Internal error with protobufs and tool use

[duccdev]

Description of the bug:

I was playing around with tool use and streaming, this was my code:

import google.generativeai as genai
import config

genai.configure(api_key=config.API_KEY)


def set_brightness(value: float) -> None:
    """Controls the brightness of all house lights. `value` is a `float` between 0 (off) and 1 (max)."""
    print("Brightness changed:", value)


model = genai.GenerativeModel(
    "gemini-1.5-flash-002",
    tools=[set_brightness],
)

chat = model.start_chat()
while True:
    stream = chat.send_message(input("User: "), stream=True)
    print("Model: ", end="", flush=True)
    for chunk in stream:
        print(chunk.text, end="", flush=True)
    print()

I got this peculiar error when I prompted it like this:

❯ python3 main.py
User: hey can you turn off the lights
Model: Traceback (most recent call last):
  File "/home/ducc/Projects/sketchbooks/apithing/main.py", line 29, in <module>
    print(chunk.text, end="", flush=True)
          ^^^^^^^^^^
  File "/home/ducc/Projects/sketchbooks/apithing/.venv/lib/python3.13/site-packages/google/generativeai/types/generation_types.py", line 536, in text
    part_type = protos.Part.pb(part).whichOneof("data")
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: whichOneof. Did you mean: 'WhichOneof'?

I am currently using Python 3.13, latest version of the SDK.
This is not an issue on my part as this seems to be a typo in the library.

Actual vs expected behavior:

Actual:

❯ python3 main.py
User: hey can you turn off the lights
Model: Traceback (most recent call last):
  File "/home/ducc/Projects/sketchbooks/apithing/main.py", line 29, in <module>
    print(chunk.text, end="", flush=True)
          ^^^^^^^^^^
  File "/home/ducc/Projects/sketchbooks/apithing/.venv/lib/python3.13/site-packages/google/generativeai/types/generation_types.py", line 536, in text
    part_type = protos.Part.pb(part).whichOneof("data")
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: whichOneof. Did you mean: 'WhichOneof'?

Expected (roughly):

❯ python3 main.py
User: hey can you turn off the lights
Brightness changed: 0.0
Model: Done!

Any other information you'd like to share?

Simply change the typo at types/generation_types.py, line 536, from .whichOneof to .WhichOneof.

[Gunand3043]

Hi @duccdev

You need to set enable_automatic_function_calling to True to make it work.

def set_brightness(value: float) -> None:
    """Controls the brightness of all house lights. `value` is a `float` between 0 (off) and 1 (max)."""
    print("Brightness changed:", value)


model = genai.GenerativeModel(
    "gemini-1.5-flash-002",
    tools=[set_brightness],
)

chat = model.start_chat(enable_automatic_function_calling=True)
while True:
    stream = chat.send_message(input("User: "))
    print("Model: ", end="", flush=True)
    print(stream.text, end="", flush=True)
    print()

[MarkDaoust]

Yes. Thanks for reporting!

[MarkDaoust]

Okay, the "WhichOneOf" error is fixed.

But this will still fail:

for chunk in stream:
        print(chunk.text, end="", flush=True)

When you start generating complex responses .text doesn't know what to do with new part types.
You need to iterate over the parts in the chunk, switch on their type, and handle FunctionCall objects yourself.