fix(openai): Check response text is present to avoid AttributeError #5081
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #5081 +/- ##
==========================================
- Coverage 83.97% 83.92% -0.06%
==========================================
Files 179 179
Lines 17951 17951
Branches 3194 3194
==========================================
- Hits 15075 15065 -10
- Misses 1905 1913 +8
- Partials 971 973 +2
|
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Bug: Nulls Undermine Token Usage Calculation
The fix for null choice.message values is incomplete. While the list comprehension on lines 246-250 now filters out null messages, the _calculate_token_usage function still only checks hasattr(choice, "message") without verifying the message is not None. This causes count_tokens to be called with None, likely triggering an error when tiktoken_encoding.encode_ordinary() is invoked.
sentry_sdk/integrations/openai.py#L155-L159
sentry-python/sentry_sdk/integrations/openai.py
Lines 155 to 159 in a9b0d86
Description
According to the user in the issue gemini can return null values for chat completion text. Verify that that the messages are not null before calling
model_dump().See the corresponding schema: https://platform.openai.com/docs/api-reference/chat/object
Issues
Closes #5071
Reminders
tox -e linters.feat:,fix:,ref:,meta:)