fix(litellm): fix gen_ai.request.messages to be as expected
#5255
Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
![sentry[bot]](https://avatars.githubusercontent.com/u/1396951?s=60&v=4){kind=small}
…inal messages and handle data URLs correctly
| messages = copy.deepcopy(messages) | ||
|
|
||
| def _map_item(item: "Dict[str, Any]") -> "Dict[str, Any]": | ||
| if item.get("type") == "image_url": |
There was a problem hiding this comment.
Missing isinstance check causes crash on non-dict items
Low Severity
The _map_item function calls item.get("type") without first verifying that item is a dict. If the content list contains non-dict elements (e.g., strings or None), this raises an AttributeError. The existing redact_blob_message_parts function in sentry_sdk/ai/utils.py follows a defensive pattern with isinstance(item, dict) and item.get("type"), but the new code doesn't match this pattern. While LiteLLM typically follows OpenAI's format where list items are dicts, malformed or unexpected input from any of the 100+ supported providers could crash the callback.
| return item | ||
|
|
||
| for message in messages: | ||
| content = message.get("content") |
There was a problem hiding this comment.
Missing isinstance check for messages causes potential crash
Low Severity
The loop iterates over messages and calls message.get("content") without checking if message is a dict. If the messages list contains a non-dict element, this raises an AttributeError. The existing similar functions normalize_message_roles and redact_blob_message_parts both include if not isinstance(message, dict): continue checks before calling .get() on messages. The new code doesn't follow this established pattern.
Issues
Closes https://linear.app/getsentry/issue/TET-1635/redact-images-litellm