RFC: installation process instructions

Based on some offline discussions with users, it may be time for some refinements to the install instructions. Specifically, we should make it clear that:

1. if they want to update mne-python, it’s usually best to just pip install -U mne
2. if they want to update everything in their environment they should create a new environment instead of using the latest environment.yml to update an existing environment
3. (optional) given item 2, perhaps a “pro tip” would be to keep a list of other packages one installs into one’s mne environment, to simplify the creation of new mne envs
4. consider revising our environment.yml to not include an env name, and update the recommended install command to conda create --name mne --file environment.yml instead. The rationale here is that the current recommended command conda env update --file environment.yml implies that it can be used for updating

cc @NeuroLaunch @larsoner was there anything else from our discussion that I’ve forgotten to include here?

[hoechenberger]

It would be wonderful if something like conda create --name my_env_name "mne-base=0.23" would “just work” and people never needed environment files.

But that’s already working … If you do conda create -n mne -c conda-forge mne, you should end up with a (not minimalist, but still) working MNE environment with 3D support.

[drammock]

@drammock No strong feeling here. I’m aiming for this to be tested & ready when we release 0.24 – for which I think we should start planning soon, the changelog has been growing quite a bit again since we released 0.23 😃

last release was end of April, which means next release should be end of october. Last year there was an “extra” december release, but personally I don’t want us to make a habit of early or extra releases (it shortens deprecation times, for one thing).

I really don’t think storage space is a good argument in 2021

Fair point, although not everybody has new-ish computers and works with large-or-huge datasets. I am also thinking more about bandwidth / download speed, perhaps moreso than actual space-on-disk.

My reconciliation proposal is this: I’ll try to finally get that mne-base conda-forge package released (i.e., MNE-Python with a minimal number of dependencies) and then build “minimal” installers based on it – “Mini-MNE”, if you will. WDYT?

It would be wonderful if something like conda create --name my_env_name "mne-base=0.23" would “just work” and people never needed environment files.