MRG, MAINT: Support only 3.6+

[larsoner]

Closes #7635

[drammock]

Questions:

1. Do you think it's worth it to start using abstract base class Collection in type checking?
2. Any parts of the codebase that would benefit from namedtuple?

[cbrnr]

👍 for type-checking against an ABC.

[larsoner]

Do you think it's worth it to start using abstract base class Collection in type checking?

Maybe but it will probably need more testing, so I'd rather let someone else tackle it in another PR

[agramfort]

would someone with python 2.5 still be able to use mne 0.21 and have silent bugs?

[larsoner]

would someone with python 2.5 still be able to use mne 0.21 and have silent bugs?

No because:

1. imports like from collections.abc will fail, and
2. we won't even allow them to install with our new python_requires line in setup.py

[agramfort]

all good then

…

[larsoner]

Any parts of the codebase that would benefit from namedtuple?

In principle anywhere we return a tuple could probably benefit from it, I guess

[codecov]

Codecov Report

Merging #7638 into master will decrease coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #7638      +/-   ##
==========================================
- Coverage   90.14%   90.08%   -0.07%     
==========================================
  Files         454      454              
  Lines       83929    83583     -346     
  Branches    13324    13230      -94     
==========================================
- Hits        75661    75298     -363     
- Misses       5406     5423      +17     
  Partials     2862     2862              

[larsoner]

Rebased, okay to merge once green @agramfort ?

[agramfort]

+1 for MRG when CIs are happy

[sappelhoff]

@larsoner why this change?

[larsoner]

Instead of being alphabetical it now should be the order on disk. I figured it shouldn't affect being able to actually use it with our montage functions...