Explicit support and tests for hg+file scheme for pip install.

[atse]

Reworking #5935 to be more complete for #4358

The hg+file scheme was already functioning implicitly (with test coverage) but inconsistently documented. This change is to make the scheme support explicit, including tests and docs.

[pradyunsg]

The linting jobs don't seem to be happy with these changes. :)

[cjerdonek]

The hg+file scheme was already functioning implicitly (with test coverage)

How could hg+file have been working if it wasn't listed explicitly in the supported schemes? That seems like something important to understand and track down first, as it could signal a bug or reveal other things..

[chrahunt]

I only see us using these schemes in a handful of places and it looks like, for editable installs, we split base on the + and then leave everything else to the backend (in InstallRequirement.update_editable).

To confirm, I ran pip under hunter like

PYTHONHASHSEED=0 PYTHONHUNTER="stdlib=False, action=CallPrinter(stream=open(f'$PWD/hunter-old-{os.getpid()}.log', 'w'))" pip install -e hg+file://$PWD/hg_repo#egg=foo

before and after the change in this PR and can see virtually no difference in behavior between the two runs.

[chrahunt]

I would save this into a variable and use it in the construction of args and in the assert. It may also make more sense to move the assert above the args declaration. Normally we assert as close as possible to the place where something could've failed.

[chrahunt]

It looks like these where alphabetized before.

[cjerdonek]

I still don't see an answer to the question I asked before. If this was already working before (without hg+file in the schemes attribute), why do we need to add it to the list of schemes -- what behavior changes by doing that? If there is no change in behavior, it seems like it shouldn't be added because it gives a false impression that it's doing something. And if there is a change, it seems like that's one of the things that should be tested (e.g. to prevent a regression).

[chrahunt]

This should be added so that Link.is_vcs returns the expected value on hg+file URLs. I was about to use it as part of some refactoring and was reminded of this PR. We should add a test covering that case here.

It currently works because we are pretty flexible when it comes to input URL parsing. In several places we just check if "+" is in the scheme to determine if something is a VCS URL, or split on + to get the VCS name and non-VCS scheme.

[chrahunt]

Thanks again for your work here @atse! I will close this since it has been superseded by #7065.