fix: align responses ids and text handling

[yyhhyyyyyy]

Aligned our OpenAI Responses integration with official docs by using input_text/output_text correctly and fixing streamed function-call item_id/call_id mapping with regression tests.

Summary by CodeRabbit

• Bug Fixes

  • Improved robustness of tool call event handling during OpenAI streaming by ensuring consistent internal call identifier mapping throughout the response stream, reducing potential mismatches.

• Tests

  • Added comprehensive test coverage for tool call mapping logic, validating streaming scenarios with varying identifier configurations to ensure correct event emission.

[coderabbitai]

Important

Review skipped

Review was skipped as selected files did not have any reviewable changes.

💤 Files selected but had no reviewable changes (1)

• resources/model-db/providers.json

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

📝 Walkthrough

Walkthrough

The PR enhances OpenAI response provider's tool call tracking by introducing state mappings to correlate item IDs, output indices, and call IDs, ensuring consistent tool_call_id references in streamed events while replacing dynamic response parsing with fixed content type assignment.

Changes

Cohort / File(s)

Summary

Tool Call ID Mapping & Event Handling src/main/presenter/llmProviderPresenter/providers/openAIResponsesProvider.ts, test/main/presenter/llmProviderPresenter/openAIResponsesProvider.test.ts

Added nativeToolCallIdByItemId and nativeToolCallIdByOutputIndex state mappings to track and resolve tool call IDs across streaming chunks. Updated tool_call_start, tool_call_chunk, and tool_call_end events to use derived callId instead of raw item.call_id. Enhanced both native and non-native function call handling with consistent ID resolution. Changed response parsing from response.output[0].content to response.output_text. Added comprehensive test suite validating ID mapping with itemId and output_index fallback scenarios.

Sequence Diagram

sequenceDiagram
    participant OpenAI as OpenAI API
    participant Provider as OpenAI Provider
    participant State as State Mappings
    participant EventBus as Event Bus

    OpenAI->>Provider: Stream: output_item.added (item.id, item.call_id)
    Provider->>State: Store nativeToolCallIdByItemId[item.id] = item.call_id
    Provider->>EventBus: Emit tool_call_start (tool_call_id = resolved callId)

    OpenAI->>Provider: Stream: function_call_arguments.delta
    Provider->>State: Lookup callId via itemId or output_index
    Provider->>EventBus: Emit tool_call_chunk (tool_call_id = resolved callId, partial args)

    OpenAI->>Provider: Stream: function_call_arguments.done
    Provider->>State: Retrieve complete arguments from mapping
    Provider->>EventBus: Emit tool_call_chunk (tool_call_id = resolved callId, complete args)

    OpenAI->>Provider: Stream: output_item.done
    Provider->>State: Store mapping for output_index to call_id
    Provider->>EventBus: Emit tool_call_end (tool_call_id = resolved callId)
    
    OpenAI->>Provider: Stream: completed
    Provider->>EventBus: Emit stop (reason = 'tool_use')

Loading

Estimated Code Review Effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly Related PRs

• fix: the message format for the role tool  #1089: Implements native/mock tool-call ID normalization and split/queue logic in openAICompatibleProvider, addressing related tool call ID mapping across provider variants.
• refactor(types): 🚀split monolithic presenter.d.ts into strict per-domain *.presenter.d.ts + typed core layer  #847: Introduces tool_call event handling contracts and streaming/event emission patterns in openAIResponsesProvider that this PR builds upon and refines.

Poem

🐰 Hop, hop! The provider now tracks every call,
With mappings precise and IDs that don't fall,
Each stream chunk knows its ID, no more missing trace,
The tests verify the flow—events land in their place! ✨

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'fix: align responses ids and text handling' directly addresses the main changes: correcting ID mapping and response text handling in OpenAI responses provider.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Merge Conflict Detection	✅ Passed	✅ No merge conflicts detected when merging into dev

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch fix/responses-api-alignment

---

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}