MANIFEST, "soi-ira-traditional-contributions-2022"), + (ROTH_MANIFEST, "soi-ira-roth-contributions-2022"), + ): + (package / name).write_text( + yaml.safe_dump( + {"source_id": "irs_soi", "package_id": package_id, "files": {}}, + sort_keys=False, + ) + ) _fetch_local( package, @@ -1573,7 +1584,7 @@ def unexpected_read(_source_url): ) assert existing.read_bytes() == before - assert not (package / requested_name).exists() + assert requested_name not in {path.name for path in package.iterdir()} def test_fetch_refuses_a_symlinked_manifest_before_publisher_io(tmp_path, monkeypatch): From 403e500ef22935e986106cc2a813033934f078e3 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:21:14 -0400 Subject: [PATCH 040/105] Record path-boundary fixes --- PROGRESS.md | 24 ++++++++++++++++++++++-- 1 file changed, 22 insertions(+), 2 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 50814a20..fef202bc 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -337,11 +337,31 @@ Running the same operations against a checkout of the previous head: artifact manifest reads to it, and shared it with source-package artifact manifest reads. Fix commit: `77e6fda`. The same focused command now exits 0 with 4 passed. +- **Findings 3, 4, and 5 reproduced and fixed.** Test-only commit: `264e46e`. + The finding 4 command covering creation beside `manifest.yml`, + `Manifest.yaml`, a mistyped named manifest, and a symlinked manifest exited + 1 with 4 failures, all reaching the publisher-I/O sentinel. The finding 5 + command covering parent traversal plus `*`, `?`, and character-class glob + selectors exited 1 with 4 failures because none raised `ManifestNameError`. + The finding 3 command covering absolute, parent-traversing, and symlinked + artifact paths exited 1 with 3 failures because inventory considered each + path valid (and the local, non-network publisher stub could read it). +- Artifact and manifest inputs are now resolved only as literal package-local + directory entries: unsafe declared filenames return the named + `non_canonical_filename` error, symlinks are refused before reads, sweep + selectors cannot contain separators or glob syntax, and no new manifest + spelling can be created beside an existing registry. Fix commit: `0017520`. + Ported the #227-shaped `is_bare_filename`, `bare_filename`, `filename_key`, + `matching_directory_entry`, and package-manifest helpers from `44e1f8d`, + `7f9bfe6`, and `c2b7722`'s corresponding safety changes; the generalized + registry-creation and exact sweep-selector guards are slice-1 additions. + The focused post-fix command exits 0 with 13 passed, and the complete + artifact module exits 0 with 132 passed. ### Next -1. Add focused failing regressions before each remaining fix and capture every red - command/observation for the external `-o out.md` report. +1. Reproduce and fix cross-manifest shared-file revision ownership (finding 2), + including the tracked USDA two-manifest shape and every same-directory owner. 2. Commit each coherent red-test and implementation step, updating this log. 3. Run focused tests, lint/format checks on changed files, and the full suite with the directly captured exit code and counts. From b0d29b777abae26ab7bdf67e7e17f6d4a2241744 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:24:13 -0400 Subject: [PATCH 041/105] Reproduce stale shared-file revision owners --- tests/test_chronicle_artifacts.py | 197 ++++++++++++++++++++++++++++++ 1 file changed, 197 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 89b995be..4f6f3100 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1179,6 +1179,203 @@ def test_record_revision_without_an_upload_records_no_current_object( ] +def _shared_archive_entry(content, *, package_id, year, filename="shared.zip"): + sha256 = hashlib.sha256(content).hexdigest() + key = f"raw/usda_snap/{package_id}/{year}/{sha256}/{filename}" + return { + "filename": filename, + "source_url": "https://example.test/shared.zip", + "sha256": sha256, + "size_bytes": len(content), + "fetched_at": "2026-05-11T11:57:29+00:00", + "storage": { + "r2": { + "provider": "r2", + "bucket": "ledger-raw", + "key": key, + "uri": f"r2://ledger-raw/{key}", + } + }, + } + + +def test_shared_archive_revision_is_refused_through_an_unregistered_owner(tmp_path): + """A selected empty vintage cannot bypass another manifest's identity.""" + package = tmp_path / "db" / "data" / "usda_snap" / "fy69_to_current" + package.mkdir(parents=True) + original = b"USDA archive, first publication" + revised = b"USDA archive, revised publication" + filename = "snap-zip-fy69tocurrent-6.zip" + (package / filename).write_bytes(original) + primary_path = package / "manifest.yaml" + primary_path.write_text( + yaml.safe_dump( + { + "source_id": "usda_snap", + "package_id": "usda-snap-fy69-to-current", + "files": {}, + }, + sort_keys=False, + ) + ) + sibling_path = package / "manifest_fy2025_monthly_source_package.yaml" + sibling_path.write_text( + yaml.safe_dump( + { + "source_id": "usda_snap", + "package_id": "usda-snap-fy2025-monthly-state-caseloads", + "files": { + 2025: _shared_archive_entry( + original, + package_id="usda-snap-fy69-to-current", + year=2024, + filename=filename, + ) + }, + }, + sort_keys=False, + ) + ) + publisher = _publish(tmp_path, filename, revised) + before = {path: path.read_bytes() for path in (primary_path, sibling_path)} + + with pytest.raises(SourceArtifactRevisionError): + fetch_source_artifact( + str(publisher), + source_id="usda_snap", + package_id="usda-snap-fy69-to-current", + year=2024, + output_dir=package, + ) + + assert (package / filename).read_bytes() == original + assert {path: path.read_bytes() for path in before} == before + + +def test_record_revision_updates_every_owner_of_usda_shared_archive(tmp_path): + """The tracked USDA two-manifest shape has one physical archive.""" + package = tmp_path / "db" / "data" / "usda_snap" / "fy69_to_current" + package.mkdir(parents=True) + original = b"USDA archive, first publication" + revised = b"USDA archive, revised publication" + revised_sha256 = hashlib.sha256(revised).hexdigest() + filename = "snap-zip-fy69tocurrent-6.zip" + (package / filename).write_bytes(original) + manifests = ( + ( + package / "manifest.yaml", + "usda-snap-fy69-to-current", + 2024, + "usda-snap-fy69-to-current", + 2024, + ), + ( + package / "manifest_fy2025_monthly_source_package.yaml", + "usda-snap-fy2025-monthly-state-caseloads", + 2025, + "usda-snap-fy69-to-current", + 2024, + ), + ) + previous_uris = {} + for path, package_id, vintage, route_package, route_year in manifests: + entry = _shared_archive_entry( + original, + package_id=route_package, + year=route_year, + filename=filename, + ) + entry["source_table"] = f"owner {vintage}" + previous_uris[path] = entry["storage"]["r2"]["uri"] + path.write_text( + yaml.safe_dump( + { + "source_id": "usda_snap", + "package_id": package_id, + "files": {vintage: entry}, + }, + sort_keys=False, + ) + ) + publisher = _publish(tmp_path, filename, revised) + + fetch_source_artifact( + str(publisher), + source_id="usda_snap", + package_id="usda-snap-fy69-to-current", + year=2024, + output_dir=package, + record_revision=True, + ) + + assert (package / filename).read_bytes() == revised + for path, _package_id, vintage, _route_package, _route_year in manifests: + entry = yaml.safe_load(path.read_text())["files"][vintage] + assert entry["sha256"] == revised_sha256 + assert entry["size_bytes"] == len(revised) + assert entry["source_table"] == f"owner {vintage}" + assert "r2" not in entry["storage"] + assert [item["uri"] for item in entry["storage"]["previous_r2"]] == [ + previous_uris[path] + ] + + +def test_record_revision_updates_every_same_manifest_owner(tmp_path): + """SSA-style semantic aliases of one file share one byte identity.""" + package = tmp_path / "db" / "data" / "ssa" / "supplement" + package.mkdir(parents=True) + original = b"SSA extracted table, first publication" + revised = b"SSA extracted table, revised publication" + revised_sha256 = hashlib.sha256(revised).hexdigest() + filename = "ssa_oasdi_ssi_2024.csv" + (package / filename).write_bytes(original) + manifest_path = package / "manifest.yaml" + entries = { + 2024: _shared_archive_entry( + original, + package_id="ssa-annual-statistical-supplement-2025", + year=2024, + filename=filename, + ), + "extracted_targets": _shared_archive_entry( + original, + package_id="ssa-annual-statistical-supplement-2025", + year="extracted_targets", + filename=filename, + ), + } + for entry in entries.values(): + entry["source_url"] = "https://example.test/ssa.csv" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "ssa", + "package_id": "ssa-annual-statistical-supplement-2025", + "files": entries, + }, + sort_keys=False, + ) + ) + publisher = _publish(tmp_path, filename, revised) + + fetch_source_artifact( + str(publisher), + source_id="ssa", + package_id="ssa-annual-statistical-supplement-2025", + year=2024, + output_dir=package, + record_revision=True, + ) + + updated = yaml.safe_load(manifest_path.read_text())["files"] + assert {entry["sha256"] for entry in updated.values()} == {revised_sha256} + assert { + item["sha256"] + for entry in updated.values() + for item in entry["storage"]["previous_r2"] + } == {hashlib.sha256(original).hexdigest()} + + def test_a_recorded_block_that_only_carries_a_uri_is_still_recognized( tmp_path, monkeypatch ): From 215cc4f8afd0c3a9d4270f7f749397c1ebfa2af5 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:29:45 -0400 Subject: [PATCH 042/105] Update every owner of a revised shared artifact --- chronicle/artifacts.py | 307 ++++++++++++++++++++++++++++++----- docs/storage-architecture.md | 15 +- 2 files changed, 280 insertions(+), 42 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index eb514c07..a9f133cd 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -13,7 +13,7 @@ import shlex import sqlite3 import subprocess -from typing import Any +from typing import Any, Mapping from urllib.parse import unquote, urlparse import httpx @@ -27,6 +27,7 @@ from chronicle.env import env_value from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain from chronicle.registration import ( + filename_key, is_bare_filename, is_manifest_filename, load_manifest_document, @@ -146,6 +147,29 @@ def _refuse_a_stray_default_manifest(output: Path, manifest_path: Path) -> None: ) +def _package_manifests( + output: Path, + manifest_path: Path, + existing_manifest: dict[str, Any], +) -> dict[str, dict[str, Any]]: + """Return every manifest the package directory keeps, by path. + + The byte boundary is the file in the directory, not whichever manifest a + fetch selected. A malformed sibling is therefore a pre-I/O refusal: until + Chronicle can read every owner, it cannot safely overwrite shared bytes. + """ + manifests: dict[str, dict[str, Any]] = {str(manifest_path): existing_manifest} + for path in package_manifest_paths(output): + if path == manifest_path or filename_key(path.name) == filename_key( + manifest_path.name + ): + continue + sibling = _read_manifest(path) + _manifest_files(sibling, path) + manifests[str(path)] = sibling + return manifests + + def _manifest_files(payload: dict[str, Any], manifest_path: Path) -> dict[str, Any]: """Return a manifest's ``files`` block, refusing one that is not a mapping. @@ -366,6 +390,16 @@ def holds(self, *, sha256: str, filename: str) -> bool: return not self.filename or self.filename == Path(filename).name +@dataclass(frozen=True) +class _ManifestFileOwner: + """One manifest entry that names a package-local artifact.""" + + manifest_path: Path + vintage: Any + spec: dict[str, Any] + identity: RecordedIdentity | None + + @dataclass(frozen=True) class ArtifactCommandResult: """Result from a storage command.""" @@ -758,6 +792,9 @@ def fetch_source_artifact( manifest_path=manifest_path, year=vintage_key, ) + manifests = _package_manifests(output, manifest_path, existing_manifest) + owners = _manifest_file_owners(manifests, filename=artifact_filename) + _assert_shared_owner_identities_agree(owners, filename=artifact_filename) fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() content, _inferred_filename = _read_artifact(source_url) @@ -777,6 +814,26 @@ def fetch_source_artifact( r2_bucket=r2_bucket, record_revision=record_revision, ) + if not record_revision: + _assert_siblings_record_these_bytes( + manifests, + manifest_path=manifest_path, + filename=artifact_filename, + sha256=sha256, + ) + for owner in owners: + if owner.manifest_path == manifest_path and owner.spec is selected_spec: + continue + _assert_recorded_identity_holds_these_bytes( + owner.identity, + manifest_path=owner.manifest_path, + year=owner.vintage, + filename=artifact_filename, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=r2_bucket, + record_revision=record_revision, + ) output.mkdir(parents=True, exist_ok=True) local_path = output / artifact_filename @@ -1713,6 +1770,106 @@ def _recorded_identity( ) +def _manifest_file_owners( + manifests: Mapping[str, dict[str, Any]], + *, + filename: str, +) -> list[_ManifestFileOwner]: + """Return every entry in a package directory that names ``filename``.""" + wanted = filename_key(filename) + owners: list[_ManifestFileOwner] = [] + for name, payload in manifests.items(): + manifest_path = Path(name) + for vintage, spec in _manifest_files(payload, manifest_path).items(): + if not isinstance(spec, dict): + raise MalformedManifestError( + f"{manifest_path} entry {vintage!r} must be a mapping; it " + f"is a {type(spec).__name__}. Chronicle cannot decide " + "whether it owns a shared package-local file." + ) + recorded_name = spec.get("filename") + if recorded_name is None: + continue + if not is_bare_filename(recorded_name): + raise MalformedManifestError( + f"{manifest_path} entry {vintage!r} filename must be a " + f"bare package-local name, not {recorded_name!r}." + ) + if filename_key(recorded_name) != wanted: + continue + identity = _recorded_identity( + spec, + manifest_path=manifest_path, + year=vintage, + ) + if identity is None: + raise MalformedManifestError( + f"{manifest_path} entry {vintage!r} names " + f"{recorded_name!r} but records no sha256 identity. " + "Chronicle cannot safely overwrite an unidentifiable " + "shared file." + ) + owners.append( + _ManifestFileOwner( + manifest_path=manifest_path, + vintage=vintage, + spec=spec, + identity=identity, + ) + ) + return owners + + +def _assert_shared_owner_identities_agree( + owners: list[_ManifestFileOwner], + *, + filename: str, +) -> None: + """Refuse an already-contradictory set of owners before publisher I/O.""" + if not owners: + return + first = owners[0] + first_identity = first.identity + assert first_identity is not None + for owner in owners[1:]: + identity = owner.identity + assert identity is not None + if identity.sha256 == first_identity.sha256 and filename_key( + identity.filename + ) == filename_key(first_identity.filename): + continue + raise SourceArtifactManifestError( + f"{first.manifest_path} entry {first.vintage!r} and " + f"{owner.manifest_path} entry {owner.vintage!r} both name " + f"{filename!r} but identify different bytes. One package-local " + "file must have one recorded identity; reconcile the manifests " + "before fetching it again." + ) + + +def _assert_siblings_record_these_bytes( + manifests: Mapping[str, dict[str, Any]], + *, + manifest_path: Path, + filename: str, + sha256: str, +) -> None: + """Refuse a default fetch that would stale another manifest's owner.""" + for owner in _manifest_file_owners(manifests, filename=filename): + if owner.manifest_path == manifest_path: + continue + identity = owner.identity + assert identity is not None + if identity.sha256 == sha256: + continue + raise SourceArtifactRevisionError( + f"{owner.manifest_path} entry {owner.vintage!r} records " + f"{filename!r} as sha256={identity.sha256}; this fetch would write " + f"sha256={sha256} to the same package-local file. Re-run with " + "--record-revision to update every owner together." + ) + + def _revision_error_message( *, manifest_path: Path, @@ -1851,6 +2008,36 @@ def _superseding_storage( ) +def _storage_for_fetched_identity( + recorded_spec: dict[str, Any], + *, + identity: RecordedIdentity | None, + filename: str, + sha256: str, + new_r2: dict[str, Any] | None, + fetched_at: str, +) -> dict[str, Any]: + """Return one owner's storage after a refetch or explicit revision.""" + recorded_storage = _recorded_storage(recorded_spec) + holds = identity is not None and identity.holds( + sha256=sha256, + filename=filename, + ) + if holds and identity.r2 is not None: + # A same-byte copy does not replace the object's recorded history. + return {**recorded_storage, "r2": _recorded_r2(recorded_spec)} + if identity is not None and not holds: + return _superseding_storage( + recorded_spec, + recorded_r2=identity.r2, + new_r2=new_r2, + superseded_at=fetched_at, + ) + if new_r2 is not None: + return {**recorded_storage, "r2": new_r2} + return dict(recorded_storage) + + def _upsert_manifest( manifest_path: Path, *, @@ -1876,6 +2063,9 @@ def _upsert_manifest( source_id=source_id, package_id=package_id, ) + manifests = _package_manifests(manifest_path.parent, manifest_path, payload) + owners = _manifest_file_owners(manifests, filename=filename) + _assert_shared_owner_identities_agree(owners, filename=filename) payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) payload.setdefault("dataset", dataset) @@ -1900,51 +2090,94 @@ def _upsert_manifest( for field, value in recorded_spec.items(): if field not in _FETCH_OWNED_FIELDS and field not in file_entry: file_entry[field] = value - recorded_storage = _recorded_storage(recorded_spec) identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=key) new_r2 = r2_location.to_dict() if r2_location is not None else None - holds = identity is not None and identity.holds(sha256=sha256, filename=filename) - if identity is not None and not holds and not record_revision: - # Different bytes under the same vintage. The guard in - # fetch_source_artifact refuses this without --record-revision; repeat - # the check here so no caller can reach a false-provenance write. - raise SourceArtifactRevisionError( - _revision_error_message( - manifest_path=manifest_path, - year=key, - filename=filename, - identity=identity, - sha256=sha256, - size_bytes=size_bytes, - r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), - ) + r2_bucket = (new_r2 or {}).get("bucket") or default_r2_raw_bucket() + _assert_recorded_identity_holds_these_bytes( + identity, + manifest_path=manifest_path, + year=key, + filename=filename, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=r2_bucket, + record_revision=record_revision, + ) + if not record_revision: + _assert_siblings_record_these_bytes( + manifests, + manifest_path=manifest_path, + filename=filename, + sha256=sha256, ) - if holds and identity.r2 is not None: - # A recorded storage.r2 block for these exact bytes is historical - # truth: archived witness records pin raw R2 URLs by hash. Re-fetching - # under a renamed bucket copies bytes; it does not restate where the - # bytes were first published (PolicyEngine/chronicle#143, mechanism 3). - storage = {**recorded_storage, "r2": _recorded_r2(recorded_spec)} - elif identity is not None and not holds: - storage = _superseding_storage( - recorded_spec, - recorded_r2=identity.r2, - new_r2=new_r2, - superseded_at=fetched_at, + for owner in owners: + if owner.manifest_path == manifest_path and owner.spec is recorded_spec: + continue + _assert_recorded_identity_holds_these_bytes( + owner.identity, + manifest_path=owner.manifest_path, + year=owner.vintage, + filename=filename, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=r2_bucket, + record_revision=record_revision, ) - elif new_r2 is not None: - storage = {**recorded_storage, "r2": new_r2} - else: - storage = dict(recorded_storage) + + storage = _storage_for_fetched_identity( + recorded_spec, + identity=identity, + filename=filename, + sha256=sha256, + new_r2=new_r2, + fetched_at=fetched_at, + ) # An entry that has no storage to record carries no empty block: a # revision over a never-published entry supersedes nothing. if storage: file_entry["storage"] = storage payload["files"][key] = file_entry - manifest_path.write_text( - yaml.safe_dump(payload, sort_keys=False), - encoding="utf-8", - ) + + revision = any( + owner.identity is not None + and not owner.identity.holds(sha256=sha256, filename=filename) + for owner in owners + ) or (identity is not None and not identity.holds(sha256=sha256, filename=filename)) + changed_paths = {manifest_path} + if record_revision and revision: + for owner in owners: + if owner.manifest_path == manifest_path and owner.spec is recorded_spec: + continue + revised_entry = dict(owner.spec) + revised_entry.update( + { + "filename": filename, + "sha256": sha256, + "size_bytes": size_bytes, + "fetched_at": fetched_at, + } + ) + owner_storage = _storage_for_fetched_identity( + owner.spec, + identity=owner.identity, + filename=filename, + sha256=sha256, + new_r2=new_r2, + fetched_at=fetched_at, + ) + if owner_storage: + revised_entry["storage"] = owner_storage + else: + revised_entry.pop("storage", None) + manifests[str(owner.manifest_path)]["files"][owner.vintage] = revised_entry + changed_paths.add(owner.manifest_path) + + rendered = { + path: yaml.safe_dump(manifests[str(path)], sort_keys=False) + for path in changed_paths + } + for path, text in sorted(rendered.items(), key=lambda item: str(item[0])): + path.write_text(text, encoding="utf-8") def _upload_r2_object( diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 4e1f47ee..8a85e1ca 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -164,11 +164,16 @@ Most packages keep one `manifest.yaml`. A publisher directory that feeds several source packages keeps one manifest each — `db/data/irs_soi/ira_contributions/` holds `manifest_traditional_source_package.yaml` beside -`manifest_roth_source_package.yaml` — and the entry being revised lives in -exactly one of them. `fetch-artifact --manifest ` selects it; -defaulting to `manifest.yaml` there would write a third manifest neither -package reads, and the recorded block would never be compared at all. The name -must be a filename inside `--out-dir`, not a path. +`manifest_roth_source_package.yaml`. `fetch-artifact --manifest ` +selects the entry whose publisher metadata the fetch updates; defaulting to +`manifest.yaml` there would write a third manifest neither package reads. A +physical artifact can also be owned by several entries in that directory (the +tracked USDA SNAP archive spans two manifests, and SSA extracts have semantic +aliases within one). Chronicle compares every such owner before overwriting the +file. A changed archive is refused by default; `--record-revision` updates every +owner to the new checksum and preserves each owner's own R2 block in +`storage.previous_r2`. The manifest selector must be a filename inside +`--out-dir`, not a path. ### What a recorded block has to say From 20b4028d8b691591357e2fd474e7c4a3e0829181 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:30:01 -0400 Subject: [PATCH 043/105] Record shared revision ownership fix --- PROGRESS.md | 24 ++++++++++++++++++++++-- 1 file changed, 22 insertions(+), 2 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index fef202bc..1affeb03 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -357,11 +357,31 @@ Running the same operations against a checkout of the previous head: registry-creation and exact sweep-selector guards are slice-1 additions. The focused post-fix command exits 0 with 13 passed, and the complete artifact module exits 0 with 132 passed. +- **Finding 2 reproduced and fixed.** Red command: + `UV_CACHE_DIR=/tmp/chronicle-uv-cache uv run pytest -q -p + no:cacheprovider tests/test_chronicle_artifacts.py::test_shared_archive_revision_is_refused_through_an_unregistered_owner + tests/test_chronicle_artifacts.py::test_record_revision_updates_every_owner_of_usda_shared_archive + tests/test_chronicle_artifacts.py::test_record_revision_updates_every_same_manifest_owner` + exited 1 with 3 failures: the empty selected vintage bypassed a sibling's + recorded identity, the USDA second manifest retained its old checksum, and + the SSA-style second key retained its old checksum. Test-only commit: + `3636395`. +- Fetch now strictly loads every manifest in the package directory before + publisher I/O, establishes one normalized byte identity for every entry + naming the physical file, and applies the revision guard to all owners. An + explicit revision rewrites every owner from payloads rendered before the + first manifest write, preserves owner-specific metadata, and archives each + owner's own R2 provenance. Ported #227's `_package_manifests` and + `_assert_siblings_record_these_bytes` names/shapes from `c0d9d74`, including + the normalized manifest-alias exclusion from `235c616`; the coordinated + all-owner rewrite is slice-1-specific. Fix/docs commit: `7da26a9`. The red + command now exits 0 with 3 passed; the artifact module exits 0 with 135 + passed. ### Next -1. Reproduce and fix cross-manifest shared-file revision ownership (finding 2), - including the tracked USDA two-manifest shape and every same-directory owner. +1. Reproduce and fix recorded-key cutover compatibility (finding 1) with the + documented non-writing sweep over a temporary copy of the tracked tree. 2. Commit each coherent red-test and implementation step, updating this log. 3. Run focused tests, lint/format checks on changed files, and the full suite with the directly captured exit code and counts. From a644be87ec308399425431e6a6c44baf7f747fdf Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:31:28 -0400 Subject: [PATCH 044/105] Reproduce tracked registry cutover failures --- tests/test_chronicle_artifacts.py | 69 +++++++++++++++++++++++++++++++ 1 file changed, 69 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 4f6f3100..6342467a 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -4,6 +4,8 @@ import hashlib import json +from pathlib import Path +import shutil import sqlite3 import pytest @@ -12,6 +14,7 @@ from chronicle.cli import main as cli_main from chronicle.artifacts import ( AmbiguousManifestError, + ArtifactCommandResult, MalformedManifestError, ManifestNameError, RecordedR2LocatorError, @@ -810,6 +813,72 @@ def test_publish_raw_skips_an_object_already_held_by_a_preserved_bucket( assert manifest_path.read_bytes() == before +def test_documented_bucket_cutover_sweep_accepts_the_tracked_registry( + tmp_path, monkeypatch, capsys +): + """The documented bucket flip is green for every recorded historical key.""" + tracked_data = Path(__file__).resolve().parents[1] / "db" / "data" + copied_data = tmp_path / "data" + shutil.copytree(tracked_data, copied_data) + manifest_bytes = { + path.relative_to(copied_data): path.read_bytes() + for path in copied_data.rglob("*") + if path.is_file() and path.name.lower().startswith("manifest") + } + uploads = [] + + def non_writing_uploader(location, local_path, *, wrangler_command): + uploads.append((location, local_path, wrangler_command)) + return ArtifactCommandResult( + command=("non-writing-uploader",), + returncode=0, + stdout="", + stderr="", + ) + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", non_writing_uploader) + + exit_code = harness_main( + [ + "publish-raw", + "--root", + str(copied_data), + "--wrangler-command", + "non-writing-uploader", + ] + ) + report = json.loads(capsys.readouterr().out) + expected_counts = { + "manifest_count": 161, + "artifact_count": 194, + "uploaded_count": 0, + "skipped_count": 194, + "failed_count": 0, + "r2_link_count": 194, + } + + observed = ( + exit_code, + report["valid"], + report["counts"], + len(report["errors"]), + ) + assert observed == ( + 0, + True, + expected_counts, + 0, + ), json.dumps(observed, sort_keys=True) + assert all(entry["skipped"] or entry["upload"] for entry in report["entries"]) + assert uploads == [] + assert { + path.relative_to(copied_data): path.read_bytes() + for path in copied_data.rglob("*") + if path.is_file() and path.name.lower().startswith("manifest") + } == manifest_bytes + + def test_fetch_artifact_keeps_an_already_recorded_bucket(tmp_path, monkeypatch): output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-1-1" source = tmp_path / "soi.xlsx" From eb4d43af966e28d3d157f1c1daca27740a5202c4 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:33:04 -0400 Subject: [PATCH 045/105] Preserve compatible historical R2 routes --- chronicle/artifacts.py | 108 ++++++++++++++---------------- tests/test_chronicle_artifacts.py | 26 ++++--- 2 files changed, 63 insertions(+), 71 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index a9f133cd..ec72a07c 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1064,24 +1064,8 @@ def publish_source_artifacts( errors.append(f"Could not read {manifest_path}: {exc}") continue - manifest_source_id = source_id or manifest.get("source_id") - manifest_package_id = package_id or manifest.get("package_id") - if not manifest_source_id: - errors.append(f"Manifest missing source_id: {manifest_path}") - continue - if not manifest_package_id: - errors.append(f"Manifest missing package_id: {manifest_path}") - continue - try: - resolved_r2_prefix = resolve_r2_prefix( - prefix=r2_prefix, - default_prefix=DEFAULT_R2_PREFIX, - source_id=str(manifest_source_id), - package_path=manifest_path, - ) - except ValueError as exc: - errors.append(f"Could not resolve R2 prefix for {manifest_path}: {exc}") - continue + manifest_source_id = str(source_id or manifest.get("source_id") or "") + manifest_package_id = str(package_id or manifest.get("package_id") or "") updated = False for year, spec in files.items(): @@ -1092,7 +1076,7 @@ def publish_source_artifacts( year, spec, r2_bucket=r2_bucket, - r2_prefix=resolved_r2_prefix, + r2_prefix=r2_prefix, wrangler_command=wrangler_command, ) entries.append(entry) @@ -1100,8 +1084,10 @@ def publish_source_artifacts( spec.update(updated_spec) updated = True if updated: - manifest.setdefault("source_id", manifest_source_id) - manifest.setdefault("package_id", manifest_package_id) + if manifest_source_id: + manifest.setdefault("source_id", manifest_source_id) + if manifest_package_id: + manifest.setdefault("package_id", manifest_package_id) manifest_path.write_text( yaml.safe_dump(manifest, sort_keys=False), encoding="utf-8", @@ -2210,7 +2196,7 @@ def _publish_raw_manifest_entry( spec: Any, *, r2_bucket: str, - r2_prefix: str, + r2_prefix: str | None, wrangler_command: str, ) -> tuple[RawArtifactPublishEntry, dict[str, Any] | None]: errors: list[str] = [] @@ -2304,37 +2290,20 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: f"local_filename={Path(filename).name}" ) - location = ArtifactStorageLocation( - provider="r2", - bucket=r2_bucket, - key=build_r2_key( - source_id=source_id, - package_id=package_id, - year=year, - sha256=sha256_actual or "", - filename=filename, - prefix=r2_prefix, - package_path=manifest_path, - ), - ) - recorded_key = recorded_r2.key if recorded_r2 is not None else None - if recorded_key and recorded_key != location.key: - # Validate the full source/package/year route before the preserved- - # bucket shortcut below. A bucket rename does not make a misrouted - # object valid history. - return refuse( - "recorded_r2_key_disagrees_with_country_prefix:" - f"recorded={recorded_key}:expected={location.key}" - ) - recorded_bucket = recorded_r2.bucket if recorded_r2 is not None else None - if recorded_r2 is not None and recorded_bucket != location.bucket: - # The recorded bucket is preserved history and, per the identity check - # above, its object holds exactly these bytes: the artifact is already - # published. Restating it under the configured bucket would rewrite - # where the bytes were first published (a backfill copy is not a - # restatement), so the entry is reported as skipped with nothing - # uploaded or rewritten. After the bucket-default flip every entry - # published before it takes this path, and the sweep stays green. + if recorded_r2 is not None: + # A recorded content-addressed object whose tail identifies the bytes + # in hand is published history. Its source/package/year route may + # predate today's country prefix or intentionally represent the + # publisher's explicit route (for example Statbel's 2023 snapshots and + # USDA's cross-manifest archive). Reconstructing a current route and + # requiring equality would rewrite that history during a bucket + # cutover. Only the checksum/filename tail decides byte identity. + skipped = "recorded_r2_already_published" + if recorded_r2.bucket != r2_bucket: + skipped = ( + "recorded_r2_bucket_is_preserved_history:" + f"recorded={recorded_r2.bucket}:requested={r2_bucket}" + ) return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), @@ -2352,13 +2321,38 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: ), upload=None, errors=(), - skipped=( - "recorded_r2_bucket_is_preserved_history:" - f"recorded={recorded_bucket}:requested={location.bucket}" - ), + skipped=skipped, ), None, ) + + if not source_id: + return refuse("missing_source_id") + if not package_id: + return refuse("missing_package_id") + try: + resolved_r2_prefix = resolve_r2_prefix( + prefix=r2_prefix, + default_prefix=DEFAULT_R2_PREFIX, + source_id=source_id, + package_path=manifest_path, + ) + except ValueError as error: + return refuse(f"r2_prefix_invalid:{error}") + + location = ArtifactStorageLocation( + provider="r2", + bucket=r2_bucket, + key=build_r2_key( + source_id=source_id, + package_id=package_id, + year=year, + sha256=sha256_actual or "", + filename=filename, + prefix=resolved_r2_prefix, + package_path=manifest_path, + ), + ) upload = _upload_r2_object( location, artifact_path, diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 6342467a..093052ac 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -331,7 +331,7 @@ def test_publish_source_artifacts_uses_country_for_each_manifest(tmp_path): assert "ledger-raw/raw/irs_soi/soi-table/2023/" in commands -def test_publish_source_artifacts_refuses_stale_country_key(tmp_path): +def test_publish_source_artifacts_preserves_a_legacy_countryless_key(tmp_path): output_dir = tmp_path / "data" / "ird" / "wff" source = tmp_path / "wff.xlsx" source.write_bytes(b"official WFF workbook") @@ -365,13 +365,11 @@ def test_publish_source_artifacts_refuses_stale_country_key(tmp_path): report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) - assert not report.valid + assert report.valid assert report.entries[0].upload is None - assert ( - report.entries[0] - .errors[0] - .startswith("recorded_r2_key_disagrees_with_country_prefix:") - ) + assert report.entries[0].errors == () + assert report.entries[0].skipped == "recorded_r2_already_published" + assert report.entries[0].r2_location.key == artifact["storage"]["r2"]["key"] assert not log.exists() @@ -2771,7 +2769,7 @@ def test_sweeps_refuse_a_symlinked_artifact_without_reading_it(tmp_path): # --------------------------------------------------------------------------- -def test_publish_checks_the_canonical_key_before_a_preserved_bucket_skip( +def test_publish_preserves_an_explicit_historical_route_during_bucket_cutover( tmp_path, monkeypatch ): package = tmp_path / "db" / "data" / "irs_soi" / "table" @@ -2803,14 +2801,14 @@ def test_publish_checks_the_canonical_key_before_a_preserved_bucket_skip( report = publish_source_artifacts(package, wrangler_command=str(wrangler)) - assert not report.valid + assert report.valid assert report.entries[0].upload is None - assert report.entries[0].skipped is None - assert ( - report.entries[0] - .errors[0] - .startswith("recorded_r2_key_disagrees_with_country_prefix:") + assert report.entries[0].errors == () + assert report.entries[0].skipped == ( + "recorded_r2_bucket_is_preserved_history:" + "recorded=ledger-raw:requested=chronicle-raw" ) + assert report.entries[0].r2_location.key == wrong_key assert not log.exists() assert manifest_path.read_bytes() == before From ec902b0b7c9df7bc313d57a5be864fc7dd7c9812 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:33:20 -0400 Subject: [PATCH 046/105] Record cutover compatibility fix --- PROGRESS.md | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 1affeb03..47ad8907 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -377,11 +377,29 @@ Running the same operations against a checkout of the previous head: all-owner rewrite is slice-1-specific. Fix/docs commit: `7da26a9`. The red command now exits 0 with 3 passed; the artifact module exits 0 with 135 passed. +- **Finding 1 reproduced and fixed.