fix: repair stale FTS on no-op updates by Ha1baraA11 · Pull Request #1105 · tirth8205/code-review-graph

Ha1baraA11 · 2026-10-08T20:34:12Z

Linked issue

Closes #1104

What & why

Leave build_state as postprocess-pending when an update skips post-processing.
Let a later update repair pending derived data without forcing a full graph rebuild.
Reconcile detectable FTS drift on no-op updates, including graphs from older versions
that incorrectly kept the complete marker.
Added regression tests for pending-state repair and legacy complete-state FTS drift.

How it was tested

uv run ruff check code_review_graph/ — passed.
uv run pytest tests/ --tb=short -q — 4,270 passed, 850 skipped, 2 xfailed, 2 xpassed.
Used watchdog's KqueueObserver because native FSEvents segfaulted on this macOS host.

Checklist

Tests added for new functionality
All tests pass: uv run pytest tests/ --tb=short -q
Linting passes: uv run ruff check code_review_graph/
Type checking passes: uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional
Lines are at most 100 characters
Docs updated where behavior changed (README, docs/, docstrings)

Ariosolzq · 2026-10-08T23:16:09Z

Thanks for working on #1104! The fix looks promising, but I noticed two potential regressions:

1. postprocess-pending may be reported as healthy.

Changing build_was_interrupted() to check only BUILD_IN_PROGRESS also affects CLI status, which uses the same function. A graph with pending post-processing may no longer report INCOMPLETE.

The existing opt-in test test_kill_after_the_anchor_is_visible_as_an_incomplete_build covers this behavior.

2. Partial parse failures may restore an incorrect complete state.

When postprocess="none" and some files fail parsing, the new branch restores previous_state. If other files were successfully updated, FTS may still be stale even though the state returns to complete.

Could we preserve the distinction between needing a full rebuild and having incomplete derived state, and add regression coverage for partial parse failures?

Thanks again for the contribution!

Ha1baraA11 · 2026-10-14T14:35:48Z

Thanks for catching these. I checked the current code and confirm both are still open:

status uses build_was_interrupted(), which now returns true only for BUILD_IN_PROGRESS; a postprocess-pending graph can therefore lose the INCOMPLETE status. The existing after-anchor interruption test still expects that status.
When an incremental update stores some files but other files fail parsing with postprocess="none", the final branch restores previous_state. If the prior state was complete, this can hide stale FTS data.

This PR is not ready to treat either case as fixed. The correction should keep the user-visible pending state distinct from “needs a full source rebuild,” and keep derived-data state pending after partial parse failures.

tirth8205 · 2026-10-15T12:39:18Z

Thank you for this PR.

I tested killed builds, update and status on your branch and on staging.

Before I can merge it, please fix these:

status no longer shows a postprocess-pending graph as incomplete (code_review_graph/tools/build.py:36-43). build_was_interrupted now checks only in-progress. Steps: build, commit a change, then update --skip-postprocess. Now status --json says "build_incomplete": false. test_kill_after_the_anchor_is_visible_as_an_incomplete_build now fails. It passes on staging.
This also breaks repair after a killed build (code_review_graph/tools/build.py:643). Kill a build after the anchor is written. Commit one changed file, then run update. It runs an incremental update, not a full rebuild. Flows and communities stay at 0. A clean build has 24 of each. The state still says complete, so status shows no problem.
A partial --skip-postprocess update still marks the graph complete (code_review_graph/tools/build.py:810-814). Example: change a.py and b.py, and make b.py fail to parse. a.py is saved, but its new name is not in search. Please keep postprocess-pending when any rows changed.

After you push the fix, I will test it again.

