Revert #7969 and fix VCS stdout/stderr capture

[sbidoul]

Fixes #8876
Supersedes #9234 and #9327
Reverts #7969
Better fix for #7545 and #7968
Reopens #7841 (which pip can't fix as it is git output) and #7711 (which is not entirely trivial and can be handled in another PR)

[sbidoul]

Wondering why these tests fail so early. They work on my machine™️.

[uranusjr]

Why not

if proc.stdin:
    proc.stdin.close()

[sbidoul]

@uranusjr I did something a little bit different in the last version. I don't quite understand why we need to pass a PIPE to stdin while we don't use it, but I don't dare to change it, and I decided to keep passing one in both modes.

[uranusjr]

I think the intent is to disallow all stdin from subprocesses. Without it, the subprocess’s stdin would be inherited from pip’s, i.e. subprocesses are able to show prompts etc.

[pradyunsg]

More fun is that it waits in CI for prompts, instead dof outright failing as well.

[Mikuana]

Confirmed this solution appears to fix #8876

[sbidoul]

@pradyunsg how do you want to proceed here ? The python 2 cleanup is creating conflicts but this PR has to go in 20.3.4.

BTW wasn't there a bot warning about merge conflicts ?

[pradyunsg]

I broke the bot, and then my laptop so, now, I can't fix the bot. :)

[pradyunsg]

@sbidoul If you (1) rebase this on top of 20.3.3, (2) create a merge commit against the current master branch, fixing conflicts and (3) force-push that.

Then, it'll be possible to have an "easy backport" of this PR -- since it's based off 20.3.3 and hence won't be difficult to cherry-pick for release, and would also be a part of 21.0, since, well, it'll be merged into master. :)

[sbidoul]

@pradyunsg ok, I did that and created the merge commit by merging the master branch into my branch.

[sbidoul]

I assign this to 21.0, as the backport to 20.3.4 is tracked via #8876