** The whole-tree regression copies tracked + `db/data/**` to `tmp_path`, configures `CHRONICLE_R2_RAW_BUCKET=chronicle-raw`, + and installs a successful non-writing uploader. Red command: + `UV_CACHE_DIR=/tmp/chronicle-uv-cache uv run pytest -q -p + no:cacheprovider tests/test_chronicle_artifacts.py::test_documented_bucket_cutover_sweep_accepts_the_tracked_registry` + exited 1 with the observed tuple `exit=1`, `valid=false`, 156 manifests, 187 + artifacts, 94 skipped, 93 failed, 94 R2-linked, zero uploaded, and five + report errors. Test-only commit: `7f0f473`. +- A syntactically valid recorded R2 locator whose content-addressed tail + matches the package-local checksum and filename is now preserved and skipped + before reconstructing today's country/source/package/year route. This + accepts 84 pre-country-prefix UK keys, eight explicit Statbel 2023 routes, + the shared USDA route, and five fully recorded Eurostat manifests that omit + package IDs, while locator contradictions and byte mismatches remain errors. + Fix commit: `25deb29`; no #227 hunk applies. The same whole-tree command now + exits 0 with 161 manifests, 194 artifacts, 194 skipped/R2-linked, zero + uploaded or failed, and no errors. The artifact module exits 0 with 136 + passed. ### Next -1. Reproduce and fix recorded-key cutover compatibility (finding 1) with the - documented non-writing sweep over a temporary copy of the tracked tree. +1. Reproduce and fix malformed `storage.previous_r2` handling (finding 7), + proving refusal before publisher I/O. 2. Commit each coherent red-test and implementation step, updating this log. 3. Run focused tests, lint/format checks on changed files, and the full suite with the directly captured exit code and counts. From 5f55184ec8c42be51dc0bc04b5616de5557a28a6 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:33:54 -0400 Subject: [PATCH 047/105] Reproduce malformed revision history loss --- tests/test_chronicle_artifacts.py | 38 +++++++++++++++++++++++++++++++ 1 file changed, 38 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 093052ac..0dba3cbf 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -2038,6 +2038,44 @@ def _other_sha256(): return hashlib.sha256(b"some other object entirely").hexdigest() +@pytest.mark.parametrize( + "previous_r2", + [ + pytest.param({}, id="mapping"), + pytest.param("not a list", id="scalar"), + pytest.param(None, id="null"), + ], +) +def test_fetch_refuses_non_list_previous_r2_before_publisher_io( + tmp_path, monkeypatch, previous_r2 +): + """Malformed archived provenance must not be replaced by a new history.""" + package, source, _report = _recorded_package(tmp_path) + manifest_path = _rewrite_recorded_r2( + package, + lambda storage: storage.__setitem__("previous_r2", previous_r2), + ) + artifact_path = package / "22in05ira.xlsx" + before = { + manifest_path: manifest_path.read_bytes(), + artifact_path: artifact_path.read_bytes(), + } + source.write_bytes(b"IRA table 5, revised publication") + + def unexpected_read(_source_url): + raise AssertionError("malformed previous_r2 reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises( + MalformedManifestError, + match=r"storage[.]previous_r2 must be a list", + ): + _fetch_local(package, source, upload_r2=False, record_revision=True) + + assert {path: path.read_bytes() for path in before} == before + + def _contradict_key(storage): key = storage["r2"]["key"] storage["r2"]["key"] = key.replace(key.split("/")[-2], _other_sha256()) From 9e26ca55be361943047909e0a4f833781ed74ed6 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:34:14 -0400 Subject: [PATCH 048/105] Refuse malformed revision history before I/O --- chronicle/artifacts.py | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index ec72a07c..adefbed9 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1598,6 +1598,13 @@ def _validated_recorded_storage( f"{manifest_path} entry {year!r} storage must be a mapping; it is a " f"{type(storage).__name__}." ) + if "previous_r2" in storage and not isinstance(storage["previous_r2"], list): + previous = storage["previous_r2"] + raise MalformedManifestError( + f"{manifest_path} entry {year!r} storage.previous_r2 must be a " + f"list; it is a {type(previous).__name__}. Chronicle will not " + "discard malformed archived provenance." + ) return storage From f5efc503cb3a6d3ccc7941701faa90f8ca90eed2 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:34:26 -0400 Subject: [PATCH 049/105] Record revision history validation --- PROGRESS.md | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 47ad8907..67992878 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -395,11 +395,21 @@ Running the same operations against a checkout of the previous head: exits 0 with 161 manifests, 194 artifacts, 194 skipped/R2-linked, zero uploaded or failed, and no errors. The artifact module exits 0 with 136 passed. +- **Finding 7 reproduced and fixed.** Red command: + `UV_CACHE_DIR=/tmp/chronicle-uv-cache uv run pytest -q -p + no:cacheprovider tests/test_chronicle_artifacts.py::test_fetch_refuses_non_list_previous_r2_before_publisher_io` + exited 1 with 3 failures: mapping, scalar, and null `previous_r2` values all + reached the publisher-I/O sentinel. Test-only commit: `e418813`. + `_validated_recorded_storage` now rejects every present non-list history + before `_read_artifact`, so `_superseding_storage` cannot replace malformed + provenance. Fix commit: `8b7ab22`; no #227 hunk applies. The focused command + now exits 0 with 3 passed (5 passed including ordinary and shared revision + history controls). ### Next -1. Reproduce and fix malformed `storage.previous_r2` handling (finding 7), - proving refusal before publisher I/O. +1. Reproduce and fix import-time Supabase compatibility aliases (finding 8) + for Chronicle/current and both legacy schema environment spellings. 2. Commit each coherent red-test and implementation step, updating this log. 3. Run focused tests, lint/format checks on changed files, and the full suite with the directly captured exit code and counts. From 27909da23c3384eda039b7120985ab5aa7da72df Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:35:11 -0400 Subject: [PATCH 050/105] Reproduce frozen Supabase schema aliases --- tests/test_chronicle_env.py | 37 +++++++++++++++++++++++++++++++++++++ 1 file changed, 37 insertions(+) diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py index 28bce190..2e1f065e 100644 --- a/tests/test_chronicle_env.py +++ b/tests/test_chronicle_env.py @@ -10,8 +10,11 @@ from __future__ import annotations import importlib +import json import os from pathlib import Path +import subprocess +import sys import pytest @@ -279,6 +282,40 @@ def test_supabase_schema_default_is_unchanged(): assert db.supabase_client.targets_schema() == "targets" +@pytest.mark.parametrize( + "schema_env", + ["CHRONICLE_SCHEMA", "POLICYENGINE_LEDGER_SCHEMA", "LEDGER_SCHEMA"], +) +def test_supabase_schema_compatibility_aliases_honor_import_time_environment( + schema_env, +): + """Deprecated exports retain the environment snapshot existing importers use.""" + environment = os.environ.copy() + for name in (*env_names("CHRONICLE_SCHEMA"), "POLICYENGINE_TARGETS_SCHEMA"): + environment.pop(name, None) + environment[schema_env] = "chronicle_import_probe" + environment["POLICYENGINE_TARGETS_SCHEMA"] = "targets_import_probe" + script = ( + "import json; " + "from db.supabase_client import LEDGER_SCHEMA, TARGETS_SCHEMA; " + "print(json.dumps([LEDGER_SCHEMA, TARGETS_SCHEMA]))" + ) + + completed = subprocess.run( + [sys.executable, "-c", script], + cwd=Path(__file__).resolve().parents[1], + env=environment, + check=True, + capture_output=True, + text=True, + ) + + assert json.loads(completed.stdout) == [ + "chronicle_import_probe", + "targets_import_probe", + ] + + def test_supabase_schema_is_not_bound_at_import(monkeypatch): """Compatibility constants do not freeze the runtime schema resolver. From cdc7968f719b91c771c869dc7858aef67c46f887 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:35:38 -0400 Subject: [PATCH 051/105] Honor schema env in compatibility exports --- db/supabase_client.py | 21 +++++++++------------ 1 file changed, 9 insertions(+), 12 deletions(-) diff --git a/db/supabase_client.py b/db/supabase_client.py index 24f89f18..a8c5dbeb 100644 --- a/db/supabase_client.py +++ b/db/supabase_client.py @@ -6,7 +6,7 @@ - Target inputs ``LEDGER_SCHEMA`` and ``TARGETS_SCHEMA`` remain as deprecated compatibility -aliases for their default schema names. Runtime code should call +snapshots of the environment at import time. Runtime code should call ``chronicle_schema()`` and ``targets_schema()`` so environment overrides are resolved at the time of use. """ @@ -20,21 +20,11 @@ from supabase import create_client, Client -from chronicle.env import ( - DEFAULT_CHRONICLE_SCHEMA, - default_chronicle_schema, - env_value, -) +from chronicle.env import default_chronicle_schema, env_value TARGETS_SCHEMA_ENV = "POLICYENGINE_TARGETS_SCHEMA" DEFAULT_TARGETS_SCHEMA = "targets" -# Deprecated import compatibility. These are deliberately aliases for the -# defaults, not environment-backed runtime values; query paths below remain on -# the lazy resolver functions. -LEDGER_SCHEMA = DEFAULT_CHRONICLE_SCHEMA -TARGETS_SCHEMA = DEFAULT_TARGETS_SCHEMA - def chronicle_schema() -> str: """Resolve the hosted Chronicle schema for a query. @@ -59,6 +49,13 @@ def targets_schema() -> str: return env_value(TARGETS_SCHEMA_ENV, default=DEFAULT_TARGETS_SCHEMA) +# Deprecated import compatibility. Preserve the historical import-time +# environment snapshot for downstream code that still imports these names; +# Chronicle's own query paths use the lazy resolvers above. +LEDGER_SCHEMA = chronicle_schema() +TARGETS_SCHEMA = targets_schema() + + @dataclass class SupabaseConfig: """Configuration for Supabase connection.""" From 9060f18f2e8bd43906c0843242877faad49c1be7 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:35:52 -0400 Subject: [PATCH 052/105] Record schema compatibility fix --- PROGRESS.md | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 67992878..c736d6a0 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -405,11 +405,25 @@ Running the same operations against a checkout of the previous head: provenance. Fix commit: `8b7ab22`; no #227 hunk applies. The focused command now exits 0 with 3 passed (5 passed including ordinary and shared revision history controls). +- **Finding 8 reproduced and fixed.** A fresh-process regression sets each of + `CHRONICLE_SCHEMA`, `POLICYENGINE_LEDGER_SCHEMA`, and `LEDGER_SCHEMA` before + importing the compatibility constants, alongside + `POLICYENGINE_TARGETS_SCHEMA`. Red command: + `UV_CACHE_DIR=/tmp/chronicle-uv-cache uv run pytest -q -p + no:cacheprovider tests/test_chronicle_env.py::test_supabase_schema_compatibility_aliases_honor_import_time_environment` + exited 1 with 3 failures; every subprocess returned `ledger` / `targets`. + Test-only commit: `e2ceedf`. +- `LEDGER_SCHEMA` and `TARGETS_SCHEMA` are now import-time snapshots of the + same lazy resolver functions runtime queries use, restoring their original + environment-backed behavior while later environment mutations remain lazy + through `chronicle_schema()` / `targets_schema()`. Fix commit: `f41ad0f`; no + #227 hunk applies. The focused regression now exits 0 with 3 passed; it and + the runtime/default namespace controls exit 0 with 11 passed. ### Next -1. Reproduce and fix import-time Supabase compatibility aliases (finding 8) - for Chronicle/current and both legacy schema environment spellings. +1. Audit the eight findings and #227 port surface for missed CLI/refusal cases, + then run focused and full verification. 2. Commit each coherent red-test and implementation step, updating this log. 3. Run focused tests, lint/format checks on changed files, and the full suite with the directly captured exit code and counts. From 1096acefc5095d4ca175e914d4077f767f0cd7cd Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:45:28 -0400 Subject: [PATCH 053/105] Reproduce residual artifact boundary escapes --- tests/test_chronicle_artifacts.py | 224 ++++++++++++++++++++++++++++++ 1 file changed, 224 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 0dba3cbf..1341746b 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -15,6 +15,7 @@ from chronicle.artifacts import ( AmbiguousManifestError, ArtifactCommandResult, + ArtifactFilenameError, MalformedManifestError, ManifestNameError, RecordedR2LocatorError, @@ -1880,6 +1881,65 @@ def unexpected_read(_source_url): assert outside_manifest.read_bytes() == before +def test_fetch_refuses_physically_distinct_normalized_manifest_aliases( + tmp_path, monkeypatch +): + package = tmp_path / "db" / "data" / "publisher" / "package" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text("source_id: publisher\npackage_id: package\nfiles: {}\n") + case_alias = package / "Manifest.yaml" + monkeypatch.setattr( + "chronicle.artifacts.package_manifest_paths", + lambda _package: [manifest_path, case_alias], + ) + + def unexpected_read(_source_url): + raise AssertionError("normalized manifest aliases reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(AmbiguousManifestError, match="normalized manifest name"): + fetch_source_artifact( + "https://example.test/table.csv", + source_id="publisher", + package_id="package", + year=2024, + output_dir=package, + ) + + +def test_fetch_refuses_a_symlinked_artifact_target_before_publisher_io( + tmp_path, monkeypatch +): + package = tmp_path / "db" / "data" / "publisher" / "package" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text("source_id: publisher\npackage_id: package\nfiles: {}\n") + outside = tmp_path / "outside.csv" + outside.write_bytes(b"outside bytes") + artifact_path = package / "table.csv" + artifact_path.symlink_to(outside) + before = {manifest_path: manifest_path.read_bytes(), outside: outside.read_bytes()} + + def unexpected_read(_source_url): + raise AssertionError("symlinked artifact target reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(ArtifactFilenameError, match="symbolic link"): + fetch_source_artifact( + "https://example.test/table.csv", + source_id="publisher", + package_id="package", + year=2024, + output_dir=package, + ) + + assert artifact_path.is_symlink() + assert {path: path.read_bytes() for path in before} == before + + @pytest.mark.parametrize( "manifest_filename", [ @@ -1932,6 +1992,34 @@ def test_sweep_manifest_selector_must_be_a_literal_supported_filename( assert not log.exists() +@pytest.mark.parametrize( + "operation", [inventory_source_artifacts, publish_source_artifacts] +) +def test_invalid_sweep_manifest_selector_is_refused_even_when_root_is_missing( + tmp_path, operation +): + with pytest.raises(ManifestNameError): + operation(tmp_path / "missing", manifest_filename="../manifest.yaml") + + +@pytest.mark.parametrize("command", ["inventory-artifacts", "publish-raw"]) +def test_sweep_cli_reports_an_invalid_manifest_selector(command, tmp_path, capsys): + exit_code = harness_main( + [ + command, + "--root", + str(tmp_path), + "--manifest", + "../manifest.yaml", + ] + ) + + captured = capsys.readouterr() + assert exit_code == 1 + assert captured.out == "" + assert captured.err.startswith("error: ") + + # --------------------------------------------------------------------------- # Identity without a recorded R2 object # --------------------------------------------------------------------------- @@ -2802,6 +2890,142 @@ def test_sweeps_refuse_a_symlinked_artifact_without_reading_it(tmp_path): assert artifact_path.is_symlink() +@pytest.mark.parametrize( + "bad_kind", + [ + pytest.param("parent", id="parent-path"), + pytest.param("symlink", id="symlink"), + pytest.param("manifest-name", id="manifest-name"), + pytest.param("previous-r2", id="malformed-history"), + ], +) +def test_publish_preflights_every_entry_before_any_upload( + tmp_path, monkeypatch, bad_kind +): + package = tmp_path / "data" / "package" + package.mkdir(parents=True) + first = b"first publisher table" + second = b"second publisher table" + (package / "one.csv").write_bytes(first) + outside = tmp_path / "data" / "outside.csv" + outside.write_bytes(second) + second_path = package / "two.csv" + bad_filename = "two.csv" + bad_storage = None + if bad_kind == "parent": + bad_filename = "../outside.csv" + elif bad_kind == "symlink": + second_path.symlink_to(outside) + elif bad_kind == "manifest-name": + bad_filename = "manifest.yaml" + else: + second_path.write_bytes(second) + bad_storage = {"previous_r2": {"not": "a list"}} + bad_entry = { + "filename": bad_filename, + "source_url": "https://example.test/two.csv", + "sha256": hashlib.sha256(second).hexdigest(), + "size_bytes": len(second), + } + if bad_storage is not None: + bad_entry["storage"] = bad_storage + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": "package", + "files": { + 2023: { + "filename": "one.csv", + "source_url": "https://example.test/one.csv", + "sha256": hashlib.sha256(first).hexdigest(), + "size_bytes": len(first), + }, + 2024: bad_entry, + }, + }, + sort_keys=False, + ) + ) + before = manifest_path.read_bytes() + uploads = [] + + def non_writing_uploader(location, local_path, *, wrangler_command): + uploads.append((location, local_path, wrangler_command)) + return ArtifactCommandResult( + command=("non-writing-uploader",), + returncode=0, + stdout="", + stderr="", + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", non_writing_uploader) + + report = publish_source_artifacts(package) + + assert not report.valid + assert uploads == [] + assert manifest_path.read_bytes() == before + + +def test_sweeps_refuse_conflicting_owners_across_package_manifests( + tmp_path, monkeypatch +): + package = tmp_path / "data" / "package" + package.mkdir(parents=True) + content = b"publisher table" + filename = "table.csv" + (package / filename).write_bytes(content) + manifest_paths = ( + package / "manifest_a.yaml", + package / "manifest_b.yaml", + ) + for path, sha256 in zip( + manifest_paths, + (hashlib.sha256(content).hexdigest(), hashlib.sha256(b"other").hexdigest()), + ): + path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": path.stem, + "files": { + 2024: { + "filename": filename, + "sha256": sha256, + "size_bytes": len(content), + } + }, + }, + sort_keys=False, + ) + ) + before = {path: path.read_bytes() for path in manifest_paths} + uploads = [] + + def non_writing_uploader(location, local_path, *, wrangler_command): + uploads.append((location, local_path, wrangler_command)) + return ArtifactCommandResult( + command=("non-writing-uploader",), + returncode=0, + stdout="", + stderr="", + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", non_writing_uploader) + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package) + + assert not inventory.valid + assert not published.valid + assert any("identify different bytes" in error for error in inventory.errors) + assert any("identify different bytes" in error for error in published.errors) + assert uploads == [] + assert {path: path.read_bytes() for path in manifest_paths} == before + + # --------------------------------------------------------------------------- # Sol gate round 3: canonical R2 locators before bucket-cutover skips # --------------------------------------------------------------------------- From f2fc086251b80cbeb6a746677958597d54dc50bb Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:50:00 -0400 Subject: [PATCH 054/105] Close residual artifact path boundaries --- chronicle/artifacts.py | 89 +++++++++++++++++++++++++++++++----- chronicle/harness.py | 34 ++++++++------ docs/storage-architecture.md | 6 ++- 3 files changed, 103 insertions(+), 26 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index adefbed9..eceaca05 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -100,13 +100,8 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: The name is a filename, not a path: it selects among the manifests a package directory keeps, and must not reach outside it. """ - name = manifest_filename.strip() - if ( - not name - or name in (".", "..") - or name != Path(name).name - or any(character in name for character in "*?[]") - ): + name = str(manifest_filename) + if not is_bare_filename(name) or any(character in name for character in "*?[]"): raise ManifestNameError( "Manifest must name a file inside the package directory, not " f"{manifest_filename!r}." @@ -158,11 +153,30 @@ def _package_manifests( fetch selected. A malformed sibling is therefore a pre-I/O refusal: until Chronicle can read every owner, it cannot safely overwrite shared bytes. """ + paths = package_manifest_paths(output) + by_name: dict[str, Path] = {} + for path in paths: + key = filename_key(path.name) + previous = by_name.get(key) + if previous is not None and previous != path: + raise AmbiguousManifestError( + f"{previous} and {path} have the same normalized manifest name " + f"{key!r}. Physically distinct manifest aliases can hide one " + "another's registrations; keep exactly one spelling." + ) + by_name[key] = path + selected_alias = by_name.get(filename_key(manifest_path.name)) + if selected_alias is not None and selected_alias != manifest_path: + raise AmbiguousManifestError( + f"{manifest_path} and existing {selected_alias} have the same " + "normalized manifest name. Selecting one spelling would hide the " + "other's registrations; address the existing manifest or remove " + "the duplicate." + ) + manifests: dict[str, dict[str, Any]] = {str(manifest_path): existing_manifest} - for path in package_manifest_paths(output): - if path == manifest_path or filename_key(path.name) == filename_key( - manifest_path.name - ): + for path in paths: + if path == manifest_path: continue sibling = _read_manifest(path) _manifest_files(sibling, path) @@ -796,6 +810,26 @@ def fetch_source_artifact( owners = _manifest_file_owners(manifests, filename=artifact_filename) _assert_shared_owner_identities_agree(owners, filename=artifact_filename) + existing_target = matching_directory_entry(output, artifact_filename) + if existing_target is not None: + if existing_target.is_symlink(): + raise ArtifactFilenameError( + f"{existing_target} is a symbolic link. Chronicle will not " + "fetch through a package-local link or overwrite its target." + ) + if existing_target.name != artifact_filename: + raise ArtifactFilenameError( + f"{existing_target} has the same normalized filename as " + f"{artifact_filename!r}. Chronicle will not create a " + "physically distinct alias; pass --filename " + f"{existing_target.name!r}." + ) + if not existing_target.is_file(): + raise ArtifactFilenameError( + f"{existing_target} exists but is not a regular file. " + "Chronicle will not overwrite it with publisher bytes." + ) + fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() content, _inferred_filename = _read_artifact(source_url) @@ -1047,6 +1081,7 @@ def publish_source_artifacts( """Upload manifest-declared raw source artifacts and record R2 locations.""" r2_bucket = r2_bucket or default_r2_raw_bucket() root_path = Path(root) + _manifest_path(Path(), manifest_filename) if not root_path.exists(): return RawArtifactPublishReport( root=str(root_path), @@ -1148,6 +1183,7 @@ def inventory_source_artifacts( ) -> ArtifactInventoryReport: """Inventory manifest-declared source artifacts under a root directory.""" root_path = Path(root) + _manifest_path(Path(), manifest_filename) errors: list[str] = [] entries: list[ArtifactInventoryEntry] = [] if not root_path.exists(): @@ -2228,6 +2264,23 @@ def _publish_raw_manifest_entry( ), None, ) + if is_manifest_filename(filename): + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(manifest_path.parent), + sha256=None, + size_bytes=None, + r2_location=None, + upload=None, + errors=(f"manifest_named_filename:{filename}",), + ), + None, + ) artifact_path = ( matching_directory_entry(manifest_path.parent, filename) or manifest_path.parent / filename @@ -2425,6 +2478,20 @@ def _inventory_entry( r2=r2, errors=(f"non_canonical_filename:{filename}",), ) + if is_manifest_filename(filename): + return ArtifactInventoryEntry( + manifest_path=str(manifest_path), + year=str(year), + filename=filename, + local_path=str(manifest_path.parent), + exists=False, + sha256_expected=spec.get("sha256"), + sha256_actual=None, + size_bytes=None, + source_url=spec.get("source_url"), + r2=r2, + errors=(f"manifest_named_filename:{filename}",), + ) artifact_path = ( matching_directory_entry(manifest_path.parent, filename) or manifest_path.parent / filename diff --git a/chronicle/harness.py b/chronicle/harness.py index ee7f6ef9..a100bea4 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -1389,22 +1389,30 @@ def main(argv: list[str] | None = None) -> int: print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 if args.command == "inventory-artifacts": - report = inventory_artifact_files( - args.root, - manifest_filename=args.manifest, - ) + try: + report = inventory_artifact_files( + args.root, + manifest_filename=args.manifest, + ) + except SourceArtifactManifestError as error: + print(f"error: {error}", file=sys.stderr) + return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 if args.command == "publish-raw": - report = publish_raw_artifact_files( - args.root, - manifest_filename=args.manifest, - source_id=args.source_id, - package_id=args.package_id, - r2_bucket=args.r2_bucket, - r2_prefix=args.r2_prefix, - wrangler_command=args.wrangler_command, - ) + try: + report = publish_raw_artifact_files( + args.root, + manifest_filename=args.manifest, + source_id=args.source_id, + package_id=args.package_id, + r2_bucket=args.r2_bucket, + r2_prefix=args.r2_prefix, + wrangler_command=args.wrangler_command, + ) + except SourceArtifactManifestError as error: + print(f"error: {error}", file=sys.stderr) + return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 if args.command == "bootstrap-r2": diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 8a85e1ca..ca770ab8 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -69,8 +69,10 @@ raw/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{sha256}/working-f The implemented country segments are `nz` and `uk`. US objects deliberately retain the legacy shape `raw/{source_id}/...`; migrating those keys requires a separate consumer audit. The fetch and raw-publish commands infer the country -from the package publisher directory. Raw publication refuses to replace a -manifest-recorded key that disagrees with the inferred country path. +from the package publisher directory for new objects. A manifest-recorded raw +object is preserved as history when its content-addressed checksum and filename +tail identify the local bytes, including legacy routes that predate the country +prefix and publisher-explicit routes such as Statbel's 2023 snapshots. New UK and New Zealand derived build artifacts use the same country segment and build-scoped keys so different builds can coexist and be audited: From 26d22e4679db51295e7431ee88bc0b08939f909b Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:50:16 -0400 Subject: [PATCH 055/105] Record residual path boundary hardening --- PROGRESS.md | 27 ++++++++++++++++++++++----- 1 file changed, 22 insertions(+), 5 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index c736d6a0..b9bcb8a7 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -419,11 +419,28 @@ Running the same operations against a checkout of the previous head: through `chronicle_schema()` / `targets_schema()`. Fix commit: `f41ad0f`; no #227 hunk applies. The focused regression now exits 0 with 3 passed; it and the runtime/default namespace controls exit 0 with 11 passed. +- **Adversarial path-boundary follow-up reproduced and fixed.** Test-only + commit `465a34c` showed normalized manifest aliases and a package-local + artifact symlink reaching the publisher-read sentinel; invalid sweep names + were accepted when the root was absent and escaped the CLI as tracebacks; + manifest-named artifacts were not refused; and publish could upload a valid + first entry before discovering an invalid later entry. The same audit also + reproduced contradictory owners across sibling manifests being ignored by + both sweeps. +- Fix commit `8e766ad` ports #227 commit `235c616`'s corrected + `_package_manifests` normalized-alias guard, validates fetch destinations + before publisher I/O, rejects manifest-named artifacts in publish and + inventory, validates sweep selectors before checking the root, and reports + those refusals cleanly from both CLIs. It also corrects the object-key docs + to describe compatible recorded history. The six focused path/CLI cases now + pass. Complete-package publish preflight and sibling-owner validation remain + the next coherent fix. ### Next -1. Audit the eight findings and #227 port surface for missed CLI/refusal cases, - then run focused and full verification. -2. Commit each coherent red-test and implementation step, updating this log. -3. Run focused tests, lint/format checks on changed files, and the full suite - with the directly captured exit code and counts. +1. Port #227's complete-manifest preflight shape and apply shared-owner + consistency checks to publish and inventory sweeps. +2. Run the complete artifact module, then the requested lint/format checks and + full suite with the directly captured exit code and counts. +3. Write the external `-o out.md` report with the per-finding command, + observation, fix, regression, commit, and port provenance. From 619d5cd07511b68914d8282ce78bf82da1c12534 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:53:49 -0400 Subject: [PATCH 056/105] Preflight complete packages before raw publish --- chronicle/artifacts.py | 122 ++++++++++++++++++++++++++++-- chronicle/registration.py | 36 ++++++++- tests/test_chronicle_artifacts.py | 56 ++++++++++++++ 3 files changed, 208 insertions(+), 6 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index eceaca05..a873b330 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -33,6 +33,7 @@ load_manifest_document, matching_directory_entry, package_manifest_paths, + validate_package_directory, ) @@ -1095,13 +1096,45 @@ def publish_source_artifacts( try: manifest = _read_manifest(manifest_path) files = _manifest_files(manifest, manifest_path) - except (OSError, MalformedManifestError) as exc: + package_manifests = _package_manifests( + manifest_path.parent, manifest_path, manifest + ) + _assert_package_file_owner_identities_agree(package_manifests) + except (OSError, SourceArtifactManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue manifest_source_id = str(source_id or manifest.get("source_id") or "") manifest_package_id = str(package_id or manifest.get("package_id") or "") + preflight_failures: list[RawArtifactPublishEntry] = [] + for package_manifest_name, package_manifest in package_manifests.items(): + package_manifest_path = Path(package_manifest_name) + package_source_id = str( + source_id or package_manifest.get("source_id") or "" + ) + package_id_value = str( + package_id or package_manifest.get("package_id") or "" + ) + package_files = _manifest_files(package_manifest, package_manifest_path) + for year, spec in package_files.items(): + entry, _updated_spec = _publish_raw_manifest_entry( + package_manifest_path, + package_source_id, + package_id_value, + year, + spec, + r2_bucket=r2_bucket, + r2_prefix=r2_prefix, + wrangler_command=wrangler_command, + preflight_only=True, + ) + if entry.errors: + preflight_failures.append(entry) + if preflight_failures: + entries.extend(preflight_failures) + continue + updated = False for year, spec in files.items(): entry, updated_spec = _publish_raw_manifest_entry( @@ -1200,19 +1233,23 @@ def inventory_source_artifacts( errors=(f"Root does not exist: {root_path}",), ) - manifests = _root_manifest_paths(root_path, manifest_filename) - for manifest_path in manifests: + manifest_paths = _root_manifest_paths(root_path, manifest_filename) + for manifest_path in manifest_paths: try: manifest = _read_manifest(manifest_path) files = _manifest_files(manifest, manifest_path) - except (OSError, MalformedManifestError) as exc: + package_manifests = _package_manifests( + manifest_path.parent, manifest_path, manifest + ) + _assert_package_file_owner_identities_agree(package_manifests) + except (OSError, SourceArtifactManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue for year, spec in files.items(): entries.append(_inventory_entry(manifest_path, year, spec)) counts = { - "manifest_count": len(manifests), + "manifest_count": len(manifest_paths), "artifact_count": len(entries), "missing_count": sum(1 for entry in entries if not entry.exists), "checksum_mismatch_count": sum( @@ -1876,6 +1913,63 @@ def _assert_shared_owner_identities_agree( ) +def _assert_package_file_owner_identities_agree( + manifests: Mapping[str, dict[str, Any]], +) -> None: + """Refuse contradictory identities for any package-local filename. + + Publish and inventory sweep a manifest at a time, but the physical byte is + shared by every manifest in its directory. Validate every identified owner + as one package boundary before a selected manifest can upload anything. + Entry-shape and local-file errors remain the per-entry preflight's job. + """ + collision_codes = validate_package_directory(manifests) + if collision_codes: + raise SourceArtifactManifestError( + "Package manifests identify different bytes for one package-local " + f"filename: {', '.join(collision_codes)}. Reconcile the manifests " + "before publishing or inventorying that directory." + ) + + owners_by_filename: dict[str, list[_ManifestFileOwner]] = {} + display_names: dict[str, str] = {} + for name, payload in manifests.items(): + manifest_path = Path(name) + for vintage, spec in _manifest_files(payload, manifest_path).items(): + if not isinstance(spec, dict): + continue + recorded_name = spec.get("filename") + if not is_bare_filename(recorded_name): + continue + try: + identity = _recorded_identity( + spec, + manifest_path=manifest_path, + year=vintage, + ) + except SourceArtifactManifestError: + # The complete per-entry preflight reports the precise locator + # or history error without letting another entry upload first. + continue + if identity is None: + continue + key = filename_key(recorded_name) + display_names.setdefault(key, str(recorded_name)) + owners_by_filename.setdefault(key, []).append( + _ManifestFileOwner( + manifest_path=manifest_path, + vintage=vintage, + spec=spec, + identity=identity, + ) + ) + for key, owners in owners_by_filename.items(): + _assert_shared_owner_identities_agree( + owners, + filename=display_names[key], + ) + + def _assert_siblings_record_these_bytes( manifests: Mapping[str, dict[str, Any]], *, @@ -2241,6 +2335,7 @@ def _publish_raw_manifest_entry( r2_bucket: str, r2_prefix: str | None, wrangler_command: str, + preflight_only: bool = False, ) -> tuple[RawArtifactPublishEntry, dict[str, Any] | None]: errors: list[str] = [] if not isinstance(spec, dict): @@ -2413,6 +2508,23 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: package_path=manifest_path, ), ) + if preflight_only: + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(artifact_path), + sha256=sha256_actual, + size_bytes=size_bytes, + r2_location=location, + upload=None, + errors=(), + ), + None, + ) upload = _upload_r2_object( location, artifact_path, diff --git a/chronicle/registration.py b/chronicle/registration.py index 2315024a..063af7d3 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -10,7 +10,7 @@ from pathlib import Path import re -from typing import Any +from typing import Any, Mapping import unicodedata import yaml @@ -76,6 +76,39 @@ def package_manifest_paths(package_dir: Path) -> list[Path]: ) +def validate_package_directory( + manifests: Mapping[str, Mapping[str, Any] | None], +) -> tuple[str, ...]: + """Return filename-identity collisions across a package's manifests. + + Two manifests may name one physical file only when they record the same + digest. A differing digest means the same package-local bytes have two + incompatible identities, so no command may act through either record. + """ + by_name: dict[str, list[tuple[str, str]]] = {} + for name, manifest in manifests.items(): + files = manifest.get("files") if isinstance(manifest, Mapping) else None + if not isinstance(files, Mapping): + continue + for entry in