- incremental update that re-parses files reports the skip, including a repo path with spaces and non-ASCII characters - repeated build/update requests report every time and never create the embeddings table - no-change update reports the skip at postprocess none/minimal/full (guards the tirth8205#1105 no-op rewrite in tools/build.py) - interrupted-build repair reports the skip - postprocess with an empty embeddings table keeps status ok - skip, then embed, then refresh reports embeddings_refreshed and no notice - has_embeddings: no table, empty table, current and legacy rows - legacy index without provider identity fails the refresh and stops watch; an empty legacy table is a skip - watch exempts only the exact notice (duplicates ok; half-pair, altered text still fail); None/non-dict results are ignored - real MCP tools and CLI build/update/postprocess/--quiet report the notice, exit cleanly and never resolve a provider Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… into pr/code-review-graph-1104 # Conflicts: # code_review_graph/tools/build.py

github-actions · 2026-10-15T16:43:39Z

code-review-graph review

Overall risk: 0.70 (HIGH) — 7 changed function(s)/class(es), 3 affected flow(s), 1 test gap(s) (0 with no tested caller found, 1 reached only through a caller)

Risk-scored changes

Risk	Level	Symbol	Location	Tested
0.70	high	code_review_graph/tools/build.py::build_was_interrupted	code_review_graph/tools/build.py:37	indirect
0.55	medium	code_review_graph/tools/build.py::build_or_update_graph	code_review_graph/tools/build.py:603	yes
0.05	low	tests/test_freshness_metadata.py::test_noop_update_repairs_fts_drift_without_rebuilding	tests/test_freshness_metadata.py:85	(test)
0.05	low	tests/test_freshness_metadata.py::test_skip_postprocess_marks_pending_and_noop_update_repairs_fts	tests/test_freshness_metadata.py:134	(test)
0.05	low	tests/test_freshness_metadata.py::test_pending_changed_file_update_rebuilds_all_derived_data	tests/test_freshness_metadata.py:176	(test)
0.05	low	tests/test_freshness_metadata.py::test_partial_skip_postprocess_update_keeps_pending_state	tests/test_freshness_metadata.py:220	(test)
0.05	low	tests/test_freshness_metadata.py::test_status_json_marks_postprocess_pending_as_incomplete	tests/test_freshness_metadata.py:246	(test)

Affected execution flows

build_or_update_graph_tool — criticality 0.73, 620 node(s) across 24 file(s)
main — criticality 0.72, 925 node(s) across 40 file(s)
run_postprocess_tool — criticality 0.63, 54 node(s) across 13 file(s)

Reached only through a caller

No test names these directly, but a tested caller reaches them along the call path shown. That is a path in the call graph, not a record of execution -- the caller's tests may never take this branch. Treat it as where to look, not as coverage: these count as gaps and are scored as untested.

code_review_graph/tools/build.py::build_was_interrupted (code_review_graph/tools/build.py:37) — via code_review_graph/tools/build.py::build_or_update_graph, 1 hop(s)

Powered by code-review-graph — local-first analysis; no code leaves the CI runner.

Ha1baraA11 · 2026-10-19T01:43:02Z

Thanks for the detailed report. The current head d099d6d addresses each case:

status treats postprocess-pending as incomplete.
An update after an interrupted build triggers a full rebuild, restoring flows and communities.
If parsing partially fails after files are written while postprocess="none", the build remains pending so stale derived data is not reported as complete.

I reran the three targeted freshness regressions (3 passed) and both opt-in interruption and recovery tests (2 passed). The required lint, type-check, security, schema-sync, and Python 3.10–3.13 checks are green on this head; Upgrade path from released versions is skipped. The full local test suite did not complete: two runs exited with code 139 at different tests, so I am not reporting a full-suite pass.

Please take another look at this head when convenient.

Copilot

🟡 Changes recommended

Pending-state recovery can skip unfinished language resolvers and mark the graph complete despite missing cross-file relationships.

3 open issues

What changed in this PR

This PR addresses #1104 by preserving pending derived-data state when post-processing is skipped and enabling later updates to repair stale full-text search (FTS) data without rebuilding source files.

Changes:

Keep skipped post-processing visibly pending.
Reconcile FTS drift on no-op updates, including older graphs marked complete.
Add regression coverage for pending-state recovery, derived-data repair, and incomplete status reporting.

File	Description
tests/test_freshness_metadata.py	Adds regression tests for FTS repair and pending-state handling.
code_review_graph/tools/build.py	Updates recovery, no-op reconciliation, and build-state transitions.

