Drop python EOLed 3.7 support (in setup.py and CI)

[adswa]

Python 3.7 EOL'ed mid-June. This PR

• bumps all python 3.7 CI environments to Python 3.8
• bumps all other CI environments' Python versions by one minor version

I think this needs to be done also for master, but for now I make only one PR to see what fails.

[codecov[bot]]

Codecov Report

All modified lines are covered by tests :white_check_mark:

Comparison is base (f989245) 91.56% compared to head (058e3b8) 91.29%.

Additional details and impacted files

@@            Coverage Diff             @@
##            maint    #7449      +/-   ##
==========================================
- Coverage   91.56%   91.29%   -0.27%     
==========================================
  Files         325      325              
  Lines       43444    43444              
  Branches     5827     5827              
==========================================
- Hits        39779    39662     -117     
- Misses       3650     3767     +117     
  Partials       15       15              

see 50 files with indirect coverage changes

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[adswa]

This PR still needs an update for Travis. I have two questions there:

• I vaguely recall issues? in another datalad-extension? that talked about removing Travis CI. Are there plans to stop using Travis for this repo?
• If we keep Travis CI, it has many Python3.7 builds to update. What I don't parse just now, though, is the following extra-arg passed to the datalad-installer: _DL_ANNEX_INSTALL_SCENARIO="miniconda=py37_23.1.0-1 --python-match minor --batch git-annex=8.20200309 -m conda". What would this change to in order to bump it to Py3.8?

[adswa]

There is one travis build that fails like this:

$ sudo pip install --user .

WARNING: The directory '/home/travis/.cache/pip' or its parent directory is not owned or is not writable by the current user. The cache has been disabled. Check the permissions and owner of that directory. If executing pip with sudo, you should use sudo's -H flag.

ERROR: Can not perform a '--user' install. User site-packages are not visible in this virtualenv.

The command "sudo pip install --user ." exited with 1.

I'm not sure whether its related to the changes here or spurious, but I can't re-trigger the build.

[yarikoptic]

• If we keep Travis CI, it has many Python3.7 builds to update. What I don't parse just now, though, is the following extra-arg passed to the datalad-installer: _DL_ANNEX_INSTALL_SCENARIO="miniconda=py37_23.1.0-1 --python-match minor --batch git-annex=8.20200309 -m conda". What would this change to in order to bump it to Py3.8?

FTR this was resolved by using py38_23.3.1-0.

[adswa]

The one travis run still fails, and I don't get why. The failure happens in the script part of the config file which is identical across runs, so maybe its something particular with ubuntu bionic and Python 3.10 ...

[yarikoptic]

well, erroring starts early with

$ sudo pip install --user .
WARNING: The directory '/home/travis/.cache/pip' or its parent directory is not owned or is not writable by the current user. The cache has been disabled. Check the permissions and owner of that directory. If executing pip with sudo, you should use sudo's -H flag.
ERROR: Can not perform a '--user' install. User site-packages are not visible in this virtualenv.
The command "sudo pip install --user ." exited with 1.

which I checked happened only in that one indeed, no other matrix runs... let's try it with focal

[adswa]

With the bump of the misbehaving travis run to Python 3.11 a very old requested git-annex conflicts with other software versions :/

[yarikoptic]

dang... I still don't know how/why it gets installed in general: https://github.com/conda-forge/datalad-feedstock/issues/109 .

[yarikoptic]

note: restarted 1 travis job which hanged in conda

[adswa]

This PR has been sleeping for a while, mostly because I couldn't figure out the setup-related travis failure described in the comments. But #7492 is planning to remove Travis entirely, so I feel like this could be the opportunity to also move forward with dropping Py3.7 support in this PR.

[adswa]

The linting has been fixed elsewhere already, and the macos and datalad-containers failures are unrelated. The travis build is almost through the board unhappy, but my suggestion would be to proceed with #7492, then this PR, and then, eventually, #7450

[yarikoptic]

@adswa would you have time to get back to this one -- we should really drop that good old 3.7 now ;)