Effectively disable the wheel cache if not writable

[sbidoul]

Move the test for cache writeability from the install command to the WheelCache object.

The test was ineffective in the install command because it was done after creating the wheel cache object. The test was also missing in the wheel and freeze commands.

The test is in WheelCache and not Cache to avoid a double warning via Cache and SimpleWheelCache.

I also considered putting the check in the option parser, but at that point the logger is not fully initialized and warning was not colored.

Fixes #7488

[pradyunsg]

Windows CI isn't happy - possibly since the test/code isn't working there as expected.

[sbidoul]

I also considered putting the check in the option parser, but at that point the logger is not fully initialized and warning was not colored.

OTOH, doing it in main() after the logger is setup looks like a better place. Disabling the cache early (as in the option value) avoids code duplication, and avoids pip warning more than once about the cache not being writable.

[sbidoul]

Windows CI isn't happy - possibly since the test/code isn't working there as expected.

Yes, apparently os.chmod is not doing what I expected on Windows. I'll need to figure out a way to create a non writable directory on Windows.

[sbidoul]

I squashed the history a little bit.

[chrahunt]

Regarding the test on Windows, we have a test helper make_unreadable_file that should work or a similar version created for "read only".

[sbidoul]

According to this comment

pip/tests/unit/test_networking_cache.py

Lines 19 to 21 in d613433

	The no_perms test are useless on Windows since SafeFileCache uses
	pip._internal.utils.filesystem.check_path_owner which is based on
	os.geteuid which is absent on Windows.

it looks like check_path_owner does not work on Windows so it would mean this whole mechanism to check for cache ownership has no effect on windows?

[sbidoul]

I think this one is good to go, because the cache permission check was never implemented on Windows, so it's ok to have the new test disabled on Windows.

[chrahunt]

Just one comment.

In a followup it may be good to separate and rename check_path_owner to actually only check whether the path is owned by the current user, and then we could use test_writable_dir to check whether it is writable like if not owned_by_current_user(...) or not is_writable_dir(...).

[pradyunsg]

Merging since there's multiple approving reviews. :)