_{Review effort: Balanced}

+        # but only rebuild source files when storage itself was interrupted.
        interrupted = build_was_interrupted(store)
-        if interrupted and not full_rebuild:
+        if not graph_contents_are_complete(read_build_state(store)) and not full_rebuild:


+                warnings = _run_postprocess(
+                    store,
+                    build_result,
+                    postprocess_level,


+                warnings = _run_postprocess(
+                    store,
+                    build_result,
+                    postprocess_level,
+                    full_rebuild=False,
+                    changed_files=None,
+                    repo_root=str(root),
+                )
+                if warnings:
+                    build_result["warnings"] = warnings
+                if build_result.get("postprocess_failed"):
+                    build_result["status"] = "partial"
+                if not build_result.get("postprocess_contended"):
+                    store.set_metadata(BUILD_STATE_KEY, BUILD_COMPLETE)
                if embedding_provider and embedding_model:


…view of tirth8205#1105) A no-op update on a complete graph now checks FTS only, and does not rewrite every edge. A pending no-op update forwards the embedding refresh. full_build runs its language resolvers before it records postprocess-pending, so a build stopped there gets a full rebuild. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts: # code_review_graph/tools/build.py

Ruby methods always carry their full lexical scope (tirth8205#933), so the leaf collision index from tirth8205#1087 now covers Python only, and the docs and tests say so. Generators are not component wrappers' functions (tirth8205#972 with tirth8205#1131). Partial graph coverage counts tracked files only (tirth8205#881 with tirth8205#1094). A pending marker recomputes all derived data (tirth8205#1105), so that test starts from a finished graph, the deleted-seed case now passes (tirth8205#1062 with tirth8205#673), and hooks.json keeps its events under a hooks key (tirth8205#1102 with tirth8205#1126). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

fix: reconcile FTS on no-op updates

bb1cf24

Ha1baraA11 added 3 commits October 16, 2026 00:12

fix: keep derived data pending after partial updates

df4a554

Merge remote-tracking branch 'refs/remotes/codex-pr-maintain/staging'…

20c2668

… into pr/code-review-graph-1104 # Conflicts: # code_review_graph/tools/build.py

fix: keep embedding refresh notice on no-op repairs

d099d6d

tirth8205 requested a balanced review from Copilot November 5, 2026 17:26

Copilot started reviewing on behalf of tirth8205 November 5, 2026 17:27

View session

Copilot AI reviewed Nov 5, 2026

View reviewed changes

tirth8205 mentioned this pull request Nov 6, 2026

Merge train: 53 reviewed PRs (2026-11-06) #1146

Merged

tirth8205 closed this pull request by merging all changes into tirth8205:staging in 0160eb0 Nov 6, 2026

gowrishacv pushed a commit to gowrishacv/code-review-graph that referenced this pull request Nov 6, 2026

Merge tirth8205#1105 into the train

e6d7699

# Conflicts: # code_review_graph/tools/build.py

Navigation Menu

Repository navigation

fix: repair stale FTS on no-op updates - #1105

fix: repair stale FTS on no-op updates#1105
4 commits merged into
tirth8205:stagingfrom
Ha1baraA11:pr/code-review-graph-1104

Ha1baraA11 commented Oct 8, 2026

Uh oh!

Ariosolzq commented Oct 8, 2026

Uh oh!

Ha1baraA11 commented Oct 14, 2026

Uh oh!

tirth8205 commented Oct 15, 2026

Uh oh!

github-actions Bot commented Oct 15, 2026

Uh oh!

Ha1baraA11 commented Oct 19, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

Ha1baraA11 commented Oct 8, 2026

Linked issue

What & why

How it was tested

Checklist

Uh oh!

Ariosolzq commented Oct 8, 2026

Uh oh!

Ha1baraA11 commented Oct 14, 2026

Uh oh!

tirth8205 commented Oct 15, 2026

Uh oh!

github-actions Bot commented Oct 15, 2026

code-review-graph review

Risk-scored changes

Affected execution flows

Reached only through a caller

Uh oh!

Ha1baraA11 commented Oct 19, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants