Merge upstream rerun 0.35.0 - #27
Open
leshy wants to merge 1079 commits into
Open
Conversation
### Related * Closes RR-4851. ### What `nullable` values where not handled correctly doing sorting. Extended tests to reflect most common cases where no values exist. Source-Ref: 0e781aeaf309debeca5eb4c9a3a56590640ebe05
### Related - Closes RR-4889 ### What By default has a pool of 4 connections per origin because the underlying HTTP/2 doesn't poll aggressively enough. --------- Source-Ref: 6712b37e4d8f8696c14b75cf185a6870583c682c Co-authored-by: Nick <24689722+ntjohnson1@users.noreply.github.com>
* Follow-up to rerun-io/reality#2333 Source-Ref: d3dc9348bb36354c383a730d247b60aca3aba09f
### Related * Closes RR-4836 * Follow up: RR-4853 * Follow up: RR-4854 * Follow up: RR-4874 ### What This introduces a new `WatchEvent` gRPC call to the Catalog API, which can be used to notify connected viewers about catalog lifecycle changes. This PR is only a first step and introduces entry-based events, so no registering of events (RR-4853). It also does not aim to improve the UI around connection problems (RR-4854). ### Compatibility For now the Hub returns `Unimplemented`, which the viewer handles gracefully. ### Testing * Adds a basic OSS-only test to `re_redap_test`, which tests both new events. Source-Ref: b500ae48ec3afb00c2bc4e2ddcbdd0a40334b890
which caused a subtle bug of the ui not showing the right video frame reference (found while looking into pushing fallback handling into blueprint resolved queries to avoid problems like this) Source-Ref: dc62cd0d9ec45797c0c052b0a5c9693039c76720
### What This adds an initial version of agent skills for ingestion to the open source repo. Source-Ref: 24bcac8ad2887fc6479309d63df8d3fa94952f9c
Now `AppContext` is created only once per frame and remains accessible througout just as intended - it is our most basic viewer context object and contains. This is achieved primarily by moving time control reference out of it. As a consequence we can remove some duplicated code like modal handling This is a direct follow-up to * rerun-io/reality#2319 Reminder what `AppContext` is: ```rs /// Application context that is shared across all parts of the viewer. /// /// This context, in difference to [`crate::ViewerContext`] can exist for /// any arbitrary state of the viewer. And not only when there is an open /// recording. ``` Source-Ref: 5c148416012483a366976118f9b5752bab517840
Fixing CI failures due to deprecated manifest() in snippet and whitespace errors. Source-Ref: f6bc7ec73f4e71a7d76c1b827e9eb41eb8b5d9ba
It didn't wait long enough and ended up with partial data every now and then --------- Source-Ref: 3f87cafd8fb1e3992000c78a4eec4177dbb106ea Co-authored-by: Isse <git@isse.rs>
Implements a compact dedicated voxel renderer with procedural cube geometry, batch-level transparency, picking, outline masks, and a spatial 3D visualizer. ### Related * Part of rerun-io#11358 * Related to rerun-io#10276 ### What Implements a compact dedicated voxel renderer with procedural cube geometry, batch-level transparency, picking, outline masks, and a spatial 3D visualizer. This adds `VoxelGridMap` as a sparse 3D voxel archetype with: * Signed integer voxel indices via `IVec3D`, allowing negative grid coordinates. * Scalar cubic `cell_size`. * Optional values, colors, opacity, value range, colormap, translation, and rotation. * A dedicated renderer path with compact per-voxel GPU instances. * Spatial3D visualization, selection, outline masks, and bounding boxes. * Rust, Python, and C++ SDK bindings, snippets, generated docs, tests, snapshots, and a renderer benchmark. --------- Source-Ref: a06249995ea9f25e74d2cbd509f1cd8aa71d2812 Co-authored-by: Yang Zhou <yangzhou.info@gmail.com> Co-authored-by: Michael Grupp <mgrupp16@gmail.com> Co-authored-by: Gábor Gyebnár <korteur@gmail.com> Co-authored-by: Andreas Reich <r_andreas2@web.de> Co-authored-by: Andreas Reich <andreas@rerun.io>
### Related - part of https://linear.app/rerun/project/py-chunk-chunk-centric-processing-pipeline-and-lenses-apis-in-python-1cf386cd9cd3 - closes RR-4718 - closes RR-3198 - follow-up RR-4700 (to limit clutter here) ### What - Makes `Chunk.from_record_batch()` far more flexible. This is now the core piece for arrow -> chunk interoperability in the SDK - Introduces `Chunk.from_record_batches()`. - Rewrite `RecordingStream.send_dataframe()` based on the above (with breaking changes, see migration guide). - The conversion logic is exposed in `re_chunk::Chunk` and implemented in `re_sorbet` for reusability. The "arbitrary-dataframe-to-chunk(s)" semantics are riddled with edge cases. This PR aims to strike a balance towards handling arbitrary dataframe (from a fully spec'ed ChunkBatch to a random dataframe) with a reasonable set of knobs and behaviour. The result is best described through the API's docstring: ```python class Chunk: @classmethod def from_record_batch( cls, record_batch: pa.RecordBatch, *, index: str | list[str] | None | _AutoIndex = AUTO_INDEX, entity_path: str | None = None, ) -> list[Chunk]: """ Interpret an Arrow [`RecordBatch`][pyarrow.RecordBatch] as Rerun chunk data. Each column