files.values(): + if not isinstance(entry, Mapping): + continue + filename = entry.get("filename") + if filename is None: + continue + digest = entry.get("sha256") + digest = digest.strip() if isinstance(digest, str) else "" + by_name.setdefault(filename_key(filename), []).append((name, digest)) + + errors: list[str] = [] + for key, records in by_name.items(): + if len({name for name, _digest in records}) < 2: + continue + if len({digest for _name, digest in records}) > 1: + errors.append(f"filename_collision_across_manifests:{key}") + return tuple(dict.fromkeys(errors)) + + def matching_directory_entry(directory: Any, filename: Any) -> Any | None: """Return the actual directory entry matching a bare filename's safe key. @@ -156,4 +189,5 @@ def load_manifest_document(text: str) -> Any: "load_manifest_document", "matching_directory_entry", "package_manifest_paths", + "validate_package_directory", ] diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 1341746b..87ef8f2d 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -2969,6 +2969,62 @@ def non_writing_uploader(location, local_path, *, wrangler_command): assert manifest_path.read_bytes() == before +def test_publish_preflights_every_sibling_manifest_before_any_upload( + tmp_path, monkeypatch +): + package = tmp_path / "data" / "package" + package.mkdir(parents=True) + content = b"publisher table" + (package / "table.csv").write_bytes(content) + manifests = { + package / "manifest_a.yaml": { + "source_id": "publisher", + "package_id": "package-a", + "files": { + 2024: { + "filename": "table.csv", + "sha256": hashlib.sha256(content).hexdigest(), + } + }, + }, + package / "manifest_b.yaml": { + "source_id": "publisher", + "package_id": "package-b", + "files": { + 2024: { + "filename": "manifest.yaml", + "sha256": hashlib.sha256(b"not a manifest").hexdigest(), + } + }, + }, + } + for path, payload in manifests.items(): + path.write_text(yaml.safe_dump(payload, sort_keys=False)) + before = {path: path.read_bytes() for path in manifests} + uploads = [] + + def non_writing_uploader(location, local_path, *, wrangler_command): + uploads.append((location, local_path, wrangler_command)) + return ArtifactCommandResult( + command=("non-writing-uploader",), + returncode=0, + stdout="", + stderr="", + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", non_writing_uploader) + + report = publish_source_artifacts(package) + + assert not report.valid + assert uploads == [] + assert any( + "manifest_named_filename:manifest.yaml" in entry.errors + for entry in report.entries + ) + assert {path: path.read_bytes() for path in manifests} == before + + def test_sweeps_refuse_conflicting_owners_across_package_manifests( tmp_path, monkeypatch ): From 3db6c852ef34085e5e8bdc537e9966f1a6343009 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:55:48 -0400 Subject: [PATCH 057/105] Reproduce source-package artifact path escapes --- tests/test_chronicle_source_package.py | 48 ++++++++++++++++++++++++++ 1 file changed, 48 insertions(+) diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index ae6f0166..bc1dd479 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -1083,6 +1083,54 @@ def test_source_artifact_loader_fetches_missing_artifact_when_enabled( assert _source_artifact_cache_path(spec).read_bytes() == content +@pytest.mark.parametrize("path_kind", ["absolute", "parent", "symlink"]) +def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( + tmp_path, monkeypatch, path_kind +): + resource_root = tmp_path / "resources" + resource_dir = resource_root / "data" / "publisher" / "package" + resource_dir.mkdir(parents=True) + outside = resource_dir.parent / "outside.csv" + outside.write_bytes(b"outside publisher bytes") + if path_kind == "absolute": + filename = str(outside) + elif path_kind == "parent": + filename = "../outside.csv" + else: + filename = "table.csv" + (resource_dir / filename).symlink_to(outside) + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "files": { + 2024: { + "filename": filename, + "source_url": outside.as_uri(), + "sha256": hashlib.sha256(outside.read_bytes()).hexdigest(), + } + } + }, + sort_keys=False, + ) + ) + monkeypatch.setattr("chronicle.source_package.files", lambda _package: resource_root) + artifact = SourceArtifactSpec( + source_name="publisher", + source_table="Table", + resource_package="test_resources", + resource_directory="data/publisher/package", + manifest="manifest.yaml", + vintage="2024", + extracted_at="2026-09-04", + extraction_method="test", + artifact_year=2024, + ) + + message = "symbolic link" if path_kind == "symlink" else "bare filename" + with pytest.raises(ValueError, match=message): + artifact._artifact_content(2024) + + def test_source_package_path_builds_valid_soi_table_1_4_facts(): package_path = REPO_ROOT / "packages" / "irs_soi" / "table_1_4" package = load_source_package(package_path) From 531de1c5b1a0a6ebedf2db8ed05ad31c4bc83162 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:56:10 -0400 Subject: [PATCH 058/105] Reproduce normalized source artifact alias --- tests/test_chronicle_source_package.py | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index bc1dd479..b169a6d7 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -1083,7 +1083,9 @@ def test_source_artifact_loader_fetches_missing_artifact_when_enabled( assert _source_artifact_cache_path(spec).read_bytes() == content -@pytest.mark.parametrize("path_kind", ["absolute", "parent", "symlink"]) +@pytest.mark.parametrize( + "path_kind", ["absolute", "parent", "symlink", "normalized-alias"] +) def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( tmp_path, monkeypatch, path_kind ): @@ -1098,7 +1100,10 @@ def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( filename = "../outside.csv" else: filename = "table.csv" - (resource_dir / filename).symlink_to(outside) + if path_kind == "symlink": + (resource_dir / filename).symlink_to(outside) + else: + (resource_dir / "TABLE.csv").write_bytes(outside.read_bytes()) (resource_dir / "manifest.yaml").write_text( yaml.safe_dump( { @@ -1126,7 +1131,12 @@ def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( artifact_year=2024, ) - message = "symbolic link" if path_kind == "symlink" else "bare filename" + if path_kind == "symlink": + message = "symbolic link" + elif path_kind == "normalized-alias": + message = "normalized filename" + else: + message = "bare filename" with pytest.raises(ValueError, match=message): artifact._artifact_content(2024) From 70b2d69689f22cd6f4bc21cf338f1a6cd529ffa7 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:56:49 -0400 Subject: [PATCH 059/105] Reproduce manifest-named source artifact read --- tests/test_chronicle_source_package.py | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index b169a6d7..53925c82 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -1084,7 +1084,8 @@ def test_source_artifact_loader_fetches_missing_artifact_when_enabled( @pytest.mark.parametrize( - "path_kind", ["absolute", "parent", "symlink", "normalized-alias"] + "path_kind", + ["absolute", "parent", "symlink", "normalized-alias", "manifest-name"], ) def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( tmp_path, monkeypatch, path_kind @@ -1102,8 +1103,11 @@ def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( filename = "table.csv" if path_kind == "symlink": (resource_dir / filename).symlink_to(outside) - else: + elif path_kind == "normalized-alias": (resource_dir / "TABLE.csv").write_bytes(outside.read_bytes()) + else: + filename = "manifest_artifact.yaml" + (resource_dir / filename).write_bytes(outside.read_bytes()) (resource_dir / "manifest.yaml").write_text( yaml.safe_dump( { @@ -1135,6 +1139,8 @@ def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( message = "symbolic link" elif path_kind == "normalized-alias": message = "normalized filename" + elif path_kind == "manifest-name": + message = "manifest name" else: message = "bare filename" with pytest.raises(ValueError, match=message): From bd49f520d5e31ac67f4e9e0e31cc8cdf82a2cb6e Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 11:57:57 -0400 Subject: [PATCH 060/105] Reproduce source-package manifest path escapes --- tests/test_chronicle_source_package.py | 59 ++++++++++++++++++++++++++ 1 file changed, 59 insertions(+) diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index 53925c82..c7ecb800 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -1147,6 +1147,65 @@ def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( artifact._artifact_content(2024) +@pytest.mark.parametrize( + "path_kind", ["absolute", "parent", "symlink", "normalized-alias", "unsupported"] +) +def test_source_artifact_spec_refuses_unsafe_manifest_path_before_artifact_read( + tmp_path, monkeypatch, path_kind +): + resource_root = tmp_path / "resources" + resource_dir = resource_root / "data" / "publisher" / "package" + resource_dir.mkdir(parents=True) + payload = { + "files": { + 2024: { + "filename": "table.csv", + "source_url": "https://example.test/table.csv", + } + } + } + outside_manifest = resource_dir.parent / "outside-manifest.yaml" + outside_manifest.write_text(yaml.safe_dump(payload, sort_keys=False)) + if path_kind == "absolute": + manifest_name = str(outside_manifest) + elif path_kind == "parent": + manifest_name = "../outside-manifest.yaml" + elif path_kind == "symlink": + manifest_name = "manifest.yaml" + (resource_dir / manifest_name).symlink_to(outside_manifest) + elif path_kind == "normalized-alias": + manifest_name = "manifest.yaml" + (resource_dir / "Manifest.yaml").write_text( + yaml.safe_dump(payload, sort_keys=False) + ) + else: + manifest_name = "registry.yaml" + (resource_dir / manifest_name).write_text(yaml.safe_dump(payload)) + monkeypatch.setattr("chronicle.source_package.files", lambda _package: resource_root) + + def unexpected_artifact_read(_artifact_path, _spec): + raise AssertionError("unsafe manifest path reached artifact I/O") + + monkeypatch.setattr( + "chronicle.source_package._read_source_artifact_content", + unexpected_artifact_read, + ) + artifact = SourceArtifactSpec( + source_name="publisher", + source_table="Table", + resource_package="test_resources", + resource_directory="data/publisher/package", + manifest=manifest_name, + vintage="2024", + extracted_at="2026-09-04", + extraction_method="test", + artifact_year=2024, + ) + + with pytest.raises(ValueError): + artifact._artifact_content(2024) + + def test_source_package_path_builds_valid_soi_table_1_4_facts(): package_path = REPO_ROOT / "packages" / "irs_soi" / "table_1_4" package = load_source_package(package_path) From f431ae718c32155547dd74ca82bb582eddb35273 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 12:10:32 -0400 Subject: [PATCH 061/105] Harden source-package artifact paths --- PROGRESS.md | 17 +++-- chronicle/source_package.py | 96 ++++++++++++++++++++++---- tests/test_chronicle_source_package.py | 20 ++++-- 3 files changed, 112 insertions(+), 21 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index b9bcb8a7..ba585198 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -435,12 +435,21 @@ Running the same operations against a checkout of the previous head: to describe compatible recorded history. The six focused path/CLI cases now pass. Complete-package publish preflight and sibling-owner validation remain the next coherent fix. +- **Source-package reader follow-up reproduced and fixed.** Test-only commits + `92af48c`, `9a51d99`, `3228238`, and `578c8b2` showed that a package spec + could read an absolute or parent-traversing artifact, follow artifact and + manifest symlinks, accept a normalized filename alias or manifest-named + artifact, and select an unsafe/unsupported manifest path before artifact + I/O. `SourceArtifactSpec` now resolves both manifest and artifact resources + through the shared #227 filename-identity helpers before opening either. + The ten focused cases pass, and the full source-package module passes with + 135 tests (13 warnings). ### Next -1. Port #227's complete-manifest preflight shape and apply shared-owner - consistency checks to publish and inventory sweeps. -2. Run the complete artifact module, then the requested lint/format checks and - full suite with the directly captured exit code and counts. +1. Reproduce and close the audit finding that a bad later package in a root + publish sweep can be discovered only after an earlier package uploads. +2. Finish the #227 port/adversarial audit, then run the complete artifact + module and requested lint/format/full-suite verification. 3. Write the external `-o out.md` report with the per-finding command, observation, fix, regression, commit, and port provenance. diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 60e4ff87..9095b612 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -33,7 +33,12 @@ ) from chronicle.env import env_flag, env_value from chronicle.epoch import SCHEMA_IDS, schema_id -from chronicle.registration import load_manifest_document +from chronicle.registration import ( + is_bare_filename, + is_manifest_filename, + load_manifest_document, + matching_directory_entry, +) from chronicle.sources.cells import ( SourceArtifactMetadata, SourceCell, @@ -869,20 +874,87 @@ def _source_artifact_metadata( raw_r2_uri=raw_r2.get("uri"), ) + def _resource_entry( + self, + value: Any, + *, + what: str, + require_manifest_name: bool = False, + forbid_manifest_name: bool = False, + ) -> Any: + """Resolve one safe file entry under the package resource directory.""" + if not is_bare_filename(value): + raise ValueError( + f"{what} must be a bare filename inside " + f"{self.resource_directory}, not {value!r}." + ) + name = str(value) + if require_manifest_name and not is_manifest_filename(name): + raise ValueError( + f"{what} must be named manifest.yaml or " + f"manifest_.yaml, not {name!r}." + ) + if forbid_manifest_name and is_manifest_filename(name): + raise ValueError( + f"{what} {name!r} is a manifest name and cannot be read as " + "source artifact bytes." + ) + + directory = files(self.resource_package).joinpath(self.resource_directory) + existing = matching_directory_entry(directory, name) + if existing is None: + return directory.joinpath(name) + is_symlink = getattr(existing, "is_symlink", None) + if callable(is_symlink) and is_symlink(): + raise ValueError( + f"{what} {existing} is a symbolic link. Chronicle will not " + "read source-package data through it." + ) + if existing.name != name: + raise ValueError( + f"{what} {existing} has the same normalized filename as " + f"{name!r}. Keep exactly one spelling in the package." + ) + return existing + + def manifest_resource(self) -> Any: + """Return the validated manifest file this package spec points at.""" + return self._resource_entry( + self.manifest, + what="Source artifact manifest", + require_manifest_name=True, + ) + + def manifest_payload(self) -> dict[str, Any]: + """Load the artifact manifest strictly as a YAML mapping.""" + with self.manifest_resource().open("r", encoding="utf-8") as file: + text = file.read() + try: + payload = load_manifest_document(text) + except yaml.YAMLError as exc: + raise ValueError( + f"{self.resource_directory}/{self.manifest} is not valid YAML: {exc}" + ) from exc + if payload is None: + return {} + if not isinstance(payload, dict): + raise ValueError( + f"{self.resource_directory}/{self.manifest} must be a YAML " + f"mapping; it parses as a {type(payload).__name__}." + ) + return payload + def _artifact_content( self, year: int, ) -> tuple[bytes, str, str, dict[str, str]]: - manifest_path = files(self.resource_package).joinpath( - self.resource_directory, - self.manifest, - ) - with manifest_path.open("r", encoding="utf-8") as file: - manifest = load_manifest_document(file.read()) + manifest = self.manifest_payload() spec = _year_mapping(manifest["files"], self.artifact_year or year) - artifact_path = files(self.resource_package).joinpath( - self.resource_directory, - spec["filename"], + filename = spec.get("filename") + artifact_path = self._resource_entry( + filename, + what="Source artifact filename", + forbid_manifest_name=True, ) content = _read_source_artifact_content(artifact_path, spec) expected_sha = spec.get("sha256") @@ -890,11 +962,11 @@ def _artifact_content( _validate_source_artifact_sha( content, expected_sha=str(expected_sha), - filename=str(spec["filename"]), + filename=str(filename), ) storage = spec.get("storage") if isinstance(spec, dict) else None raw_r2 = storage.get("r2") if isinstance(storage, dict) else {} - return content, spec["filename"], spec["source_url"], raw_r2 or {} + return content, str(filename), spec["source_url"], raw_r2 or {} def _sheet_name(self, filename: str, *, year: int) -> str: if self.sheet_name: diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index c7ecb800..a6ef90da 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -244,9 +244,7 @@ def test_hmrc_cgt_reuses_one_publisher_series_across_definition_years(): package = load_source_package("hmrc-cgt-statistics-2026") package_facts = package.build_facts(2026) facts = [ - fact - for fact in package_facts - if fact.measure.concept == "hmrc.cgt_tax_total" + fact for fact in package_facts if fact.measure.concept == "hmrc.cgt_tax_total" ] assert {fact.entity.name for fact in package_facts} == {"tax_unit"} @@ -1122,7 +1120,17 @@ def test_source_artifact_spec_refuses_unsafe_manifest_filename_before_read( sort_keys=False, ) ) - monkeypatch.setattr("chronicle.source_package.files", lambda _package: resource_root) + monkeypatch.setattr( + "chronicle.source_package.files", lambda _package: resource_root + ) + + def unexpected_artifact_read(_artifact_path, _spec): + raise AssertionError("unsafe artifact path reached artifact I/O") + + monkeypatch.setattr( + "chronicle.source_package._read_source_artifact_content", + unexpected_artifact_read, + ) artifact = SourceArtifactSpec( source_name="publisher", source_table="Table", @@ -1181,7 +1189,9 @@ def test_source_artifact_spec_refuses_unsafe_manifest_path_before_artifact_read( else: manifest_name = "registry.yaml" (resource_dir / manifest_name).write_text(yaml.safe_dump(payload)) - monkeypatch.setattr("chronicle.source_package.files", lambda _package: resource_root) + monkeypatch.setattr( + "chronicle.source_package.files", lambda _package: resource_root + ) def unexpected_artifact_read(_artifact_path, _spec): raise AssertionError("unsafe manifest path reached artifact I/O") From 446bd17983ad652a5ecab2980802f02c6af6e162 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 12:13:37 -0400 Subject: [PATCH 062/105] Reproduce residual artifact side effects --- PROGRESS.md | 12 +++- tests/test_chronicle_artifacts.py | 92 +++++++++++++++++++++++++++++++ 2 files changed, 101 insertions(+), 3 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index ba585198..958c6641 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -443,12 +443,18 @@ Running the same operations against a checkout of the previous head: I/O. `SourceArtifactSpec` now resolves both manifest and artifact resources through the shared #227 filename-identity helpers before opening either. The ten focused cases pass, and the full source-package module passes with - 135 tests (13 warnings). + 135 tests (13 warnings). Fix/journal commit: `64fe94b`. +- **Residual side-effect ordering reproduced.** The focused command covering + `test_fetch_refuses_invalid_r2_identity_before_publisher_io` and + `test_publish_preflights_entire_root_before_any_upload` exited 1 with three + failures. Both slash-only R2 identity fields reached the publisher-read + sentinel; a root sweep uploaded and rewrote the valid `a_good` package before + reporting the unsafe filename in `z_bad`. ### Next -1. Reproduce and close the audit finding that a bad later package in a root - publish sweep can be discovered only after an earlier package uploads. +1. Validate raw-key identity before publisher I/O and make publish's preflight + cover the complete selected root before any upload or manifest rewrite. 2. Finish the #227 port/adversarial audit, then run the complete artifact module and requested lint/format/full-suite verification. 3. Write the external `-o out.md` report with the per-finding command, diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 87ef8f2d..ece20809 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1734,6 +1734,39 @@ def unexpected_read(_source_url): assert manifest_path.read_bytes() == before +@pytest.mark.parametrize( + ("source_id", "package_id"), + [ + pytest.param("/", "package", id="source-id"), + pytest.param("publisher", "/", id="package-id"), + ], +) +def test_fetch_refuses_invalid_r2_identity_before_publisher_io( + tmp_path, monkeypatch, source_id, package_id +): + package = tmp_path / "db" / "data" / "publisher" / "package" + package.mkdir(parents=True) + artifact_path = package / "table.csv" + artifact_path.write_bytes(b"registered publisher bytes") + + def unexpected_read(_source_url): + raise AssertionError("an invalid R2 identity reached publisher I/O") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(ValueError, match="R2 key parts cannot be empty"): + fetch_source_artifact( + "https://publisher.test/table.csv", + source_id=source_id, + package_id=package_id, + year=2024, + output_dir=package, + ) + + assert artifact_path.read_bytes() == b"registered publisher bytes" + assert not (package / "manifest.yaml").exists() + + def test_manifest_name_must_be_discoverable_before_publisher_io(tmp_path, monkeypatch): package = tmp_path / "db" / "data" / "irs_soi" / "soi-table" source = _publish(tmp_path, "table.csv", b"publisher table") @@ -3025,6 +3058,65 @@ def non_writing_uploader(location, local_path, *, wrangler_command): assert {path: path.read_bytes() for path in manifests} == before +def test_publish_preflights_entire_root_before_any_upload(tmp_path, monkeypatch): + root = tmp_path / "data" + good_package = root / "a_good" + bad_package = root / "z_bad" + good_package.mkdir(parents=True) + bad_package.mkdir(parents=True) + good_content = b"good publisher table" + bad_content = b"bad publisher table" + (good_package / "good.csv").write_bytes(good_content) + (bad_package / "bad.csv").write_bytes(bad_content) + manifests = { + good_package / "manifest.yaml": { + "source_id": "publisher", + "package_id": "good-package", + "files": { + 2024: { + "filename": "good.csv", + "sha256": hashlib.sha256(good_content).hexdigest(), + } + }, + }, + bad_package / "manifest.yaml": { + "source_id": "publisher", + "package_id": "bad-package", + "files": { + 2024: { + "filename": "../bad.csv", + "sha256": hashlib.sha256(bad_content).hexdigest(), + } + }, + }, + } + for path, payload in manifests.items(): + path.write_text(yaml.safe_dump(payload, sort_keys=False)) + before = {path: path.read_bytes() for path in manifests} + uploads = [] + + def non_writing_uploader(location, local_path, *, wrangler_command): + uploads.append((location, local_path, wrangler_command)) + return ArtifactCommandResult( + command=("non-writing-uploader",), + returncode=0, + stdout="", + stderr="", + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", non_writing_uploader) + + report = publish_source_artifacts(root) + + assert not report.valid + assert uploads == [] + assert any( + "non_canonical_filename:../bad.csv" in entry.errors + for entry in report.entries + ) + assert {path: path.read_bytes() for path in manifests} == before + + def test_sweeps_refuse_conflicting_owners_across_package_manifests( tmp_path, monkeypatch ): From 6f74c4f08bd9babb8ab415001db3275dd61fa806 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 12:14:39 -0400 Subject: [PATCH 063/105] Preflight artifact operations before side effects --- PROGRESS.md | 11 ++++++--- chronicle/artifacts.py | 40 +++++++++++++++++++++++++++---- tests/test_chronicle_artifacts.py | 3 +-- 3 files changed, 45 insertions(+), 9 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 958c6641..26418d19 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -450,12 +450,17 @@ Running the same operations against a checkout of the previous head: failures. Both slash-only R2 identity fields reached the publisher-read sentinel; a root sweep uploaded and rewrote the valid `a_good` package before reporting the unsafe filename in `z_bad`. +- Fetch now validates the source/package object-key components before reading + publisher bytes. Raw publish now completes one root-wide read/identity/entry + preflight and returns every refusal before its first uploader call or + manifest rewrite. The three red cases plus the existing entry-, sibling-, + and tracked-cutover preflight controls pass (9 tests). ### Next -1. Validate raw-key identity before publisher I/O and make publish's preflight - cover the complete selected root before any upload or manifest rewrite. -2. Finish the #227 port/adversarial audit, then run the complete artifact +1. Finish the #227 port/adversarial audit, including normalized-alias edge + cases and multi-owner write consistency. +2. Run the complete artifact module and requested lint/format/full-suite verification. 3. Write the external `-o out.md` report with the per-finding command, observation, fix, regression, commit, and port provenance. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index a873b330..f180e0c1 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -779,6 +779,12 @@ def fetch_source_artifact( "not be named like a manifest, which it would overwrite; pass " "--filename with the publisher's name for the bytes." ) + # These fields become object-key path segments even when this fetch does + # not upload. Validate them before reading the publisher so a malformed + # registration identity cannot overwrite package-local bytes and fail only + # when the prospective R2 key is constructed below. + _clean_key_part(source_id) + _clean_key_part(package_id) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, @@ -1092,6 +1098,8 @@ def publish_source_artifacts( entries: list[RawArtifactPublishEntry] = [] errors: list[str] = [] + prepared: list[tuple[Path, dict[str, Any], dict[str, Any], str, str]] = [] + preflight_failures: list[RawArtifactPublishEntry] = [] for manifest_path in _root_manifest_paths(root_path, manifest_filename): try: manifest = _read_manifest(manifest_path) @@ -1107,7 +1115,6 @@ def publish_source_artifacts( manifest_source_id = str(source_id or manifest.get("source_id") or "") manifest_package_id = str(package_id or manifest.get("package_id") or "") - preflight_failures: list[RawArtifactPublishEntry] = [] for package_manifest_name, package_manifest in package_manifests.items(): package_manifest_path = Path(package_manifest_name) package_source_id = str( @@ -1131,10 +1138,35 @@ def publish_source_artifacts( ) if entry.errors: preflight_failures.append(entry) - if preflight_failures: - entries.extend(preflight_failures) - continue + prepared.append( + ( + manifest_path, + manifest, + files, + manifest_source_id, + manifest_package_id, + ) + ) + # A root sweep is one requested publish operation. Validate every selected + # package before the first uploader call or manifest rewrite, otherwise a + # malformed later package can make the command fail after earlier packages + # have already changed external and local state. + if errors or preflight_failures: + entries.extend(preflight_failures) + return RawArtifactPublishReport( + root=str(root_path), + entries=tuple(entries), + errors=tuple(errors), + ) + + for ( + manifest_path, + manifest, + files, + manifest_source_id, + manifest_package_id, + ) in prepared: updated = False for year, spec in files.items(): entry, updated_spec = _publish_raw_manifest_entry( diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index ece20809..7202dc0c 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3111,8 +3111,7 @@ def non_writing_uploader(location, local_path, *, wrangler_command): assert not report.valid assert uploads == [] assert any( - "non_canonical_filename:../bad.csv" in entry.errors - for entry in report.entries + "non_canonical_filename:../bad.csv" in entry.errors for entry in report.entries ) assert {path: path.read_bytes() for path in manifests} == before From c4f59fe33a03ac03fedd1f4436cd4ca8d9c0b931 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 12:19:41 -0400 Subject: [PATCH 064/105] Record artifact preflight verification --- PROGRESS.md | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 26418d19..ac7eece3 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -454,13 +454,11 @@ Running the same operations against a checkout of the previous head: publisher bytes. Raw publish now completes one root-wide read/identity/entry preflight and returns every refusal before its first uploader call or manifest rewrite. The three red cases plus the existing entry-, sibling-, - and tracked-cutover preflight controls pass (9 tests). + and tracked-cutover preflight controls pass (9 tests). Fix commit: + `4991e8f`. The complete artifact module passes with 154 tests (12 warnings). ### Next -1. Finish the #227 port/adversarial audit, including normalized-alias edge - cases and multi-owner write consistency. -2. Run the complete artifact - module and requested lint/format/full-suite verification. -3. Write the external `-o out.md` report with the per-finding command, +1. Run the requested lint/format/full-suite verification. +2. Write the external `-o out.md` report with the per-finding command, observation, fix, regression, commit, and port provenance. From 7e62a1ebba39fb22a9b938129c406e011bb06a59 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 12:42:53 -0400 Subject: [PATCH 065/105] Record final eight-finding verification --- PROGRESS.md | 41 ++++++++++++++++++++++++++++++++++++++--- 1 file changed, 38 insertions(+), 3 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index ac7eece3..3ee219c6 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -459,6 +459,41 @@ Running the same operations against a checkout of the previous head: ### Next -1. Run the requested lint/format/full-suite verification. -2. Write the external `-o out.md` report with the per-finding command, - observation, fix, regression, commit, and port provenance. +None. The eight findings, required #227 ports, adversarial preflight follow-ups, +and final verification are complete. The external `-o out.md` report contains +the per-finding reproduction/fix/test/commit/port map. + +### Final verification (eight-finding round) + +- The first bare `uv run` lint invocations exited 2 before Ruff started because + the sandbox denied access to `/Users/maxghenis/.cache/uv`. Re-running through + the permitted existing cache (`UV_CACHE_DIR=/tmp/chronicle-uv-cache`) gave: + `uv run ruff check .` exit 0 (`All checks passed!`) and `uv run ruff format + --check` on the eight changed Python files exit 0 (`8 files already + formatted`). +- `UV_CACHE_DIR=/tmp/chronicle-uv-cache uv run pytest -q -p no:cacheprovider`: + direct exit 0, 1,039 passed, 7 skipped, 18 warnings in 1,323.73 seconds. +- Focused artifact module: direct exit 0, 154 passed, 12 warnings in 7.15 + seconds. Focused source-package module: direct exit 0, 135 passed, 13 warnings + in 193.89 seconds. +- `git diff c36f3fc8..HEAD -- db/data` is empty: no tracked source manifest was + modified. + +### Deliberate boundaries + +- Did not add crash-atomic multi-file transactions for an unexpected write or + process failure midway through a coordinated revision. Every deterministic + refusal is preflighted before writes/uploads and the successful path updates + all owners; true cross-file crash atomicity needs a separate staging/rollback + design and is not one of the eight findings. +- Did not port #227's `iter_directory_entries`: its exact implementation depends + on #227's microdata list-entry machinery, which this slice explicitly + excludes. Also retained the current artifact-local `bare_filename` exception + wrapper so filename refusals remain `SourceArtifactManifestError` and the + existing CLI reports them cleanly; #227's later rebase can adopt its broader + exception hierarchy together. +- Did not add a new rule for two physical artifact files whose names normalize + to the same key when the exact requested spelling sorts first. That is + defense-in-depth outside findings 3/4; there are no such collisions in the + tracked tree. Manifest-name collisions are already refused independent of + sort order. From 29336c2aba698624734e56c370f36c164a597ad1 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 17:11:07 -0400 Subject: [PATCH 066/105] Reproduce Sol round-3 findings: identity segments, alias enumeration, resource-directory containment, non-regular manifests --- tests/test_chronicle_artifacts.py | 127 +++++++++++++++++++++++++ tests/test_chronicle_source_package.py | 56 +++++++++++ 2 files changed, 183 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 7202dc0c..5eb4e721 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3289,3 +3289,130 @@ def test_publish_refuses_a_self_consistent_non_r2_locator(tmp_path, monkeypatch) assert "provider" in report.entries[0].errors[0] assert not log.exists() assert manifest_path.read_bytes() == before + + +# --------------------------------------------------------------------------- +# Sol gate round 3: identity segments, alias enumeration, non-regular manifests +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + "bad_id", + ["irs soi", "a/b", "..", " irs_soi", "irs_soi ", "a\\b", "a\tb"], +) +@pytest.mark.parametrize("field", ["source_id", "package_id"]) +def test_fetch_refuses_noncanonical_identity_segments_before_io( + tmp_path, monkeypatch, bad_id, field +): + """A registration identity that _clean_key_part would rewrite (or that + embeds separators) must be refused, never normalized into a different + R2 namespace.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "table.xlsx", b"table") + + def unexpected_read(_url): + raise AssertionError("publisher read reached with a bad identity") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + kwargs = {"source_id": "irs_soi", "package_id": "soi-table-5"} + kwargs[field] = bad_id + + with pytest.raises(SourceArtifactManifestError, match="segment"): + fetch_source_artifact( + str(source), + year=2022, + output_dir=package, + **kwargs, + ) + + assert not package.exists() + + +def test_matching_directory_entry_refuses_multiple_normalized_aliases(): + """Two physical entries sharing one normalized key are a package defect; + returning the first spelling would silently ignore the other bytes.""" + from types import SimpleNamespace + + entries = [ + SimpleNamespace(name="TABLE.CSV"), + SimpleNamespace(name="other.csv"), + SimpleNamespace(name="table.csv"), + ] + directory = SimpleNamespace( + is_dir=lambda: True, iterdir=lambda: iter(entries) + ) + + with pytest.raises(ValueError, match="TABLE.CSV.*table.csv|table.csv.*TABLE.CSV"): + matching_directory_entry(directory, "table.csv") + + assert ( + matching_directory_entry(directory, "other.csv").name == "other.csv" + ) + + +def test_publish_and_inventory_report_duplicate_artifact_aliases( + tmp_path, monkeypatch +): + """A duplicate-alias defect surfaces as an entry error, not a crash and + not a silent first-match read.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5") + _fetch_local(output_dir, source, upload_r2=False) + + def duplicate_alias(_directory, filename): + raise ValueError( + f"{filename!r} matches two physical spellings in the package." + ) + + monkeypatch.setattr( + "chronicle.artifacts.matching_directory_entry", duplicate_alias + ) + + inventory = inventory_source_artifacts(output_dir) + published = publish_source_artifacts(output_dir) + + assert not inventory.valid + assert any( + "duplicate_artifact_spellings" in error + for entry in inventory.entries + for error in entry.errors + ) + assert not published.valid + assert any( + "duplicate_artifact_spellings" in error + for entry in published.entries + for error in entry.errors + ) + + +@pytest.mark.parametrize("shape", ["dangling", "directory"]) +def test_sweeps_refuse_non_regular_manifest_entries(tmp_path, shape): + """A manifest-named entry that is not a regular file must fail the sweep + loudly instead of vanishing from it.