Fix pip installs

[orlnub123]

When readthedocs pip installs a previously installed core component the previous version doesn't get fully uninstalled. This is caused by the --ignore-installed flag and the results can be disastrous: https://stackoverflow.com/a/51916623

[stsewd]

Makes sense for me, not sure if there was a reason for this in the past.

[humitos]

Thanks!

Seems a safer and robust way to handle this.

[vidartf]

This commit broke my build hard. I'm relying on some dependencies for my build that are only available via conda. The --force-reinstall basically makes avoiding pip >= 10 messages like these for deep dependencies unavoidable:

Cannot uninstall 'certifi'. It is a distutils installed project and thus we cannot accurately determine which files belong to it which would lead to only a partial uninstall.

Any workarounds or other hints would be very welcome!

Example failure: https://readthedocs.org/projects/ipyscales/builds/8821876/

[vidartf]

Xref after diggin through blame history: 433f198

[stsewd]

hmm, what about just removing the flag? What would be the default behaviour of pip? Is this only happening with conda?

[orlnub123]

I'm guessing the reason the --ignore-installed flag was added was to mimic what setup.py install --force does. The --force flag was added in 1617d96 but has no explanation as to why. All I could find about why you might want to use it is in this question. According to this post distutils/setuptools compares the file modification times to see if it should re-build the package but pip doesn't seem to have the same behavior and re-builds it regardless.

The only place the --ignore-installed and --force-reinstall flags get used is here (CTRL-F ignore_installed and force_reinstall). Outright removing it would probably cause some packages to get stale. Replacing it with --upgrade would be more sensible and shouldn't cause any problems.

[humitos]

Thanks @orlnub123 for your explanation and help here! I suggest you to open a new issue because I feel that this conversation will be lost here in a merged pull request.

I supposed that the change to --force-reinstall won't bring any issue, but it seems that we didn't consider this case with conda, unfortunately.

At this point, I'm not sure if we can have a "general pip command that works for all the cases" :(

[humitos]

Oh, I mixed usernames! The one with the problem is @vidartf :/

(so, please @vidartf, open a new issue :) )

[vidartf]

Thanks for the context! Using it, I've opened #5545.