fix(responses): bound streaming citation spans - #418
Conversation
release: promote dev into main for v2.32.1
# Conflicts: # package.json
[WRONG BRANCH] merge dev into main for the v2.33.0 release
Promotes the dev integration line onto main. The resulting tree is byte-identical to origin/dev, including package.json at 2.34.0. The package.json conflict is resolved to dev's side, NOT to main's stale 2.33.0. Earlier promotions (lidge-jun#2553, lidge-jun#2507) kept the target's version so the release bump would land on its own "release: vX.Y.Z" commit. That is no longer legal: this very delta adds tests/release-version-line.test.ts, which fails when the in-tree version sits behind the highest release tag. With v2.34.0-preview.20260827 now published, 2.33.0 orders behind it, so a promotion carrying the stale line turns CI red on every shard that runs the suite. The consequence for the release step is that scripts/release.ts skips the bump (release.ts:568, currentVersion === version), so v2.34.0 gets tagged on this merge commit rather than on a separate release commit. The workflow creates the tag itself after publishing and validates expected-sha against the checked-out commit, so the tag still names exactly the audited tree.
[WRONG BRANCH] promote dev onto main for v2.34.0
[WRONG BRANCH] promote dev onto main for v2.35.0
[WRONG BRANCH] promote dev onto main for v2.36.0
[WRONG BRANCH] promote dev to main for the v2.37.0 release
[WRONG BRANCH] promote dev onto main for v2.38.0
[WRONG BRANCH] promote dev onto main for v2.39.0
release: promote dev to main for v2.40.0
…rkflow call (lidge-jun#3262) Both v2.40.0 release dispatches (33615174183 preview, 33615177849 main) died at startup_failure: a workflow_call cannot grant its callee more than the calling job holds, and dev-version-bump.yml's job declares contents+pull- requests write. lidge-jun#3129 wired the call but never dispatched a release, so this is its first live run. The caller job now declares exactly the callee's two permissions; no other job in release.yml gains anything. Co-authored-by: jun <jun@lidge.dev> (cherry picked from commit 7ce0ba5)
…400-relfix release: carry the release.yml permissions fix onto main for v2.40.0
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Team Run ID: 📒 Files selected for processing (3)
Included review availability: Your plan provides up to 1 included review per hour; 0 remain after this review. 📝 WalkthroughWalkthroughThe citation marker filter now releases oversized unterminated spans incrementally and removes later valid spans correctly. The release workflow job now declares the write permissions required by its reusable workflow. ChangesStreaming citation handling
Release workflow permissions
Estimated code review effort: 3 (Moderate) | ~20 minutes Merge Risk: ⚪ Minimal · up to The citation filtering and release-workflow permission changes have no remaining concrete merge-blocking risk. Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
Full details: Title checkExplanation The title clearly and concisely describes the main change: bounding streaming citation spans to prevent unbounded buffering and rescanning. It matches the citation-marker implementation and regression tests, while the workflow permission update is secondary. Full details: Docstring CoverageExplanation Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 2 files. (1 skipped: 1 unsupported.) ✨ Finishing Touches 💡 1🛠️ Fix failing CI checks 💡
📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
github-actions
Bot
changed the title
✅ READY
Hygiene✅ Deterministic PR hygiene checks passed. |
![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: 8f7d5763a3
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".
| const end = delta.indexOf(CITATION_MARKER_END, index); | ||
| const nextStart = delta.indexOf(CITATION_MARKER_START, index); |
There was a problem hiding this comment.
When one text delta contains many START characters and no END (for example, a model or upstream emits START.repeat(n)), every iteration calls indexOf(END, index) over the entire remaining suffix before advancing to the next START, making push() O(n²). This creates another event-loop denial-of-service path in the code intended to eliminate one; scan forward with parser state so each character is examined only a bounded number of times.
Useful? React with 👍 / 👎.
| // An implausibly large unterminated span is malformed ordinary text. Releasing | ||
| // it bounds both retained memory and work per delta; subsequent STARTs can still | ||
| // begin valid citation spans. | ||
| out.push(takeHeld()); |
There was a problem hiding this comment.
When a marker remains open beyond 4,096 characters but an END arrives later, this branch releases the opener and content as ordinary streaming deltas, and the later END is emitted too. However, closeCurrentMessage() in src/bridge.ts still passes the accumulated raw text through the unbounded stripCitationMarkers(), so response.output_text.done and response.content_part.done omit the entire span. The concatenated deltas therefore disagree with the terminal text; preserve the oversized-span decision when producing the final text, or make both paths use the same filtering state.
Useful? React with 👍 / 👎.
luvs01
changed the title
2 participants
Motivation
Description
heldbuffer with a small-chunk buffer andheldLength, and add a linear-time streaming parser increateCitationMarkerFilter()to avoid repeated full-string scans.MAX_STREAMING_MARKER_SPAN_LENGTH = 4096so implausibly large unterminated spans are released as ordinary text rather than retained indefinitely.flush()join and emit any retained pieces once, preserving user-visible bytes.tests/citation-markers.test.tsthat verify large unterminated spans are released incrementally and that a later valid citation span is still removed.Testing
bun test tests/citation-markers.test.ts, which passed (all streaming and whole-string filter cases succeeded).bun run typecheck, which completed successfully against the changed files.bun run privacy:scanandgit diff --checkas repository hygiene checks; they passed for the touched files.bun run test) for context but it exposed unrelated pre-existing integration failures; the focused citation-marker tests passed within that run.Codex Task
Summary by CodeRabbit