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + target = package / "manifest.yaml" + if shape == "dangling": + target.symlink_to(package / "nowhere.yaml") + else: + target.mkdir() + + with pytest.raises(SourceArtifactManifestError, match="regular file"): + inventory_source_artifacts(tmp_path / "db" / "data") + with pytest.raises(SourceArtifactManifestError, match="regular file"): + publish_source_artifacts(tmp_path / "db" / "data") + + +def test_fetch_refuses_a_dangling_manifest_symlink_instead_of_creating_one( + tmp_path, +): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + (package / "manifest.yaml").symlink_to(package / "nowhere.yaml") + source = _publish(tmp_path, "table.xlsx", b"table") + + with pytest.raises(SourceArtifactManifestError, match="regular file"): + _fetch_local(package, source, upload_r2=False) + + assert (package / "manifest.yaml").is_symlink() + assert not (package / "table.xlsx").exists() diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index a6ef90da..41994caf 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -4417,3 +4417,59 @@ def test_build_facts_with_label_year_does_not_crash(): # Record-set periods are literal 2024, so a label build still resolves them. assert facts assert all(fact.period.value == 2024 for fact in facts) + + +@pytest.mark.parametrize( + "bad_directory", + ["/etc", "../outside", "data/../../outside", "data/./publisher"], +) +def test_resource_directory_must_stay_inside_the_resource_package( + tmp_path, monkeypatch, bad_directory +): + """`resource_directory` is joined under the resource package root; an + absolute or parent-traversing value escapes it and must be refused + before any read.""" + resource_root = tmp_path / "pkg" + resource_root.mkdir() + monkeypatch.setattr( + "chronicle.source_package.files", lambda _package: resource_root + ) + artifact = SourceArtifactSpec( + source_name="publisher", + source_table="Table", + resource_package="test_resources", + resource_directory=bad_directory, + manifest="manifest.yaml", + vintage="2024", + extracted_at="2026-09-04", + extraction_method="test", + artifact_year=2024, + ) + + with pytest.raises(ValueError, match="resource_directory"): + artifact._artifact_content(2024) + + +def test_resource_directory_refuses_a_symlinked_ancestor(tmp_path, monkeypatch): + resource_root = tmp_path / "pkg" + real = tmp_path / "elsewhere" / "package" + real.mkdir(parents=True) + (resource_root / "data").mkdir(parents=True) + (resource_root / "data" / "publisher").symlink_to(real.parent) + monkeypatch.setattr( + "chronicle.source_package.files", lambda _package: resource_root + ) + artifact = SourceArtifactSpec( + source_name="publisher", + source_table="Table", + resource_package="test_resources", + resource_directory="data/publisher/package", + manifest="manifest.yaml", + vintage="2024", + extracted_at="2026-09-04", + extraction_method="test", + artifact_year=2024, + ) + + with pytest.raises(ValueError, match="symbolic link|resource_directory"): + artifact._artifact_content(2024) From 768bed889a08228439fdcbfa991beb415e5d2890 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 17:17:39 -0400 Subject: [PATCH 067/105] Close Sol round 3: canonical identity segments, alias enumeration, resource-directory containment, non-regular manifests refused - A new registration identity (source_id, package_id) must already be one canonical R2 key segment; anything _clean_key_part would rewrite, or that embeds whitespace, a separator, or '..', is refused (IdentitySegmentError, still a ValueError) before the publisher is read, so two spellings can no longer collide in one namespace or shift the key's path shape. - matching_directory_entry enumerates every physical entry sharing the normalized key and refuses when there is more than one; fetch surfaces it as an ArtifactFilenameError, publish and inventory as the entry error duplicate_artifact_spellings, so a second alias is never silently ignored. - SourceArtifactSpec resolves resource_directory through one containment check: relative, plain segments only, no symlinked ancestor, and the resolved directory stays inside the resource package root; every manifest and artifact read goes through it. - Root sweeps and package discovery refuse a manifest-named entry that is not a regular non-symlink file (dangling symlink, directory) instead of letting it vanish; fetch refuses to create a manifest beside one. --- chronicle/artifacts.py | 106 ++++++++++++++++++++++++------ chronicle/registration.py | 37 +++++++---- chronicle/source_package.py | 47 ++++++++++++- tests/test_chronicle_artifacts.py | 24 +++---- 4 files changed, 163 insertions(+), 51 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index f180e0c1..2e2fde79 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -83,16 +83,29 @@ def _root_manifest_paths(root: Path, manifest_filename: str) -> list[Path]: """ selected_name = _manifest_path(Path(), manifest_filename).name if selected_name != DEFAULT_MANIFEST_FILENAME: - return sorted( - path - for path in root.rglob("*") - if path.is_file() and path.name == selected_name + candidates = [path for path in root.rglob("*") if path.name == selected_name] + else: + candidates = [ + path for path in root.rglob("*") if is_manifest_filename(path.name) + ] + for path in candidates: + _require_regular_manifest_file(path) + return sorted(candidates) + + +def _require_regular_manifest_file(path: Path) -> None: + """Refuse a manifest-named entry that is not a regular, non-symlink file. + + ``is_file`` follows symlinks, so a dangling symlink, a symlink to a + directory, or any other non-regular entry would silently vanish from a + sweep and from sibling-registry checks; a registry entry that cannot be + read as a manifest is a defect to surface, never to skip. + """ + if path.is_symlink() or not path.is_file(): + raise MalformedManifestError( + f"{path} carries a manifest name but is not a regular file; " + "Chronicle will not sweep past it or register beside it." ) - return sorted( - path - for path in root.rglob("*") - if path.is_file() and is_manifest_filename(path.name) - ) def _manifest_path(output: Path, manifest_filename: str) -> Path: @@ -276,6 +289,11 @@ class ArtifactFilenameError(SourceArtifactManifestError, ValueError): """An artifact filename is not a bare, non-manifest package filename.""" +class IdentitySegmentError(SourceArtifactManifestError, ValueError): + """A registration identity (source_id / package_id) is not one canonical + R2 key segment.""" + + class AmbiguousManifestError(SourceArtifactManifestError): """The default manifest name would create a manifest beside the ones a package already keeps (PolicyEngine/chronicle#225).""" @@ -783,8 +801,8 @@ def fetch_source_artifact( # not upload. Validate them before reading the publisher so a malformed # registration identity cannot overwrite package-local bytes and fail only # when the prospective R2 key is constructed below. - _clean_key_part(source_id) - _clean_key_part(package_id) + _require_identity_segment(source_id, what="source_id") + _require_identity_segment(package_id, what="package_id") resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, @@ -794,7 +812,15 @@ def fetch_source_artifact( # Read and validate the entry being written before anything is fetched: a # manifest Chronicle cannot read, or a recorded block that names two # different objects, is a refusal that need not touch the publisher. - _refuse_a_stray_default_manifest(output, manifest_path) + try: + _refuse_a_stray_default_manifest(output, manifest_path) + except SourceArtifactManifestError: + raise + except ValueError as error: + # package_manifest_paths refuses non-regular manifest-named entries + # with a plain ValueError; surface it as the manifest error the CLI + # reports rather than a traceback. + raise MalformedManifestError(str(error)) from error existing_manifest = _read_manifest(manifest_path) _manifest_files(existing_manifest, manifest_path) _assert_manifest_identifies( @@ -817,7 +843,10 @@ def fetch_source_artifact( owners = _manifest_file_owners(manifests, filename=artifact_filename) _assert_shared_owner_identities_agree(owners, filename=artifact_filename) - existing_target = matching_directory_entry(output, artifact_filename) + try: + existing_target = matching_directory_entry(output, artifact_filename) + except ValueError as error: + raise ArtifactFilenameError(str(error)) from error if existing_target is not None: if existing_target.is_symlink(): raise ArtifactFilenameError( @@ -2408,10 +2437,14 @@ def _publish_raw_manifest_entry( ), None, ) - artifact_path = ( - matching_directory_entry(manifest_path.parent, filename) - or manifest_path.parent / filename - ) + try: + artifact_path = ( + matching_directory_entry(manifest_path.parent, filename) + or manifest_path.parent / filename + ) + except ValueError: + errors.append(f"duplicate_artifact_spellings:{filename}") + artifact_path = manifest_path.parent / filename sha256_expected = spec.get("sha256") sha256_actual = None size_bytes = None @@ -2636,10 +2669,14 @@ def _inventory_entry( r2=r2, errors=(f"manifest_named_filename:{filename}",), ) - artifact_path = ( - matching_directory_entry(manifest_path.parent, filename) - or manifest_path.parent / filename - ) + try: + artifact_path = ( + matching_directory_entry(manifest_path.parent, filename) + or manifest_path.parent / filename + ) + except ValueError: + errors.append(f"duplicate_artifact_spellings:{filename}") + artifact_path = manifest_path.parent / filename symlink = bool(filename) and artifact_path.is_symlink() exists = bool(filename) and not symlink and artifact_path.is_file() sha256_expected = spec.get("sha256") @@ -2706,6 +2743,33 @@ def _clean_key_part(value: str) -> str: return cleaned.replace(" ", "_") +def _require_identity_segment(value: Any, *, what: str) -> str: + """Require a registration identity to be one canonical key segment. + + ``_clean_key_part`` normalizes what it is given (strips, folds spaces to + underscores) because it also renders legacy recorded values; a NEW + registration identity must already be canonical, or two spellings such as + ``foo bar`` and ``foo_bar`` would collide in one R2 namespace and a + separator would shift the key's path shape. + """ + if ( + not isinstance(value, str) + or not value + or value in (".", "..") + or value != value.strip() + or any(character.isspace() for character in value) + or "/" in value + or "\\" in value + or _clean_key_part(value) != value + ): + raise IdentitySegmentError( + f"{what} must be one canonical R2 key segment (no whitespace, " + f"slashes, or '..'), not {value!r}; R2 key parts cannot be empty " + "or rewritten." + ) + return value + + def _clean_relative_key_parts(value: str) -> tuple[str, ...]: path = Path(value) if path.is_absolute(): diff --git a/chronicle/registration.py b/chronicle/registration.py index 063af7d3..5b28b6d1 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -69,11 +69,17 @@ def package_manifest_paths(package_dir: Path) -> list[Path]: directory = Path(package_dir) if not directory.is_dir(): return [] - return sorted( - path - for path in directory.iterdir() - if path.is_file() and is_manifest_filename(path.name) - ) + manifests = [] + for path in sorted(directory.iterdir()): + if not is_manifest_filename(path.name): + continue + if path.is_symlink() or not path.is_file(): + raise ValueError( + f"{path} carries a manifest name but is not a regular file; " + "Chronicle will not register beside it or sweep past it." + ) + manifests.append(path) + return manifests def validate_package_directory( @@ -118,14 +124,19 @@ def matching_directory_entry(directory: Any, filename: Any) -> Any | None: if not is_bare_filename(filename) or not directory.is_dir(): return None wanted = filename_key(filename) - return next( - ( - path - for path in sorted(directory.iterdir(), key=lambda item: item.name) - if filename_key(path.name) == wanted - ), - None, - ) + matches = [ + path + for path in sorted(directory.iterdir(), key=lambda item: item.name) + if filename_key(path.name) == wanted + ] + if len(matches) > 1: + names = ", ".join(repr(path.name) for path in matches) + raise ValueError( + f"{filename!r} matches more than one physical entry ({names}); " + "the package holds conflicting spellings of one artifact identity " + "and must be repaired by hand." + ) + return matches[0] if matches else None class StrictManifestLoader(yaml.SafeLoader): diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 9095b612..ebf8179d 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -874,6 +874,51 @@ def _source_artifact_metadata( raw_r2_uri=raw_r2.get("uri"), ) + def _resource_root(self) -> Any: + """Resolve the resource directory, refusing an escape from the package. + + ``resource_directory`` is joined under ``files(resource_package)``; an + absolute value would discard that root entirely, a ``..`` or ``.`` + component would step outside it, and a symlinked ancestor would follow + the link out of the package tree. Every byte and manifest read goes + through here, so the containment check runs before any I/O. + """ + raw = self.resource_directory + parts = str(raw).split("/") + if ( + not isinstance(raw, str) + or not raw + or raw.startswith("/") + or "\\" in raw + or any( + not part or part in (".", "..") or part != part.strip() + for part in parts + ) + ): + raise ValueError( + f"resource_directory must be a relative path of plain segments " + f"inside the resource package, not {raw!r}." + ) + root = files(self.resource_package) + directory = root.joinpath(raw) + if isinstance(root, Path): + current = root + for part in parts: + current = current / part + if current.is_symlink(): + raise ValueError( + f"resource_directory component {current} is a symbolic " + "link. Chronicle will not read source-package data " + "through it." + ) + resolved_root = root.resolve() + if not Path(directory).resolve().is_relative_to(resolved_root): + raise ValueError( + f"resource_directory {raw!r} escapes the resource package " + f"root {resolved_root}." + ) + return directory + def _resource_entry( self, value: Any, @@ -900,7 +945,7 @@ def _resource_entry( "source artifact bytes." ) - directory = files(self.resource_package).joinpath(self.resource_directory) + directory = self._resource_root() existing = matching_directory_entry(directory, name) if existing is None: return directory.joinpath(name) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 5eb4e721..f32d2d1f 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1754,7 +1754,7 @@ def unexpected_read(_source_url): monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) - with pytest.raises(ValueError, match="R2 key parts cannot be empty"): + with pytest.raises(ValueError, match="canonical R2 key segment"): fetch_source_artifact( "https://publisher.test/table.csv", source_id=source_id, @@ -3333,26 +3333,22 @@ def test_matching_directory_entry_refuses_multiple_normalized_aliases(): returning the first spelling would silently ignore the other bytes.""" from types import SimpleNamespace + from chronicle.registration import matching_directory_entry + entries = [ SimpleNamespace(name="TABLE.CSV"), SimpleNamespace(name="other.csv"), SimpleNamespace(name="table.csv"), ] - directory = SimpleNamespace( - is_dir=lambda: True, iterdir=lambda: iter(entries) - ) + directory = SimpleNamespace(is_dir=lambda: True, iterdir=lambda: iter(entries)) with pytest.raises(ValueError, match="TABLE.CSV.*table.csv|table.csv.*TABLE.CSV"): matching_directory_entry(directory, "table.csv") - assert ( - matching_directory_entry(directory, "other.csv").name == "other.csv" - ) + assert matching_directory_entry(directory, "other.csv").name == "other.csv" -def test_publish_and_inventory_report_duplicate_artifact_aliases( - tmp_path, monkeypatch -): +def test_publish_and_inventory_report_duplicate_artifact_aliases(tmp_path, monkeypatch): """A duplicate-alias defect surfaces as an entry error, not a crash and not a silent first-match read.""" output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" @@ -3360,13 +3356,9 @@ def test_publish_and_inventory_report_duplicate_artifact_aliases( _fetch_local(output_dir, source, upload_r2=False) def duplicate_alias(_directory, filename): - raise ValueError( - f"{filename!r} matches two physical spellings in the package." - ) + raise ValueError(f"{filename!r} matches two physical spellings in the package.") - monkeypatch.setattr( - "chronicle.artifacts.matching_directory_entry", duplicate_alias - ) + monkeypatch.setattr("chronicle.artifacts.matching_directory_entry", duplicate_alias) inventory = inventory_source_artifacts(output_dir) published = publish_source_artifacts(output_dir) From 098135434745fa384b739a2b08b84a5464c50c37 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:04:35 -0400 Subject: [PATCH 068/105] docs: start peer round 4 fix journal --- PROGRESS.md | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 3ee219c6..4a9ef573 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -497,3 +497,32 @@ the per-finding reproduction/fix/test/commit/port map. defense-in-depth outside findings 3/4; there are no such collisions in the tracked tree. Manifest-name collisions are already refused independent of sort order. + +## Peer round 4 (nine findings) + +### State + +- Started from detached HEAD `fa98993a` with a clean worktree; no branches, + pushes, stashes, GitHub access, or tracked `db/data/**` changes are authorized. +- Scope: the nine supplied provenance, publication preflight, canonical + identity, manifest-vintage, filename, and regular-file findings. +- Preserve the shared helper names used by stacked PR #227; its worktree is + read-only and will not be modified. +- Report path: `/tmp/chronicle-226-round4/out.md` (no explicit runner `-o` + path was provided in the visible request). + +### Done + +- Read the existing journal, storage architecture, role rules, and named code + surfaces. Began reviewing the named test modules and shared registration + helpers. GitNexus debugging guidance is available, but no graph tools are + exposed; use repository search and hermetic regression tests. +- Established this committed state/done/next journal before implementation. + +### Next + +- Add and run failing regressions for each finding before its implementation. +- Fix and commit coherent steps, recording exact red commands and observations + in the external report. +- Run full Ruff lint, formatting checks for changed Python files, and the full + pytest suite with direct exit codes; record counts and final commit map. From 317f051c492f341a4b482b1bc77b9c03e6025a40 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:07:47 -0400 Subject: [PATCH 069/105] test: reproduce publication preflight and identity findings --- PROGRESS.md | 16 ++ tests/test_chronicle_artifact_peer4.py | 228 +++++++++++++++++++++++++ 2 files changed, 244 insertions(+) create mode 100644 tests/test_chronicle_artifact_peer4.py diff --git a/PROGRESS.md b/PROGRESS.md index 4a9ef573..4d139874 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -526,3 +526,19 @@ the per-finding reproduction/fix/test/commit/port map. in the external report. - Run full Ruff lint, formatting checks for changed Python files, and the full pytest suite with direct exit codes; record counts and final commit map. + +### Round 4 reproduction checkpoint + +- Added publication/tree/identity/alias/sibling regression coverage in + `tests/test_chronicle_artifact_peer4.py` before any associated fix. +- Red command: `UV_CACHE_DIR=/tmp/chronicle-uv-cache uv run pytest -q -p + no:cacheprovider tests/test_chronicle_artifact_peer4.py` (stdout/stderr saved + to `/tmp/chronicle-226-round4/main-red.log`) exited 1: 60 failed, 12 warnings. +- Finding 3: all 4 non-regular tree cases reached build-ID inference before + refusal. Finding 4: all 24 invalid publication identity cases reached an + artifact read or build-ID inference. Finding 5: all 10 noncanonical manifest + declarations reached the publisher read. Finding 6: all 4 single case/Unicode + aliases reached artifact reads. Finding 8: all 18 explicit-selector sibling + cases escaped as plain `ValueError`, including both CLI entry points. +- Consumer guard, source resolver, and shared vintage regressions are being + developed independently; all implementations remain gated on observed red. diff --git a/tests/test_chronicle_artifact_peer4.py b/tests/test_chronicle_artifact_peer4.py new file mode 100644 index 00000000..78f9359b --- /dev/null +++ b/tests/test_chronicle_artifact_peer4.py @@ -0,0 +1,228 @@ +"""Failing-first regressions for PR #226 peer round 4.""" + +import hashlib +import json +import os +from pathlib import Path + +import pytest +import yaml + +from chronicle.artifacts import ( + ArtifactCommandResult, + SourceArtifactManifestError, + fetch_source_artifact, + inventory_source_artifacts, + publish_derived_artifacts, + publish_source_artifacts, +) +from chronicle.cli import main as cli_main +from chronicle.harness import main as harness_main + + +def _package(tmp_path, *, filename="table.csv", manifest_name="manifest.yaml"): + package = tmp_path / "package" + package.mkdir() + content = b"publisher,value\nexample,123\n" + (package / filename).write_bytes(content) + manifest_path = package / manifest_name + manifest = { + "source_id": "publisher", + "package_id": "package", + "files": { + 2024: { + "filename": filename, + "source_url": "https://example.test/table.csv", + "sha256": hashlib.sha256(content).hexdigest(), + "size_bytes": len(content), + } + }, + } + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + return package, manifest_path, manifest + + +def _no_upload(monkeypatch): + def unexpected_upload(*args, **kwargs): + pytest.fail("uploader reached before refusal") + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", unexpected_upload) + + +@pytest.mark.parametrize("shape", ["file-link", "directory-link", "dangling", "fifo"]) +def test_derived_preflights_complete_tree_before_reads_or_uploads( + tmp_path, monkeypatch, shape +): + suite = tmp_path / "suite" + suite.mkdir() + (suite / "a_good.jsonl").write_bytes(b"{}\n") + outside = tmp_path / "outside" + outside.mkdir() + (outside / "secret.csv").write_bytes(b"outside bytes") + unsafe = suite / "z_unsafe" + if shape == "file-link": + unsafe.symlink_to(outside / "secret.csv") + elif shape == "directory-link": + unsafe.symlink_to(outside, target_is_directory=True) + elif shape == "dangling": + unsafe.symlink_to(outside / "missing") + else: + os.mkfifo(unsafe) + output = tmp_path / "registry.jsonl" + output.write_text("sentinel\n") + _no_upload(monkeypatch) + + def unexpected_read(*args, **kwargs): + pytest.fail("build read reached before tree refusal") + + monkeypatch.setattr(Path, "read_bytes", unexpected_read) + monkeypatch.setattr("chronicle.artifacts.infer_build_id", unexpected_read) + report = publish_derived_artifacts( + suite, + source_id="publisher", + package_id="package", + year=2024, + build_artifacts_output=output, + ) + assert not report.valid + assert report.entries == () + assert any("regular" in error or "symlink" in error for error in report.errors) + assert output.read_text() == "sentinel\n" + + +@pytest.mark.parametrize("field", ["source_id", "package_id"]) +@pytest.mark.parametrize("bad_id", ["foo bar", "a/b", "..", 123, None, ""]) +@pytest.mark.parametrize("operation", ["raw", "derived"]) +def test_new_publication_refuses_noncanonical_identity_before_reads( + tmp_path, monkeypatch, field, bad_id, operation +): + package, manifest_path, manifest = _package(tmp_path) + kwargs = {"source_id": "publisher", "package_id": "package"} + kwargs[field] = bad_id + if operation == "raw": + manifest[field] = bad_id + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_text() + _no_upload(monkeypatch) + + def unexpected_read(*args, **kwargs): + pytest.fail("artifact read reached with noncanonical publication identity") + + monkeypatch.setattr(Path, "read_bytes", unexpected_read) + monkeypatch.setattr("chronicle.artifacts.infer_build_id", unexpected_read) + if operation == "raw": + report = publish_source_artifacts(package) + else: + report = publish_derived_artifacts(package, year=2024, **kwargs) + assert not report.valid + assert any( + "identity" in error or "source_id" in error or "package_id" in error + for error in ( + *report.errors, + *(e for entry in report.entries for e in entry.errors), + ) + ) + assert manifest_path.read_text() == before + + +@pytest.mark.parametrize("field", ["source_id", "package_id"]) +@pytest.mark.parametrize("declaration", [" padded ", 123, None, " ", ""]) +def test_fetch_refuses_noncanonical_present_manifest_identity_before_io( + tmp_path, monkeypatch, field, declaration +): + package, manifest_path, manifest = _package(tmp_path) + args = {"source_id": "publisher", "package_id": "package"} + args[field] = "padded" if declaration == " padded " else "123" + manifest[field] = declaration + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_text() + _no_upload(monkeypatch) + + def unexpected_read(*args, **kwargs): + pytest.fail("publisher read reached with noncanonical manifest declaration") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + with pytest.raises(SourceArtifactManifestError, match=field): + fetch_source_artifact( + "https://example.test/table.csv", + year=2024, + output_dir=package, + filename="table.csv", + **args, + ) + assert manifest_path.read_text() == before + + +@pytest.mark.parametrize("operation", ["publish", "inventory"]) +@pytest.mark.parametrize( + ("physical", "declared"), + [("table.csv", "TABLE.csv"), ("café.csv", "cafe\u0301.csv")], +) +def test_sweeps_refuse_single_normalized_artifact_alias_before_read( + tmp_path, monkeypatch, operation, physical, declared +): + package, manifest_path, manifest = _package(tmp_path, filename=physical) + # On filesystems that normalize Unicode on creation, choose the other + # logical spelling from the actual directory entry returned by iterdir. + actual = next( + path.name for path in package.iterdir() if path.name != manifest_path.name + ) + if actual == declared: + declared = physical + manifest["files"][2024]["filename"] = declared + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_text() + _no_upload(monkeypatch) + + def unexpected_read(*args, **kwargs): + pytest.fail("artifact read reached through a normalized alias") + + monkeypatch.setattr(Path, "read_bytes", unexpected_read) + function = ( + publish_source_artifacts + if operation == "publish" + else inventory_source_artifacts + ) + report = function(package) + assert not report.valid + assert any( + "spelling" in error for entry in report.entries for error in entry.errors + ) + assert manifest_path.read_text() == before + + +@pytest.mark.parametrize("operation", ["publish-raw", "inventory-artifacts"]) +@pytest.mark.parametrize("shape", ["directory", "dangling", "fifo"]) +@pytest.mark.parametrize("entrypoint", ["function", "harness", "cli"]) +def test_explicit_sweep_reports_nonregular_manifest_sibling( + tmp_path, monkeypatch, capsys, operation, shape, entrypoint +): + package, manifest_path, _manifest = _package( + tmp_path, manifest_name="manifest_selected.yaml" + ) + sibling = package / "manifest_sibling.yaml" + if shape == "directory": + sibling.mkdir() + elif shape == "dangling": + sibling.symlink_to(package / "missing") + else: + os.mkfifo(sibling) + before = manifest_path.read_text() + _no_upload(monkeypatch) + if entrypoint == "function": + function = ( + publish_source_artifacts + if operation == "publish-raw" + else inventory_source_artifacts + ) + report = function(package, manifest_filename=manifest_path.name) + assert not report.valid + assert "regular file" in " ".join(report.errors) + else: + main = harness_main if entrypoint == "harness" else cli_main + assert ( + main([operation, "--root", str(package), "--manifest", manifest_path.name]) + == 1 + ) + assert "regular file" in json.dumps(json.loads(capsys.readouterr().out)) + assert manifest_path.read_text() == before From c88b8c7db5c816b2c8d331126596f95e49634876 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:08:06 -0400 Subject: [PATCH 070/105] fix: validate immutable R2 provenance in source artifact loading --- PROGRESS.md | 12 +++ chronicle/source_package.py | 37 ++++++- tests/test_chronicle_source_package.py | 130 +++++++++++++++++++++++++ 3 files changed, 175 insertions(+), 4 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 4d139874..ac637d74 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -542,3 +542,15 @@ the per-finding reproduction/fix/test/commit/port map. cases escaped as plain `ValueError`, including both CLI entry points. - Consumer guard, source resolver, and shared vintage regressions are being developed independently; all implementations remain gated on observed red. + +### Finding 2: source provenance reader + +- Reproduced malformed/contradictory/non-R2 locators and checksum/filename + mismatches with 10 failures and 1 passing control before implementation; + exact command and output are in the external report's source evidence. +- `SourceArtifactSpec` now reuses `_validated_recorded_r2`, checks the manifest + identity, and uses the immutable object's digest to validate local/fetched + bytes even without a separately declared checksum. Invalid metadata is + refused before artifact/cache I/O; bad fetched bytes before cache writes. +- The same focused command now exits 0: 11 passed, 12 warnings. Inventory's + half of finding 2 remains next; no source package or source data was changed. diff --git a/chronicle/source_package.py b/chronicle/source_package.py index ebf8179d..ffbe70ff 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -15,6 +15,7 @@ import httpx import yaml +from chronicle.artifacts import SourceArtifactManifestError, _validated_recorded_r2 from chronicle.core import ( ALLOWED_AGGREGATIONS, ALLOWED_ASSERTIONS, @@ -1001,17 +1002,45 @@ def _artifact_content( what="Source artifact filename", forbid_manifest_name=True, ) - content = _read_source_artifact_content(artifact_path, spec) + try: + recorded_r2 = _validated_recorded_r2( + spec, + manifest_path=Path(self.resource_directory) / self.manifest, + year=self.artifact_year or year, + ) + except SourceArtifactManifestError as exc: + raise ValueError(str(exc)) from exc expected_sha = spec.get("sha256") + raw_r2 = {} + if recorded_r2 is not None: + if recorded_r2.filename != filename or ( + expected_sha is not None and expected_sha != recorded_r2.sha256 + ): + raise ValueError( + f"Source artifact {filename!r} disagrees with its recorded " + f"R2 identity: storage.r2 names {recorded_r2.filename!r} " + f"with sha256={recorded_r2.sha256}, while the manifest " + f"declares sha256={expected_sha!r}." + ) + expected_sha = recorded_r2.sha256 + # The immutable object also supplies the checksum for a manifest + # without a separate sha256 field. Pass it into the fetch/cache + # reader so wrong publisher bytes are refused before cache writes. + spec = {**spec, "sha256": expected_sha} + raw_r2 = { + "provider": recorded_r2.provider, + "bucket": recorded_r2.bucket, + "key": recorded_r2.key, + "uri": recorded_r2.uri, + } + content = _read_source_artifact_content(artifact_path, spec) if expected_sha: _validate_source_artifact_sha( content, expected_sha=str(expected_sha), filename=str(filename), ) - storage = spec.get("storage") if isinstance(spec, dict) else None - raw_r2 = storage.get("r2") if isinstance(storage, dict) else {} - return content, str(filename), spec["source_url"], raw_r2 or {} + return content, str(filename), spec["source_url"], raw_r2 def _sheet_name(self, filename: str, *, year: int) -> str: if self.sheet_name: diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index 41994caf..a6438a78 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -1081,6 +1081,136 @@ def test_source_artifact_loader_fetches_missing_artifact_when_enabled( assert _source_artifact_cache_path(spec).read_bytes() == content +@pytest.fixture +def recorded_r2_artifact(tmp_path, monkeypatch): + resource_root = tmp_path / "resources" + resource_dir = resource_root / "data" / "publisher" / "package" + resource_dir.mkdir(parents=True) + content = b"publisher source artifact" + digest = hashlib.sha256(content).hexdigest() + key = f"raw/publisher/package/2024/{digest}/table.csv" + entry = { + "filename": "table.csv", + "source_url": "https://example.test/table.csv", + "sha256": digest, + "storage": { + "r2": { + "provider": "r2", + "bucket": "ledger-raw", + "key": key, + "uri": f"r2://ledger-raw/{key}", + } + }, + } + artifact = SourceArtifactSpec( + source_name="publisher", + source_table="Table", + resource_package="test_resources", + resource_directory="data/publisher/package", + manifest="manifest.yaml", + vintage="2024", + extracted_at="2026-09-04", + extraction_method="test", + artifact_year=2024, + ) + monkeypatch.setattr( + "chronicle.source_package.files", lambda _package: resource_root + ) + return artifact, resource_dir, content, entry + + +@pytest.mark.parametrize( + "invalid_locator", + [ + "contradictory-key", + "contradictory-bucket", + "missing-provider", + "missing-uri", + "non-r2", + "null-block", + "checksum-identity", + "filename-identity", + ], +) +def test_source_artifact_spec_refuses_invalid_recorded_r2_before_read( + recorded_r2_artifact, monkeypatch, invalid_locator +): + artifact, resource_dir, content, entry = recorded_r2_artifact + recorded = entry["storage"]["r2"] + if invalid_locator == "contradictory-key": + recorded["key"] = recorded["key"].replace("table.csv", "other.csv") + elif invalid_locator == "contradictory-bucket": + recorded["bucket"] = "other-raw" + elif invalid_locator == "missing-provider": + del recorded["provider"] + elif invalid_locator == "missing-uri": + del recorded["uri"] + elif invalid_locator == "non-r2": + recorded["provider"] = "s3" + recorded["uri"] = recorded["uri"].replace("r2://", "s3://") + elif invalid_locator == "null-block": + entry["storage"]["r2"] = None + elif invalid_locator == "checksum-identity": + entry["sha256"] = "0" * 64 + else: + recorded["key"] = recorded["key"].replace("table.csv", "other.csv") + recorded["uri"] = recorded["uri"].replace("table.csv", "other.csv") + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump({"files": {2024: entry}}) + ) + (resource_dir / "table.csv").write_bytes(content) + + def unexpected_read(_artifact_path, _spec): + raise AssertionError("invalid R2 provenance reached artifact I/O") + + monkeypatch.setattr( + "chronicle.source_package._read_source_artifact_content", unexpected_read + ) + with pytest.raises(ValueError, match="storage.r2|recorded R2"): + artifact._artifact_content(2024) + + +@pytest.mark.parametrize("location", ["local", "fetch"]) +def test_source_artifact_spec_checks_recorded_r2_digest_without_declared_checksum( + recorded_r2_artifact, tmp_path, monkeypatch, location +): + artifact, resource_dir, _content, entry = recorded_r2_artifact + del entry["sha256"] + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump({"files": {2024: entry}}) + ) + changed_content = b"different publisher bytes" + cache = tmp_path / "cache" + monkeypatch.setenv(SOURCE_ARTIFACT_CACHE_ENV, str(cache)) + if location == "local": + (resource_dir / "table.csv").write_bytes(changed_content) + else: + monkeypatch.setenv(SOURCE_ARTIFACT_FETCH_ENV, "1") + monkeypatch.setattr( + "chronicle.source_package._fetch_source_artifact_content", + lambda _url: changed_content, + ) + + with pytest.raises(ValueError, match="checksum mismatch"): + artifact._artifact_content(2024) + assert not cache.exists() + + +def test_source_artifact_spec_accepts_consistent_recorded_r2(recorded_r2_artifact): + artifact, resource_dir, content, entry = recorded_r2_artifact + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump({"files": {2024: entry}}) + ) + (resource_dir / "table.csv").write_bytes(content) + + assert artifact._artifact_content(2024) == ( + content, + "table.csv", + entry["source_url"], + entry["storage"]["r2"], + ) + + @pytest.mark.parametrize( "path_kind", ["absolute", "parent", "symlink", "normalized-alias", "manifest-name"], From 1727737e979c564081b845b82866a38fbb81bfcd Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:08:38 -0400 Subject: [PATCH 071/105] fix: require exact canonical manifest identity declarations --- PROGRESS.md | 10 ++++++++++ chronicle/artifacts.py | 9 ++++++--- 2 files changed, 16 insertions(+), 3 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index ac637d74..6c53d058 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -554,3 +554,13 @@ the per-finding reproduction/fix/test/commit/port map. refused before artifact/cache I/O; bad fetched bytes before cache writes. - The same focused command now exits 0: 11 passed, 12 warnings. Inventory's half of finding 2 remains next; no source package or source data was changed. + +### Finding 5: exact manifest declarations + +- Present `source_id` and `package_id` now pass `_require_identity_segment` + without stripping or stringification and must equal the fetch argument. + Absent fields remain eligible for first registration; null/empty fields do + not impersonate absence. +- Focused regressions and existing other-package refusal controls passed: + direct exit 0, 12 passed. The red checkpoint above recorded all 10 new + declaration cases reaching publisher I/O before this fix. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 2e2fde79..077bddb3 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -234,9 +234,12 @@ def _assert_manifest_identifies( leave one entry making two incompatible provenance claims. """ for field, value in (("source_id", source_id), ("package_id", package_id)): - declared = existing_manifest.get(field) - declared = declared.strip() if isinstance(declared, str) else declared - if declared not in (None, "") and str(declared) != value: + if field not in existing_manifest: + continue + declared = _require_identity_segment( + existing_manifest[field], what=f"{manifest_path} {field}" + ) + if declared != value: raise SourceArtifactManifestError( f"{manifest_path} declares {field}={declared!r}; refusing to " f"fetch {field}={value!r} into it. Fetch into the package the " From 12ef53d35941fd9e90bd19301cbeebbbf4f1262a Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:09:16 -0400 Subject: [PATCH 072/105] fix: validate logical vintages in every manifest consumer --- PROGRESS.md | 11 ++ chronicle/registration.py | 38 ++++++- tests/test_chronicle_manifest_vintages.py | 132 ++++++++++++++++++++++ 3 files changed, 177 insertions(+), 4 deletions(-) create mode 100644 tests/test_chronicle_manifest_vintages.py diff --git a/PROGRESS.md b/PROGRESS.md index 6c53d058..42fb5b6d 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -564,3 +564,14 @@ the per-finding reproduction/fix/test/commit/port map. - Focused regressions and existing other-package refusal controls passed: direct exit 0, 12 passed. The red checkpoint above recorded all 10 new declaration cases reaching publisher I/O before this fix. + +### Finding 7: shared logical vintage validation + +- Before implementation, the new manifest-vintage tests exited 1: 5 failed, + 1 passed. Duplicate keys `2024` and `"2024"` reached artifact I/O through + publish, inventory, source loading, and fetch of another selected vintage. +- `load_manifest_document` now invokes shared `validate_manifest_vintages` + over the entire manifest, so every consumer refuses a logical duplicate + through its existing controlled YAML error path. Shared names are unchanged. +- Six regressions plus existing quoted-year compatibility tests exit 0: + 8 passed, 12 warnings. Scoped Ruff lint/format checks both pass. diff --git a/chronicle/registration.py b/chronicle/registration.py index 5b28b6d1..c580d7c2 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -181,13 +181,42 @@ def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: return mapping +def validate_manifest_vintages(payload: Any) -> None: + """Refuse different keys that identify one logical ``files`` vintage. + + YAML distinguishes integer ``2024`` from quoted ``"2024"``, but manifest + consumers select or report them as the same vintage. Validate the entire + manifest, including vintages other than the one a caller requested, before + any consumer can read artifact bytes or construct publication routes. + + Leave non-mapping documents and ``files`` blocks to the consumers' existing + shape checks. Labels retain their spelling, including leading zeroes. + """ + files = payload.get("files") if isinstance(payload, Mapping) else None + if not isinstance(files, Mapping): + return + seen: dict[str, Any] = {} + for vintage in files: + identity = str(vintage) + if identity in seen: + raise yaml.YAMLError( + f"Vintage {identity!r} is recorded under both keys " + f"{seen[identity]!r} and {vintage!r}; one vintage has one key. " + "Merge the entries by hand first. Chronicle will not choose " + "which entry is the record." + ) + seen[identity] = vintage + + def load_manifest_document(text: str) -> Any: - """Parse a manifest document, refusing duplicate keys. + """Parse a manifest document, refusing duplicate keys and vintages. - Raises :class:`yaml.YAMLError` (a ``ConstructorError`` naming the - duplicate key) for a document YAML would otherwise silently collapse. + Raises :class:`yaml.YAMLError` for keys YAML would silently collapse or + for distinct YAML keys that manifest consumers treat as one vintage. """ - return yaml.load(text, Loader=StrictManifestLoader) # noqa: S506 + payload = yaml.load(text, Loader=StrictManifestLoader) # noqa: S506 + validate_manifest_vintages(payload) + return payload __all__ = [ @@ -200,5 +229,6 @@ def load_manifest_document(text: str) -> Any: "load_manifest_document", "matching_directory_entry", "package_manifest_paths", + "validate_manifest_vintages", "validate_package_directory", ] diff --git a/tests/test_chronicle_manifest_vintages.py b/tests/test_chronicle_manifest_vintages.py new file mode 100644 index 00000000..11e4764c --- /dev/null +++ b/tests/test_chronicle_manifest_vintages.py @@ -0,0 +1,132 @@ +"""Every artifact boundary refuses duplicate logical manifest vintages.""" + +from __future__ import annotations + +import hashlib +from pathlib import Path + +import pytest +import yaml + +from chronicle.artifacts import ( + MalformedManifestError, + fetch_source_artifact, + inventory_source_artifacts, + publish_source_artifacts, +) +from chronicle.source_package import SourceArtifactSpec + + +@pytest.fixture +def duplicate_vintage_package(tmp_path): + package = tmp_path / "data" / "irs_soi" / "table" + package.mkdir(parents=True) + files = {} + for vintage, filename in ( + (2023, "selected.csv"), + (2024, "numeric.csv"), + ("2024", "quoted.csv"), + ): + content = f"publisher bytes for {filename}\n".encode() + (package / filename).write_bytes(content) + files[vintage] = { + "filename": filename, + "source_url": f"https://example.test/{filename}", + "sha256": hashlib.sha256(content).hexdigest(), + "size_bytes": len(content), + } + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "irs_soi", + "package_id": "soi-table", + "files": files, + }, + sort_keys=False, + ) + ) + return package + + +def _unexpected_artifact_io(*_args, **_kwargs): + raise AssertionError("duplicate logical vintages reached artifact I/O") + + +@pytest.mark.parametrize("year", [2023, 2024]) +def test_fetch_refuses_logical_vintage_duplicates_anywhere_before_io( + duplicate_vintage_package, monkeypatch, year +): + package = duplicate_vintage_package + before = {path.name: path.read_bytes() for path in package.iterdir()} + monkeypatch.setattr("chronicle.artifacts._read_artifact", _unexpected_artifact_io) + monkeypatch.setattr( + "chronicle.artifacts._upload_r2_object", _unexpected_artifact_io + ) + + with pytest.raises(MalformedManifestError, match="both keys"): + fetch_source_artifact( + "https://example.test/selected.csv", + source_id="irs_soi", + package_id="soi-table", + year=year, + output_dir=package, + upload_r2=True, + ) + + assert {path.name: path.read_bytes() for path in package.iterdir()} == before + + +@pytest.mark.parametrize( + "operation", [publish_source_artifacts, inventory_source_artifacts] +) +def test_sweeps_refuse_logical_vintage_duplicates_before_artifact_io( + duplicate_vintage_package, monkeypatch, operation +): + package = duplicate_vintage_package + before = {path.name: path.read_bytes() for path in package.iterdir()} + read_bytes = Path.read_bytes + + def checked_read_bytes(path): + if path.parent == package and path.suffix == ".csv": + _unexpected_artifact_io() + return read_bytes(path) + + with monkeypatch.context() as guarded: + guarded.setattr(Path, "read_bytes", checked_read_bytes) + guarded.setattr( + "chronicle.artifacts._upload_r2_object", _unexpected_artifact_io + ) + report = operation(package) + + assert not report.valid + assert report.entries == () + assert any("both keys" in error for error in report.errors) + assert {path.name: path.read_bytes() for path in package.iterdir()} == before + + +@pytest.mark.parametrize("year", [2023, 2024]) +def test_source_loader_refuses_logical_vintage_duplicates_anywhere_before_io( + duplicate_vintage_package, tmp_path, monkeypatch, year +): + package = duplicate_vintage_package + before = {path.name: path.read_bytes() for path in package.iterdir()} + monkeypatch.setattr("chronicle.source_package.files", lambda _package: tmp_path) + monkeypatch.setattr( + "chronicle.source_package._read_source_artifact_content", + _unexpected_artifact_io, + ) + artifact = SourceArtifactSpec( + source_name="irs_soi", + source_table="Table", + resource_package="test_resources", + resource_directory="data/irs_soi/table", + manifest="manifest.yaml", + vintage="2024", + extracted_at="2026-09-04", + extraction_method="test", + ) + + with pytest.raises(ValueError, match="both keys"): + artifact._artifact_content(year) + + assert {path.name: path.read_bytes() for path in package.iterdir()} == before From ce083e7ccb544230bec0ba4a1dd4a91c9f330d97 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:10:00 -0400 Subject: [PATCH 073/105] fix: refuse facts citing configured derived storage routes --- PROGRESS.md | 13 ++++ chronicle/artifacts.py | 5 ++ chronicle/consumer_contract.py | 45 +++++++++--- tests/test_chronicle_consumer_contract.py | 87 +++++++++++++++++++++++ 4 files changed, 142 insertions(+), 8 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 42fb5b6d..cb9a02e2 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -575,3 +575,16 @@ the per-finding reproduction/fix/test/commit/port map. through its existing controlled YAML error path. Shared names are unchanged. - Six regressions plus existing quoted-year compatibility tests exit 0: 8 passed, 12 warnings. Scoped Ruff lint/format checks both pass. + +### Finding 1: configured derived provenance + +- Consumer guard regressions ran red before implementation: direct exit 1, + 35 failed and 1 passed; configured bucket/prefix provenance did not produce + `derived_fact_provenance`. Exact evidence is in the external report. +- The guard resolves derived bucket/prefix configuration lazily, retains exact + archived routes, and checks raw locator fields, source-file locators, and + source URL. Added `default_r2_derived_prefix` with the standard environment + lookup ladder so publication and the guard can share it. +- Full consumer-contract module passes: direct exit 0, 95 passed, 32 warnings; + scoped Ruff checks pass. Wiring configured prefixes into publication and + refusing unrecognizable explicit routes remains a publication follow-up. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 077bddb3..3ed4bf9d 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -257,6 +257,11 @@ def default_r2_derived_bucket() -> str: return env_value(R2_DERIVED_BUCKET_ENV, default=DEFAULT_R2_DERIVED_BUCKET) +def default_r2_derived_prefix() -> str: + """Resolve the derived route shared by publication and fact refusals.""" + return env_value("CHRONICLE_R2_DERIVED_PREFIX", default=DEFAULT_R2_DERIVED_PREFIX) + + class SourceArtifactManifestError(RuntimeError): """A manifest refuses the write a fetch is about to make. diff --git a/chronicle/consumer_contract.py b/chronicle/consumer_contract.py index 17315a13..a4766193 100644 --- a/chronicle/consumer_contract.py +++ b/chronicle/consumer_contract.py @@ -15,6 +15,7 @@ from pathlib import Path from typing import Any +from chronicle import artifacts from chronicle.core import ( ALLOWED_PROVENANCE_CLASSES, DEFAULT_ASSERTION, @@ -493,11 +494,30 @@ def _is_derived_source_record_id(source_record_id: str) -> bool: def _points_at_derived(bucket: str, key: str) -> bool: """Whether an R2 bucket/key pair addresses derived build output. - Matched on shape rather than on the ledger-era bucket names, so the guard - keeps firing once the buckets are renamed (PolicyEngine/chronicle#143, - mechanism 3). + Resolve publication configuration at validation time: an operator may use + a bucket or prefix with no ``derived`` marker in its spelling. Archived + rename-window routes remain derived after the active destination changes. """ - return bucket.endswith("-derived") or key.startswith("derived/") + derived_buckets = { + "ledger-derived", + "chronicle-derived", + artifacts.DEFAULT_R2_DERIVED_BUCKET, + artifacts.default_r2_derived_bucket(), + } + derived_prefixes = { + "derived", + artifacts.resolve_r2_prefix( + prefix=None, + default_prefix=artifacts.DEFAULT_R2_DERIVED_PREFIX, + ), + artifacts.resolve_r2_prefix( + prefix=None, + default_prefix=artifacts.default_r2_derived_prefix(), + ), + } + return bucket in derived_buckets or any( + key == prefix or key.startswith(f"{prefix}/") for prefix in derived_prefixes + ) def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: @@ -521,14 +541,23 @@ def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: "itself. Target construction, aging, and reconciliation belong in " "Microcosm." ) - source_file_bucket, bucket_separator, _ = source_file.partition(":") - if bucket_separator and source_file_bucket.endswith("-derived"): + if source_file.startswith("r2://"): + source_file_bucket, source_file_key = _r2_uri_parts(source_file) + else: + source_file_bucket, bucket_separator, source_file_key = source_file.partition( + ":" + ) + if not bucket_separator: + source_file_bucket = source_file_key = "" + if _points_at_derived(source_file_bucket, source_file_key): return ( "Chronicle consumer facts must cite raw publisher artifacts. Derived " "target-construction artifacts belong in Microcosm." ) - if _points_at_derived(raw_r2_bucket, raw_r2_key) or _points_at_derived( - *_r2_uri_parts(raw_r2_uri) + if ( + _points_at_derived(raw_r2_bucket, raw_r2_key) + or _points_at_derived(*_r2_uri_parts(raw_r2_uri)) + or _points_at_derived(*_r2_uri_parts(source.url or "")) ): return ( "Chronicle consumer facts must point at raw source artifacts, not " diff --git a/tests/test_chronicle_consumer_contract.py b/tests/test_chronicle_consumer_contract.py index 9cc730c1..60e0844e 100644 --- a/tests/test_chronicle_consumer_contract.py +++ b/tests/test_chronicle_consumer_contract.py @@ -11,6 +11,7 @@ import pytest +import chronicle.artifacts as artifacts import chronicle.consumer_contract as consumer_contract from chronicle.consumer_contract import ( CONSUMER_FACT_SCHEMA_VERSION, @@ -940,6 +941,92 @@ def test_consumer_contract_rejects_downstream_derived_target_facts( assert "derived_fact_provenance" in {error.code for error in report.errors} +@pytest.mark.parametrize( + "bucket_env", + [ + "CHRONICLE_R2_DERIVED_BUCKET", + "POLICYENGINE_LEDGER_R2_DERIVED_BUCKET", + "LEDGER_R2_DERIVED_BUCKET", + ], +) +@pytest.mark.parametrize( + ("field", "value"), + [ + ("raw_r2_bucket", "chronicle-builds"), + ("raw_r2_uri", "r2://chronicle-builds/builds/source/fact.json"), + ("source_file", "chronicle-builds:builds/source/fact.json"), + ("source_file", "r2://chronicle-builds/builds/source/fact.json"), + ("url", "r2://chronicle-builds/builds/source/fact.json"), + ], +) +def test_consumer_contract_rejects_configured_derived_bucket( + monkeypatch, tmp_path, bucket_env, field, value +): + monkeypatch.setenv(bucket_env, "chronicle-builds") + fact = _soi_agi_fact() + derived = replace(fact, source=replace(fact.source, **{field: value})) + + report = validate_consumer_fact_contract([derived]) + + assert "derived_fact_provenance" in {error.code for error in report.errors} + output = tmp_path / "new-directory" / "consumer_facts.jsonl" + with pytest.raises(ValueError, match="consumer-contract"): + write_consumer_facts_jsonl([derived], output) + assert not output.parent.exists() + + +@pytest.mark.parametrize( + "prefix_env", + [ + None, + "CHRONICLE_R2_DERIVED_PREFIX", + "POLICYENGINE_LEDGER_R2_DERIVED_PREFIX", + "LEDGER_R2_DERIVED_PREFIX", + ], +) +@pytest.mark.parametrize( + ("field", "value"), + [ + ("raw_r2_key", "builds/source/fact.json"), + ("raw_r2_uri", "r2://publisher-archive/builds/source/fact.json"), + ("source_file", "publisher-archive:builds/source/fact.json"), + ("source_file", "r2://publisher-archive/builds/source/fact.json"), + ("url", "r2://publisher-archive/builds/source/fact.json"), + ], +) +def test_consumer_contract_rejects_configured_derived_prefix( + monkeypatch, prefix_env, field, value +): + if prefix_env is None: + monkeypatch.setattr(artifacts, "DEFAULT_R2_DERIVED_PREFIX", "builds") + else: + monkeypatch.setenv(prefix_env, "builds") + fact = _soi_agi_fact() + derived = replace(fact, source=replace(fact.source, **{field: value})) + + report = validate_consumer_fact_contract([derived]) + + assert "derived_fact_provenance" in {error.code for error in report.errors} + + +def test_consumer_contract_derived_routes_match_complete_names(monkeypatch): + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-builds") + monkeypatch.setattr(artifacts, "DEFAULT_R2_DERIVED_PREFIX", "builds") + fact = _soi_agi_fact() + publisher = replace( + fact, + source=replace( + fact.source, + source_file="chronicle-builds-raw:buildstats/source/publisher.csv", + raw_r2_bucket="chronicle-builds-raw", + raw_r2_key="buildstats/source/publisher.csv", + raw_r2_uri="r2://chronicle-builds-raw/buildstats/source/publisher.csv", + ), + ) + + assert validate_consumer_fact_contract([publisher]).valid + + def test_derived_record_marker_is_rejected_in_either_spelling(): """Both rename-window spellings produce the identical boundary error.""" fact = _soi_agi_fact() From 8bb8219391bd3a4c24d6d7b358b78af08edadf45 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:10:11 -0400 Subject: [PATCH 074/105] fix: refuse nonregular source package resources before opening --- PROGRESS.md | 11 ++++ chronicle/source_package.py | 5 ++ tests/test_chronicle_source_package.py | 70 +++++++++++++++++++++++++- 3 files changed, 85 insertions(+), 1 deletion(-) diff --git a/PROGRESS.md b/PROGRESS.md index cb9a02e2..791a86b6 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -588,3 +588,14 @@ the per-finding reproduction/fix/test/commit/port map. - Full consumer-contract module passes: direct exit 0, 95 passed, 32 warnings; scoped Ruff checks pass. Wiring configured prefixes into publication and refusing unrecognizable explicit routes remains a publication follow-up. + +### Finding 9: regular source resources + +- Directory and FIFO manifest/artifact regressions ran red: direct exit 1, + 4 failed, 2 passed. The resolver returned non-regular manifests and reached + artifact-read sentinels; no test opened a FIFO. +- The source resolver now requires an existing selected entry to be a regular + file after symlink and spelling checks. Missing artifact entries still use + the existing checksum-validated cache/fetch path; zip-backed resources work. +- Same focused command exits 0: 6 passed, 12 warnings. The full source-package + module is running. Exact red/green commands are in the external evidence. diff --git a/chronicle/source_package.py b/chronicle/source_package.py index ffbe70ff..0738b08b 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -961,6 +961,11 @@ def _resource_entry( f"{what} {existing} has the same normalized filename as " f"{name!r}. Keep exactly one spelling in the package." ) + if not existing.is_file(): + raise ValueError( + f"{what} {existing} is not a regular file. Chronicle will " + "not open non-regular source-package resources." + ) return existing def manifest_resource(self) -> Any: diff --git a/tests/test_chronicle_source_package.py b/tests/test_chronicle_source_package.py index a6438a78..aeda3df0 100644 --- a/tests/test_chronicle_source_package.py +++ b/tests/test_chronicle_source_package.py @@ -5,8 +5,9 @@ from copy import deepcopy import hashlib from io import BytesIO +import os from pathlib import Path -from zipfile import ZipFile +from zipfile import Path as ZipPath, ZipFile import openpyxl import pytest @@ -1211,6 +1212,73 @@ def test_source_artifact_spec_accepts_consistent_recorded_r2(recorded_r2_artifac ) +@pytest.mark.parametrize("entry_kind", ["directory", "fifo"]) +@pytest.mark.parametrize("resource_kind", ["manifest", "artifact"]) +def test_source_artifact_spec_refuses_non_regular_resource_before_open( + recorded_r2_artifact, monkeypatch, entry_kind, resource_kind +): + artifact, resource_dir, _content, entry = recorded_r2_artifact + name = "manifest.yaml" if resource_kind == "manifest" else "table.csv" + resource = resource_dir / name + if entry_kind == "directory": + resource.mkdir() + else: + os.mkfifo(resource) + if resource_kind == "artifact": + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump({"files": {2024: entry}}) + ) + + def unexpected_read(_artifact_path, _spec): + raise AssertionError("non-regular artifact reached artifact I/O") + + monkeypatch.setattr( + "chronicle.source_package._read_source_artifact_content", unexpected_read + ) + with pytest.raises(ValueError, match="not a regular file"): + if resource_kind == "manifest": + # Resolve without opening: a broken guard must fail promptly even + # for a FIFO, whose actual open would block the test process. + artifact.manifest_resource() + else: + artifact._artifact_content(2024) + + +def test_source_artifact_spec_reads_regular_importlib_zip_resources( + recorded_r2_artifact, monkeypatch +): + artifact, _resource_dir, content, entry = recorded_r2_artifact + buffer = BytesIO() + with ZipFile(buffer, "w") as archive: + archive.writestr("data/publisher/package/table.csv", content) + archive.writestr( + "data/publisher/package/manifest.yaml", + yaml.safe_dump({"files": {2024: entry}}), + ) + with ZipFile(buffer) as archive: + monkeypatch.setattr( + "chronicle.source_package.files", lambda _package: ZipPath(archive) + ) + assert artifact._artifact_content(2024)[0] == content + + +def test_source_artifact_spec_fetches_absent_regular_resource( + recorded_r2_artifact, tmp_path, monkeypatch +): + artifact, resource_dir, content, entry = recorded_r2_artifact + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump({"files": {2024: entry}}) + ) + monkeypatch.setenv(SOURCE_ARTIFACT_CACHE_ENV, str(tmp_path / "cache")) + monkeypatch.setenv(SOURCE_ARTIFACT_FETCH_ENV, "1") + monkeypatch.setattr( + "chronicle.source_package._fetch_source_artifact_content", lambda _url: content + ) + + assert artifact._artifact_content(2024)[0] == content + assert _source_artifact_cache_path(entry).read_bytes() == content + + @pytest.mark.parametrize( "path_kind", ["absolute", "parent", "symlink", "normalized-alias", "manifest-name"], From 2c683a46c7624c1acf0e5e111ed88c1d325cd370 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:10:58 -0400 Subject: [PATCH 075/105] fix: report nonregular manifest siblings through shared errors --- PROGRESS.md | 8 ++++++++ chronicle/artifacts.py | 7 ++++++- 2 files changed, 14 insertions(+), 1 deletion(-) diff --git a/PROGRESS.md b/PROGRESS.md index 791a86b6..f537a8bd 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -599,3 +599,11 @@ the per-finding reproduction/fix/test/commit/port map. the existing checksum-validated cache/fetch path; zip-backed resources work. - Same focused command exits 0: 6 passed, 12 warnings. The full source-package module is running. Exact red/green commands are in the external evidence. + +### Finding 8: controlled sibling-manifest refusals + +- `_package_manifests` now translates discovery's `ValueError` into + `MalformedManifestError`, so publish/inventory and both CLI entry points + use the shared controlled refusal path even with an explicit selector. +- The same 18 cases that previously trace-backed now pass: direct exit 0, + 18 passed, 12 warnings; no uploads or manifest rewrites occur. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 3ed4bf9d..10f1d91c 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -167,7 +167,12 @@ def _package_manifests( fetch selected. A malformed sibling is therefore a pre-I/O refusal: until Chronicle can read every owner, it cannot safely overwrite shared bytes. """ - paths = package_manifest_paths(output) + try: + paths = package_manifest_paths(output) + except ValueError as error: + # Explicit sweep selectors still inspect every sibling registry. Keep + # discovery refusals in the shared exception family both CLIs report. + raise MalformedManifestError(str(error)) from error by_name: dict[str, Path] = {} for path in paths: key = filename_key(path.name) From 2f7cb31d75a114aaa1b40f527deaa998b45dcc36 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:11:53 -0400 Subject: [PATCH 076/105] test: exercise sibling refusal through the top-level CLI --- PROGRESS.md | 12 ++++++++---- tests/test_chronicle_artifact_peer4.py | 13 ++++++++----- 2 files changed, 16 insertions(+), 9 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index f537a8bd..c4c5ae64 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -538,8 +538,9 @@ the per-finding reproduction/fix/test/commit/port map. refusal. Finding 4: all 24 invalid publication identity cases reached an artifact read or build-ID inference. Finding 5: all 10 noncanonical manifest declarations reached the publisher read. Finding 6: all 4 single case/Unicode - aliases reached artifact reads. Finding 8: all 18 explicit-selector sibling - cases escaped as plain `ValueError`, including both CLI entry points. + aliases reached artifact reads. Finding 8: 12 explicit-selector sibling cases escaped as plain + `ValueError` through the functions and harness; 6 top-level CLI tests had + an invocation error (corrected and verified below). - Consumer guard, source resolver, and shared vintage regressions are being developed independently; all implementations remain gated on observed red. @@ -605,5 +606,8 @@ the per-finding reproduction/fix/test/commit/port map. - `_package_manifests` now translates discovery's `ValueError` into `MalformedManifestError`, so publish/inventory and both CLI entry points use the shared controlled refusal path even with an explicit selector. -- The same 18 cases that previously trace-backed now pass: direct exit 0, - 18 passed, 12 warnings; no uploads or manifest rewrites occur. +- The first post-fix command actually exited 1: 12 passed and 6 top-level + CLI cases failed because the test passed argv to a zero-argument entry point. + Corrected the test to set sys.argv and assert SystemExit. The corrected + command exits 0: 18 passed, 12 warnings; no uploads or rewrites occur. + The earlier premature passing-count journal entry is corrected here. diff --git a/tests/test_chronicle_artifact_peer4.py b/tests/test_chronicle_artifact_peer4.py index 78f9359b..3e62580f 100644 --- a/tests/test_chronicle_artifact_peer4.py +++ b/tests/test_chronicle_artifact_peer4.py @@ -219,10 +219,13 @@ def test_explicit_sweep_reports_nonregular_manifest_sibling( assert not report.valid assert "regular file" in " ".join(report.errors) else: - main = harness_main if entrypoint == "harness" else cli_main - assert ( - main([operation, "--root", str(package), "--manifest", manifest_path.name]) - == 1 - ) + args = [operation, "--root", str(package), "--manifest", manifest_path.name] + if entrypoint == "harness": + assert harness_main(args) == 1 + else: + monkeypatch.setattr("sys.argv", ["chronicle", *args]) + with pytest.raises(SystemExit) as exit_info: + cli_main() + assert exit_info.value.code == 1 assert "regular file" in json.dumps(json.loads(capsys.readouterr().out)) assert manifest_path.read_text() == before From 4cecc72545380706ddc0eab8c7fe774a5ebf7941 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:13:26 -0400 Subject: [PATCH 077/105] test: cover locator validation and remaining publication routes --- PROGRESS.md | 16 +++ tests/test_chronicle_artifact_peer4.py | 179 +++++++++++++++++++++++++ 2 files changed, 195 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index c4c5ae64..761fe0a0 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -611,3 +611,19 @@ the per-finding reproduction/fix/test/commit/port map. Corrected the test to set sys.argv and assert SystemExit. The corrected command exits 0: 18 passed, 12 warnings; no uploads or rewrites occur. The earlier premature passing-count journal entry is corrected here. + +### Additional round 4 red checkpoint + +- Added inventory locator/identity regressions, explicit raw identity overrides, + root/nested/excluded-registry derived symlinks, and derived prefix publication + integration before fixing those paths. Focused `additional-red.log` command + exited 1: 24 failed, 1 passed, 60 deselected, 14 warnings. +- Finding 2 inventory: six malformed/contradictory identity cases reached + artifact reads; absent separate checksum allowed wrong local bytes as a valid + R2 link. Valid URI-only locator control passed. +- Finding 3 additional tree cases reached upload or silently skipped the link. + Finding 4 explicit overrides reached reads. Finding 1 both-custom explicit + route reached inference; all three prefix environment names were ignored by + publication. Exact command and observations are in the external report. +- Full source-package module completed: direct exit 0, 157 passed, 13 warnings + in 192.07 seconds, including both source-reader fixes. diff --git a/tests/test_chronicle_artifact_peer4.py b/tests/test_chronicle_artifact_peer4.py index 3e62580f..14e3d524 100644 --- a/tests/test_chronicle_artifact_peer4.py +++ b/tests/test_chronicle_artifact_peer4.py @@ -229,3 +229,182 @@ def test_explicit_sweep_reports_nonregular_manifest_sibling( assert exit_info.value.code == 1 assert "regular file" in json.dumps(json.loads(capsys.readouterr().out)) assert manifest_path.read_text() == before + + +@pytest.mark.parametrize("location", ["root", "nested", "excluded-registry"]) +def test_derived_preflight_rejects_symlinks_at_every_tree_boundary( + tmp_path, monkeypatch, location +): + suite = tmp_path / "suite" + outside = tmp_path / "outside" + outside.mkdir() + (outside / "secret.csv").write_bytes(b"outside") + if location == "root": + suite.symlink_to(outside, target_is_directory=True) + else: + suite.mkdir() + target = suite / "build_artifacts.jsonl" + if location == "nested": + (suite / "reports").mkdir() + target = suite / "reports" / "database.json" + target.symlink_to(outside / "secret.csv") + _no_upload(monkeypatch) + report = publish_derived_artifacts( + suite, + source_id="publisher", + package_id="package", + year=2024, + build_id="ledger.build.v1:peer4", + ) + assert not report.valid + assert report.entries == () + + +@pytest.mark.parametrize("field", ["source_id", "package_id"]) +@pytest.mark.parametrize("bad_id", ["foo bar", "a/b", "..", 123, ""]) +def test_raw_publication_refuses_noncanonical_explicit_identity_before_read( + tmp_path, monkeypatch, field, bad_id +): + package, manifest_path, _manifest = _package(tmp_path) + before = manifest_path.read_text() + _no_upload(monkeypatch) + + def unexpected_read(*args, **kwargs): + pytest.fail("artifact read reached with invalid explicit identity") + + monkeypatch.setattr(Path, "read_bytes", unexpected_read) + report = publish_source_artifacts(package, **{field: bad_id}) + assert not report.valid + assert manifest_path.read_text() == before + + +@pytest.mark.parametrize( + "defect", + [ + "contradictory", + "incomplete", + "non-r2", + "empty", + "sha256", + "filename", + "local-digest", + ], +) +def test_inventory_refuses_invalid_recorded_r2_and_does_not_count_link( + tmp_path, monkeypatch, defect +): + package, manifest_path, manifest = _package(tmp_path) + spec = manifest["files"][2024] + key = f"raw/publisher/package/2024/{spec['sha256']}/table.csv" + block = { + "provider": "r2", + "bucket": "archive", + "key": key, + "uri": f"r2://archive/{key}", + } + if defect == "contradictory": + block["bucket"] = "different" + elif defect == "incomplete": + del block["provider"] + elif defect == "non-r2": + block["provider"] = "s3" + block["uri"] = f"s3://archive/{key}" + elif defect == "empty": + block = {} + elif defect == "sha256": + spec["sha256"] = "0" * 64 + elif defect == "filename": + block["key"] = key.replace("table.csv", "other.csv") + block["uri"] = f"r2://archive/{block['key']}" + else: + del spec["sha256"] + (package / "table.csv").write_bytes(b"different local bytes") + spec["storage"] = {"r2": block} + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_text() + if defect != "local-digest": + + def unexpected_read(*args, **kwargs): + pytest.fail("inventory read bytes before refusing invalid R2 identity") + + monkeypatch.setattr(Path, "read_bytes", unexpected_read) + report = inventory_source_artifacts(package) + assert not report.valid + assert report.counts["r2_link_count"] == 0 + assert report.entries[0].r2 is None + assert any("recorded_r2" in error for error in report.entries[0].errors) + assert manifest_path.read_text() == before + + +def test_inventory_accepts_consistent_uri_only_r2_locator(tmp_path): + package, manifest_path, manifest = _package(tmp_path) + spec = manifest["files"][2024] + spec["storage"] = { + "r2": { + "provider": "r2", + "uri": f"r2://archive/raw/publisher/package/2024/{spec['sha256']}/table.csv", + } + } + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + report = inventory_source_artifacts(package) + assert report.valid + assert report.counts["r2_link_count"] == 1 + + +def test_derived_publication_refuses_unrecognized_explicit_route_before_read( + tmp_path, monkeypatch +): + suite = tmp_path / "suite" + suite.mkdir() + (suite / "facts.jsonl").write_bytes(b"{}\n") + _no_upload(monkeypatch) + + def unexpected_read(*args, **kwargs): + pytest.fail("build read reached for an unrecognizable derived route") + + monkeypatch.setattr("chronicle.artifacts.infer_build_id", unexpected_read) + report = publish_derived_artifacts( + suite, + source_id="publisher", + package_id="package", + year=2024, + r2_bucket="chronicle-builds", + r2_prefix="builds", + ) + assert not report.valid + assert "derived_route" in " ".join(report.errors) + + +@pytest.mark.parametrize( + "env_prefix", ["CHRONICLE_", "POLICYENGINE_LEDGER_", "LEDGER_"] +) +def test_derived_publication_propagates_configured_prefix( + tmp_path, monkeypatch, env_prefix +): + from chronicle.consumer_contract import _points_at_derived + + suite = tmp_path / "suite" + suite.mkdir() + (suite / "facts.jsonl").write_bytes(b"{}\n") + monkeypatch.setenv(f"{env_prefix}R2_DERIVED_BUCKET", "chronicle-builds") + monkeypatch.setenv(f"{env_prefix}R2_DERIVED_PREFIX", "builds") + uploaded = [] + + def upload(location, *_args, **_kwargs): + uploaded.append(location) + return ArtifactCommandResult( + command=("test",), returncode=0, stdout="", stderr="" + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", upload) + report = publish_derived_artifacts( + suite, + source_id="publisher", + package_id="package", + year=2024, + build_id="ledger.build.v1:peer4", + ) + assert report.valid + assert len(uploaded) == 1 + assert uploaded[0].key.startswith("builds/") + assert _points_at_derived(uploaded[0].bucket, uploaded[0].key) From 1c3a223eda34c2de61b57a4617edf64eec6eafd9 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:14:05 -0400 Subject: [PATCH 078/105] fix: preflight the complete derived tree before publication --- PROGRESS.md | 9 +++++++++ chronicle/artifacts.py | 38 +++++++++++++++++++++++++++++++++++++- 2 files changed, 46 insertions(+), 1 deletion(-) diff --git a/PROGRESS.md b/PROGRESS.md index 761fe0a0..3881d2df 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -627,3 +627,12 @@ the per-finding reproduction/fix/test/commit/port map. publication. Exact command and observations are in the external report. - Full source-package module completed: direct exit 0, 157 passed, 13 warnings in 192.07 seconds, including both source-reader fixes. + +### Finding 3: complete derived tree preflight + +- Derived publication now walks the root and every descendant with `lstat`, + allowing only directories and regular files. It refuses symlinks/FIFOs, + including an excluded registry filename, before build-ID inference, file + reads, uploads, or registry writes. +- Both failing-first tree groups now pass: direct exit 0, 7 passed, + 78 deselected, 12 warnings. Safe nested directories remain publishable. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 10f1d91c..bd198c62 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -12,6 +12,7 @@ import re import shlex import sqlite3 +import stat import subprocess from typing import Any, Mapping from urllib.parse import unquote, urlparse @@ -980,6 +981,26 @@ def fetch_source_artifact( ) +def _derived_artifact_paths(input_path: Path) -> list[Path]: + """Preflight every build-tree entry before reading or publishing any file.""" + if not stat.S_ISDIR(input_path.lstat().st_mode): + raise ValueError(f"derived_root_not_regular_directory:{input_path}") + artifacts: list[Path] = [] + + def visit(directory: Path) -> None: + for path in sorted(directory.iterdir()): + mode = path.lstat().st_mode + if stat.S_ISDIR(mode): + visit(path) + elif stat.S_ISREG(mode): + artifacts.append(path) + else: + raise ValueError(f"derived_entry_not_regular_file:{path}") + + visit(input_path) + return sorted(artifacts) + + def publish_derived_artifacts( input_dir: str | Path, *, @@ -1022,6 +1043,22 @@ def publish_derived_artifacts( errors=(f"input_dir_is_not_directory:{input_path}",), ) + try: + artifact_paths = _derived_artifact_paths(input_path) + except (OSError, ValueError) as error: + return DerivedArtifactPublishReport( + input_dir=str(input_path), + source_id=source_id, + package_id=package_id, + year=year, + build_id=build_id or "", + entries=(), + build_artifacts_path=str(build_artifacts_output) + if build_artifacts_output + else None, + errors=(str(error),), + ) + resolved_build_id = build_id or infer_build_id(input_path) if not resolved_build_id: return DerivedArtifactPublishReport( @@ -1063,7 +1100,6 @@ def publish_derived_artifacts( ) entries: list[DerivedArtifactUploadEntry] = [] errors: list[str] = [] - artifact_paths = sorted(path for path in input_path.rglob("*") if path.is_file()) for artifact_path in artifact_paths: relative_path = artifact_path.relative_to(input_path).as_posix() if relative_path == "build_artifacts.jsonl": From b223066a2a49df2a3ce482ddf4a704ff2be48bc4 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:14:44 -0400 Subject: [PATCH 079/105] fix: refuse normalized artifact aliases in publication and inventory --- PROGRESS.md | 8 ++++++++ chronicle/artifacts.py | 10 +++++++++- 2 files changed, 17 insertions(+), 1 deletion(-) diff --git a/PROGRESS.md b/PROGRESS.md index 3881d2df..5fd95abc 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -636,3 +636,11 @@ the per-finding reproduction/fix/test/commit/port map. reads, uploads, or registry writes. - Both failing-first tree groups now pass: direct exit 0, 7 passed, 78 deselected, 12 warnings. Safe nested directories remain publishable. + +### Finding 6: exact physical artifact spelling + +- Publish and inventory now turn a single normalized alias with different + physical spelling into `artifact_spelling_mismatch`, and neither reads + bytes after alias errors. Duplicate-alias refusal also stops before reads. +- Case/Unicode regressions plus duplicate/symlink controls exit 0: 6 passed, + 12 warnings. Missing filenames retain their separate missing-name error. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index bd198c62..75a1d359 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -2491,6 +2491,8 @@ def _publish_raw_manifest_entry( matching_directory_entry(manifest_path.parent, filename) or manifest_path.parent / filename ) + if filename and artifact_path.name != filename: + errors.append(f"artifact_spelling_mismatch:{filename}:{artifact_path.name}") except ValueError: errors.append(f"duplicate_artifact_spellings:{filename}") artifact_path = manifest_path.parent / filename @@ -2499,6 +2501,8 @@ def _publish_raw_manifest_entry( size_bytes = None if not filename: errors.append("missing_filename") + elif errors: + pass elif artifact_path.is_symlink(): errors.append(f"artifact_path_is_symlink:{filename}") elif not artifact_path.is_file(): @@ -2723,16 +2727,20 @@ def _inventory_entry( matching_directory_entry(manifest_path.parent, filename) or manifest_path.parent / filename ) + if filename and artifact_path.name != filename: + errors.append(f"artifact_spelling_mismatch:{filename}:{artifact_path.name}") except ValueError: errors.append(f"duplicate_artifact_spellings:{filename}") artifact_path = manifest_path.parent / filename symlink = bool(filename) and artifact_path.is_symlink() - exists = bool(filename) and not symlink and artifact_path.is_file() + exists = bool(filename) and not errors and not symlink and artifact_path.is_file() sha256_expected = spec.get("sha256") sha256_actual = None size_bytes = None if not filename: errors.append("missing_filename") + elif errors: + pass elif symlink: errors.append(f"artifact_path_is_symlink:{filename}") elif not exists: From 1570fecf322d3aef0c25f02d08cbae38825ea88d Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:16:29 -0400 Subject: [PATCH 080/105] fix: validate new publication identities before artifact reads --- PROGRESS.md | 11 +++ chronicle/artifacts.py | 114 +++++++++++++++++-------- tests/test_chronicle_artifact_peer4.py | 20 +++++ 3 files changed, 110 insertions(+), 35 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 5fd95abc..fd232489 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -644,3 +644,14 @@ the per-finding reproduction/fix/test/commit/port map. bytes after alias errors. Duplicate-alias refusal also stops before reads. - Case/Unicode regressions plus duplicate/symlink controls exit 0: 6 passed, 12 warnings. Missing filenames retain their separate missing-name error. + +### Finding 4: canonical identities on new publication paths + +- Derived publication validates source/package identifiers before inference or + artifact reads. Raw publication preserves original argument/declaration types + until new-key validation, rejects noncanonical identifiers and contradictory + declarations before reading artifacts, and retains root-wide preflight. +- Identity regressions plus declaration controls exit 0: 44 passed, + 41 deselected, 12 warnings. Six historical-identity controls and the full + tracked-registry cutover test also pass: 7 passed, 12 warnings. Recorded + objects keep their original routes even when old manifests omit package_id. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 75a1d359..1029ec2f 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1016,6 +1016,22 @@ def publish_derived_artifacts( """Upload a deterministic build output directory to the derived R2 bucket.""" r2_bucket = r2_bucket or default_r2_derived_bucket() input_path = Path(input_dir) + try: + _require_identity_segment(source_id, what="source_id") + _require_identity_segment(package_id, what="package_id") + except SourceArtifactManifestError as error: + return DerivedArtifactPublishReport( + input_dir=str(input_path), + source_id=source_id, + package_id=package_id, + year=year, + build_id=build_id or "", + entries=(), + build_artifacts_path=str(build_artifacts_output) + if build_artifacts_output + else None, + errors=(f"r2_identity_invalid:{error}",), + ) if not input_path.exists(): return DerivedArtifactPublishReport( input_dir=str(input_path), @@ -1190,16 +1206,24 @@ def publish_source_artifacts( errors.append(f"Could not read {manifest_path}: {exc}") continue - manifest_source_id = str(source_id or manifest.get("source_id") or "") - manifest_package_id = str(package_id or manifest.get("package_id") or "") + manifest_source_id = ( + source_id if source_id is not None else manifest.get("source_id") + ) + manifest_package_id = ( + package_id if package_id is not None else manifest.get("package_id") + ) for package_manifest_name, package_manifest in package_manifests.items(): package_manifest_path = Path(package_manifest_name) - package_source_id = str( - source_id or package_manifest.get("source_id") or "" + package_source_id = ( + source_id + if source_id is not None + else package_manifest.get("source_id") ) - package_id_value = str( - package_id or package_manifest.get("package_id") or "" + package_id_value = ( + package_id + if package_id is not None + else package_manifest.get("package_id") ) package_files = _manifest_files(package_manifest, package_manifest_path) for year, spec in package_files.items(): @@ -1213,6 +1237,7 @@ def publish_source_artifacts( r2_prefix=r2_prefix, wrangler_command=wrangler_command, preflight_only=True, + manifest_identity=package_manifest, ) if entry.errors: preflight_failures.append(entry) @@ -1256,6 +1281,7 @@ def publish_source_artifacts( r2_bucket=r2_bucket, r2_prefix=r2_prefix, wrangler_command=wrangler_command, + manifest_identity=manifest, ) entries.append(entry) if updated_spec is not None and isinstance(spec, dict): @@ -2446,12 +2472,38 @@ def _publish_raw_manifest_entry( r2_prefix: str | None, wrangler_command: str, preflight_only: bool = False, + manifest_identity: dict[str, Any] | None = None, ) -> tuple[RawArtifactPublishEntry, dict[str, Any] | None]: errors: list[str] = [] if not isinstance(spec, dict): spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") + artifact_path = manifest_path.parent + sha256_actual = None + size_bytes = None + + def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: + """Report the entry unpublished, with nothing uploaded or rewritten.""" + if reason is not None: + errors.append(reason) + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(artifact_path), + sha256=sha256_actual, + size_bytes=size_bytes, + r2_location=None, + upload=None, + errors=tuple(errors), + ), + None, + ) + if filename and not is_bare_filename(filename): return ( RawArtifactPublishEntry( @@ -2486,6 +2538,27 @@ def _publish_raw_manifest_entry( ), None, ) + try: + recorded_r2 = _validated_recorded_r2( + spec, manifest_path=manifest_path, year=year + ) + except SourceArtifactManifestError as error: + # A block that does not name one object cannot be treated as history, + # and publishing under it would ship whichever field was read. + return refuse(f"recorded_r2_locator_invalid:{error}") + if recorded_r2 is None: + try: + _require_identity_segment(source_id, what="source_id") + _require_identity_segment(package_id, what="package_id") + _assert_manifest_identifies( + manifest_identity or {}, + manifest_path, + source_id=source_id, + package_id=package_id, + ) + except SourceArtifactManifestError as error: + return refuse(f"r2_identity_invalid:{error}") + try: artifact_path = ( matching_directory_entry(manifest_path.parent, filename) @@ -2514,38 +2587,9 @@ def _publish_raw_manifest_entry( if sha256_expected and sha256_actual != sha256_expected: errors.append("checksum_mismatch") - def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: - """Report the entry unpublished, with nothing uploaded or rewritten.""" - if reason is not None: - errors.append(reason) - return ( - RawArtifactPublishEntry( - manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, - year=str(year), - filename=filename, - local_path=str(artifact_path), - sha256=sha256_actual, - size_bytes=size_bytes, - r2_location=None, - upload=None, - errors=tuple(errors), - ), - None, - ) - if errors: return refuse() - try: - recorded_r2 = _validated_recorded_r2( - spec, manifest_path=manifest_path, year=year - ) - except SourceArtifactManifestError as error: - # A block that does not name one object cannot be treated as history, - # and publishing under it would ship whichever field was read. - return refuse(f"recorded_r2_locator_invalid:{error}") if recorded_r2 is not None and (recorded_r2.sha256, recorded_r2.filename) != ( sha256_actual or "", Path(filename).name, diff --git a/tests/test_chronicle_artifact_peer4.py b/tests/test_chronicle_artifact_peer4.py index 14e3d524..eeb89f3a 100644 --- a/tests/test_chronicle_artifact_peer4.py +++ b/tests/test_chronicle_artifact_peer4.py @@ -408,3 +408,23 @@ def upload(location, *_args, **_kwargs): assert len(uploaded) == 1 assert uploaded[0].key.startswith("builds/") assert _points_at_derived(uploaded[0].bucket, uploaded[0].key) + + +@pytest.mark.parametrize("field", ["source_id", "package_id"]) +@pytest.mark.parametrize("declaration", [None, 123, "historical name"]) +def test_raw_publication_preserves_history_without_new_identity_requirements( + tmp_path, monkeypatch, field, declaration +): + package, manifest_path, manifest = _package(tmp_path) + spec = manifest["files"][2024] + key = f"historical/route/{spec['sha256']}/table.csv" + spec["storage"] = {"r2": {"provider": "r2", "uri": f"r2://archive/{key}"}} + manifest[field] = declaration + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_text() + _no_upload(monkeypatch) + report = publish_source_artifacts(package) + assert report.valid + assert report.entries[0].skipped + assert report.entries[0].r2_location.uri == f"r2://archive/{key}" + assert manifest_path.read_text() == before From b7c88a1648ca55c5dcadf34950407cec1e2e313e Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:17:26 -0400 Subject: [PATCH 081/105] fix: validate recorded R2 identity before counting inventory links --- PROGRESS.md | 10 ++++++++++ chronicle/artifacts.py | 42 ++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 50 insertions(+), 2 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index fd232489..df2fba1b 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -655,3 +655,13 @@ the per-finding reproduction/fix/test/commit/port map. 41 deselected, 12 warnings. Six historical-identity controls and the full tracked-registry cutover test also pass: 7 passed, 12 warnings. Recorded objects keep their original routes even when old manifests omit package_id. + +### Finding 2: inventory provenance + +- Inventory now reuses `_validated_recorded_r2` before artifact reads, rejects + contradictory/incomplete/non-R2 locators and mismatched checksum/filename + declarations, and checks actual bytes against the recorded digest even when + a separate checksum is absent. Invalid entries expose no R2 link. +- Seven failing-first defect cases plus the valid URI-only control now pass: + direct exit 0, 8 passed, 83 deselected, 12 warnings. Together with `d5e9952`, + both consumers named in finding 2 now enforce immutable provenance. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 1029ec2f..5f64c9d6 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -2736,8 +2736,36 @@ def _inventory_entry( spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") - storage = spec.get("storage") if isinstance(spec, dict) else None - r2 = storage.get("r2") if isinstance(storage, dict) else None + r2 = None + recorded_r2 = None + try: + recorded_r2 = _validated_recorded_r2( + spec, manifest_path=manifest_path, year=year + ) + except SourceArtifactManifestError as error: + errors.append(f"recorded_r2_locator_invalid:{error}") + if recorded_r2 is not None and ( + recorded_r2.filename != filename + or ( + spec.get("sha256") is not None + and recorded_r2.sha256 != spec["sha256"] + ) + ): + errors.append("recorded_r2_identity_mismatch") + if errors: + return ArtifactInventoryEntry( + manifest_path=str(manifest_path), + year=str(year), + filename=filename, + local_path=str(manifest_path.parent), + exists=False, + sha256_expected=spec.get("sha256"), + sha256_actual=None, + size_bytes=None, + source_url=spec.get("source_url"), + r2=None, + errors=tuple(errors), + ) if filename and not is_bare_filename(filename): return ArtifactInventoryEntry( manifest_path=str(manifest_path), @@ -2795,6 +2823,16 @@ def _inventory_entry( size_bytes = len(content) if sha256_expected and sha256_actual != sha256_expected: errors.append("checksum_mismatch") + if recorded_r2 is not None: + if sha256_actual is not None and sha256_actual != recorded_r2.sha256: + errors.append("recorded_r2_identity_mismatch") + if not errors: + r2 = { + "provider": recorded_r2.provider, + "bucket": recorded_r2.bucket, + "key": recorded_r2.key, + "uri": recorded_r2.uri, + } return ArtifactInventoryEntry( manifest_path=str(manifest_path), year=str(year), From 9a6345fd2d26d904a1e179285c0aebb8884c49dc Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:18:38 -0400 Subject: [PATCH 082/105] fix: bind custom derived publication to recognized provenance routes --- PROGRESS.md | 12 +++++++ chronicle/artifacts.py | 65 ++++++++++++++++++++++++++++------ chronicle/consumer_contract.py | 28 ++------------- docs/storage-architecture.md | 10 ++++++ 4 files changed, 79 insertions(+), 36 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index df2fba1b..5d24afae 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -665,3 +665,15 @@ the per-finding reproduction/fix/test/commit/port map. - Seven failing-first defect cases plus the valid URI-only control now pass: direct exit 0, 8 passed, 83 deselected, 12 warnings. Together with `d5e9952`, both consumers named in finding 2 now enforce immutable provenance. + +### Finding 1: publication route enforcement + +- Publication and the consumer guard now share `is_derived_r2_route` and the + lazy prefix resolver. Derived key generation honors all three environment + spellings for the configured prefix. +- A both-custom explicit bucket/prefix combination must identify a configured + or archived derived route, otherwise publication refuses it before build + reads. The storage architecture documents the shared route configuration. +- Four failing-first route integration cases now pass: direct exit 0, + 4 passed, 87 deselected, 16 warnings. All nine findings are implemented; + focused integration verification and full final checks remain. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 5f64c9d6..53804ce3 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -268,6 +268,35 @@ def default_r2_derived_prefix() -> str: return env_value("CHRONICLE_R2_DERIVED_PREFIX", default=DEFAULT_R2_DERIVED_PREFIX) +def is_derived_r2_route(bucket: str, key: str) -> bool: + """Whether an R2 bucket/key pair addresses derived build output. + + Resolve publication configuration at validation time: an operator may use + a bucket or prefix with no ``derived`` marker in its spelling. Archived + rename-window routes remain derived after the active destination changes. + """ + derived_buckets = { + "ledger-derived", + "chronicle-derived", + DEFAULT_R2_DERIVED_BUCKET, + default_r2_derived_bucket(), + } + derived_prefixes = { + "derived", + resolve_r2_prefix( + prefix=None, + default_prefix=DEFAULT_R2_DERIVED_PREFIX, + ), + resolve_r2_prefix( + prefix=None, + default_prefix=default_r2_derived_prefix(), + ), + } + return bucket in derived_buckets or any( + key == prefix or key.startswith(f"{prefix}/") for prefix in derived_prefixes + ) + + class SourceArtifactManifestError(RuntimeError): """A manifest refuses the write a fetch is about to make. @@ -1032,6 +1061,30 @@ def publish_derived_artifacts( else None, errors=(f"r2_identity_invalid:{error}",), ) + try: + resolved_r2_prefix = resolve_r2_prefix( + prefix=r2_prefix, + default_prefix=default_r2_derived_prefix(), + source_id=source_id, + ) + if not is_derived_r2_route(r2_bucket, resolved_r2_prefix): + raise ValueError( + "Set CHRONICLE_R2_DERIVED_BUCKET or CHRONICLE_R2_DERIVED_PREFIX " + "to identify a custom derived route before publishing to it." + ) + except ValueError as error: + return DerivedArtifactPublishReport( + input_dir=str(input_path), + source_id=source_id, + package_id=package_id, + year=year, + build_id=build_id or "", + entries=(), + build_artifacts_path=str(build_artifacts_output) + if build_artifacts_output + else None, + errors=(f"derived_route_invalid:{error}",), + ) if not input_path.exists(): return DerivedArtifactPublishReport( input_dir=str(input_path), @@ -1109,11 +1162,6 @@ def publish_derived_artifacts( errors=("malformed_build_id",), ) - resolved_r2_prefix = resolve_r2_prefix( - prefix=r2_prefix, - default_prefix=DEFAULT_R2_DERIVED_PREFIX, - source_id=source_id, - ) entries: list[DerivedArtifactUploadEntry] = [] errors: list[str] = [] for artifact_path in artifact_paths: @@ -1574,7 +1622,7 @@ def build_derived_r2_key( """Build the canonical R2 key for a derived build artifact.""" resolved_prefix = resolve_r2_prefix( prefix=prefix, - default_prefix=DEFAULT_R2_DERIVED_PREFIX, + default_prefix=default_r2_derived_prefix(), source_id=source_id, ) return posixpath.join( @@ -2746,10 +2794,7 @@ def _inventory_entry( errors.append(f"recorded_r2_locator_invalid:{error}") if recorded_r2 is not None and ( recorded_r2.filename != filename - or ( - spec.get("sha256") is not None - and recorded_r2.sha256 != spec["sha256"] - ) + or (spec.get("sha256") is not None and recorded_r2.sha256 != spec["sha256"]) ): errors.append("recorded_r2_identity_mismatch") if errors: diff --git a/chronicle/consumer_contract.py b/chronicle/consumer_contract.py index a4766193..846bde51 100644 --- a/chronicle/consumer_contract.py +++ b/chronicle/consumer_contract.py @@ -492,32 +492,8 @@ def _is_derived_source_record_id(source_record_id: str) -> bool: def _points_at_derived(bucket: str, key: str) -> bool: - """Whether an R2 bucket/key pair addresses derived build output. - - Resolve publication configuration at validation time: an operator may use - a bucket or prefix with no ``derived`` marker in its spelling. Archived - rename-window routes remain derived after the active destination changes. - """ - derived_buckets = { - "ledger-derived", - "chronicle-derived", - artifacts.DEFAULT_R2_DERIVED_BUCKET, - artifacts.default_r2_derived_bucket(), - } - derived_prefixes = { - "derived", - artifacts.resolve_r2_prefix( - prefix=None, - default_prefix=artifacts.DEFAULT_R2_DERIVED_PREFIX, - ), - artifacts.resolve_r2_prefix( - prefix=None, - default_prefix=artifacts.default_r2_derived_prefix(), - ), - } - return bucket in derived_buckets or any( - key == prefix or key.startswith(f"{prefix}/") for prefix in derived_prefixes - ) + """Use the same derived routes publication is permitted to address.""" + return artifacts.is_derived_r2_route(bucket, key) def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index ca770ab8..91cacbb1 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -90,6 +90,16 @@ derived/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{build_id}/chr Legacy US derived keys likewise remain `derived/{source_id}/...`. +The derived prefix defaults to `derived` and can be configured with +`CHRONICLE_R2_DERIVED_PREFIX`, using the same legacy environment fallback as +the bucket. Publisher and consumer validation must share this route configuration: +facts citing a configured derived bucket or prefix are refused. The archived +`ledger-derived`, `chronicle-derived`, and `derived/` routes remain derived. +An explicit `publish-derived --r2-bucket ... --r2-prefix ...` combination must +use a recognized derived bucket or prefix; configure a custom route through +the environment before publishing it. This keeps custom build locations +identifiable at the publisher-fact boundary. + Derived artifacts are reproducible and may be replaced by a new build, but a specific `{build_id}` path should be immutable once published. From 62cc35101aac2a7da59b764d80fac75b263b1381 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:23:18 -0400 Subject: [PATCH 083/105] fix: recognize derived R2 provenance regardless of URI scheme case --- PROGRESS.md | 15 ++++++++++++++ chronicle/consumer_contract.py | 4 ++-- tests/test_chronicle_consumer_contract.py | 25 +++++++++++++++++++++++ 3 files changed, 42 insertions(+), 2 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 5d24afae..49501436 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -677,3 +677,18 @@ the per-finding reproduction/fix/test/commit/port map. - Four failing-first route integration cases now pass: direct exit 0, 4 passed, 87 deselected, 16 warnings. All nine findings are implemented; focused integration verification and full final checks remain. + +### Final review follow-ups + +- Focused integration passed: direct exit 0, 409 passed, 40 warnings in 15.10s. + Ruff lint passed and all 8 changed Python files passed formatting checks. +- Interrupted the first full pytest run (direct exit 130) after peer review + found that date/set YAML identity refusals could fail CLI JSON serialization. + It is not a completed full-suite verification and will be restarted. +- Reproduced a remaining configured-route bypass with uppercase `R2://` in + raw URI, source_file, or source URL: 3 failed, 3 lowercase controls passed. + URI scheme matching is now case-insensitive. Full consumer module now passes: + direct exit 0, 101 passed, 32 warnings; scoped Ruff checks pass. +- Additional failing-first checks cover date/set report serialization (8 failed), + vintage namespace escapes and build-ID separators (8 failed), and fetch + vintage namespace escapes. These publication fixes precede the full restart. diff --git a/chronicle/consumer_contract.py b/chronicle/consumer_contract.py index 846bde51..a98256d8 100644 --- a/chronicle/consumer_contract.py +++ b/chronicle/consumer_contract.py @@ -471,7 +471,7 @@ def validate_consumer_fact_contract( def _r2_uri_parts(uri: str) -> tuple[str, str]: """Split an ``r2://bucket/key`` URI into its bucket and key.""" - if not uri.startswith("r2://"): + if uri[: len("r2://")].lower() != "r2://": return "", "" bucket, _, key = uri[len("r2://") :].partition("/") return bucket, key @@ -517,7 +517,7 @@ def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: "itself. Target construction, aging, and reconciliation belong in " "Microcosm." ) - if source_file.startswith("r2://"): + if source_file[: len("r2://")].lower() == "r2://": source_file_bucket, source_file_key = _r2_uri_parts(source_file) else: source_file_bucket, bucket_separator, source_file_key = source_file.partition( diff --git a/tests/test_chronicle_consumer_contract.py b/tests/test_chronicle_consumer_contract.py index 60e0844e..08a26c51 100644 --- a/tests/test_chronicle_consumer_contract.py +++ b/tests/test_chronicle_consumer_contract.py @@ -1027,6 +1027,31 @@ def test_consumer_contract_derived_routes_match_complete_names(monkeypatch): assert validate_consumer_fact_contract([publisher]).valid +@pytest.mark.parametrize("field", ["raw_r2_uri", "source_file", "url"]) +@pytest.mark.parametrize("scheme", ["R2", "r2"]) +def test_consumer_contract_rejects_derived_uri_scheme_case( + monkeypatch, tmp_path, field, scheme +): + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-builds") + monkeypatch.setenv("CHRONICLE_R2_DERIVED_PREFIX", "builds") + fact = _soi_agi_fact() + derived = replace( + fact, + source=replace( + fact.source, + **{field: f"{scheme}://chronicle-builds/builds/source/fact.json"}, + ), + ) + + report = validate_consumer_fact_contract([derived]) + + assert "derived_fact_provenance" in {error.code for error in report.errors} + output = tmp_path / "new-directory" / "consumer_facts.jsonl" + with pytest.raises(ValueError, match="consumer-contract"): + write_consumer_facts_jsonl([derived], output) + assert not output.parent.exists() + + def test_derived_record_marker_is_rejected_in_either_spelling(): """Both rename-window spellings produce the identical boundary error.""" fact = _soi_agi_fact() From b3511ed4b5782f25084a9559b5c6df89be51a3c8 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:23:40 -0400 Subject: [PATCH 084/105] test: cover noncanonical report identities and namespace escapes --- PROGRESS.md | 7 ++ tests/test_chronicle_artifact_peer4.py | 95 ++++++++++++++++++++++++++ 2 files changed, 102 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 49501436..70f4301e 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -692,3 +692,10 @@ the per-finding reproduction/fix/test/commit/port map. - Additional failing-first checks cover date/set report serialization (8 failed), vintage namespace escapes and build-ID separators (8 failed), and fetch vintage namespace escapes. These publication fixes precede the full restart. + +- Committed the publication follow-up regressions before their fixes. Red + commands/logs: `report-identity-red.log` (8 failed), + `namespace-followup-red.log` (8 failed, 99 deselected), and + `fetch-vintage-red.log` (3 failed); all direct exits 1. CLI refusals hit + non-JSON date/set values; malformed vintage/build segments reached reads or + upload sentinels. Exact commands will be included in the final report. diff --git a/tests/test_chronicle_artifact_peer4.py b/tests/test_chronicle_artifact_peer4.py index eeb89f3a..f63f5935 100644 --- a/tests/test_chronicle_artifact_peer4.py +++ b/tests/test_chronicle_artifact_peer4.py @@ -428,3 +428,98 @@ def test_raw_publication_preserves_history_without_new_identity_requirements( assert report.entries[0].skipped assert report.entries[0].r2_location.uri == f"r2://archive/{key}" assert manifest_path.read_text() == before + + +@pytest.mark.parametrize("field", ["source_id", "package_id"]) +@pytest.mark.parametrize("yaml_identity", ["2024-01-01", "!!set {legacy: null}"]) +@pytest.mark.parametrize("entrypoint", ["harness", "cli"]) +def test_raw_cli_serializes_noncanonical_yaml_identity_refusals( + tmp_path, monkeypatch, capsys, field, yaml_identity, entrypoint +): + package, manifest_path, manifest = _package(tmp_path) + manifest[field] = yaml.safe_load(yaml_identity) + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + before = manifest_path.read_text() + _no_upload(monkeypatch) + args = ["publish-raw", "--root", str(package)] + if entrypoint == "harness": + assert harness_main(args) == 1 + else: + monkeypatch.setattr("sys.argv", ["chronicle", *args]) + with pytest.raises(SystemExit) as exit_info: + cli_main() + assert exit_info.value.code == 1 + payload = json.loads(capsys.readouterr().out) + assert not payload["valid"] + assert isinstance(payload["entries"][0][field], str) + assert "r2_identity_invalid" in " ".join(payload["entries"][0]["errors"]) + assert manifest_path.read_text() == before + + +@pytest.mark.parametrize("bad_year", ["/2024", "..", "2024/elsewhere"]) +@pytest.mark.parametrize("operation", ["raw", "derived"]) +def test_publication_refuses_vintage_namespace_escape_before_reads( + tmp_path, monkeypatch, bad_year, operation +): + package, manifest_path, manifest = _package(tmp_path) + manifest["files"][bad_year] = manifest["files"].pop(2024) + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + _no_upload(monkeypatch) + + def unexpected_read(*args, **kwargs): + pytest.fail("publication read reached with a vintage namespace escape") + + monkeypatch.setattr(Path, "read_bytes", unexpected_read) + if operation == "raw": + report = publish_source_artifacts(package) + else: + report = publish_derived_artifacts( + package, + source_id="publisher", + package_id="package", + year=bad_year, + build_id="ledger.build.v1:peer4", + r2_bucket="custom-store", + ) + assert not report.valid + + +@pytest.mark.parametrize( + "bad_build_id", ["ledger.build.v1:bad/id", "ledger.build.v1:bad id"] +) +def test_derived_publication_refuses_noncanonical_build_segment( + tmp_path, monkeypatch, bad_build_id +): + suite = tmp_path / "suite" + suite.mkdir() + (suite / "facts.jsonl").write_bytes(b"{}\n") + _no_upload(monkeypatch) + report = publish_derived_artifacts( + suite, + source_id="publisher", + package_id="package", + year=2024, + build_id=bad_build_id, + ) + assert not report.valid + assert report.errors == ("malformed_build_id",) + + +@pytest.mark.parametrize("bad_year", ["/2024", "..", "2024/elsewhere"]) +def test_fetch_refuses_vintage_namespace_escape_before_publisher_io( + tmp_path, monkeypatch, bad_year +): + package, _manifest_path, _manifest = _package(tmp_path) + + def unexpected_read(*args, **kwargs): + pytest.fail("fetch reached publisher with a vintage namespace escape") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + with pytest.raises(SourceArtifactManifestError, match="year"): + fetch_source_artifact( + "https://example.test/table.csv", + source_id="publisher", + package_id="package", + year=bad_year, + output_dir=package, + ) From b87231fe96ff5a7ef2d4ee6c48670f300caf9239 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:24:13 -0400 Subject: [PATCH 085/105] fix: serialize raw identity refusals without weakening validation --- PROGRESS.md | 5 +++++ chronicle/artifacts.py | 28 ++++++++++++++++------------ 2 files changed, 21 insertions(+), 12 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 70f4301e..9a3ef769 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -699,3 +699,8 @@ the per-finding reproduction/fix/test/commit/port map. `fetch-vintage-red.log` (3 failed); all direct exits 1. CLI refusals hit non-JSON date/set values; malformed vintage/build segments reached reads or upload sentinels. Exact commands will be included in the final report. + +- Raw publication now keeps original identity values for validation and uses + separate string fields only in reports. Date/set refusals serialize cleanly + through both CLIs; historical skips also retain serializable identity fields. + The regression and history controls exit 0: 14 passed, 12 warnings. diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 53804ce3..15527983 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -2523,6 +2523,10 @@ def _publish_raw_manifest_entry( manifest_identity: dict[str, Any] | None = None, ) -> tuple[RawArtifactPublishEntry, dict[str, Any] | None]: errors: list[str] = [] + # Validate original values below; reports must still serialize refusals of + # YAML dates, sets, and other non-string declarations through the CLI. + reported_source_id = str(source_id) if source_id is not None else "" + reported_package_id = str(package_id) if package_id is not None else "" if not isinstance(spec, dict): spec = {} errors.append("malformed_file_spec") @@ -2538,8 +2542,8 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, + source_id=reported_source_id, + package_id=reported_package_id, year=str(year), filename=filename, local_path=str(artifact_path), @@ -2556,8 +2560,8 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, + source_id=reported_source_id, + package_id=reported_package_id, year=str(year), filename=filename, local_path=str(manifest_path.parent), @@ -2573,8 +2577,8 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, + source_id=reported_source_id, + package_id=reported_package_id, year=str(year), filename=filename, local_path=str(manifest_path.parent), @@ -2672,8 +2676,8 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, + source_id=reported_source_id, + package_id=reported_package_id, year=str(year), filename=filename, local_path=str(artifact_path), @@ -2722,8 +2726,8 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, + source_id=reported_source_id, + package_id=reported_package_id, year=str(year), filename=filename, local_path=str(artifact_path), @@ -2759,8 +2763,8 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, + source_id=reported_source_id, + package_id=reported_package_id, year=str(year), filename=filename, local_path=str(artifact_path), From 20663fbe074b2d20b823081708d53e2a859fdd3c Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:24:57 -0400 Subject: [PATCH 086/105] fix: keep publication vintage and build segments inside their routes --- PROGRESS.md | 6 ++++++ chronicle/artifacts.py | 4 ++++ 2 files changed, 10 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 9a3ef769..c5c29078 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -704,3 +704,9 @@ the per-finding reproduction/fix/test/commit/port map. separate string fields only in reports. Date/set refusals serialize cleanly through both CLIs; historical skips also retain serializable identity fields. The regression and history controls exit 0: 14 passed, 12 warnings. + +- New publication vintage segments and derived build-ID segments now pass the + same canonical segment validator, preventing API-provided slashes or '..' + from moving an otherwise recognized route. Fetch applies the vintage check + before publisher I/O; historical raw objects retain their existing routes. + All 11 failing-first namespace cases pass (99 deselected, 12 warnings). diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 15527983..fe767245 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -846,6 +846,7 @@ def fetch_source_artifact( # when the prospective R2 key is constructed below. _require_identity_segment(source_id, what="source_id") _require_identity_segment(package_id, what="package_id") + _require_identity_segment(str(year), what="year") resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, @@ -1048,6 +1049,7 @@ def publish_derived_artifacts( try: _require_identity_segment(source_id, what="source_id") _require_identity_segment(package_id, what="package_id") + _require_identity_segment(str(year), what="year") except SourceArtifactManifestError as error: return DerivedArtifactPublishReport( input_dir=str(input_path), @@ -1148,6 +1150,7 @@ def publish_derived_artifacts( # input failure like the ones above, reported rather than raised. try: canonicalize_key("build", resolved_build_id) + _require_identity_segment(resolved_build_id, what="build_id") except ValueError: return DerivedArtifactPublishReport( input_dir=str(input_path), @@ -2602,6 +2605,7 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: try: _require_identity_segment(source_id, what="source_id") _require_identity_segment(package_id, what="package_id") + _require_identity_segment(str(year), what="year") _assert_manifest_identifies( manifest_identity or {}, manifest_path, From d6d9fad12cb444d33f8368afa22a6af8cb80a70d Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:25:37 -0400 Subject: [PATCH 087/105] docs: record final focused checks and full-suite restart --- PROGRESS.md | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index c5c29078..7e4b28c5 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -710,3 +710,15 @@ the per-finding reproduction/fix/test/commit/port map. from moving an otherwise recognized route. Fetch applies the vintage check before publisher I/O; historical raw objects retain their existing routes. All 11 failing-first namespace cases pass (99 deselected, 12 warnings). + +### Final verification in progress + +- Final focused integration exits 0: 434 passed, 40 warnings in 11.11s. +- Final `uv run ruff check .