Changed casting rule in np.clip to allow reading of raw GDF files

[rasgaard]

When reading in raw GDF files I got the following error

UFuncTypeError: Cannot cast ufunc 'clip' output from dtype('float64') to dtype('uint32') with casting rule 'same_kind'

Referring to the usage of np.clip here:

mne-python/mne/io/edf/edf.py

Line 1446 in 87d6be9

np.clip(dur, 1, np.inf, out=dur)

This error can reproduced noting that np.inf has dtype of float64.

import numpy as np
dur = np.ones(10, dtype=np.uint32)
np.clip(dur, 1, np.inf, out=dur)
>>> UFuncTypeError: Cannot cast ufunc 'clip' output from dtype('float64') to dtype('uint32') with casting rule 'same_kind'

Changing aforementioned line 1446 to np.clip(dur, 1, np.inf, out=dur, casting="unsafe") seemingly fixes this issue.

Additionally, there is a typo when using the read_raw_gdf-function on files that does not have the gdf-file extenion

mne-python/mne/io/edf/edf.py

Line 1881 in 87d6be9

raise NotImplementedError(f"Only BDF files are supported, got {ext}.")

[welcome]

Hello! 👋 Thanks for opening your first pull request here! ❤️ We will try to get back to you soon. 🚴

[larsoner]

Can you also add an entry to doc/changes/devel.rst using the :newcontrib: role and also a name-url pair to doc/changes/names.inc?

[larsoner]

I'm not really sure what purpose the upper bound is meant to serve, but to me this is a cleaner / more explicit fix

                np.clip(dur, 1, np.iinfo(dur.dtype).max, out=dur, casting="unsafe")

But really there is no point in checking the max here since it will always be satisfied. So really we should probably do:

Suggested change

	np.clip(dur, 1, np.inf, out=dur, casting="unsafe")
	np.maximum(dur, 1, out=dur)

[larsoner]

... and actually this np.maximum / np.clip step can live in the first branch of the conditional above if you want. There is no point in running this unless etmode == 3 because in the else we construct it to satisfy the maximum check with dur = np.ones(n_events, dtype=np.uint32).

[rasgaard]

I noticed that

mne-python/mne/io/edf/edf.py

Line 1194 in 87d6be9

np.maximum(dur, 1, out=dur)

uses np.maximum in the exact way you deduced your way through. I changed my commit to mimic that instead :)

[larsoner]

Marking for merge-when-green, thanks in advance @rasgaard !

[welcome]

🎉 Congrats on merging your first pull request! 🥳 Looking forward to seeing more from you in the future! 💪