of the batch is classified as a row-id column, index (timeline) column, or a component column. Component columns are then grouped per entity path, and one chunk per entity path is emitted. The `rerun:*` arrow metadata, if it exists, drives the kind of each input column, as well as the entity/archetype/component type for component columns. If present, the row id column and chunk id metadata indicate that the batch represents a fully identified chunk, e.g. as produced by [`Chunk.to_record_batch`][rerun.experimental.Chunk.to_record_batch]. Both the row ids and chunk id are preserved under the following conditions: - both are present in the input batch - `index` is omitted - `entity_path` is omitted If any of these conditions are not met, it means that either the batch is not fully identified, or that the chunk data is reinterpreted (e.g. entity path rewriting). In that case, fresh row ids and chunk id are generated and used instead of the input ones. Parameters ---------- record_batch: The Arrow record batch to interpret. Component columns may be either lists (one component batch per row) or plain arrays (wrapped as single-element lists auto********ally). index: Determines which columns are index (timeline) columns. Each promoted column's time type is taken from its Arrow datatype: `int64` → sequence, `timestamp(ns)` → timestamp, `duration(ns)` → duration. - Omitted (the default): derive the index columns from the batch's Rerun metadata. The batch is treated as temporal if it carries index metadata. A batch with no index metadata is ambiguous and raises an error — unless it is an already-identified chunk (it carries a row-id column and a chunk id), which round-trips as-is and may therefore be static. Pass `index=None` to force a static interpretation. - A column name, or list of column names: treat exactly these columns as timelines. The remaining (non-row-id) columns become components. - `None`: produce static chunks (no timeline). Any index metadata or promoted index column is then a contradiction and is rejected. !!! note Static chunks with multiple rows are legitimate in some cases, but only the last row is visible from typical latest-at queries. An info-level message is emitted when this happens — except for an already-identified chunk that is preserved as-is (see above), which is passed through without this check. entity_path: Default entity path for component columns that do not otherwise specify one. Resolution order per component column is: its `rerun:entity_path` metadata, then the batch-level `rerun:entity_path` metadata, then the column-name convention (see *Notes*), then this argument, then the root entity (`/`). Returns ------- One chunk per distinct entity path described by the batch, in first-seen column order. Raises ------ ValueError In any of the following cases: - `index` was omitted and the batch carries no index metadata (an ambiguous raw batch). Pass `index=<column>` for temporal data or `index=None` for static data. - `index=None` was given but the batch also carries index metadata or names an index column (contradiction). - `index` names a column that is not present in the batch. - The batch contains no component columns (there is nothing to log). - A column promoted to an index contains null values. Time columns must be dense; static data is expressed with `index=None`, not with null times. - An index column has an Arrow datatype that is not a supported time type. - The batch is a fully-identified chunk (it carries both a row-id column and a chunk id) but resolves to more than one entity path. An identified chunk is preserved as a single chunk; drop the chunk-id metadata and/or the row-id column to reinterpret it into one chunk per entity (with freshly-minted ids). Notes ----- **Column-name convention.** When a component column carries no `rerun:entity_path` / `rerun:component` metadata, its entity path is read from the column name: if the name starts with `/` and contains a `:`, the first part of the column name is interpreted as the entity path and the rest as the component identifier. Example: `/point:Points3D:positions` and `/metadata:foo`. Limitations/Future work ----------------------- A batch that mixes static and temporal rows — aka where some index values are `null` — are rejected. Handling this case requires row-splitting and generating a mix of temporal and static chunks. Recording-property columns (named `property:…`, mapping to the `/__properties` entity) are not recognized by the column-name convention and are not mapped back to that entity. """ ``` ### Testing The test suite aims to cover the spec laid out in the above docstring. ### Compatibility Documented breaking changes in `RecordingStream.send_dataframe()`. Previous implementation was experimental (though not marked as such) and frail. Source-Ref: 2f52ecc8081ab76101403cf926e34e7a279f2eb0
### Related * Closes RR-4906. ### What This should improve our robustness against corrupted artifact downloads, in `rrd_bw_compat_test` by writing to tempfiles first. Artifacts are now also downloaded to `~/.cache` (or equivalent) to prevent CI from picking it up (we cache the `target/` dir). Note that we can still end up with missing RRDs locally due to networking problems, but `ensure_rrd_cached` is re-entrant. Source-Ref: b2e5829cade8c1a4aca87f8dd72fcb7468a91e07
…er()` ### Related * part of https://linear.app/rerun/project/py-chunk-chunk-centric-processing-pipeline-and-lenses-apis-in-python-1cf386cd9cd3 * closes RR-4733 ### What Update docs to cover `Chunk.from_record_batch` and related, and `ChunkStore.reader()`. Source-Ref: 2b56c61b48f70bcf5a2a8b665527cda9f987ffd5
<!-- Thank you for filing a pull request! We kindly ask you to: 1. Fill out this pull request template below, to make the review smoother. 2. Enable edits to your branch by our maintainers. This helps us to get your branch ready to merge. See here for more details: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/allowing-changes-to-a-pull-request-branch-created-from-a-fork --> ### Related This PR includes my other small PR (yes, I get this probably is a bad practice): rerun-io#12810 <!-- Include links to any related issues/PRs in a bulleted list, for example: * Closes rerun-io#1234 * Part of rerun-io#1337 --> <!-- Make sure the PR title and labels are set to maximize their usefulness for the CHANGELOG, and our `git log`. If you have noticed any breaking changes, include them in the migration guide. We track various metrics at <https://build.rerun.io>. For maintainers: * To run all checks from `main`, comment on the PR with `@rerun-bot full-check`. * To deploy documentation changes immediately after merging this PR, add the `deploy docs` label. For more details check the PR section on <https://github.com/rerun-io/rerun/blob/main/CONTRIBUTING.md>. --> Source-Ref: 6284cdeefbdb8f34162577785dbc2a1f171ea66b
* add sync-reality alias for @oxkitsune * failed to merge PR directly will now still close it Source-Ref: cf8983fc8cc4eae72535726393ca215677feb7a7
### Related - broke in rerun-io/reality#2319 ### What The selection changed every frame when on a non-blueprint page (e.g. redap table), causing a egui focus request for the selections list item every frame, causing typing/keyboard focus to be broken on those page. ### Testing tested on native viewer Source-Ref: 0163ff9d28a01fa22e4a95076feb922f83aab4a9
…ests Final step of the quiver PR chain (splitting up rerun-io#2263). Stacked on rerun-io#2281. Pure cleanup, no functional change: * Shorter `ext::` import paths throughout * `RegisterWithDatasetDataframe::try_from` destructuring instead of per-column extracts * `TaskId::from` / `into_string` conversions instead of manual struct building * `.unique()` instead of a BTreeSet round-trip for dedup Deliberate deviations from the prototype branch (kept from main, not regressed): * `Option<&[SegmentId]>`-based segment filters in `re_server` (newer than the prototype's empty-list convention) * `QueryTasksDataframe`'s strongly-typed `Column<TaskId>` from rerun-io#2274 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- ### 🚂 Quiver PR train 1. rerun-io#2274 — Introduce quiver + migrate QueryTasks ✅ 2. rerun-io#2278 — RegisterWithDataset response ✅ 3. rerun-io#2280 — ScanSegmentTable / ScanDatasetManifest ✅ 4. rerun-io#2281 — QueryDataset 5. rerun-io#2283 — Polish: idio******** imports & typed test extracts 6. rerun-io#2287 — Catalog entries table + Tuid serialization 7. rerun-io#2288 — manifest-registry column helpers 8. rerun-io#2331 — DatasetManifest read _(combined into rerun-io#2334)_ 9. rerun-io#2334 — DatasetManifest read + write paths 10. rerun-io#2335 — Strongly-typed `DatasetManifestBatch` via `#[derive(Quiver)]` _(Related: rerun-io#2314 — update to quiver 0.2.0)_ Source-Ref: 624799f63a0beb419aee9a20a673f792838e2d3d Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
### Related - Closes rerun-io#9464 - Closes RR-4760 ### What Instead of navigating to some arbitrary open recording when closing a recording, go through navigation history and find the last visited one that's still open, and navigate there. Source-Ref: 41fd7790a0eb86c7ded8ae293fe44abbf4b40b91
### Related * new Rerun `VoxelGridMap` archetype: rerun-io@fa40ec1 * Foxglove voxel grid: https://docs.foxglove.dev/docs/sdk/schemas/voxel-grid * rerun-io#11358 ### What Foxglove `VoxelGrid` stores voxels in a dense byte array with row/column/slice stride and RGBA colors. The Rerun type uses sparse voxel indices, but these can be easily mapped. ### Testing Snapshot test with minimal MCAP file. <img width="762" height="442" alt="image" src="https://github.com/user-attachments/assets/d66e8b61-e264-423d-b2b8-f26eb6840043" /> ### Compatibility Just adds a new lens. Source-Ref: 098205d410a54eafda88570723148be70837fe62
slipped through the cracks since wgsl is compiled at runtime only. ... so we're also fixing that here and add a test that validates all wgsl shaders as part of re_renderer's test suite! Source-Ref: 70f5f0ea869906cac65adea5ee21d880adede2cd
Follow-up to the quiver PR chain (rerun-io#2274 → rerun-io#2278 → rerun-io#2280 → rerun-io#2281 → rerun-io#2283), applying quiver beyond the wire dataframes. * New `EntriesTableDataframe` (`#[derive(Quiver)]`) replaces the hand-written schema and manual array construction in `redap_catalog`'s `EntriesTable`. The existing `entries_table_schema` snapshot pins that the schema is unchanged. * As a side effect, `redap_catalog` no longer needs its `re_types` and `re_types_core` dependencies. * `Loggable for Tuid`'s `to_arrow` now builds through `quiver::Column<Tuid>` instead of a manual `FixedSizeBinaryBuilder`. * `Tuid::from_arrow` is deliberately unchanged: it ignores the validity mask (per its datatype contract) and reads bulk zero-copy — quiver currently supports neither (noted as quiver improvement suggestions in `quiver_todo.md`). 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- ### 🚂 Quiver PR train 1. rerun-io#2274 — Introduce quiver + migrate QueryTasks ✅ 2. rerun-io#2278 — RegisterWithDataset response ✅ 3. rerun-io#2280 — ScanSegmentTable / ScanDatasetManifest ✅ 4. rerun-io#2281 — QueryDataset 5. rerun-io#2283 — Polish: idio******** imports & typed test extracts 6. rerun-io#2287 — Catalog entries table + Tuid serialization 7. rerun-io#2288 — manifest-registry column helpers 8. rerun-io#2331 — DatasetManifest read _(combined into rerun-io#2334)_ 9. rerun-io#2334 — DatasetManifest read + write paths 10. rerun-io#2335 — Strongly-typed `DatasetManifestBatch` via `#[derive(Quiver)]` _(Related: rerun-io#2314 — update to quiver 0.2.0)_ --------- Source-Ref: ec23a72f41c5ea74446a16e499970dd51931c833 Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Asked claude to ensure all our internal architecture docs are up-to-date Source-Ref: 7b6cce3b0d4b021bed22ebe596862521b751b700
### Related - Part of RR-4091 ### What Buffer time playing if video indicates that it's buffering. This does not affect overall play behavior on what happens when buffering, just when buffering happens. https://github.com/user-attachments/assets/000462e1-3565-47ab-b935-97e73d99fedf Source-Ref: 699a73699610d786e57f0e45eb222dbf016103a3
slipped through - this was always wrong despite the comment correctly explaining it Source-Ref: eb55e603d2b4981ff5bc00d05c7c30f36e986c63
### Related * Small things identified while running ROS 2 bag tests for (unlanded) bagfile plugin work ### What Three fixes to ROS 2 reflection decoding caught while running ROS 2 bag tests. * **`byte`/`char` → uint8.** ROS 2 `byte` and `char` are both unsigned 8-bit; `char` was wrongly decoded as `int8`. * **Pad empty message specs like rosidl.** Field-less specs now gets `structure_needs_at_least_one_member`. * **Reject `wstring`** `wstring` is UTF-16 and decoding it as a narrow string corrupted the rest of the message. The reflection decoder now skips `wstring` channels (raw fallback), and de/serialization errors. ### Testing Unit tests in `re_ros_msg` and `re_mcap` (byte/char aliasing, padding, wide/mixed enum specs, `wstring` detection). `pixi run cargo test -p re_ros_msg -p re_mcap` green. ### Compatibility No format or API changes. Affects only how ROS 2 MCAP channels decode (`wstring` channels now fall back to lossless raw bytes instead of producing corrupt data.) Source-Ref: 0d3f1889c48eba97b4044b28145ffa1fadfedee9
Builds on rerun-io#2334. Replaces the naked `RecordBatch` inside `DatasetManifestBatch` with a `#[derive(Quiver)]` struct, so the dataset-manifest columns are validated and strongly typed. ### What - New `DatasetManifestBatchQuiver` (`#[derive(Quiver)]`): every manifest column as `quiver::Column<L>` (`Column<Option<L>>` for nullable). `segment`-named Rust fields map to the legacy on-disk names via `#[quiver(name = "rerun_partition_*")]`; `#[quiver(extra_columns)]` collects the runtime `property:*` / index-range columns. - `DatasetManifestBatch { handle, data: DatasetManifestBatchQuiver }`. Construction via fallible `try_new` (validates the full schema once); `num_rows()` covers the remaining need. - Dropped the per-column accessors (`col_*_raw`, `col_rerun_*`, `col_rerun_*_parsed`) — full-schema call sites read `batch.data.<field>` directly. Kept `col_property_raw` for the runtime `property:*` columns. The URL / segment-kind parsing now lives in the free, per-row helpers `parse_storage_url` / `parse_segment_kind`. - Projected (partial-schema) scans — drop-segments, schema-sha256 lookup, the dedup paths, and the patcher — deliberately bypass `try_new` and extract individual columns via the generated `DatasetManifestBatchQuiver::COLUMN_*` descriptors, since a partial batch can't pass whole-schema validation. - The drop-segments path (the only consumer of gRPC-schema scan output) reads ids/layers via the public `ScanDatasetManifestDataframe`, keeping `DatasetManifestBatchQuiver` purely on-disk-schema. - `DatasetManifestInserter`/`DatasetManifestPatcher::apply` now return the written `RecordBatch` instead of a `DatasetManifestBatch`: patch payloads are partial, so wrapping them in a fully-typed batch would fail validation. All callers discard the value anyway. ### Testing `cargo check --all-targets`, clippy (deny-warnings), and `cargo nextest run` — all clean for `redap_manifest_registry` (109 tests pass). `cargo fmt`. CI integration tests exercise the manifest read/write/scan paths. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- ### 🚂 Quiver PR train 1. rerun-io#2274 — Introduce quiver + migrate QueryTasks ✅ 2. rerun-io#2278 — RegisterWithDataset response ✅ 3. rerun-io#2280 — ScanSegmentTable / ScanDatasetManifest ✅ 4. rerun-io#2281 — QueryDataset 5. rerun-io#2283 — Polish: idio******** imports & typed test extracts 6. rerun-io#2287 — Catalog entries table + Tuid serialization 7. rerun-io#2288 — manifest-registry column helpers 8. rerun-io#2331 — DatasetManifest read _(combined into rerun-io#2334)_ 9. rerun-io#2334 — DatasetManifest read + write paths 10. rerun-io#2335 — Strongly-typed `DatasetManifestBatch` via `#[derive(Quiver)]` _(Related: rerun-io#2314 — update to quiver 0.2.0)_ --------- Source-Ref: 8fafa5f01211f6a41b0565112593a05fd1d91c3a Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
### Related * Subset of the typed-parser → lens migration draft rerun-io#1730 * Follows rerun-io#1555 (occupancy_grid lens), rerun-io#1991 (pose_stamped lens), and the log/string lens PRs ### What * Adds a lens for `sensor_msgs/msg/MagneticField` on top of the `ros2_reflection` decoder, slotted in next to `occupancy_grid`. * Deletes the typed `MagneticFieldMessageParser` and the now-unused `MagneticField` struct definition. ### Testing * New snapshot test `test_magnetic_field` using `tests/assets/ros_magnetic_field.mcap`. * Tested for regression: ran the same mcap through the old typed parser and the new lens, the results were identical. <screenshot placeholder> ### Compatibility No shape changes. The lens emits the same `Arrows3D:vectors` (FixedSizeList<f32, 3>) and `CoordinateFrame:frame` (Utf8) as the previous typed parser, on the same `ros2_timestamp` timeline derived from `.header.stamp`. `magnetic_field_covariance` is intentionally ignored, matching the previous parser. ### Out of scope This PR is a 1:1 logic move from parser to lens. It does not change NaN handling for the `magnetic_field` vector (the ROS spec allows NaN in any component if the sensor does not report that axis). --------- Source-Ref: ce7e2f184795970593d4a0b575b2191ea67e2e4f Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
### What This makes it possible to populate the `Selector` runtime that is used in our SDKs with custom functions. This makes it easier to create reusable transformations for data pipelines. All functions should be registered and defined in `re_lenses` so that we always share a common runtime across all products. Source-Ref: 08ed1cad2dcedcda8ae8c1ccf5d239d77ce2029e
### Why Some Hub calls return `ResaurceExhausted` when concurrency limits are hit. They normally should be interpreted as suggestions for the client to retry. We want to take some of those retries off the customers' shoulders, so we do them in the SDK. ### What * Use `re_backoff` for exponential backoff with full jitter retry ### Compatibility * No compatibility issues. --------- Signed-off-by: Andrea Reale <andrea@rerun.io> Source-Ref: bcfbb98afe1ac919f46134072710b109d517baec Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com> Co-authored-by: Andrea Reale <andrea-reale@users.noreply.github.com>
* Alternative to rerun-io/reality#2410 * Fixes RR-4944 Make extra sure that things are already loaded before taking screenshot Source-Ref: 80f6441705b07500e2f20f0241457fc40e106369
### Related The new start/end time and recover options are available in the Python binding but were not yet exposed on the CLI. ### What Adds 3 args to `rerun mcap convert`: * `--recover` * `--start-time` & `--end-time` The time arguments allow strings with different time units / representation for CLI ergonomics. See `--help` strings / `cli.md` for all details. ### Testing Manual test with an MCAP file ``` rerun mcap convert example.mcap --start-time 2025-06-02T11:26:50.16879979Z --end-time 1748863630s … INFO re_mcap::decoders: Filtering MCAP messages to log_time range [1748863610168799790, 1748863630000000000) (nanoseconds) INFO rerun::commands::mcap: Processed 2261 messages. ``` vs ``` rerun mcap convert example.mcap … INFO rerun::commands::mcap: Processed 4187 messages. ``` ### Compatibility Just new CLI args. Source-Ref: 2d5726db2ae8f11759c4772ebac2001a7215dde0
### Related * Closes RR-5033 * Closes RR-5034 ### What This makes the internal Viewer catalog present unconditional. With this PR, it is always started and always reachable via gRPC. The settings dialog governs if RRD files get loaded into the Viewer catalog or take the legacy path. Also gets rid of the `internal_catalog` feature flag again. Source-Ref: 58e3fac59a53699629fbc3d7067f0f94a52aac70
### Related - part of https://linear.app/rerun/project/py-chunk-chunk-centric-processing-pipeline-and-lenses-apis-in-python-1cf386cd9cd3 - closes RR-4667 - follow-up to rerun-io/reality#2646 ### What Introduce `Hdf5Reader` to the Python SDK. Short-term future work at least includes: - RR-5249 ### Testing Include python integration tests ### Compatibility New API surface. --------- Source-Ref: 62525ea9a67988afd85194b20fc7728216dcaf41 Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> Co-authored-by: Nick <24689722+ntjohnson1@users.noreply.github.com>
On top of rerun-io#2683 ### What Exposes `re_video`'s GOP/keyframe detection and Annex B conversion utilities to Python (through `rerun.experimental.video` for now) and switches the dataloader's `VideoFrameDecoder` to use them instead of its own hand-rolled H.264/H.265/AV1 keyframe heuristics in python. --------- Source-Ref: 92ea101c846a821962b817705313386bd07da010 Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
…e's a time selection ### Related * Fixes RR-5193 ### What ... and if we _somehow_ get there, simply select the entire time range. Previously, we'd sometimes create very small selections, typically 0s-1s, which in a larger recording one wouldn't find. Therefore, just by misclicking the loop button you end up with a time selection that you then might later share on links which can be very confusing. (the previous `initial_time_selection` tried to be clever about `segments` (runs of continuous recording) but judging from the behavior it was broken. Didn't spend time going deep there because a) with the behavior change it's not important b) either way it may produce a selection that you don't see in a large recording!) https://github.com/user-attachments/assets/69f969fa-77a6-4bd6-a48b-d6fc383a3145 Source-Ref: 692326234cb43acf37e0be3af0c9a36cdd8024eb
Reduces boilerplate a bit by taking appropriate context objects for the job. There's gonna be a follow-up that makes `ViewContext` more readily available in tests which builds on top of this! --------- Source-Ref: eff1a0aed1431705b061dcb7f4794be8032c3773 Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
### Related * Follow-up to rerun-io#2592 ### What Replace the generic SQL filter fields on segment-table and dataset-manifest scans with an explicit `SegmentIdFilter` protobuf message. The filter uses a `oneof` strategy: `scan_only` scans rows for the listed segment IDs, while `skip` scans every segment except the listed IDs. I wrote the protobuf definition like this to be very clear and explicit, because an earlier version suggested putting "include" and "exclude" on the filter and I thought it was ambiguous about "are we including in the things to filter or including in the final results?" The DataFusion providers normalize supported `=`, `!=`, `IN`, `NOT IN`, and compatible boolean expressions into these strategies. Pushdown remains `Inexact`, so DataFusion re-applies the original expression after receiving the scan output. The code documents that this can become `Exact` after capability negotiation or after compatibility with servers that predate the new field is no longer required. We can do this after ALL servers have been upgraded or we add some kind of versioning information to the connection. Remove generic SQL parsing, post-aggregation filtering, and filter-specific public-to-internal column translation from both server implementations. ### Testing * Focused `cargo nextest` coverage for DataFusion normalization, manifest-registry expression construction, and both scan endpoints against the OSS server and dataplatform frontend ### Compatibility The previous `sql_filter` field names and numbers are reserved, and the new message uses new field numbers. Old servers ignore the new field, while new servers ignore scan filter fields sent by old clients. Because pushed filters remain `Inexact`, downstream DataFusion re-applies them and preserves query correctness during mixed-version deployments. Since we're still marking everything as `Inexact` we can guarantee correctness by the follow on filters. Source-Ref: 1782581f8117f3ef27364785aaacfeb3d8491ad0
### What Ideally, coding agents should self-disclose themselves, especially when creating issues and PRs auto********ally without a human in the loop. These changes might help nudging the LLMs into doing that. --------- Source-Ref: fdf08c3673bde074ac072629382bc0cd8c82e8a1 Co-authored-by: Michael Grupp <michael@rerun.io>
### Related Follow-up to: rerun-io/reality#2771 ### What rerun-io#2771 introduced a small `IndexColumn` dataclass to describe an index column (name, kind, input unit). It's also used in `ParquetReader`. We previously were using a tuple of str that is eminently unergonomic. ### Testing Updated test ### Compatibility Breaking change without deprecation (`rerun.experimental`) Source-Ref: d3f2b6d0b67c638d991bfd5258671479d8875f88
### Related * Part of RR-5086 * Follow up RR-5248 ### What RRDs uploaded to OPFS for the Viewer catalog are persisted and take up user's disk space. We want to be mindful about not poluting their disk, so we use a introduce `RrdFingerprint` as a form of "content-based" hash of an RRD. For RRDs with footer, the hash is computed solely over the bytes of the footer, for footerless RRDs, the hash is computed over the entire RRD. A file is only uploaded to OPFS if it does not exist there yet and we stop clearing files from OPFS on internal catalog startup. ### Compatibility This PR should not add any new allocations to pre-existing code paths. Source-Ref: 9a6e2ac7c36d9950e01fae550daff4cdb59e960a
Source-Ref: 5c94ad6fe73bc08fa7a4a21558279c84c5fb1241 Co-authored-by: Michael Grupp <michael@rerun.io> Co-authored-by: Nick <24689722+ntjohnson1@users.noreply.github.com> Co-authored-by: Lucas Meurer <hi@lucasmerlin.me>
### Related * RR-4595 ### What Add pixi tasks to measure code coverage: ```bash cd rerun pixi run rs-coverage [crate] pixi run rs-coverage-html [crate] ``` <img width="1539" height="693" alt="image" src="https://github.com/user-attachments/assets/13329c6c-9ea9-49a5-89f3-8474ac58c0b7" /> --------- Source-Ref: 9ad6e7c24dc7351bc2e9b31f0ffa8d5af5571623 Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
…erun-io#12858) ### What Loading a Draco- or meshopt-compressed glTF/GLB currently surfaces the raw `gltf` crate validation error: ``` invalid glTF: extensionsRequired[0] = "KHR_draco_mesh_compression": Unsupported extension ``` which names the extension but doesn't tell the user what to do. This maps that case to an actionable error instead: ``` This file requires glTF extension(s) that Rerun does not support: KHR_draco_mesh_compression. If this is Draco or meshopt compression, re-export the model uncompressed. ``` ### How `gltf::import_slice` rejects unsupported `extensionsRequired` during validation, *before* any geometry is read (verified against gltf 1.4.1), so the previously-hypothesized "mesh has no triangles" path never triggers. On the `import_slice` error we re-parse with `Gltf::from_slice_without_validation` purely to recover and name the required extensions. This is the "give a decent error message" slice of rerun-io#6365 — it does **not** add Draco/meshopt decoding. ### Confidence / disclosure Written with the help of a coding agent (Claude), reviewed by me. Confidence is high: the change is small and localized to the import error path, the underlying `gltf` behavior was verified empirically, and there's a unit test covering the extension-naming. I'm less familiar with how this error ultimately renders in the viewer UI, so pointers there are welcome. Part of rerun-io#6365. Source-Ref: 506227337ac77bf76c1b78e092757a60c5a91682
- broke in rerun-io/reality#1879 Source-Ref: be0eeef65004f2ac6b8fa8e8458f3a90bc144178
Merges three releases of upstream work (0.33, 0.34, 0.35) onto a fork based on 0.32.0-alpha.1. The previous sync landed as a squash, so git's merge base was stale by five releases. The parent commit grafts the real base (8e9635b) back into the ancestry, which cut this merge from thousands of spurious conflicts down to 43. Conflicts resolved: * Version strings, lockfiles and packaging — took upstream, preserving the "dimos" workspace member and the dimos-viewer console script. * re_video player — kept the fork's live-stream GOP-continuity fix and ported it onto upstream's renamed `video_source` API. * Window title — kept "DimOS Viewer". Behaviour changes: * `--follow` is removed. Upstream deleted file-tailing in 14f4d32 along with `FromUriOptions.follow`, the field the flag fed into. Nothing in DimOS passes the flag. * `start_native_viewer_with_wrapper` is rewritten against upstream's current `start_native_viewer`, keeping the `app_wrapper` and `startup_patch` hooks. The fork's copy had drifted and was missing 0.35's internal-catalog service and gRPC memory accounting, and still used the dropped `cfg-if` dependency. Upstream moved rust-version 1.92 to 1.95 and datafusion 52.3 to 53.0. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
|
|
||
| import rerun as rr | ||
|
|
||
| output_path = Path(tempfile.mktemp(suffix=".rrd")) |
gilrs-core's libudev-sys needs libudev.pc on Linux: libudev-dev on the ubuntu check job, systemd-devel in the manylinux wheel containers. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
leshy
added
the
exclude from changelog
Newline before the start_native_viewer_with_wrapper doc comment, and redirects for the Status -> State doc pages the rename moved. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… decoder The GOP-rollover guard added in 2b473d1 required the enqueued samples to lead *exactly* into the next GOP's keyframe (`last_enqueued + 1 >= gop_start`). At the live edge that holds only when the tail of the old GOP has fully arrived by the time its successor's IDR lands, which is not reliable: on a 1080p/53fps drone stream (GOP 192) the gap is regularly 2-3 samples, and every one of those fell through to the reset branch — killing and respawning the ffmpeg CLI process, then re-decoding from the keyframe. A gap of a few samples is ingest jitter, not a seek, and the reset is unnecessary regardless: the target is an IDR, which resets reference state by definition, so it can be fed to the running decoder whether or not the frames before it made it in. Skips of more than one GOP never reach this check — handle_errors_and_reset_decoder_if_needed has already reset by then — so the tolerance only has to cover jitter. It stays well under a GOP so that a genuine forward skip inside the previous GOP still resets rather than decoding the whole remainder for nothing. Measured on a 75s replay of that stream, before -> after: 5 -> 0 gap resets, 7 -> 2 ffmpeg spawns (both at startup, 0.5ms apart), 8 -> 14 rollovers decoded straight through. One long-lived ffmpeg process for the whole run. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Merges upstream rerun 0.35.0 into the fork, from a base of 0.32.0-alpha.1 — three releases of upstream work (0.33, 0.34, 0.35).
Why this merge looks different from the last one
The 0.32.0a1 sync landed as a squash (
5e4e46c94, 3391 files in one commit), so git had no ancestry link to upstream past1f628f78c(0.30.0-alpha.1).A plain
git merge 0.35.0would therefore have picked a merge base five releases stale and produced thousands of bogus conflicts.This PR fixes that first.
The upstream commit the fork actually sits on was identified as
8e9635b322by minimizing the tree diff againstmain, and a no-op-s oursmerge records it as an ancestor.The tree is provably unchanged by that commit; it exists purely to repair the merge base.
With the base repaired, the real merge produced 43 conflicts instead of thousands, and the next sync to 0.36 will be routine.
Conflict resolution
The three real merges
re_video/src/player/mod.rs— the fork's live-stream GOP-continuity fix (don't respawn ffmpeg on GOP rollover) sat exactly on the code upstream rewrote, and upstream renamedget_video_buffer→video_source.Kept the fork's contiguity
if/elseand ported it to the new API.entrypoint.rs— kept the fork's--ws-urlflag and therun_with_app_wrapperentry point.re_viewer/src/native.rs— kept the"DimOS Viewer"window title;APP_ID = "dimos-viewer"was untouched.Behaviour change:
--followis goneUpstream removed file-tailing in
14f4d32843, deletingFromUriOptions.follow— the field the fork's--followflag was plumbed into.Upstream's note: "This feature was also broken for quite some time now."
Their replacement is sink teeing (log to viewer and
.rrdfrom the producing process).The flag is dropped rather than reimplemented, because the machinery behind it no longer exists upstream.
Verified nothing in the DimOS repo passes it — the
--followindimos/robot/cli/dimos.pyis an unrelated log-tailing flag on a different CLI.start_native_viewer_with_wrapperrewrittenThe fork's copy of
start_native_viewerhad drifted from upstream (126 lines vs 169) and silently missed 0.35's internal-catalog gRPC service and gRPC-server memory accounting.Rewritten to mirror upstream's current implementation, keeping only the two fork hooks (
startup_patch,app_wrapper), which also restores--headlesssupport.Upstream's
cfg_if::cfg_if!→cfg_select!migration was applied here too — upstream dropped thecfg-ifdependency, which is what broke the first build.Breaking changes checked against DimOS
Every breaking change in the 0.33/0.34/0.35 migration guides was grepped against the DimOS repo — zero hits in application code:
rerun.recordingmodule removed,rr.send_dataframestricter,log_tickno longer default (0.34)rerun-sdk[dataplatform]/[datafusion]→[catalog](0.33)SaveScreenshotmoved toViewerControlService(0.34)ParquetReaderindex columns now useIndexColumn,StateChange.statenow an array (0.35)Toolchain
Upstream moved
rust-version1.92 → 1.95 (0.35 uses the built-incfg_select!) anddatafusion52.3 → 53.0.Pinned toolchains and CI images need to match.
dimos-viewerbumped to0.35.0a1/0.35.0-alpha.1.Verification
cargo check -p dimos-viewer— cleancargo build -p dimos-viewer --bin dimos-viewer— linksrun_with_app_wrapper,AppWrapper,StartupOptionsPatch,RerunArgs,--ws-url/DIMOS_VIEWER_WS_URL,APP_ID, thedimosworkspace member, thedimos-viewerconsole script, anddocs/websockets.mdrerunanddimos/src/interaction/ws.rsare pre-existing onmain, not introduced hereNot run: the full test suite and web/wasm build.