Add genai.errors handling in GoogleModel

[timothy-jeong]

• Add google.genai.errors. APIError handling at GoogleModel
• Add test case tests/models/test_googe.py: test_model_status_error

Resolves: #3088

[Kludex]

I think we can do what we do in the other models, and just have a single except.

[timothy-jeong]

I initially believed I needed to explicitly handle all of the more specific exceptions defined in google.genai.errors (e.g., UnknownFunctionCallArgumentError, UnknownApiResponseError) to provide more detailed error messages.

However, after reviewing the implementation for other models (like Groq and OpenAI, which only handle their respective APIStatusError subclasses to catch HTTP status codes ≥400), I realize I can simplify the approach.

I'll only focus on the main google.genai.errors.APIError to align with the standard pattern for re-raising a ModelHTTPError.

[timothy-jeong]

Hi @Kludex ~
thank you for your review
I've changed the code to only return a ModelHttpError for errors that have an explicit Status Code, similar to how other models handle it.

[timothy-jeong]

should I make MockGoogle and inject mock client like groq test code do?

def test_model_status_error(allow_model_requests: None) -> None:
    mock_client = MockGroq.create_mock(
        APIStatusError(
            'test error',
            response=httpx.Response(status_code=500, request=httpx.Request('POST', 'https://example.com/v1')),
            body={'error': 'test error'},
        )
    )
    m = GroqModel('llama-3.3-70b-versatile', provider=GroqProvider(groq_client=mock_client))
    agent = Agent(m)
    with pytest.raises(ModelHTTPError) as exc_info:
        agent.run_sync('hello')
    assert str(exc_info.value) == snapshot(
        "status_code: 500, model_name: llama-3.3-70b-versatile, body: {'error': 'test error'}"
    )

[Kludex]

For testing, I think for http exceptions, mocking an HTTP response would be okay. Otherwise, you can't consistently reproduce this.

[cursor]

This PR is being reviewed by Cursor Bugbot

Details

You are on the Bugbot Free tier. On this plan, Bugbot will review limited PRs each billing cycle.

To receive Bugbot reviews on all of your PRs, visit the Cursor dashboard to activate Pro and start your 14-day free trial.

[cursor]

Bug: Unhandled Exception in Error Handling

The try...except block can raise a TypeError if e.code is None for an APIError. The comparison (status_code := e.code) >= 400 attempts to compare None with an integer, causing an unhandled exception when an APIError lacks a status code.

 

[timothy-jeong]

Hi @Kludex

I identified that the test code was causing the CI failures, so I removed it entirely. Now, it seems the coverage test is failing as a result.

Should I add the test code back?

[Kludex]

Should I add the test code back?

Yes, that's what I meant with my last comment.