fix: support reasoning field in OpenAiHandler for OpenAI-compatible providers

[roomote]

Related GitHub Issue

Closes: #11495

Description

This PR attempts to address Issue #11495. Feedback and guidance are welcome.

Problem: OpenAiHandler only checked for reasoning_content field in streaming responses and had no reasoning extraction at all in non-streaming responses. Some OpenAI-compatible providers (e.g., vLLM after v0.15.1) return reasoning in the reasoning field instead, following OpenAI's own recommendations.

Changes (all in src/api/providers/openai.ts):

1. Streaming path: Replaced the single reasoning_content check with a loop that checks both reasoning_content and reasoning fields, matching the pattern already used in BaseOpenAiCompatibleProvider. Also added whitespace-trimming filter to avoid yielding empty reasoning chunks.

2. Non-streaming path: Added reasoning extraction from the response message for both reasoning_content and reasoning fields before yielding the text content.

3. Tests: Added 7 new test cases in src/api/providers/__tests__/openai.spec.ts covering:

   • Streaming with reasoning_content field
   • Streaming with reasoning field
   • Priority when both fields present (prefers reasoning_content)
   • Empty/whitespace-only reasoning is filtered out
   • Non-streaming with reasoning_content field
   • Non-streaming with reasoning field

Test Procedure

• Run: cd src && npx vitest run api/providers/__tests__/openai.spec.ts
• All 54 tests pass (including 7 new reasoning field tests)

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue
• Scope: Changes are focused on the linked issue
• Self-Review: Performed self-review
• Testing: New tests added to cover changes
• Documentation Impact: No documentation updates required
• Contribution Guidelines: Read and agree

Documentation Updates

• No documentation updates are required.

Start a new Roo Code Cloud session on this branch

[roomote]

Rooviewer   See task

The changes correctly align OpenAiHandler with the existing pattern in BaseOpenAiCompatibleProvider for both streaming and non-streaming paths. One minor inconsistency flagged:

• Non-streaming reasoning extraction missing whitespace-trim filter (inconsistent with streaming path)

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

[roomote]

The non-streaming path doesn't apply the same whitespace-trimming filter as the streaming path (line 205). A whitespace-only string like " " is truthy, so it would be yielded as a reasoning chunk here but filtered out in streaming mode. This is unlikely to cause real problems since providers rarely return whitespace-only reasoning in complete responses, but it's an inconsistency that could be worth aligning.

Suggested change

	for (const key of ["reasoning_content", "reasoning"] as const) {
	if (message && key in message && (message as any)[key]) {
	yield { type: "reasoning", text: (message as any)[key] as string }
	break
	}
	}
	for (const key of ["reasoning_content", "reasoning"] as const) {
	if (message && key in message && (message as any)[key]) {
	const reasoning = (message as any)[key] as string
	if (reasoning.trim()) {
	yield { type: "reasoning", text: reasoning }
	}
	break
	}
	}

_{Fix it with Roo Code or mention @roomote and request a fix.}