Streamline x fmt and improve its output

[nnethercote]

• Removes the ability to pass paths to x fmt, because it's complicated and not useful, and adds --all.
• Improves x fmt output.
• Improves x fmt's internal code.

r? @GuillaumeGomez

[rustbot]

This PR modifies src/bootstrap/src/core/config.

If appropriate, please update CONFIG_CHANGE_HISTORY in src/bootstrap/src/utils/change_tracker.rs.

[GuillaumeGomez]

Small nit:

Suggested change

	if let Ok(cwd) = cwd {
	if let Ok(path2) = path.strip_prefix(cwd) {
	path = path2.to_path_buf();
	}
	}
	if let Ok(path2) = cwd.and_then(|cwd| path.strip_prefix(cwd)) {
	path = path2.to_path_buf();
	}

[nnethercote]

It doesn't work because the error types in the two Results don't match:

error[E0308]: mismatched types
   --> src/core/build_steps/format.rs:292:45
    |
292 |             if let Ok(path2) = cwd.and_then(|cwd| path.strip_prefix(cwd)) {
    |                                                   ^^^^^^^^^^^^^^^^^^^^^^ expected `Result<_, Error>`, found `Result<&Path, ...>`
    |
    = note: expected enum `Result<_, std::io::Error>`
               found enum `Result<&Path, StripPrefixError>`

[nnethercote]

I'm looking forward to let-chains reaching stable :)

[GuillaumeGomez]

Me too...

[GuillaumeGomez]

Just a small nit but otherwise looks great to me, thanks!

r=me once nit fixed.

[nnethercote]

@bors r=GuillaumeGomez

[bors]

📌 Commit 4dec0a0 has been approved by GuillaumeGomez

It is now in the queue for this repository.

[nnethercote]

@bors rollup

[ChrisDenton]

Removes the ability to pass paths to x fmt, because it's complicated and not useful

I'll just add that I found this useful and was surprised when I rebased and was met with a terse and slightly ambiguous error instead.

Passing paths was especially useful for generated code when you only want to automatically format the generated file. But I also fairly often used ./x fmt library/std to only format std and not the whole code base.

[nnethercote]

Can you give me more details about your use case, especially the generated code?

And with the x fmt library/std case, does running x fmt --all give a different outcome?

[ChrisDenton]

Specifically in my case std generates the Windows API bindings using src/tools/generate-windows-sys, which writes to library/std/src/sys/pal/windows/c/windows_sys.rs that then needs to be formatted according to std's formatting config.

And with the x fmt library/std case, does running x fmt --all give a different outcome?

That seems to work. Though that is noticeably slower instead of being near enough instant.

[nnethercote]

x fmt --all is slower, to be sure. On my machine x fmt with very few changes takes about 2.2 seconds, and x fmt --all takes about 3.3 seconds. So bootstrap stuff before formatting takes more time than the actual formatting. But I do have a fast machine, so your results may be different. How often do you generate the windows bindings?

[ChrisDenton]

Not that often. And to be clear, I'm not saying it's a big deal but I wouldn't quite characterise it as useless either.

[nnethercote]

Thanks for the info. The overall calculus was:

• Path support is rarely needed, --all subsumes it and is only marginally slower.
• Path support is surprisingly complicated. It was completely broken for a while until rustfmt fixes #125637 fixed it again, and a number of non-formatted files slipped into the repository in this time. The ignore library has some surprising behaviours when you use combinations of explicitly excluded and explicitly included paths.
• So overall, it's not worth keeping.

I have filed #126051 to improve the error message given when x fmt is given a path. I agree the current one is not clear.