fix: combine tool call sync fix with GLM model detection for issue #11071

[roomote]

Related GitHub Issue

Closes: #11071

Roo Code Task Context (Optional)

N/A

Description

This PR attempts to address Issue #11071 by combining the fix from PR #11093 with GLM-related improvements (points 1-5 from the discussion), excluding point 6 (path-based repetition detection).

Changes included:

1. PR fix: prevent tool_call_end events for unstarted tool calls #11093 Fix - Tool Call Sync: Fixed NativeToolCallParser.processFinishReason() to only emit tool_call_end events for tool calls that have actually started (i.e., where tool_call_start was emitted). This prevents finalizeStreamingToolCall from receiving IDs that were never registered, which was causing tool results to be silently dropped and triggering infinite retry loops.

2. GLM Model Detection (glm-model-detection.ts): New utility that detects GLM models (GLM-4.5, GLM-4.6, GLM-4.7 and their variants) from model IDs, supporting various formats from LM Studio and OpenAI-compatible endpoints.

3. mergeToolResultText Optimization: For detected GLM models, the providers now use convertToZAiFormat with mergeToolResultText: true to prevent conversation flow disruption and reasoning_content loss.

4. Disable parallel_tool_calls: GLM models have parallel_tool_calls disabled as they may not support it properly.

5. GLM-4.7 Thinking Parameters: For GLM-4.7+ models, the providers send the thinking: { type: "enabled" | "disabled" } parameter based on user settings.

6. Diagnostic Logging: Console logging shows GLM detection results including version, variant, and applied optimizations.

Test Procedure

• Added 5 new test cases in NativeToolCallParser.spec.ts covering the processFinishReason fix
• Added 34 test cases in glm-model-detection.spec.ts covering GLM detection

Run tests:

cd src && npx vitest run core/assistant-message/__tests__/NativeToolCallParser.spec.ts
cd src && npx vitest run api/providers/utils/__tests__/glm-model-detection.spec.ts

All 51 tests pass.

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes (if applicable).
• Documentation Impact: I have considered if my changes require documentation updates (see "Documentation Updates" section below).
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Screenshots / Videos

N/A - Backend fix, no UI changes.

Documentation Updates

• No documentation updates are required.

Additional Notes

This PR combines solutions from prior PRs (#11093, #11092) as requested by @mark-ucalgary in the issue discussion. Feedback and guidance are welcome!

---

Important

Fix tool call sync issue and enhance GLM model detection, adding support for GLM-4.5, 4.6, 4.7 models and optimizing message handling.

• Behavior:
  • Fix processFinishReason() in NativeToolCallParser to emit tool_call_end only for started tool calls, preventing unregistered IDs from causing issues.
  • Add glm-model-detection.ts to detect GLM models (GLM-4.5, GLM-4.6, GLM-4.7) and handle various ID formats.
  • Optimize mergeToolResultText for GLM models to prevent conversation flow disruption.
  • Disable parallel_tool_calls for GLM models.
  • Add thinking parameter for GLM-4.7 models based on user settings.
• Tests:
  • Add 5 test cases in NativeToolCallParser.spec.ts for processFinishReason fix.
  • Add 34 test cases in glm-model-detection.spec.ts for GLM detection.
• Misc:
  • Log GLM detection results including version and variant.

^{This description was created by for 9e96578. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Rooviewer   See task on Roo Cloud

Review complete. No issues found.

The PR correctly addresses the reported issue by:

1. Fixing NativeToolCallParser.processFinishReason() to only emit tool_call_end events for tool calls that have started (preventing infinite retry loops)
2. Adding comprehensive GLM model detection for LM Studio and OpenAI-compatible endpoints
3. Applying GLM-specific optimizations (mergeToolResultText, disabled parallel_tool_calls, thinking parameters for GLM-4.7)

The latest commit (9e96578) adds GLM detection to OpenAiHandler (the "OpenAI Compatible" provider), following the same patterns already established in BaseOpenAiCompatibleProvider. The implementation is consistent between streaming and non-streaming paths.

The changes are well-tested with 39 new test cases.

Previous reviews

• 96b6845: Review #1

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

[mark-ucalgary]

@roomote tested with OpenAI-compatible endpoint and GLM-4.5-UD-Q8_K_XL-00001-of-00008.gguf as the model. I didnt see any logging about model detection in the console.