MLFlowLogger fails to log artifact on Windows

[niander]

Bug description

Error when training with "MLFlowLogger" and with log_models="all", running on Windows:

mlflow.exceptions.MlflowException: Invalid artifact path: 'epoch=0-step=43654'. Names may be treated as files in certain cases, and must not resolve to other names when treated as such. This name would resolve to 'epoch=0-step=43654'.

I was able to find the reason for this error:

MLFlowLogger is calling MLflowClient.log_artifact(...) internally -- inside method MLFlowLogger._scan_and_log_checkpoints(...) -- passing the artifact path as a pathlib.Path object. However, MLFlow expects paths in the POSIX format, whereas pathlib.Path will use the current filesystem format.

Now, the actual reason for the error is due to an internal check that MLFlow does. It tries to verify that the path is already "normalized", that is, doesn't contain "." or "..". The way it does that is by calling posixpath.normpath(...) and checking if the output is the same as the original path. Because the original path is a "pathlib.Path" object, both paths do not match. I am assuming that if the platform was different from Windows, the comparison would match because the internal string representation of pathlib.Path would be the same.

What version are you seeing the problem on?

v2.5

How to reproduce the bug

Error messages and logs

Environment

Current environment

#- PyTorch Lightning Version (e.g., 2.5.0): 2.5.1
#- PyTorch Version (e.g., 2.5): 2.6.0
#- Python version (e.g., 3.12): 3.12.9
#- OS (e.g., Linux): Windows
#- CUDA/cuDNN version: 12.6
#- GPU models and configuration:
#- How you installed Lightning(`conda`, `pip`, source): pip

More info

No response

[Alexandre-SCHOEPP]

To add more information, in mlflow.utils.validation, I added a print statement:

def path_not_unique(name):
    norm = posixpath.normpath(name)
    print(
        f"Examining {name}:\n\t{(norm != name)=}\n\t{norm == '.'=}\n\t{norm.startswith('..')=}\n\t{norm.startswith('/')=}\n\t{(norm, name)=}"
    )
    return norm != name or norm == "." or norm.startswith("..") or norm.startswith("/")

And got:

Examining <CENSORED>
        (norm != name)=True
        norm == '.'=False
        norm.startswith('..')=False
        norm.startswith('/')=False
       (norm, name)=('<CENSORED>', WindowsPath('<CENSORED>'))

This issue is breaking my software. Please fix ASAP