Handle function calls without text

[ethanabrooks]

This addresses the following error:

Traceback (most recent call last):
  File ".venv/lib/python3.13/site-packages/opentelemetry/trace/__init__.py", line 589, in use_span
    yield span
  File ".venv/lib/python3.13/site-packages/opentelemetry/sdk/trace/__init__.py", line 1105, in start_as_current_span
    yield span
  File ".venv/lib/python3.13/site-packages/temporalio/contrib/opentelemetry.py", line 295, in execute_activity
    return await super().execute_activity(input)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.13/site-packages/temporalio/worker/_interceptor.py", line 123, in execute_activity
    return await self.next.execute_activity(input)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".venv/lib/python3.13/site-packages/temporalio/worker/_activity.py", line 784, in execute_activity
    return await input.fn(*input.args)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "app/temporal/heartbeater.py", line 26, in wrapper
    return await fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "app/temporal/activities.py", line 160, in wrapper
    raise ApplicationError(
    ...<3 lines>...
    ) from e
temporalio.exceptions.ApplicationError: ClientError: ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'Unable to submit request because it must include at least one parts field, which describes the prompt input. Learn more: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini', 'status': 'INVALID_ARGUMENT'}}

[DouweM]

@ethanabrooks Can you show me the error you'd get without this? It's odd that Google could generate a model response with only tool calls and no text, that it then wouldn't accept when sent back on a subsequent request.

[DouweM]

Does this need text or could be empty?

[DouweM]

Could we do this inside _content_model_response based on the ModelResponse's parts rather than parsing the returned ContentDict? We already iterate over all items there, so we can keep track of what we've seen.

[ethanabrooks]

Updated the description to include the full error. Yes your suggestion sounds like a good idea.

[ethanabrooks]

Does this need text or could be empty?

Gave that a shot, but still got:

google.genai.errors.ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'Unable to submit request because it must have a text parameter. Add a text parameter and try again. Learn more: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini', 'status': 'INVALID_ARGUMENT'}}

Could do something more minimal like 'Function call complete'.

[ethanabrooks]

Could we do this inside _content_model_response based on the ModelResponse's parts rather than parsing the returned ContentDict? We already iterate over all items there, so we can keep track of what we've seen.

Just pushed a new commit that does this. Might need to iterate on the tests.

[DouweM]

This is only a Vertex AI issue right? Could we gate it by checking if self.system so that we don't send unnecessary tokens to the non-Vertex Gemini API?

Does it accept ' '? :) I wonder how strict their check is, and I'd rather not waste any tokens

[DouweM]

Unable to submit request because it must include at least one parts field, which describes the prompt input. Learn more: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini

@ethanabrooks The error is odd, because there is in fact a parts field with the function_call, right?

We already have similar logic for the user role message, but that's only for when parts is actually empty:

pydantic-ai/pydantic_ai_slim/pydantic_ai/models/google.py

Lines 455 to 457 in eeeb32d

	# Google GenAI requires at least one part in the message.
	if not message_parts:
	message_parts = [{'text': ''}]

This looks like a bug on Google's side, but I agree we should work around it.

[DouweM]

@ethanabrooks By the way nice to see you using Pydantic AI with Temporal! Is that already using our brand-new integration? I'd love feedback on Slack!

[ethanabrooks]

@ethanabrooks By the way nice to see you using Pydantic AI with Temporal! Is that already using our brand-new integration? I'd love feedback on Slack!

We just had a big planning meeting where talked about upgrading to v0.7 and using the new temporal features. Will keep you posted!

[ethanabrooks]

Might need some help with the test-coverage issue. It seems to be referencing lines that I didn't change... Also I get the test-coverage error on main as well.

[DouweM]

Might need some help with the test-coverage issue. It seems to be referencing lines that I didn't change... Also I get the test-coverage error on main as well.

How strange, line 584 doesn't even have a pragma: no cover :/ I can look into that once the code here is ready

[DouweM]

@ethanabrooks Just a heads-up that I'll be out this coming week and will be back the 25th. Assuming this is not urgent I'll review it then. If it is, please ping Kludex! Appreciate the patience :)

[ethanabrooks]

@ethanabrooks Just a heads-up that I'll be out this coming week and will be back the 25th. Assuming this is not urgent I'll review it then. If it is, please ping Kludex! Appreciate the patience :)

Not a problem. I'll check again next week.