fix(ai): re-open force-completed tool calls when more args arrive - #1019
fix(ai): re-open force-completed tool calls when more args arrive#1019Mohith26 wants to merge 2 commits into
Conversation
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (3)
🚧 Files skipped from review as they are similar to previous changes (3)
📝 WalkthroughWalkthroughThe stream processor reopens tool calls marked complete when later argument deltas arrive. Authoritative or safety-net completion can then rebuild ChangesTool-call input repair
Estimated code review effort: 2 (Simple) | ~15 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Warning There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure. 🔧 ESLint
ESLint install failed. For unrecoverable errors, disable the tool in CodeRabbit configuration. 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 |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
packages/ai/tests/stream-processor.test.ts (1)
829-855: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winMake the canonical input intentionally differ from streamed JSON.
The supplied
TOOL_CALL_END.inputequalsJSON.parse(FULL_ARGS), so this test cannot prove the canonical override path ran. Use a distinct canonical value and assert bothpart.inputandparsedArgumentsuse it whileargumentsremainsFULL_ARGS.Proposed regression strengthening
ev.toolEnd('tc-1', 'offerTemplates', { - input: { templateIds: ['mock-gsk-efimosfermin'] }, + input: { templateIds: ['canonical-template-id'] }, }), @@ + expect(part?.arguments).toBe(FULL_ARGS) - expect(part?.input).toEqual({ templateIds: ['mock-gsk-efimosfermin'] }) + expect(part?.input).toEqual({ + templateIds: ['canonical-template-id'], + }) @@ - ).toEqual({ templateIds: ['mock-gsk-efimosfermin'] }) + ).toEqual({ templateIds: ['canonical-template-id'] })🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@packages/ai/tests/stream-processor.test.ts` around lines 829 - 855, Strengthen the test around StreamProcessor by making TOOL_CALL_END.input differ from the streamed FULL_ARGS value. Update the canonical input fixture in the toolEnd call, then assert part.input and the tc-1 parsedArguments both equal that distinct canonical value while part.arguments remains FULL_ARGS.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@packages/ai/tests/stream-processor.test.ts`:
- Around line 776-915: Move the regression test block “text delta interleaved
between TOOL_CALL_ARGS deltas (`#1017`)” from the package-level tests into a
*.test.ts file alongside the source containing StreamProcessor. Preserve all
five test cases and their existing assertions unchanged.
---
Nitpick comments:
In `@packages/ai/tests/stream-processor.test.ts`:
- Around line 829-855: Strengthen the test around StreamProcessor by making
TOOL_CALL_END.input differ from the streamed FULL_ARGS value. Update the
canonical input fixture in the toolEnd call, then assert part.input and the tc-1
parsedArguments both equal that distinct canonical value while part.arguments
remains FULL_ARGS.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 49951d96-af20-4710-9e97-9068fdfd5912
📒 Files selected for processing (3)
.changeset/interleaved-text-tool-input.mdpackages/ai/src/activities/chat/stream/processor.tspackages/ai/tests/stream-processor.test.ts
A TEXT_MESSAGE_CONTENT event between two TOOL_CALL_ARGS events force-completes the in-flight tool call, populating part.input from a lenient partial-JSON parse of the arguments received so far. The subsequent authoritative TOOL_CALL_END was then skipped because the call was already input-complete, so the truncated input was never corrected even though the arguments string held the complete JSON. Re-open such an inferred-complete call back to input-streaming when a further TOOL_CALL_ARGS delta arrives, so TOOL_CALL_END (or the RUN_FINISHED safety net) re-finalizes input from the complete arguments string. Fixes TanStack#1017
Regression tests for TanStack#1017: interleaving orders with TOOL_CALL_END, TOOL_CALL_END.input as canonical, the RUN_FINISHED safety net with no TOOL_CALL_END, multiple parallel tool calls, and the internal input-complete -> input-streaming re-open transition.
6a4b458 to
d4e99f9
Compare
|
Note GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer. |
|
Maintainer sweep: rebased onto |
|
View your CI Pipeline Execution ↗ for commit d4e99f9
☁️ Nx Cloud last updated this comment at |
2 participants
Fixes #1017, reported by @andrewmcgivery, matching the diagnosis in the issue exactly.
handleTextMessageContentEventcallscompleteAllToolCallsForMessage()on every text delta, which finalizes in-flight tool calls with a lenient partial-JSON parse intoinput.handleToolCallEndEventthen early-skips calls already ininput-complete, so when a TEXT_MESSAGE_CONTENT delta interleaves between two TOOL_CALL_ARGS deltas, the truncated parse becomes permanent even thoughargumentsholds the full correct JSON.The fix is 8 lines in
handleToolCallArgsEvent: a further args delta on aninput-completecall re-opens it toinput-streaming, letting the authoritative TOOL_CALL_END (or the RUN_FINISHED safety net) re-finalizeinputfrom the complete arguments string. Self-correcting across interleaving orders; the existing inferred-completion heuristic is preserved.Patch changeset included. Five regression tests (END-finalized ordering, END-input-canonical, RUN_FINISHED safety net, multiple parallel tool calls, the internal re-open transition) were written fail-first and all fail without the fix, reproducing the issue's exact truncated output. Package suite: 1291 passed, zero new failures; oxlint and tsc clean.
Summary by CodeRabbit
Bug Fixes
Tests