Use std::process::Command throughout compile-test

[euclio]

Fixes#43762.

r? @alexcrichton

Fixed up what didn't require a major refactor. I'd like to remove ProcArgs, but there's no way to get the current program name and args from a Command builder.

• What do you think about compose_and_run? I wonder if we could simplify that.
• Still a little unhappy with the number of expects in the output. Maybe it's worth threading the errors up through.

[alexcrichton]

I wonder if maybe we could remove these arguments as well? It looks like not all tests need to pass this so we could have a helper which works with these variables and otherwise these aren't passed and the helper isn't called when not needed.

[alexcrichton]

One thing that may be worth testing is to take &mut Command here instead of Command. I'm not 100% sure which is more ergonomic, though!

[alexcrichton]

Great to see all these FIXME annotations removed!

[alexcrichton]

This looks great to me, thanks!

We can always delegate future refactorings to later PRs, this seems great enough to land in the meantime.

[euclio]

Great. Happy to address the comments in the future.

[alexcrichton]

Oh so to be clear, @euclio mind trying to remove the aux_path argument from the compose_and_run function before landing? Working with Command vs &mut Command doesn't necessarily have to block this.

[euclio]

Ah, sorry for the misunderstanding. I'll see if I can find the time this weekend.

[euclio]

Which invocations actually need the library path munging? Just rustc?

[alexcrichton]

Oh I figured we'd just preserve the existing behavior which is that wherever Some is passed we continue to munge the search path, I don't think it's just rustc but also the test binaries themselves.

[bors]

☔ The latest upstream changes (presumably #43710) made this pull request unmergeable. Please resolve the merge conflicts.

[euclio]

@alexcrichton Ok, do the commands ever need just "" added to their search path? I was a little wary of doing this because I wasn't sure how to tell which invocations required an empty path in the library path.

[euclio]

In other words I'm wondering if there are any invocations that have a None aux path and still need their path munged to include an empty path.

[alexcrichton]

@euclio did you mean to close this? I don't think anything needs "" added to the search path, but that's just my hope sort of!

If this takes awhile to do though I don't want to block this! I'm fine landing this ahead of time if you'd like to take some more time to game out that change.

[euclio]

Ah, whoops! Reopened.

So, I think there's still a little question on how to handle the make_cmdline function if we remove the libpath argument, and I'd like to take the time to experiment with it. I'd rather land this now and fix it up later since this rebase is no fun.

[euclio]

Ugh, looks like my rebase borked the branch. Will reopen shortly.

[alexcrichton]

Ok sounds good to me! Feel free to cc me on the new PR and I'll r+