Rewrite testing code that makes use of `ernoise` FIF file

[hoechenberger]

EDIT

please see: #1406 (comment)

The affected code must be re-written to not save data into mne datasets.

---

I'm on main, I deleted my entire ~/mne_data directory and downloaded the latest testing dataset. I want to run all tests in test_write.py.

All of them complete successfully, except for test_anonymize_dataset():

Running pytest with args: ['-p', 'vscode_pytest', '--rootdir=/Users/richardhochenberger/Development/mne-bids', '/Users/richardhochenberger/Development/mne-bids/mne_bids/tests/test_write.py::test_anonymize_dataset']
============================= test session starts ==============================
platform darwin -- Python 3.12.10, pytest-8.3.5, pluggy-1.6.0
rootdir: /Users/richardhochenberger/Development/mne-bids
configfile: pyproject.toml
plugins: anyio-4.9.0, sugar-1.0.0, cov-6.1.1
collected 1 item

mne_bids/tests/test_write.py F                                           [100%]

=================================== FAILURES ===================================
____________________________ test_anonymize_dataset ____________________________
mne_bids/tests/test_write.py:3795: in test_anonymize_dataset
    raw_er = _read_raw_fif(raw_er_path, verbose=False)
mne_bids/tests/test_write.py:105: in fn
    raw = read_raw(fname, *args, **kwargs)
../mne-python/mne/io/fiff/raw.py:543: in read_raw_fif
    return Raw(
<decorator-gen-273>:10: in __init__
    ???
../mne-python/mne/io/fiff/raw.py:103: in __init__
    raw, next_fname, buffer_size_sec = self._read_raw_file(
<decorator-gen-274>:12: in _read_raw_file
    ???
../mne-python/mne/io/fiff/raw.py:198: in _read_raw_file
    fname = _check_fname(fname, "read", True, "fname")
<decorator-gen-0>:12: in _check_fname
    ???
../mne-python/mne/utils/check.py:302: in _check_fname
    raise FileNotFoundError(f'{name} does not exist: "{fname}"')
E   FileNotFoundError: fname does not exist: "/Users/richardhochenberger/mne_data/MNE-testing-data/MEG/sample/ernoise_raw.fif"
- generated xml file: /Users/richardhochenberger/Development/mne-bids/junit-results.xml -
============================= slowest 20 durations =============================
0.04s call     mne_bids/tests/test_write.py::test_anonymize_dataset

(2 durations < 0.005s hidden.  Use -vv to show these durations.)
=========================== short test summary info ============================
FAILED mne_bids/tests/test_write.py::test_anonymize_dataset - FileNotFoundErr...
============================== 1 failed in 0.60s ===============================
Finished running tests!

And indeed, MNE-testing-data/MEG/sample/ernoise_raw.fif doesn't exist because it's not even supposed to be included in the dataset, at least I cannot find it in the repo:

https://github.com/mne-tools/mne-testing-data/tree/master/MEG/sample

So … what gives?

[sappelhoff]

Interesting, do you have a hypothesis why the tests are not failing on the CI? I pushed a commit to see how the CI behaves without the cache: #1406

Also, for some reason I do have this file 🤣

[hoechenberger]

Do you have the latest version of the dataset? maybe @larsoner knows more

the file does exist in the sample dataset

[sappelhoff]

Do you have the latest version of the dataset?

good point, I did not check. That was on my personal PC.

On my work laptop I am one minor version behind the latest version (0.161 rather than current 0.162), and here, I do not have the file:

[sappelhoff]

also interesting:

mne-bids/mne_bids/tests/test_path.py

Lines 1266 to 1273 in 4f12d19

	# Now create data resembling an empty-room recording.
	# The testing data has no "noise" recording, so save the actual data
	# as named as if it were noise. We first need to write the FIFF file
	# before reading it back in.
	tmp_dir = tmp_path / "tmp"
	tmp_dir.mkdir()
	er_raw_fname = op.join(tmp_dir, "ernoise_raw.fif")
	raw.copy().crop(0, 10).save(er_raw_fname, overwrite=True)

[sappelhoff]

I have edited the title and OP to reflect what we found out. Thanks for alerting us, Richard.

[hoechenberger]

@sappelhoff Should we advocate for an upstream inclusion of empty-room data in the testing dataset?

[sappelhoff]

I think that would make sense! @larsoner what do you think?

[larsoner]

We already have this would it work?

https://github.com/mne-tools/mne-testing-data/blob/master/SSS/test_move_anon_erm_raw.fif

[sappelhoff]

We already have this would it work?

given that this is probably a "real" empty room recording that would probably be an improvement, but still not ideal. We have a lot of examples with the "sample" dataset, and are referring to the "fake" empty room recording in several places. It would likely seem a bit odd and would disrupt the reading flow to be like "we don't have an empty room for this, so we take this completely unrelated data as a replacement". WDYT? 🤔

To see what I mean, you can do a global search of ernoise_raw.fif in the mne-bids repo and have a quick look at the few spots it shows up.

[larsoner]

It would likely seem a bit odd and would disrupt the reading flow to be like "we don't have an empty room for this, so we take this completely unrelated data as a replacement". WDYT? 🤔

In what sense? If needed you could hide this detail in a pytest fixture that even read the original data, renamed the channels to the old format "MEG XYZW" (with the space), dropped channels, etc. then wrote it to a tempdir. As long as the person authoring the test doesn't care that it lives in a different folder then it should be fine...

I don't know if there are empty room data for the sample dataset actually. Would require some digging to figure out. So to me the pytest fixture (or just point to the test_* file directly and live with channel names being different if okay) is the path of least resistance here