fix: pickle exceptions with custom __reduce__
#2747
Conversation
srilman
changed the title
__reduce__
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| @classmethod | ||
| def _reconstruct(cls, message: str, request: httpx.Request) -> Self: | ||
| return cls(message=message, request=request) | ||
|
|
||
| @override | ||
| def __reduce__(self) -> tuple[Callable[..., Self], tuple[Any, ...]]: | ||
| return (self.__class__._reconstruct, (self.message, self.request)) |
There was a problem hiding this comment.
The new APIConnectionError.__reduce__ returns _reconstruct(self.message, self.request), which reconstructs the object via cls(message=message, request=request). APITimeoutError subclasses APIConnectionError but its __init__ takes only a request parameter and hard‑codes the message. Attempting to pickle an APITimeoutError will therefore raise TypeError: APITimeoutError.__init__() got an unexpected keyword argument 'message', defeating the goal of making these exceptions picklable. APITimeoutError needs its own __reduce__/reconstructor or _reconstruct must handle subclasses without a message parameter.
Useful? React with 👍 / 👎.
Showmick119
left a comment
Showmick119
left a comment
There was a problem hiding this comment.
The APIConnectionError class passes an unsupported message arg, which will raise a TypeError. Kindly address this, before next steps.
2 participants
Changes being requested
Closes #2629. When working in a distributed environment, its useful to pickle and unpickle exceptions between workers. The default
__reduce__in Python does not account for keyword-only args, which is why we need to handle it ourselves.Alternatively, we could modify the constructor of the exception classes to not use keyword-only args, but that was a larger change.