` exits 0 (`All checks passed!`). Final + `uv run ruff format --check` on the 8 changed Python files exits 0 + (`8 files already formatted`). Both use the permitted UV cache. +- `git diff --check fa98993a..HEAD` and the tracked `db/data` no-change check + both exit 0. Worktree clean before this journal update. +- Restart the full suite at code commit `19d037f` with UV offline, offline + OpenTimestamps client selection, and live Supabase credentials removed from + the child environment. This keeps the required full run free of network. From a00e274fe9dc039a4f7fad0ba9eb1a6c061c8bde Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 18:50:44 -0400 Subject: [PATCH 088/105] docs: record completed peer round 4 verification --- PROGRESS.md | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 7e4b28c5..9d1938ee 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -722,3 +722,29 @@ the per-finding reproduction/fix/test/commit/port map. - Restart the full suite at code commit `19d037f` with UV offline, offline OpenTimestamps client selection, and live Supabase credentials removed from the child environment. This keeps the required full run free of network. + +### Final state, done, and next (peer round 4) + +**State:** complete. All nine findings and the peer-review follow-ups are fixed +in small detached-HEAD commits. Final code commit: `19d037f`; the completed +full run included the subsequent journal commit `9fa6a1d`. + +**Done:** + +- Full `uv run pytest -q -p no:cacheprovider` returned direct exit 0: + **1,238 passed, 7 skipped, 42 warnings in 1,468.40s (24m 28s)**. + The child environment used the permitted UV cache, `UV_OFFLINE=1`, an offline + OpenTimestamps client command, and no live Supabase URL/service/secret keys. + Exact invocation and output are in `/tmp/chronicle-226-round4/out.md` and + `/tmp/chronicle-226-round4/full-pytest.log`. +- Final whole-repository Ruff lint exited 0. Formatting checks on every changed + Python file exited 0: 8 files already formatted. Final focused integration + also exited 0: 434 passed, 40 warnings. +- No tracked `db/data/**` changes, GitHub/network access, pushes, branches, + stashes, or edits to PR #227's worktree. Shared helper names remain intact. +- The external report contains each finding's red command and observed failure, + fix, regression names, and commit SHA, plus the final verification and + deliberate boundaries. It records the corrected CLI-test invocation and the + interrupted first full run without counting either as passing verification. + +**Next:** none in this fix lane. No push or branch operation is authorized. From ba8147a7305683db3e73d61395d0bc0a7ec2efba Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 19:16:34 -0400 Subject: [PATCH 089/105] Pin the cutover sweep test by its invariants, not today's registry counts Main gained packages (#239) between the round-4 fixes and the rebase, so the exact 161/194 counts no longer hold; the sweep's contract is that every tracked artifact is a preserved-bucket skip with an R2 link, nothing uploads or fails, and the command exits 0. Floors keep the test from passing on an empty tree. --- tests/test_chronicle_artifacts.py | 38 ++++++++++++++----------------- 1 file changed, 17 insertions(+), 21 deletions(-) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index f32d2d1f..5397f2f2 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -848,27 +848,23 @@ def non_writing_uploader(location, local_path, *, wrangler_command): ] ) report = json.loads(capsys.readouterr().out) - expected_counts = { - "manifest_count": 161, - "artifact_count": 194, - "uploaded_count": 0, - "skipped_count": 194, - "failed_count": 0, - "r2_link_count": 194, - } - - observed = ( - exit_code, - report["valid"], - report["counts"], - len(report["errors"]), - ) - assert observed == ( - 0, - True, - expected_counts, - 0, - ), json.dumps(observed, sort_keys=True) + counts = report["counts"] + + # The tracked registry grows as packages land, so the sweep is pinned by + # its invariants rather than by today's exact counts: every artifact is a + # preserved-bucket skip with an R2 link, nothing uploads or fails, no + # manifest-level error, exit 0. The floors keep the test meaningful. + observed = (exit_code, report["valid"], len(report["errors"])) + assert observed == (0, True, 0), json.dumps( + {"observed": observed, "counts": counts}, sort_keys=True + ) + assert counts["uploaded_count"] == 0, counts + assert counts["failed_count"] == 0, counts + assert ( + counts["skipped_count"] == counts["artifact_count"] == counts["r2_link_count"] + ), counts + assert counts["artifact_count"] >= 194, counts + assert counts["manifest_count"] >= 161, counts assert all(entry["skipped"] or entry["upload"] for entry in report["entries"]) assert uploads == [] assert { From 49a3ad737245e7bd9341ddcda705e4feec6b4e0c Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 19:23:41 -0400 Subject: [PATCH 090/105] Reproduce Astra round-1 findings: legacy -derived routes accepted; YAML-aliased owners skipped on revision --- tests/test_chronicle_artifacts.py | 40 +++++++++++++++++++++++ tests/test_chronicle_consumer_contract.py | 29 ++++++++++++++++ 2 files changed, 69 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 5397f2f2..ac7ce0d1 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3404,3 +3404,43 @@ def test_fetch_refuses_a_dangling_manifest_symlink_instead_of_creating_one( assert (package / "manifest.yaml").is_symlink() assert not (package / "table.xlsx").exists() + + + +def test_record_revision_updates_every_owner_even_when_yaml_aliases_share_one_entry( + tmp_path, +): + """Two vintages sharing one physical file are two owners even when the + manifest spelled them with a YAML anchor and alias (one dict object); + identifying the selected owner by object identity skipped both.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "archive.zip", b"first publication") + _fetch_local(package, source, upload_r2=False) + manifest_path = package / "manifest.yaml" + entry = _entry(manifest_path) + lines = ["source_id: irs_soi", "package_id: soi-table-5", "files:", " 2022: &shared"] + for field, value in entry.items(): + lines.append(f" {field}: {json.dumps(value)}") + lines.append(" 2023: *shared") + manifest_path.write_text("\n".join(lines) + "\n") + old_sha = entry["sha256"] + + source.write_bytes(b"second publication, revised rows") + report = fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-5", + year=2023, + output_dir=package, + upload_r2=False, + record_revision=True, + ) + + assert report.valid + manifest = yaml.safe_load(manifest_path.read_text()) + new_sha = hashlib.sha256(b"second publication, revised rows").hexdigest() + assert manifest["files"][2023]["sha256"] == new_sha + assert manifest["files"][2022]["sha256"] == new_sha, ( + "the aliased owner kept the superseded checksum" + ) + assert old_sha != new_sha diff --git a/tests/test_chronicle_consumer_contract.py b/tests/test_chronicle_consumer_contract.py index 08a26c51..30dd0056 100644 --- a/tests/test_chronicle_consumer_contract.py +++ b/tests/test_chronicle_consumer_contract.py @@ -1127,3 +1127,32 @@ def test_contract_reports_malformed_lineage_keys_instead_of_raising(tmp_path): ValueError, match="Cannot export invalid Chronicle consumer-contract facts" ): write_consumer_facts_jsonl([fact], tmp_path / "facts.jsonl") + + +@pytest.mark.parametrize( + ("bucket", "key"), + [ + ("publisher-derived", "exports/facts.jsonl"), + ("PUBLISHER-Derived", "exports/facts.jsonl"), + ("some-archive", "derived/exports/facts.jsonl"), + ], +) +def test_consumer_contract_keeps_rejecting_legacy_derived_routes(bucket, key): + """Configured routes extend the derived boundary; they never narrow it. + A bucket ending in ``-derived`` or a ``derived/`` key was rejected before + routes became configurable and must still be, under default config.""" + fact = _soi_agi_fact() + derived = replace( + fact, + source=replace( + fact.source, + source_file=f"{bucket}:{key}", + raw_r2_bucket=bucket, + raw_r2_key=key, + raw_r2_uri=f"r2://{bucket}/{key}", + ), + ) + + report = validate_consumer_fact_contract([derived]) + + assert "derived_fact_provenance" in {error.code for error in report.errors} From e72f194ab0eff3a8c19c8d272fe53354b9df4f9b Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 19:27:49 -0400 Subject: [PATCH 091/105] Keep the legacy -derived route rule beside configured routes; identify revision owners by manifest and vintage - is_derived_r2_route treats any bucket ending in "-derived" (case-folded) as derived alongside the configured bucket/prefix set, so configuring a route can extend the consumer boundary but never narrow it; archived facts citing publisher-derived buckets keep failing derived_fact_provenance. - The selected revision owner is identified by (manifest path, vintage key) at both the preflight and the write, not by dict identity: a YAML anchor can make two vintages share one object, which skipped every owner and left the aliased vintage on the superseded checksum. --- chronicle/artifacts.py | 21 +++++++++++++++++---- tests/test_chronicle_artifacts.py | 8 ++++++-- 2 files changed, 23 insertions(+), 6 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index fe767245..7e9b0965 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -292,8 +292,16 @@ def is_derived_r2_route(bucket: str, key: str) -> bool: default_prefix=default_r2_derived_prefix(), ), } - return bucket in derived_buckets or any( - key == prefix or key.startswith(f"{prefix}/") for prefix in derived_prefixes + # Configured routes extend the boundary; they never narrow it. A bucket + # ending in ``-derived`` (any case) or a ``derived/`` key was derived + # before routes became configurable, and archived facts still cite such + # routes, so the legacy spelling rule stays alongside the configured set. + return ( + bucket in derived_buckets + or bucket.casefold().endswith("-derived") + or any( + key == prefix or key.startswith(f"{prefix}/") for prefix in derived_prefixes + ) ) @@ -936,7 +944,12 @@ def fetch_source_artifact( sha256=sha256, ) for owner in owners: - if owner.manifest_path == manifest_path and owner.spec is selected_spec: + # The selected owner is (manifest, vintage key): a YAML anchor can make + # two vintages share one dict object, and object identity would skip + # both. + if owner.manifest_path == manifest_path and str(owner.vintage) == str( + vintage_key + ): continue _assert_recorded_identity_holds_these_bytes( owner.identity, @@ -2456,7 +2469,7 @@ def _upsert_manifest( changed_paths = {manifest_path} if record_revision and revision: for owner in owners: - if owner.manifest_path == manifest_path and owner.spec is recorded_spec: + if owner.manifest_path == manifest_path and str(owner.vintage) == str(key): continue revised_entry = dict(owner.spec) revised_entry.update( diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index ac7ce0d1..1f6e5196 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3406,7 +3406,6 @@ def test_fetch_refuses_a_dangling_manifest_symlink_instead_of_creating_one( assert not (package / "table.xlsx").exists() - def test_record_revision_updates_every_owner_even_when_yaml_aliases_share_one_entry( tmp_path, ): @@ -3418,7 +3417,12 @@ def test_record_revision_updates_every_owner_even_when_yaml_aliases_share_one_en _fetch_local(package, source, upload_r2=False) manifest_path = package / "manifest.yaml" entry = _entry(manifest_path) - lines = ["source_id: irs_soi", "package_id: soi-table-5", "files:", " 2022: &shared"] + lines = [ + "source_id: irs_soi", + "package_id: soi-table-5", + "files:", + " 2022: &shared", + ] for field, value in entry.items(): lines.append(f" {field}: {json.dumps(value)}") lines.append(" 2023: *shared") From ff99d8375770524a1038e9082f7123a1166ade89 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 19:52:23 -0400 Subject: [PATCH 092/105] Judge duplicate manifest keys before YAML merges expand The strict loader flattened `<<` merges first, so an entry that overrode a merged default with its own explicit key was refused as a duplicate. Explicit keys are now checked among themselves before merge expansion; merges then apply with YAML precedence (the explicit key wins), while two explicit spellings of one key are still refused. --- chronicle/registration.py | 18 ++++++++--- tests/test_chronicle_manifest_reading.py | 38 ++++++++++++++++++++++++ 2 files changed, 52 insertions(+), 4 deletions(-) create mode 100644 tests/test_chronicle_manifest_reading.py diff --git a/chronicle/registration.py b/chronicle/registration.py index c580d7c2..c9514f56 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -157,9 +157,14 @@ def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: f"expected a mapping node, but found {node.id}", node.start_mark, ) - self.flatten_mapping(node) - mapping: dict[Any, Any] = {} - for key_node, value_node in node.value: + # Duplicates are judged among the keys the document spells out, before + # ``<<`` merges are expanded: an explicit key that overrides a merged + # default is valid YAML (the explicit key wins), while two explicit + # spellings of one key are the silent shadowing this loader refuses. + explicit: set[Any] = set() + for key_node, _value_node in node.value: + if key_node.tag == "tag:yaml.org,2002:merge": + continue key = self.construct_object(key_node, deep=deep) try: hash(key) @@ -170,13 +175,18 @@ def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: f"found unhashable key ({exc})", key_node.start_mark, ) from exc - if key in mapping: + if key in explicit: raise yaml.constructor.ConstructorError( "while constructing a mapping", node.start_mark, f"found duplicate key {key!r}", key_node.start_mark, ) + explicit.add(key) + self.flatten_mapping(node) + mapping: dict[Any, Any] = {} + for key_node, value_node in node.value: + key = self.construct_object(key_node, deep=deep) mapping[key] = self.construct_object(value_node, deep=deep) return mapping diff --git a/tests/test_chronicle_manifest_reading.py b/tests/test_chronicle_manifest_reading.py new file mode 100644 index 00000000..bff87035 --- /dev/null +++ b/tests/test_chronicle_manifest_reading.py @@ -0,0 +1,38 @@ +"""Manifest reading contracts shared by every consumer on this branch.""" + +from __future__ import annotations + +import pytest +import yaml + + +def test_strict_loader_keeps_yaml_merge_overrides_and_refuses_explicit_duplicates(): + """A ``<<: *defaults`` merge followed by an explicit override is valid YAML + (the explicit key wins); only two explicit spellings of one key are a + duplicate the loader must refuse.""" + from chronicle.registration import load_manifest_document + + merged = load_manifest_document( + "defaults: &defaults\n" + " source_url: https://publisher.test/a\n" + " filename: table.csv\n" + "files:\n" + " 2024:\n" + " <<: *defaults\n" + " source_url: https://publisher.test/b\n" + " sha256: " + "ab" * 32 + "\n" + ) + + assert merged["files"][2024]["source_url"] == "https://publisher.test/b" + assert merged["files"][2024]["filename"] == "table.csv" + + with pytest.raises(yaml.YAMLError, match="duplicate key 'source_url'"): + load_manifest_document( + "defaults: &defaults\n" + " filename: table.csv\n" + "files:\n" + " 2024:\n" + " <<: *defaults\n" + " source_url: https://publisher.test/a\n" + " source_url: https://publisher.test/b\n" + ) From 9379a54c7d276819c16c8bbf769266a7ae938f1f Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 20:19:02 -0400 Subject: [PATCH 093/105] Expand YAML merges without mutating nodes and validate every merged mapping PyYAML's flatten_mapping rewrites a node in place and construction is lazy, so a later `<<: *entry` merge turned an anchored entry's inherited keys into apparent explicit duplicates, and an inline merge mapping's own duplicate keys were consumed without any check. The strict loader now splits each mapping into merge sources and explicit pairs (refusing explicit duplicates), expands merge sources recursively into a fresh pair list with the same check applied to every mapping reached through a merge, and builds the result with YAML precedence, leaving every node untouched. --- chronicle/registration.py | 88 +++++++++++++++++++----- tests/test_chronicle_manifest_reading.py | 32 +++++++++ 2 files changed, 101 insertions(+), 19 deletions(-) diff --git a/chronicle/registration.py b/chronicle/registration.py index c9514f56..b6c39931 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -147,23 +147,30 @@ class StrictManifestLoader(yaml.SafeLoader): would read as one entry and the shadowed entry would be dropped by the next write. A manifest is the record the byte boundary is decided from, so a document the loader cannot represent faithfully is malformed. + + ``<<`` merges are honoured with YAML precedence (an explicit key overrides + a merged one) but never by mutating a node: PyYAML's ``flatten_mapping`` + rewrites the node in place, and because construction is lazy a later + merge of an anchored entry would turn that entry's inherited keys into + apparent explicit duplicates. Merge sources are expanded into a fresh + pair list instead, and every mapping reached through a merge gets the + same duplicate check as a mapping the document spells out directly. """ - def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: - if not isinstance(node, yaml.MappingNode): - raise yaml.constructor.ConstructorError( - None, - None, - f"expected a mapping node, but found {node.id}", - node.start_mark, - ) - # Duplicates are judged among the keys the document spells out, before - # ``<<`` merges are expanded: an explicit key that overrides a merged - # default is valid YAML (the explicit key wins), while two explicit - # spellings of one key are the silent shadowing this loader refuses. - explicit: set[Any] = set() - for key_node, _value_node in node.value: - if key_node.tag == "tag:yaml.org,2002:merge": + _MERGE_TAG = "tag:yaml.org,2002:merge" + + def _explicit_pairs(self, node: Any, deep: bool) -> tuple[list[Any], list[Any]]: + """Split a mapping node into merge sources and its explicit pairs. + + Refuses duplicate explicit keys on the node's own pair list, before any + merge is consulted. + """ + merge_sources: list[Any] = [] + explicit_pairs: list[Any] = [] + seen: set[Any] = set() + for key_node, value_node in node.value: + if key_node.tag == self._MERGE_TAG: + merge_sources.append(value_node) continue key = self.construct_object(key_node, deep=deep) try: @@ -175,17 +182,60 @@ def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: f"found unhashable key ({exc})", key_node.start_mark, ) from exc - if key in explicit: + if key in seen: raise yaml.constructor.ConstructorError( "while constructing a mapping", node.start_mark, f"found duplicate key {key!r}", key_node.start_mark, ) - explicit.add(key) - self.flatten_mapping(node) + seen.add(key) + explicit_pairs.append((key_node, value_node)) + return merge_sources, explicit_pairs + + def _merged_pairs(self, source: Any, deep: bool) -> list[Any]: + """Return the pairs a ``<<`` source contributes, validated, unmutated.""" + if isinstance(source, yaml.MappingNode): + nested_sources, explicit = self._explicit_pairs(source, deep) + pairs: list[Any] = [] + for nested in nested_sources: + pairs.extend(self._merged_pairs(nested, deep)) + pairs.extend(explicit) + return pairs + if isinstance(source, yaml.SequenceNode): + pairs = [] + for subnode in source.value: + if not isinstance(subnode, yaml.MappingNode): + raise yaml.constructor.ConstructorError( + "while constructing a mapping", + source.start_mark, + f"expected a mapping for merging, but found {subnode.id}", + subnode.start_mark, + ) + pairs.extend(self._merged_pairs(subnode, deep)) + return pairs + raise yaml.constructor.ConstructorError( + "while constructing a mapping", + source.start_mark, + f"expected a mapping or list of mappings for merging, but found {source.id}", + source.start_mark, + ) + + def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: + if not isinstance(node, yaml.MappingNode): + raise yaml.constructor.ConstructorError( + None, + None, + f"expected a mapping node, but found {node.id}", + node.start_mark, + ) + merge_sources, explicit_pairs = self._explicit_pairs(node, deep) + merged_pairs: list[Any] = [] + for source in merge_sources: + merged_pairs.extend(self._merged_pairs(source, deep)) mapping: dict[Any, Any] = {} - for key_node, value_node in node.value: + # Merged pairs first, explicit pairs last: YAML precedence, last wins. + for key_node, value_node in [*merged_pairs, *explicit_pairs]: key = self.construct_object(key_node, deep=deep) mapping[key] = self.construct_object(value_node, deep=deep) return mapping diff --git a/tests/test_chronicle_manifest_reading.py b/tests/test_chronicle_manifest_reading.py index bff87035..0f53deca 100644 --- a/tests/test_chronicle_manifest_reading.py +++ b/tests/test_chronicle_manifest_reading.py @@ -36,3 +36,35 @@ def test_strict_loader_keeps_yaml_merge_overrides_and_refuses_explicit_duplicate " source_url: https://publisher.test/a\n" " source_url: https://publisher.test/b\n" ) + + +def test_strict_loader_does_not_mutate_anchored_entries_when_merged_later(): + """Constructing a later merge of an anchored entry must not turn that + entry's inherited keys into 'explicit' ones: PyYAML constructs lazily and + flattens merged nodes in place.""" + from chronicle.registration import load_manifest_document + + document = load_manifest_document( + "defaults: &d\n" + " source_url: old\n" + "files:\n" + " 2024: &e\n" + " <<: *d\n" + " source_url: new\n" + "latest:\n" + " <<: *e\n" + ) + + assert document["files"][2024]["source_url"] == "new" + assert document["latest"]["source_url"] == "new" + + +def test_strict_loader_validates_duplicate_keys_inside_inline_merges(): + """A mapping reached through ``<<`` is still a mapping the document + spells out; duplicate keys inside it must be refused, not collapsed.""" + from chronicle.registration import load_manifest_document + + with pytest.raises(yaml.YAMLError, match="duplicate key 'files'"): + load_manifest_document( + "<<: {files: {2023: {filename: old.csv}}, files: {2024: {filename: new.csv}}}\n" + ) From 4be3854fd27bd4834f032ae631200619a05a0eaf Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 20:27:08 -0400 Subject: [PATCH 094/105] test: pin YAML merge-sequence precedence for artifact selection Failing-first regression for Astra gate round 4: a <<: [first, second] merge must select first's filename/sha256 (YAML merge-key precedence, matching yaml.safe_load), with explicit keys still overriding every merged source, including nested merge sequences. --- tests/test_chronicle_manifest_reading.py | 55 ++++++++++++++++++++++++ 1 file changed, 55 insertions(+) diff --git a/tests/test_chronicle_manifest_reading.py b/tests/test_chronicle_manifest_reading.py index 0f53deca..04b8db63 100644 --- a/tests/test_chronicle_manifest_reading.py +++ b/tests/test_chronicle_manifest_reading.py @@ -38,6 +38,61 @@ def test_strict_loader_keeps_yaml_merge_overrides_and_refuses_explicit_duplicate ) +def test_strict_loader_keeps_merge_sequence_precedence_for_artifact_selection(): + """``<<: [first, second]`` selects ``first``'s values: earlier mappings in a + merge sequence take precedence over later ones (YAML merge-key semantics, + and what ``yaml.safe_load`` does), while an explicit key in the entry still + overrides every merged source. The selected ``filename``/``sha256`` pair is + what fetch, publish, and the source-package reader use to pick bytes, so + reversing the precedence silently selects a different artifact.""" + from chronicle.registration import load_manifest_document + + document = ( + "first: &first\n" + " filename: first.csv\n" + " sha256: " + "aa" * 32 + "\n" + " source_url: https://publisher.test/first\n" + "second: &second\n" + " filename: second.csv\n" + " sha256: " + "bb" * 32 + "\n" + " source_url: https://publisher.test/second\n" + " licence: second-only\n" + "files:\n" + " 2024:\n" + " <<: [*first, *second]\n" + " 2023:\n" + " <<: [*first, *second]\n" + " filename: explicit.csv\n" + " 2022:\n" + " <<: [{filename: inline-first.csv}, {filename: inline-second.csv}]\n" + " 2021:\n" + " <<:\n" + " - <<: [*second, *first]\n" + " source_url: https://publisher.test/nested\n" + " - *first\n" + ) + + strict = load_manifest_document(document) + reference = yaml.safe_load(document) + assert strict["files"] == reference["files"] + + selected = strict["files"][2024] + assert selected["filename"] == "first.csv" + assert selected["sha256"] == "aa" * 32 + assert selected["source_url"] == "https://publisher.test/first" + # Keys only the later source carries are still merged in. + assert selected["licence"] == "second-only" + # An explicit key beats every merged source; the rest still follow first. + assert strict["files"][2023]["filename"] == "explicit.csv" + assert strict["files"][2023]["sha256"] == "aa" * 32 + assert strict["files"][2022]["filename"] == "inline-first.csv" + # Nested: the first sequence entry is itself a merge whose own explicit + # key wins inside it, and whose [second, first] order selects second. + nested = strict["files"][2021] + assert nested["filename"] == "second.csv" + assert nested["source_url"] == "https://publisher.test/nested" + + def test_strict_loader_does_not_mutate_anchored_entries_when_merged_later(): """Constructing a later merge of an anchored entry must not turn that entry's inherited keys into 'explicit' ones: PyYAML constructs lazily and From c41567ef2546ae95ed2f6244f071e53ba7f44422 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 20:27:08 -0400 Subject: [PATCH 095/105] fix(registration): restore merge-sequence precedence in the strict loader The non-mutating merge expansion appended a <<: [a, b] sequence's mappings in document order and assigned last-wins, so the LAST mapping won; YAML gives precedence to the FIRST. Contribute later mappings first so the first mapping's pairs are assigned last. Validation of every sequence entry still happens before any pair is contributed. --- chronicle/registration.py | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/chronicle/registration.py b/chronicle/registration.py index b6c39931..2d44e798 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -203,7 +203,6 @@ def _merged_pairs(self, source: Any, deep: bool) -> list[Any]: pairs.extend(explicit) return pairs if isinstance(source, yaml.SequenceNode): - pairs = [] for subnode in source.value: if not isinstance(subnode, yaml.MappingNode): raise yaml.constructor.ConstructorError( @@ -212,6 +211,11 @@ def _merged_pairs(self, source: Any, deep: bool) -> list[Any]: f"expected a mapping for merging, but found {subnode.id}", subnode.start_mark, ) + # YAML merge-key precedence: earlier mappings in a ``<<`` sequence + # win over later ones. Pairs are assigned last-wins, so contribute + # the later mappings first and the first mapping last. + pairs = [] + for subnode in reversed(source.value): pairs.extend(self._merged_pairs(subnode, deep)) return pairs raise yaml.constructor.ConstructorError( From 3cc4738878f307328ea430ff777ce1df9cff30a0 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 20:44:31 -0400 Subject: [PATCH 096/105] test: reproduce recursive YAML merges and a false shared-file collision Failing-first regressions for Astra gate 6aa29716 round 1: - a mapping that merges itself (directly, through a merge sequence, or through a merged mapping's own merge) must be refused as a yaml.YAMLError, not escape as RecursionError; - two manifests identifying one package-local file through identical content-addressed R2 locators without a declared sha256 must stay valid for every sweep after an identical-byte refetch records sha256 on the selected manifest only. --- tests/test_chronicle_artifacts.py | 90 ++++++++++++++++++++++++ tests/test_chronicle_manifest_reading.py | 24 +++++++ 2 files changed, 114 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 1f6e5196..d423e391 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3448,3 +3448,93 @@ def test_record_revision_updates_every_owner_even_when_yaml_aliases_share_one_en "the aliased owner kept the superseded checksum" ) assert old_sha != new_sha + + +def test_identical_byte_refetch_keeps_r2_identified_shared_file_valid(tmp_path): + """Two manifests may identify one package-local file through identical + content-addressed R2 locators without declaring ``sha256``. Refetching + the same bytes records ``sha256`` on the selected manifest only; the + sibling's effective identity (its recorded R2 key) still agrees, so the + package directory must stay valid for every sweep.""" + from chronicle.artifacts import ( + _effective_recorded_digest, + default_r2_raw_bucket, + ) + from chronicle.registration import validate_package_directory + + def collisions(manifests): + # The sweeps resolve each entry's effective identity (its recorded + # content-addressed R2 key when no ``sha256`` is declared) exactly + # like this before comparing owners. + return validate_package_directory( + manifests, entry_digest=_effective_recorded_digest + ) + + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + content = b"shared publisher table" + filename = "shared.csv" + sha256 = hashlib.sha256(content).hexdigest() + (package / filename).write_bytes(content) + source = tmp_path / "publisher-download.csv" + source.write_bytes(content) + bucket = default_r2_raw_bucket() + manifest_paths = (package / "manifest_a.yaml", package / "manifest_b.yaml") + for path in manifest_paths: + key = build_r2_key( + source_id="publisher", + package_id=path.stem, + year=2024, + sha256=sha256, + filename=filename, + ) + path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": path.stem, + "files": { + 2024: { + "filename": filename, + "source_url": str(source), + "storage": { + "r2": { + "provider": "r2", + "bucket": bucket, + "key": key, + "uri": f"r2://{bucket}/{key}", + } + }, + } + }, + }, + sort_keys=False, + ) + ) + manifests = {str(path): yaml.safe_load(path.read_text()) for path in manifest_paths} + assert collisions(manifests) == () + assert inventory_source_artifacts(package).valid + + report = fetch_source_artifact( + str(source), + source_id="publisher", + package_id="manifest_a", + year=2024, + output_dir=package, + filename=filename, + manifest_filename="manifest_a.yaml", + ) + assert report.valid + assert report.sha256 == sha256 + + manifests = {str(path): yaml.safe_load(path.read_text()) for path in manifest_paths} + assert manifests[str(manifest_paths[0])]["files"][2024]["sha256"] == sha256 + assert "sha256" not in manifests[str(manifest_paths[1])]["files"][2024] + assert collisions(manifests) == () + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package) + for sweep in (inventory, published): + assert not any("filename_collision" in error for error in sweep.errors) + assert not any("identify different bytes" in error for error in sweep.errors) + assert sweep.valid diff --git a/tests/test_chronicle_manifest_reading.py b/tests/test_chronicle_manifest_reading.py index 04b8db63..d6589c48 100644 --- a/tests/test_chronicle_manifest_reading.py +++ b/tests/test_chronicle_manifest_reading.py @@ -93,6 +93,30 @@ def test_strict_loader_keeps_merge_sequence_precedence_for_artifact_selection(): assert nested["source_url"] == "https://publisher.test/nested" +def test_strict_loader_refuses_recursive_merges_as_yaml_errors(): + """A mapping that merges itself (directly or through a nested merge) has + no expansion. The loader must refuse it with a ``yaml.YAMLError`` that the + manifest-reading commands already handle, never a ``RecursionError``.""" + from chronicle.registration import load_manifest_document + + for document in ( + # Direct self-merge. + "files: &f\n 2024:\n filename: table.csv\n <<: *f\n", + # Self-merge through a nested merge sequence. + "files: &f\n 2024:\n filename: table.csv\n <<:\n - {filename: other.csv}\n - *f\n", + # Self-merge through a merged mapping's own merge. + "a: &a\n filename: table.csv\n <<: {<<: *a}\n", + ): + with pytest.raises(yaml.YAMLError, match="recursive"): + load_manifest_document(document) + + # A merge that only *repeats* a source is not a cycle. + repeated = load_manifest_document( + "d: &d\n filename: table.csv\nfiles:\n 2024:\n <<: [*d, *d]\n" + ) + assert repeated["files"][2024] == {"filename": "table.csv"} + + def test_strict_loader_does_not_mutate_anchored_entries_when_merged_later(): """Constructing a later merge of an anchored entry must not turn that entry's inherited keys into 'explicit' ones: PyYAML constructs lazily and From 9681c1c25032be813459e768dafaba79d99f37ed Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 20:44:31 -0400 Subject: [PATCH 097/105] fix: refuse recursive merges; compare effective identities across manifests registration: the strict loader tracks the mappings whose merges are being expanded and refuses a source that is already active with a ConstructorError ("found a recursive merge"), so self-merging manifests fail through the readers' existing yaml.YAMLError handling instead of a RecursionError traceback. Repeated (non-cyclic) sources still merge. artifacts: validate_package_directory now accepts an entry_digest resolver and the owner-agreement preflight passes _effective_recorded_digest, which resolves the digest an entry's recorded content-addressed R2 key encodes (the identity _recorded_identity uses) before falling back to the declared sha256 field. A sibling that only carries the locator no longer reads as an empty digest after an identical-byte refetch, so inventory and publication stop reporting a false filename_collision_across_manifests. --- chronicle/artifacts.py | 27 +++++++++++++++++++++++- chronicle/registration.py | 43 +++++++++++++++++++++++++++++++-------- 2 files changed, 61 insertions(+), 9 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 7e9b0965..e8834e9b 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -2113,6 +2113,29 @@ def _assert_shared_owner_identities_agree( ) +def _effective_recorded_digest( + manifest_name: str, vintage: Any, entry: Mapping[str, Any] +) -> str | None: + """Return the digest an entry's recorded R2 key encodes, if it has one. + + Two manifests may identify one package-local file through identical + content-addressed R2 locators without declaring ``sha256``; refetching + those bytes records ``sha256`` on the selected manifest only. The + directory-level collision check must compare the identities the entries + *effectively* record -- the same identity :func:`_recorded_identity` + resolves -- so a sibling that only carries the locator does not read as + an empty digest. A malformed locator is the per-entry preflight's error, + not a collision: fall back to the declared field for it. + """ + try: + recorded = _validated_recorded_r2( + entry, manifest_path=Path(manifest_name), year=vintage + ) + except SourceArtifactManifestError: + return None + return None if recorded is None else recorded.sha256 + + def _assert_package_file_owner_identities_agree( manifests: Mapping[str, dict[str, Any]], ) -> None: @@ -2123,7 +2146,9 @@ def _assert_package_file_owner_identities_agree( as one package boundary before a selected manifest can upload anything. Entry-shape and local-file errors remain the