[BUG] Provider ended the request: Gemini generate context stream error: No output generated. Check the stream for errors.

[rakaki]

Problem (one or two sentences)

Date/time: 2026-02-12T05:39:36.585Z
Extension version: 3.47.3
Provider: gemini
Model: gemini-3-pro-preview

Provider ended the request: Gemini generate context stream error: No output generated. Check the stream for errors.

Context (who is affected and when)

immediately API Request Failed without sending request to api server.

Reproduction steps

can not resolve by reseting roo code in vscode

Expected result

normaly send request to api server

Actual result

immediately API Request Failed without sending request to api server.

Variations tried (optional)

No response

App Version

3.47.3 (ec2d110)

API Provider (optional)

None

Model Used (optional)

No response

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote-v0]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause Analysis:

The error message "Provider ended the request: Gemini generate context stream error: No output generated. Check the stream for errors." traces through this code path:

1. GeminiHandler.createMessage() in src/api/providers/gemini.ts calls streamText() and processes the stream
2. When Gemini returns no output (e.g., due to content filtering, thinking-only response, or transient API issues), the AI SDK rejects both result.providerMetadata and result.usage with NoOutputGeneratedError
3. The code already handles NoOutputGeneratedError gracefully in the result.usage catch block (lines 194-206) by suppressing it and yielding a user-friendly "Model returned an empty response..." message
4. However, result.providerMetadata is awaited before result.usage (line 162), and its catch block (lines 163-168) does NOT handle NoOutputGeneratedError -- it simply rethrows it
5. This unhandled error reaches the outer catch (line 210), which formats it as "Gemini generate context stream error: No output generated. Check the stream for errors." and shows the scary error to the user

Proposed Fix:

1. Modify src/api/providers/gemini.ts: Add NoOutputGeneratedError handling to the providerMetadata catch block (lines 163-168), mirroring the usage catch block logic. When !hasContent and the error is NoOutputGeneratedError, set providerMetadata to undefined instead of rethrowing.

2. Add a test in src/api/providers/__tests__/gemini.spec.ts: Add a test case where both providerMetadata AND usage reject with NoOutputGeneratedError, verifying the stream completes gracefully with the informative empty-response message instead of throwing.

3. Update the existing test: The current "should suppress NoOutputGeneratedError when no text content was yielded" test (line 183) only rejects usage while keeping providerMetadata resolved -- this doesn't match the real AI SDK behavior. Update it to reject both promises.

This approach will ensure that when Gemini returns an empty response, users see the user-friendly message ("Model returned an empty response...") instead of the raw SDK error.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.