minor issue: MNE naming convention enforcement

I would like to save my ICAs in a “BIDS-like” fashion, such as sub-002_task-bandit_ica.fif.gz, where “bandit” is the task … However, this triggers a warning from check_fname:

https://github.com/mne-tools/mne-python/blob/9865f9275f596443668a06e84e54f2e2c645bddb/mne/preprocessing/ica.py#L1385

While I see the benefit of using ica at the end of the file name, is it crucial to enforce -ica as the convention? I suggest to drop the - before ica.

[larsoner]

From chatting with @jasmainak, the BIDS naming convention uses <key1>-<value1>_<key2>-<value2>..._modality.extension. Underscore is used to separate key-value pairs, so _ica.fif and -ica.fif are both incompatible.

There will be a meeting in August to discuss derivative filenames like you’d use for ICA for example. My sense is that we should scrap the filename check altogether (no warn or logger.info), aside from looking for .fif or .fif.gz extensions for now. Otherwise we are pushing people toward a standard that will be obsolete soon.

The one place this is problematic is Report, which expects the current -ave.fif-style filenames. I think we can add to the Notes section of Report to say what it currently will parse properly, and update it once the BIDS standard is fully usable.

@jasmainak said he’s fine with this, @sappelhoff @cbrnr it sounds like from above/before you agree with eliminating the warning (though perhaps not putting it back eventually, but we can decide that later). Let’s see if @agramfort agrees.

[larsoner]

… but at least _ is incremental progress, so we can do that for now