per-entry preflight's job. """ - collision_codes = validate_package_directory(manifests) + collision_codes = validate_package_directory( + manifests, entry_digest=_effective_recorded_digest + ) if collision_codes: raise SourceArtifactManifestError( "Package manifests identify different bytes for one package-local " diff --git a/chronicle/registration.py b/chronicle/registration.py index 2d44e798..6e9c555a 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -10,7 +10,7 @@ from pathlib import Path import re -from typing import Any, Mapping +from typing import Any, Callable, Mapping import unicodedata import yaml @@ -84,25 +84,35 @@ def package_manifest_paths(package_dir: Path) -> list[Path]: def validate_package_directory( manifests: Mapping[str, Mapping[str, Any] | None], + *, + entry_digest: Callable[[str, Any, Mapping[str, Any]], str | None] | None = None, ) -> tuple[str, ...]: """Return filename-identity collisions across a package's manifests. Two manifests may name one physical file only when they record the same digest. A differing digest means the same package-local bytes have two incompatible identities, so no command may act through either record. + + ``entry_digest(manifest_name, vintage, entry)`` resolves an entry's + *effective* recorded digest -- for example the one its content-addressed + R2 key encodes when the entry declares no ``sha256`` -- and returns + ``None`` to fall back to the declared ``sha256`` field. Without it only the + declared field is compared. """ by_name: dict[str, list[tuple[str, str]]] = {} for name, manifest in manifests.items(): files = manifest.get("files") if isinstance(manifest, Mapping) else None if not isinstance(files, Mapping): continue - for entry in files.values(): + for vintage, entry in files.items(): if not isinstance(entry, Mapping): continue filename = entry.get("filename") if filename is None: continue - digest = entry.get("sha256") + digest = entry_digest(name, vintage, entry) if entry_digest else None + if digest is None: + digest = entry.get("sha256") digest = digest.strip() if isinstance(digest, str) else "" by_name.setdefault(filename_key(filename), []).append((name, digest)) @@ -193,13 +203,30 @@ def _explicit_pairs(self, node: Any, deep: bool) -> tuple[list[Any], list[Any]]: explicit_pairs.append((key_node, value_node)) return merge_sources, explicit_pairs - def _merged_pairs(self, source: Any, deep: bool) -> list[Any]: - """Return the pairs a ``<<`` source contributes, validated, unmutated.""" + def _merged_pairs( + self, source: Any, deep: bool, active: tuple[int, ...] + ) -> list[Any]: + """Return the pairs a ``<<`` source contributes, validated, unmutated. + + ``active`` holds the mappings whose merges are being expanded on the + way to ``source``. A source that is already active merges itself, + directly or through nested merges, which has no expansion: refuse it + as a ``ConstructorError`` (a ``yaml.YAMLError`` every manifest reader + handles) rather than recursing until the interpreter gives up. + """ if isinstance(source, yaml.MappingNode): + if id(source) in active: + raise yaml.constructor.ConstructorError( + "while constructing a mapping", + source.start_mark, + "found a recursive merge: a mapping merges itself", + source.start_mark, + ) nested_sources, explicit = self._explicit_pairs(source, deep) + nested_active = (*active, id(source)) pairs: list[Any] = [] for nested in nested_sources: - pairs.extend(self._merged_pairs(nested, deep)) + pairs.extend(self._merged_pairs(nested, deep, nested_active)) pairs.extend(explicit) return pairs if isinstance(source, yaml.SequenceNode): @@ -216,7 +243,7 @@ def _merged_pairs(self, source: Any, deep: bool) -> list[Any]: # the later mappings first and the first mapping last. pairs = [] for subnode in reversed(source.value): - pairs.extend(self._merged_pairs(subnode, deep)) + pairs.extend(self._merged_pairs(subnode, deep, active)) return pairs raise yaml.constructor.ConstructorError( "while constructing a mapping", @@ -236,7 +263,7 @@ def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: merge_sources, explicit_pairs = self._explicit_pairs(node, deep) merged_pairs: list[Any] = [] for source in merge_sources: - merged_pairs.extend(self._merged_pairs(source, deep)) + merged_pairs.extend(self._merged_pairs(source, deep, (id(node),))) mapping: dict[Any, Any] = {} # Merged pairs first, explicit pairs last: YAML precedence, last wins. for key_node, value_node in [*merged_pairs, *explicit_pairs]: From d62fbe9a517fac60208969db1e98c8fe53519c7a Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 20:58:47 -0400 Subject: [PATCH 098/105] test: reproduce false collisions from selected and partially failed publication Failing-first regressions for Astra gate 6aa29716 round 2: two manifests naming one package-local file before either records an identity must stay valid for every sweep after publishing only one of them, and after a full sweep whose second upload fails; the sibling's later publication records the same identity. Identifying a sibling whose package-local bytes no longer match the identified owner must be refused before any upload. --- tests/test_chronicle_artifacts.py | 177 ++++++++++++++++++++++++++++++ 1 file changed, 177 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index d423e391..a1b15451 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3538,3 +3538,180 @@ def collisions(manifests): assert not any("filename_collision" in error for error in sweep.errors) assert not any("identify different bytes" in error for error in sweep.errors) assert sweep.valid + + +def _write_unidentified_shared_manifests(package, source, *, filename="shared.csv"): + """Two manifests naming one package-local file with no identity yet.""" + manifest_paths = (package / "manifest_a.yaml", package / "manifest_b.yaml") + for path in manifest_paths: + path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": path.stem, + "files": { + 2024: { + "filename": filename, + "source_url": str(source), + } + }, + }, + sort_keys=False, + ) + ) + return manifest_paths + + +def _fake_wrangler(tmp_path): + log = tmp_path / "wrangler.log" + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + return wrangler, log + + +def test_selected_publication_of_a_shared_unidentified_file_keeps_siblings_valid( + tmp_path, +): + """Two manifests may name one package-local file before either records an + identity. Publishing only one of them records its checksum and locator; + the sibling, which still records nothing, must not become a false + collision, and its own later publication must record the same identity.""" + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + content = b"shared publisher table" + (package / "shared.csv").write_bytes(content) + source = tmp_path / "publisher-download.csv" + source.write_bytes(content) + manifest_a, manifest_b = _write_unidentified_shared_manifests(package, source) + wrangler, log = _fake_wrangler(tmp_path) + expected_sha256 = hashlib.sha256(content).hexdigest() + + assert inventory_source_artifacts(package).valid + + selected = publish_source_artifacts( + package, manifest_filename="manifest_a.yaml", wrangler_command=str(wrangler) + ) + assert selected.valid + assert selected.counts["uploaded_count"] == 1 + assert yaml.safe_load(manifest_a.read_text())["files"][2024]["sha256"] == ( + expected_sha256 + ) + assert "sha256" not in yaml.safe_load(manifest_b.read_text())["files"][2024] + + inventory = inventory_source_artifacts(package) + assert inventory.valid, inventory.errors + assert not any("filename_collision" in error for error in inventory.errors) + + sibling = publish_source_artifacts( + package, manifest_filename="manifest_b.yaml", wrangler_command=str(wrangler) + ) + assert sibling.valid, sibling.errors + assert sibling.counts["uploaded_count"] == 1 + recorded_b = yaml.safe_load(manifest_b.read_text())["files"][2024] + recorded_a = yaml.safe_load(manifest_a.read_text())["files"][2024] + assert recorded_b["sha256"] == expected_sha256 + assert recorded_b["storage"]["r2"]["key"].endswith(f"/{expected_sha256}/shared.csv") + assert recorded_a["storage"]["r2"]["key"].endswith(f"/{expected_sha256}/shared.csv") + assert len(log.read_text().splitlines()) == 2 + + everything = publish_source_artifacts(package, wrangler_command=str(wrangler)) + assert everything.valid, everything.errors + assert everything.counts["uploaded_count"] == 0 + assert everything.counts["skipped_count"] == 2 + + +def test_partial_upload_failure_of_a_shared_file_stays_retryable(tmp_path, monkeypatch): + """A full sweep whose second upload fails records the first manifest's + identity only. The retry must not report a collision for the sibling that + still records nothing, and must finish recording the same identity.""" + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + content = b"shared publisher table" + (package / "shared.csv").write_bytes(content) + source = tmp_path / "publisher-download.csv" + source.write_bytes(content) + manifest_a, manifest_b = _write_unidentified_shared_manifests(package, source) + expected_sha256 = hashlib.sha256(content).hexdigest() + uploads = [] + failures = {"remaining": 1} + + def uploader(location, local_path, *, wrangler_command): + uploads.append(location) + if len(uploads) == 2 and failures["remaining"]: + failures["remaining"] -= 1 + return ArtifactCommandResult( + command=("failing-uploader",), returncode=1, stdout="", stderr="boom" + ) + return ArtifactCommandResult( + command=("uploader",), returncode=0, stdout="", stderr="" + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", uploader) + + first = publish_source_artifacts(package) + assert not first.valid + assert first.counts["uploaded_count"] == 1 + assert first.counts["failed_count"] == 1 + recorded = { + path.name: yaml.safe_load(path.read_text())["files"][2024] + for path in (manifest_a, manifest_b) + } + identified = [name for name, entry in recorded.items() if "sha256" in entry] + assert len(identified) == 1 + + inventory = inventory_source_artifacts(package) + assert inventory.valid, inventory.errors + assert not any("filename_collision" in error for error in inventory.errors) + + retry = publish_source_artifacts(package) + assert retry.valid, retry.errors + assert retry.counts["uploaded_count"] == 1 + assert retry.counts["skipped_count"] == 1 + assert retry.counts["failed_count"] == 0 + for path in (manifest_a, manifest_b): + entry = yaml.safe_load(path.read_text())["files"][2024] + assert entry["sha256"] == expected_sha256 + assert entry["storage"]["r2"]["key"].endswith(f"/{expected_sha256}/shared.csv") + assert len(uploads) == 3 + + +def test_sweeps_refuse_identifying_a_shared_file_whose_bytes_changed( + tmp_path, monkeypatch +): + """An unidentified sibling is not a collision, but identifying it would + hash the package-local bytes. When those bytes no longer match what the + identified owner records, every sweep refuses before any upload.""" + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + old_content = b"shared publisher table" + new_content = b"shared publisher table, revised" + (package / "shared.csv").write_bytes(new_content) + source = tmp_path / "publisher-download.csv" + source.write_bytes(new_content) + manifest_a, manifest_b = _write_unidentified_shared_manifests(package, source) + identified = yaml.safe_load(manifest_a.read_text()) + identified["files"][2024]["sha256"] = hashlib.sha256(old_content).hexdigest() + identified["files"][2024]["size_bytes"] = len(old_content) + manifest_a.write_text(yaml.safe_dump(identified, sort_keys=False)) + uploads = [] + + def unexpected_uploader(location, local_path, *, wrangler_command): + uploads.append(location) + return ArtifactCommandResult( + command=("uploader",), returncode=0, stdout="", stderr="" + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", unexpected_uploader) + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package) + sibling_only = publish_source_artifacts( + package, manifest_filename="manifest_b.yaml" + ) + + for sweep in (inventory, published, sibling_only): + assert not sweep.valid + assert any("two identities" in error for error in sweep.errors), sweep.errors + assert uploads == [] + assert "sha256" not in yaml.safe_load(manifest_b.read_text())["files"][2024] From cd14372f2915bed2a9a7868d3727a32da7194d06 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 20:58:47 -0400 Subject: [PATCH 099/105] fix: unidentified sibling entries are not collisions; guard their bytes before upload validate_package_directory no longer counts an entry that records no identity (no sha256, no locator) against identified owners: it has nothing to contradict, so a selected publication or a partially failed sweep that identifies one manifest cannot strand its siblings behind a false filename_collision_across_manifests. The owner-agreement preflight now also hashes the package-local bytes of each unidentified entry that shares a filename with identified owners and refuses, before any upload or manifest rewrite, when they differ from the recorded identity: identifying that entry would give one file two identities. --- chronicle/artifacts.py | 39 +++++++++++++++++++++++++++++++++++++-- chronicle/registration.py | 15 ++++++++++++--- 2 files changed, 49 insertions(+), 5 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index e8834e9b..f8401583 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -2158,6 +2158,7 @@ def _assert_package_file_owner_identities_agree( owners_by_filename: dict[str, list[_ManifestFileOwner]] = {} display_names: dict[str, str] = {} + unidentified: dict[str, list[tuple[Path, Any, str]]] = {} for name, payload in manifests.items(): manifest_path = Path(name) for vintage, spec in _manifest_files(payload, manifest_path).items(): @@ -2176,10 +2177,13 @@ def _assert_package_file_owner_identities_agree( # The complete per-entry preflight reports the precise locator # or history error without letting another entry upload first. continue - if identity is None: - continue key = filename_key(recorded_name) display_names.setdefault(key, str(recorded_name)) + if identity is None: + unidentified.setdefault(key, []).append( + (manifest_path, vintage, str(recorded_name)) + ) + continue owners_by_filename.setdefault(key, []).append( _ManifestFileOwner( manifest_path=manifest_path, @@ -2193,6 +2197,37 @@ def _assert_package_file_owner_identities_agree( owners, filename=display_names[key], ) + # An entry that records no identity yet is not a collision (nothing to + # contradict), but the command that identifies it will hash the shared + # bytes. Those bytes must already be what the identified owners record, + # otherwise identifying it would split one package-local file into two + # identities. Check before any upload or manifest rewrite. + for key, pending in unidentified.items(): + owners = owners_by_filename.get(key) + if not owners: + continue + expected = owners[0].identity + assert expected is not None + for manifest_path, vintage, recorded_name in pending: + try: + local = matching_directory_entry(manifest_path.parent, recorded_name) + except ValueError: + # Conflicting spellings are the per-entry preflight's refusal. + continue + if local is None or local.is_symlink() or not local.is_file(): + continue + actual = hashlib.sha256(local.read_bytes()).hexdigest() + if actual == expected.sha256: + continue + raise SourceArtifactManifestError( + f"{manifest_path} entry {vintage!r} names {recorded_name!r} " + f"without a recorded identity, and the package-local bytes " + f"(sha256={actual}) are not what {owners[0].manifest_path} entry " + f"{owners[0].vintage!r} records for it (sha256=" + f"{expected.sha256}). Identifying this entry would give one " + "package-local file two identities; reconcile the manifests " + "or the file before publishing or inventorying the directory." + ) def _assert_siblings_record_these_bytes( diff --git a/chronicle/registration.py b/chronicle/registration.py index 6e9c555a..f9f5393e 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -89,9 +89,15 @@ def validate_package_directory( ) -> tuple[str, ...]: """Return filename-identity collisions across a package's manifests. - Two manifests may name one physical file only when they record the same - digest. A differing digest means the same package-local bytes have two - incompatible identities, so no command may act through either record. + Two manifests may name one physical file only when every identity they + record for it agrees. A differing digest means the same package-local + bytes have two incompatible identities, so no command may act through + either record. An entry that records no identity yet -- no ``sha256`` and + no locator, as a manifest looks before its first fetch or publication -- + cannot contradict an identified owner and is not a collision: the command + that eventually identifies it hashes the shared bytes, so a selected or + partially failed publication never strands its siblings behind a false + collision. ``entry_digest(manifest_name, vintage, entry)`` resolves an entry's *effective* recorded digest -- for example the one its content-addressed @@ -114,6 +120,9 @@ def validate_package_directory( if digest is None: digest = entry.get("sha256") digest = digest.strip() if isinstance(digest, str) else "" + if not digest: + # No recorded identity yet: nothing to contradict. + continue by_name.setdefault(filename_key(filename), []).append((name, digest)) errors: list[str] = [] From 9f7cdf26245c26b9edbc70176c5c00f94a515ee4 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 21:12:14 -0400 Subject: [PATCH 100/105] test: reproduce first fetch of a predeclared entry without identity Failing-first regression for Astra gate 6aa29716 round 3: a manifest entry predeclared with only filename and source_url must acquire its initial bytes on first fetch (default and --record-revision), while another manifest's unidentified entry naming the same file keeps refusing the fetch before any byte is written. --- tests/test_chronicle_artifacts.py | 87 +++++++++++++++++++++++++++++++ 1 file changed, 87 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index a1b15451..7dd80f5b 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3715,3 +3715,90 @@ def unexpected_uploader(location, local_path, *, wrangler_command): assert any("two identities" in error for error in sweep.errors), sweep.errors assert uploads == [] assert "sha256" not in yaml.safe_load(manifest_b.read_text())["files"][2024] + + +@pytest.mark.parametrize("record_revision", [False, True]) +def test_first_fetch_initializes_a_predeclared_entry_without_identity( + tmp_path, record_revision +): + """A manifest may predeclare an entry with only ``filename`` and + ``source_url``. Its first fetch identifies it: the entry is the one being + initialized, not an unidentifiable owner, so both the fetch preflight + and the manifest writer must accept it (default and --record-revision).""" + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + source = tmp_path / "publisher-download.csv" + content = b"first publisher bytes" + source.write_bytes(content) + manifest_path = package / "manifest.yaml" + manifest_path.write_text( + yaml.safe_dump( + { + "source_id": "publisher", + "package_id": "package", + "files": {2024: {"filename": "table.csv", "source_url": str(source)}}, + }, + sort_keys=False, + ) + ) + assert not (package / "table.csv").exists() + + report = fetch_source_artifact( + str(source), + source_id="publisher", + package_id="package", + year=2024, + output_dir=package, + filename="table.csv", + record_revision=record_revision, + ) + + assert report.valid, report.errors + expected_sha256 = hashlib.sha256(content).hexdigest() + assert report.sha256 == expected_sha256 + assert (package / "table.csv").read_bytes() == content + entry = yaml.safe_load(manifest_path.read_text())["files"][2024] + assert entry["filename"] == "table.csv" + assert entry["sha256"] == expected_sha256 + assert entry["size_bytes"] == len(content) + assert inventory_source_artifacts(package).valid + + again = fetch_source_artifact( + str(source), + source_id="publisher", + package_id="package", + year=2024, + output_dir=package, + filename="table.csv", + ) + assert again.valid, again.errors + assert again.sha256 == expected_sha256 + + +def test_fetch_still_refuses_an_unidentified_owner_in_another_manifest(tmp_path): + """Only the entry being initialized is exempt: another manifest naming the + same package-local file without an identity keeps refusing the fetch + before any byte is written, because the fetched bytes would silently + define what that sibling means.""" + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + source = tmp_path / "publisher-download.csv" + source.write_bytes(b"first publisher bytes") + manifest_a, manifest_b = _write_unidentified_shared_manifests( + package, source, filename="table.csv" + ) + before = {path: path.read_text() for path in (manifest_a, manifest_b)} + + with pytest.raises(MalformedManifestError, match="records no sha256 identity"): + fetch_source_artifact( + str(source), + source_id="publisher", + package_id="manifest_a", + year=2024, + output_dir=package, + filename="table.csv", + manifest_filename="manifest_a.yaml", + ) + + assert not (package / "table.csv").exists() + assert {path: path.read_text() for path in (manifest_a, manifest_b)} == before From 7d9fd068161dc7d62eea0d98bb5d174e3bad5647 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 21:12:14 -0400 Subject: [PATCH 101/105] fix(artifacts): let a first fetch initialize the predeclared entry it identifies _manifest_file_owners refused every entry naming the artifact that records no identity, including the selected entry on its first fetch, so a manifest predeclaring files: {2024: {filename, source_url}} could never acquire its bytes (MalformedManifestError before publisher I/O; --record-revision too). The fetch preflight, the sibling-bytes check, and the manifest writer now name the (manifest, vintage) entry they are initializing; that entry is not an owner yet and is skipped, while every other unidentified entry naming the file keeps the existing refusal. --- chronicle/artifacts.py | 35 +++++++++++++++++++++++++++++++---- 1 file changed, 31 insertions(+), 4 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index f8401583..ea701863 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -892,7 +892,11 @@ def fetch_source_artifact( year=vintage_key, ) manifests = _package_manifests(output, manifest_path, existing_manifest) - owners = _manifest_file_owners(manifests, filename=artifact_filename) + owners = _manifest_file_owners( + manifests, + filename=artifact_filename, + initializing=(manifest_path, vintage_key), + ) _assert_shared_owner_identities_agree(owners, filename=artifact_filename) try: @@ -940,6 +944,7 @@ def fetch_source_artifact( _assert_siblings_record_these_bytes( manifests, manifest_path=manifest_path, + vintage=vintage_key, filename=artifact_filename, sha256=sha256, ) @@ -2040,8 +2045,17 @@ def _manifest_file_owners( manifests: Mapping[str, dict[str, Any]], *, filename: str, + initializing: tuple[Path, Any] | None = None, ) -> list[_ManifestFileOwner]: - """Return every entry in a package directory that names ``filename``.""" + """Return every entry in a package directory that names ``filename``. + + ``initializing`` names the ``(manifest_path, vintage)`` entry the calling + command is about to identify -- a predeclared entry (``filename`` and + ``source_url`` only) on its first fetch. That entry has no identity yet + and is not an owner; every other entry naming the file must already + record one, because Chronicle cannot tell whether overwriting the shared + bytes would change what an unidentified sibling means. + """ wanted = filename_key(filename) owners: list[_ManifestFileOwner] = [] for name, payload in manifests.items(): @@ -2069,6 +2083,13 @@ def _manifest_file_owners( year=vintage, ) if identity is None: + if ( + initializing is not None + and manifest_path == initializing[0] + and str(vintage) == str(initializing[1]) + ): + # The entry this command identifies: not an owner yet. + continue raise MalformedManifestError( f"{manifest_path} entry {vintage!r} names " f"{recorded_name!r} but records no sha256 identity. " @@ -2234,11 +2255,14 @@ def _assert_siblings_record_these_bytes( manifests: Mapping[str, dict[str, Any]], *, manifest_path: Path, + vintage: Any, filename: str, sha256: str, ) -> None: """Refuse a default fetch that would stale another manifest's owner.""" - for owner in _manifest_file_owners(manifests, filename=filename): + for owner in _manifest_file_owners( + manifests, filename=filename, initializing=(manifest_path, vintage) + ): if owner.manifest_path == manifest_path: continue identity = owner.identity @@ -2447,7 +2471,9 @@ def _upsert_manifest( package_id=package_id, ) manifests = _package_manifests(manifest_path.parent, manifest_path, payload) - owners = _manifest_file_owners(manifests, filename=filename) + owners = _manifest_file_owners( + manifests, filename=filename, initializing=(manifest_path, year) + ) _assert_shared_owner_identities_agree(owners, filename=filename) payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) @@ -2490,6 +2516,7 @@ def _upsert_manifest( _assert_siblings_record_these_bytes( manifests, manifest_path=manifest_path, + vintage=key, filename=filename, sha256=sha256, ) From 2938b2a73273d4ca4f2d1c4c1c7a077667d5c9ca Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 21:22:06 -0400 Subject: [PATCH 102/105] test: reproduce selected publication overrides leaking onto sibling manifests Failing-first regression for Astra gate 6aa29716 round 4: publishing one manifest with --source-id/--package-id supplying its missing identifier must not preflight an unselected sibling that declares a different identifier against the override; the selected publication succeeds and the sibling is left untouched. --- tests/test_chronicle_artifacts.py | 74 +++++++++++++++++++++++++++++++ 1 file changed, 74 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 7dd80f5b..0506160a 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3802,3 +3802,77 @@ def test_fetch_still_refuses_an_unidentified_owner_in_another_manifest(tmp_path) assert not (package / "table.csv").exists() assert {path: path.read_text() for path in (manifest_a, manifest_b)} == before + + +@pytest.mark.parametrize("missing", ["source_id", "package_id"]) +def test_selected_publication_overrides_apply_only_to_the_selected_manifest( + tmp_path, missing +): + """``--source-id`` / ``--package-id`` complete the selected manifest's + identity. An unselected sibling that declares a different identifier is + preflighted with its own identifiers, not the override, so the selected + publication succeeds and the sibling is left untouched.""" + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + content_a = b"table a" + content_b = b"table b" + (package / "table_a.csv").write_bytes(content_a) + (package / "table_b.csv").write_bytes(content_b) + identity_a = {"source_id": "publisher_a", "package_id": "package_a"} + identity_b = {"source_id": "publisher_b", "package_id": "package_b"} + override = {missing: identity_a[missing]} + declared_a = {key: value for key, value in identity_a.items() if key != missing} + manifest_a = package / "manifest_a.yaml" + manifest_b = package / "manifest_b.yaml" + manifest_a.write_text( + yaml.safe_dump( + { + **declared_a, + "files": { + 2024: { + "filename": "table_a.csv", + "source_url": "https://publisher.test/a", + "sha256": hashlib.sha256(content_a).hexdigest(), + "size_bytes": len(content_a), + } + }, + }, + sort_keys=False, + ) + ) + manifest_b.write_text( + yaml.safe_dump( + { + **identity_b, + "files": { + 2024: { + "filename": "table_b.csv", + "source_url": "https://publisher.test/b", + "sha256": hashlib.sha256(content_b).hexdigest(), + "size_bytes": len(content_b), + } + }, + }, + sort_keys=False, + ) + ) + before_b = manifest_b.read_text() + wrangler, log = _fake_wrangler(tmp_path) + + report = publish_source_artifacts( + package, + manifest_filename="manifest_a.yaml", + wrangler_command=str(wrangler), + **override, + ) + + assert report.valid, report.errors + assert report.counts["uploaded_count"] == 1 + assert report.counts["failed_count"] == 0 + recorded = yaml.safe_load(manifest_a.read_text()) + assert recorded[missing] == identity_a[missing] + assert recorded["files"][2024]["storage"]["r2"]["key"].startswith( + "raw/publisher_a/package_a/2024/" + ) + assert manifest_b.read_text() == before_b + assert len(log.read_text().splitlines()) == 1 From d544ffe77ffd0d2cb3551fb6f2ec378979e5fa09 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 21:22:06 -0400 Subject: [PATCH 103/105] fix(artifacts): apply publish identity overrides to the selected manifest only The package-wide publish preflight applied --source-id/--package-id to every sibling manifest in the directory, so a selected publication that supplied the selected manifest's missing identifier was refused before upload when a sibling declared a different one. Unselected siblings are now preflighted with their own identifiers; the override completes or confirms only the selected manifest. --- chronicle/artifacts.py | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index ea701863..b70e9f0d 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1284,16 +1284,16 @@ def publish_source_artifacts( for package_manifest_name, package_manifest in package_manifests.items(): package_manifest_path = Path(package_manifest_name) - package_source_id = ( - source_id - if source_id is not None - else package_manifest.get("source_id") - ) - package_id_value = ( - package_id - if package_id is not None - else package_manifest.get("package_id") - ) + # ``--source-id`` / ``--package-id`` complete or confirm the + # *selected* manifest's identity. An unselected sibling is only + # preflighted so the package boundary holds; it keeps its own + # identifiers, which may legitimately differ. + if package_manifest_path == manifest_path: + package_source_id = manifest_source_id + package_id_value = manifest_package_id + else: + package_source_id = package_manifest.get("source_id") + package_id_value = package_manifest.get("package_id") package_files = _manifest_files(package_manifest, package_manifest_path) for year, spec in package_files.items(): entry, _updated_spec = _publish_raw_manifest_entry( From b3251a31bacae2def896dafbf9501772f50938e9 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 21:32:42 -0400 Subject: [PATCH 104/105] test: reproduce a default sweep withholding overrides from a selected sibling Failing-first regression for Astra gate cc680cc1 round 1: a default sweep selects every manifest in the directory, so --source-id/--package-id must complete the manifest lacking the identifier and confirm the sibling that declares the same value regardless of processing order. --- tests/test_chronicle_artifacts.py | 69 +++++++++++++++++++++++++++++++ 1 file changed, 69 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 0506160a..c72d737c 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -3876,3 +3876,72 @@ def test_selected_publication_overrides_apply_only_to_the_selected_manifest( ) assert manifest_b.read_text() == before_b assert len(log.read_text().splitlines()) == 1 + + +@pytest.mark.parametrize("missing", ["source_id", "package_id"]) +def test_default_sweep_overrides_apply_to_every_selected_sibling(tmp_path, missing): + """A default sweep selects every manifest in the directory. The override + must complete the manifest that lacks the identifier and confirm the + sibling that declares the same value, whichever order they are processed + in -- a selected sibling met through another selected manifest's package + preflight is not an unselected one.""" + package = tmp_path / "data" / "publisher" / "package" + package.mkdir(parents=True) + content_a = b"table a" + content_b = b"table b" + (package / "table_a.csv").write_bytes(content_a) + (package / "table_b.csv").write_bytes(content_b) + identity = {"source_id": "publisher", "package_id": "package"} + override = {missing: identity[missing]} + declared_a = {key: value for key, value in identity.items() if key != missing} + manifest_a = package / "manifest_a.yaml" + manifest_b = package / "manifest_b.yaml" + manifest_a.write_text( + yaml.safe_dump( + { + **declared_a, + "files": { + 2024: { + "filename": "table_a.csv", + "source_url": "https://publisher.test/a", + "sha256": hashlib.sha256(content_a).hexdigest(), + "size_bytes": len(content_a), + } + }, + }, + sort_keys=False, + ) + ) + manifest_b.write_text( + yaml.safe_dump( + { + **identity, + "files": { + 2024: { + "filename": "table_b.csv", + "source_url": "https://publisher.test/b", + "sha256": hashlib.sha256(content_b).hexdigest(), + "size_bytes": len(content_b), + } + }, + }, + sort_keys=False, + ) + ) + wrangler, log = _fake_wrangler(tmp_path) + + report = publish_source_artifacts( + package, wrangler_command=str(wrangler), **override + ) + + assert report.valid, report.errors + assert report.counts["uploaded_count"] == 2 + assert report.counts["failed_count"] == 0 + assert not any("r2_identity_invalid" in error for error in report.errors) + for path in (manifest_a, manifest_b): + recorded = yaml.safe_load(path.read_text()) + assert recorded[missing] == identity[missing] + assert recorded["files"][2024]["storage"]["r2"]["key"].startswith( + "raw/publisher/package/2024/" + ) + assert len(log.read_text().splitlines()) == 2 From 3eea393a78a71632d863c89d2300140ec9b8a44a Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 21:32:42 -0400 Subject: [PATCH 105/105] fix(artifacts): decide publish override eligibility from the whole selection The package preflight applied --source-id/--package-id only to the manifest currently being processed, so in a default sweep (every manifest selected) a selected sibling met through another manifest's package preflight was treated as unselected, preflighted without the override, and the whole publication aborted with r2_identity_invalid. The selected set is now computed up front from the root sweep; selected siblings take the overrides, unselected siblings keep their own identifiers. --- chronicle/artifacts.py | 28 +++++++++++++++++++++++----- 1 file changed, 23 insertions(+), 5 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index b70e9f0d..876d9730 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1263,7 +1263,14 @@ def publish_source_artifacts( errors: list[str] = [] prepared: list[tuple[Path, dict[str, Any], dict[str, Any], str, str]] = [] preflight_failures: list[RawArtifactPublishEntry] = [] - for manifest_path in _root_manifest_paths(root_path, manifest_filename): + # ``--source-id`` / ``--package-id`` complete or confirm the identity of + # every manifest this sweep selects (one with an explicit selector, all of + # a directory's manifests by default). Decide eligibility from the whole + # selection up front, so a selected sibling met through another selected + # manifest's package preflight is not mistaken for an unselected one. + selected_manifest_paths = list(_root_manifest_paths(root_path, manifest_filename)) + selected_manifests = set(selected_manifest_paths) + for manifest_path in selected_manifest_paths: try: manifest = _read_manifest(manifest_path) files = _manifest_files(manifest, manifest_path) @@ -1284,13 +1291,24 @@ def publish_source_artifacts( for package_manifest_name, package_manifest in package_manifests.items(): package_manifest_path = Path(package_manifest_name) - # ``--source-id`` / ``--package-id`` complete or confirm the - # *selected* manifest's identity. An unselected sibling is only - # preflighted so the package boundary holds; it keeps its own - # identifiers, which may legitimately differ. + # A selected sibling takes the overrides like the selected manifest + # itself; an unselected sibling is only preflighted so the package + # boundary holds and keeps its own identifiers, which may + # legitimately differ. if package_manifest_path == manifest_path: package_source_id = manifest_source_id package_id_value = manifest_package_id + elif package_manifest_path in selected_manifests: + package_source_id = ( + source_id + if source_id is not None + else package_manifest.get("source_id") + ) + package_id_value = ( + package_id + if package_id is not None + else package_manifest.get("package_id") + ) else: package_source_id = package_manifest.get("source_id") package_id_value = package_manifest.get("package_id")