Capturing warnings with the correct source location

[ibro45]

https://loguru.readthedocs.io/en/stable/resources/recipes.html#capturing-standard-stdout-stderr-and-warnings

When logging the path from which the warning message occurred, it will show the exact location of where you patched showwarning instead of where the warning message was actually called from.

For example, the first and last warnings are coming from PyTorch Lightning, but here it says that they come from my logging.py module, where I did what's described in the link above.

Is there any viable solution to this? Thanks!

[Delgan]

You can certainly use the depth argument of the .opt() method:

def showwarning(message, *args, **kwargs):
    logger.opt(depth=2).warning(message)
    showwarning_(message, *args, **kwargs)

If you can confirm that works, I will update the documentation.

[ibro45]

Yes, that did it, thanks for such a quick reply 😁

[Delgan]

Great, you're welcome. 👍

[ibro45]

Hi, sorry to bother you again. Have you ever passed the warnings.warn's stacklevel argument to loguru as depth? I don't see a clean way of doing it currently.

For example: https://github.com/Lightning-AI/utilities/blob/275edd4fdd7612faa9be2c7cf962780cf618d082/src/lightning_utilities/core/rank_zero.py#L74

Unfortunately seems like warnings.warn() doesn't pass stacklevel to warnings.showwarning https://github.com/python/cpython/blob/01dceba13e872e9ca24b8e00a2b75db3d0d6c1a3/Lib/warnings.py#L303

[Delgan]

Hum, that's a pitty... I don't think there's any clean way of doing it, unfortunately.

Technically, you could iterate the stack to retrieve the depth where warns.warn() was emitted. More realistically, I would just advice to add the filename and lineno to your logged message:

def showwarning(message, category, filename, lineno, file=None, line=None):
    logger.opt(depth=2).warning("Warning emitted from '{}:{}': {}", message, filename, lineno)
    showwarning_(message, category, filename, lineno, file, line)