[Assistants API] Run Step Tool call element changes type depending on run status

[jordan-barrett-jm]

Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

When polling run steps to read tool calls, I've observed that when the run is not in status 'completed' the function object in the tool_calls array is of type dict, however, once the run is completed it becomes a regular object. This has implications for how the tool call attributes can be accessed, leading to redundant code checking the element type. I have specifically observed this with function tool calls. I haven't done any work with the other types of Assistant tools.

To Reproduce

1. Initiate a run that requires the use of a function
2. Retrieve the latest run step associated with the run
3. Attempt to read the tool calls when the run is of status completed versus when it's in an in_progress or requires_action state.

Code snippets

No response

OS

Windows 11

Python version

Python v3.9,3

Library version

openai v1.3.5

[rattrayalex]

cc @athyuttamre do you know why this might be happening? (my guess is the object shape changes in ways that aren't documented properly in the openapi spec)

[athyuttamre]

Thanks for the report! I think it's because of a bug in our OpenAPI spec, attempting to fix here: openai/openai-openapi#187.