I have to work a bit with EEGLAB, and I'm not very familiar with it (yet) but I already noticed a couple of improvements that could be made to the I/O in my opinion.

Reader

Channel type is hard-coded to 'eeg' here while EEGLAB seems to support a larger variety of channel types.
From the documentation:

‘EEG’, ‘MEG’, ‘EMG’, ‘ECG’, ‘Events’, etc.

Proposition: Try to retrieve the channel types saved by EEGLAB and convert it to one of MNE channel types (e.g. 'EEG' -> 'eeg', 'Events' -> 'stim').

Writer / Export

Looking at export_set in eeglabio, it has an argument ref_channels that is not used by the export code. If I'm not mistaken, the referencing information is present in Info, so it could be passed to the .set file via this argument.

eeglabio.raw.export_set(
        fname, data=raw.get_data(picks=ch_names), sfreq=raw.info['sfreq'],
        ch_names=ch_names, ch_locs=cart_coords, annotations=annotations)

Support for channel type is not included in eeglabio, I would like to add it jackz314/eeglabio#4
Once added to eeglabio, the exporter could make use of this additional argument.

TODO:

• Add support for MNE channel types in the reader (MRG: Add support to retrieve channel type from ['chanlocs']['type'] with EEGLAB reader #9990)
• Add a warning when exporting with unapplied projectors + warning in documentation (MRG: Issue warning when exporting with unapplied projectors and improve docstring #9994)
• Add support to export MNE channel types (depends on jackz314/eeglabio#4)