Conversation
…g/feature move Commands like `bug move` modify the HTML file in-place without committing to git. The incremental reindex only checked `git diff <lastCommit>..HEAD` (committed changes) and `git ls-files --others` (untracked new files), so dirty (modified-but-uncommitted) files were never picked up — leaving track_id stale in SQLite until `--full` reindex. Fix: add `appendDirtyHTMLFiles` which runs `git diff --name-only` (unstaged) and `git diff --cached --name-only` (staged) to include working-tree dirty .htmlgraph files in the incremental reindex file list. `deduplicatePaths` prevents double-processing when the same file appears in both the commit diff and the working-tree diff. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
![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: 000688c9ad
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".
| {"diff", "--name-only", "--", relHg}, | ||
| {"diff", "--cached", "--name-only", "--", relHg}, |
There was a problem hiding this comment.
Using git diff --name-only for dirty files includes deleted .html paths as well as modified ones, but this helper appends every returned path to the added list. In incremental reindex, those deleted files are then parsed as if they exist, producing parse errors and leaving stale feature/track rows because they never enter the deleted flow; this occurs whenever a tracked .htmlgraph/*.html file is removed (staged or unstaged) before commit.
Useful? React with 👍 / 👎.
git diff --name-only includes deleted files. The previous appendDirtyHTMLFiles routed all dirty paths to the upsert list, so a dirty-deleted .htmlgraph/*.html would either error during parse or leave a stale SQLite row instead of being removed. Fix: switch to git diff --name-status and route by status code: A, M → added (upsert) D → deleted (remove from SQLite) R → old path to deleted, new path to added Caller updated to receive both lists. Added regression test TestIncrementalReindex_DirtyDeletion: create + index → os.Remove → incremental reindex → assert SQLite row is gone.
1 participant
Motivation
Found while moving bug-92690d5b to the Yolo Mode track: after running
htmlgraph bug move, the context-pack tool showed empty track membership because the SQLite index still had the old track_id. Onlyhtmlgraph reindex --fullfixed it. This blocked lineage queries for the entire incremental session window.Root Cause
gitChangedFiles(called byrunIncrementalReindex) collects changed files from two sources:git diff <lastCommit>..HEAD— committed changes since last reindexgit ls-files --others— newly untracked filesCommands like
bug move/feature movemodify the HTML file in-place viaworkitem.Edit().SetTrack().Save()without committing. This means after a move, the changed.htmlgraph/*.htmlfile is a working-tree dirty tracked file — invisible to both sources above. Incremental reindex skips it entirely, leavingtrack_idstale in SQLite until a--fullreindex is run.The
UpsertFeatureupsert SQL does includetrack_id = excluded.track_idin theON CONFLICTclause, so the upsert path itself is correct — the bug is that the incremental path never reaches the upsert for dirty files.Fix
Added two functions in
cmd/htmlgraph/reindex.go:appendDirtyHTMLFiles: runsgit diff --name-only -- <htmlgraphDir>(unstaged) andgit diff --cached --name-only -- <htmlgraphDir>(staged) to include working-tree dirty files in the incremental file list. This is the same pattern already used for committed changes, extended to the working-tree layer.deduplicatePaths: prevents double-processing when the same file appears in both the commit diff and the working-tree diff (e.g., a file modified and staged in the same session).No changes to
internal/htmlparse/orinternal/db/feature_repo.gowere needed — the parsing and upsert paths were already correct.Tests
TestIncrementalReindex_PropagatesTrackIDAfterMove: seeds a feature with old track, writes updated HTML with new track (simulatingbug move), runsreindexFromFileLists, asserts SQLitetrack_idmatches HTML — without needing a git commit.TestDeduplicatePaths: unit-tests the deduplication helper.Acceptance
no --full) now picks uptrack_idchanges frombug move/feature movewithout requiring a git commit--fullreindex unaffected (uses the separatereindexFeatureDirpath)go build ./... && go vet ./... && go test ./...all passhtmlgraph buildsucceedsMigration Note
Existing stale rows in user DBs are not auto-migrated. A one-time
htmlgraph reindex --fullon next upgrade is sufficient to correct any staletrack_idvalues.🤖 Generated with Claude Code