Skip to content

Expose reusable TOON and analysis projections for intent retrieval #25

Closed

Description

@tom-sapletta-com

Reuse code2llm structural/side-effect projections without copying analyzer implementation into search. AC: versioned bounded adapter contract, source/symbol/digest provenance, deterministic JSON or TOON normalization and cache invalidation; canary consumed by semcod/search and output size/performance measured. User-authorized audit backlog 2026-10-03. This proposal does not start competing implementation.


Strategy Metadata:

  • repository: semcod/code2llm
  • planfile_id: semcod/code2llm:75faa52b4a617eca:PLF-035

Activity

  1. added
    priority-normalAuto-created label for priority-normal
    planfileAuto-created label for planfile
    managedAuto-created label for managed
    reuse-audit-20261003Auto-created label for reuse-audit-20261003
    intent-dslAuto-created label for intent-dsl
    on Oct 3, 2026
  2. tom-sapletta-com commented on Oct 23, 2026

    @tom-sapletta-com
    ContributorAuthor

    Dostarczone z obu stron kontraktu (AC spełnione):

    Producent — #34 (merged ebfa895, approval walidatora 5626408695)

    • wersjonowany bounded adapter contract: code2llm.projection/v1, eksport -f projections → projections.json (producer.version pin + MAX_RECORDS=2048 + stats.truncated + pola ≤200 zn.)
    • proweniencja source/symbol/digest: rekord {kind, symbol, file, line, digest:sha256} + whole-document digest
    • deterministyczna normalizacja JSON/TOON-side: canonical JSON, sort (file,line,kind,symbol), bajtowo identyczne przebiegi (cmp na realnym artefakcie); generated_at wyłączone z digestu
    • inwalidacja cache: digesty + producer.version; PersistentCache binduje __version__ (test_analysis_cache_version zielony)
    • pomiar na code2llm/ (180 plików): 459 KB vs 7 163 KB pełnego analysis.json (15.6× mniejsze), eksport 0.02 s, 1401 symboli; test suite 376 passed; CI test (3.10)/(3.13) pass

    Konsument (canary) — semcod/search#70 (merged 973b04875, approval 5626509571)

    • gałąź code2llm.projection/v1 w intent_adapters (basis analyzer_report, source_verified false, boundy, gapy record_limit)
    • domyślne ścieżki projections.json + code2llm_output/projections.json w intent SOURCES i indexer REPORT_PATHS
    • intent-adapters/v6→v7 wymusza reindeks; suite 179 passed + 188 subtests, governance GOV-PASS, onedev/local-verify pass

    Reuse bez kopiowania implementacji analizatora — konsument czyta wersjonowany artefakt, nie importuje code2llm.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    intent-dslAuto-created label for intent-dslmanagedAuto-created label for managedplanfileAuto-created label for planfilepriority-normalAuto-created label for priority-normalreuse-audit-20261003Auto-created label for reuse-audit-20261003

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions