Feature/tracer #113
Feature/tracer #113
Conversation
…tures-from-events.jsonl Guard planner_input dependency when emitting replay_point requires
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…tures-and-migrate-paths Clarify fixture help defaults in Makefile
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…tures-and-migrate-paths Codex/reorganize fixtures and migrate paths
|
@codex, please review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d9076ec05a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
tests/test_replay_fixtures.py
Outdated
| bucket_dir = FIXTURES_ROOT / bucket | ||
| if not bucket_dir.exists(): | ||
| continue | ||
| for path in bucket_dir.rglob("*.json"): | ||
| paths.append((bucket, path)) |
There was a problem hiding this comment.
When a replay fixture is exported with --with-requires, export_replay_fixture/write_bundle copies dependent resource files into the same output directory, and those resources can be .json files. _iter_fixture_paths currently glob/rglob’s every *.json under tests/fixtures/replay_points/{fixed,known_bad}, so resource payloads get parameterized as fixtures and then test_replay_fixture fails at assert event.get("type") == "replay_point" because resource files don’t contain a replay payload. This breaks tests as soon as any fixture includes JSON resources; consider filtering by type in _iter_fixture_paths or storing resources in a subdirectory that isn’t globbed.
Useful? React with 👍 / 👎.
…er-v2-features-and-export Tracer: observed-first `replay_case` v2, bundle export/runtime, CLI, and remove generated fixtures
|
@codex, please review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: be0ea0f3e2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
examples/demo_qa/runner.py
Outdated
| events_path = run_dir / "events.jsonl" | ||
| case_logger = event_logger.for_case(case.id, events_path) if event_logger else EventLogger(events_path, run_id, case.id) |
There was a problem hiding this comment.
Because run_one always instantiates an EventLogger when event_logger is None, per-case events.jsonl emission becomes unconditional. That means batch.handle_batch’s --events off path (which intentionally passes event_logger=None) no longer disables event logging, and the agent still records events to disk; this can unexpectedly write large trace files or capture data that users explicitly opted out of logging. Consider only creating a case logger when event_logger is provided (or add an explicit flag) so the CLI semantics for --events still work as advertised.
Useful? React with 👍 / 👎.
…ing-with-events-off Respect events flag when creating case loggers
This reverts commit 2685c7dc10e1586321b13073d8a586b631aab156.
AlexanderOnischenko
force-pushed
the
feature/tracer
branch
from
764aa87 to
be0ea0f
Compare
…een-to-ensure-validation fixture-green: validate by default, add --expected-from, rollback on mismatch with improved diffs
* Allow explicit disabling of event logging * Refine event logger sentinel typing
* Fix relational selector validation * Add replay validator checks for fixtures * Use shared replay validators in known_bad tests * Require replay validators in fixture-green and fixed tests
|
@codex, please review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7091ab385c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| log_replay_case( | ||
| replay_logger, | ||
| id="plan_normalize.spec_v1", | ||
| meta={ | ||
| "provider": spec.provider, |
There was a problem hiding this comment.
The new replay handler falls back to ctx.extras["planner_input_v1"] when provider_info_snapshot is missing (see src/fetchgraph/replay/handlers/plan_normalize.py), but this log_replay_case call never populates requires, so export_replay_case_bundle will not carry the planner_input_v1 extra (or schema resource) into the case bundle. That means replays for providers not present in the catalog/registry will silently use a minimal ProviderInfo and can normalize selectors differently from the original run. Consider adding a requires entry for planner_input_v1 (and schema resource if needed) when emitting the replay_case.
Useful? React with 👍 / 👎.
* Improve replay requires resolution for plan normalization * Refine replay dependency heuristics
|
@codex, please review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 45eb6bef44
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
src/fetchgraph/tracer/cli.py
Outdated
| def _resolve_case_dir_from_run_id(*, data_dir: Path, runs_subdir: str, run_id: str, case_id: str) -> Path: | ||
| runs_root = data_dir / runs_subdir / run_id / "cases" | ||
| if not runs_root.exists(): | ||
| raise FileNotFoundError(f"Run directory does not exist: {runs_root}") | ||
| case_dirs = sorted(runs_root.glob(f"{case_id}_*"), key=lambda p: p.stat().st_mtime, reverse=True) |
There was a problem hiding this comment.
The --run-id path resolution assumes the run_id is the directory name by building data_dir/<runs_subdir>/<run_id>/cases, but batch runs create directories named like <timestamp>_<cases_stem> while storing the real run_id only in run_meta.json/events, so --run-id will raise “Run directory does not exist” even when a valid run_id is supplied from metadata. Consider scanning run directories for a matching run_meta.json run_id (or clarifying that --run-id expects the run folder name) to make this flag work for batch runs.
Useful? React with 👍 / 👎.
…cs (#131) * Resolve run ids from history * Fix run and case dir resolution * Tighten events/run-dir resolve output * Clarify events run_dir requirements
|
@codex, please review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d3bf518b65
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| rel = _safe_resource_path(file_name, stem=stem) | ||
| if rel.parts[:3] == ("resources", stem, resource_id): | ||
| continue | ||
| if rel.parts[:1] == ("resources",) and len(rel.parts) >= 2: | ||
| rel_tail = Path(*rel.parts[2:]) | ||
| else: |
There was a problem hiding this comment.
When a bundle’s data_ref.file already includes a resource-id segment (e.g., resources/<old_stem>/<resource_id>/file.txt), the current migration logic only drops the first two path parts and then re-prefixes resource_id, producing resources/<stem>/<resource_id>/<resource_id>/file.txt. That yields a non-canonical layout and updates data_ref.file to the doubled path, so resources end up in an unexpected location and tooling that assumes resources/<stem>/<resource_id>/... (per tracer docs) won’t find them. Consider checking rel.parts[2] == resource_id and dropping three parts in that case.
Useful? React with 👍 / 👎.
No description provided.