perf: Delay formatting of lint messages until they are know to be used

[Marwes]

Found that (touch+) check builds in one of my crates spent 5% of the time just
formatting types for the sake of the BoxPointers lint, only for the
strings to get thrown away immediately since it is an allow lint.

This does the bare minimum to instead pass a fmt::Display instance to
the lint so that msg may only be created if it is needed.

Unsure if this is the desired way to solve this problem so I didn't go
through the lint system throughly to see if there were more places that
should take/pass format_args! instances instead of using format!.

[rust-highfive]

r? @eddyb

(rust_highfive has picked a reviewer for you, use r? to override)

[jonas-schievink]

@bors try

@rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 055ce62 with merge 228ed11...

[Centril]

Hmm... this does nothing though to prevent similar mistakes in the future.

Perhaps we should:

• Set the initial message to "" (so there should be no allocation).

• Invert control: Instead of returning a DiagnosticBuilder<'_>, we take in a function impl Fn(&mut DiagnosticBuilder<'_>) which allows the caller to make modifications, but ultimately this is never called on Level::Allow, thereby avoiding allocations that follow in e.g.

  rust/src/librustc_lint/builtin.rs

  Lines 190 to 200 in 055ce62

  	let binding = match binding_annot {
  	hir::BindingAnnotation::Unannotated => None,
  	hir::BindingAnnotation::Mutable => Some("mut"),
  	hir::BindingAnnotation::Ref => Some("ref"),
  	hir::BindingAnnotation::RefMut => Some("ref mut"),
  	};
  	let ident = if let Some(binding) = binding {
  	format!("{} {}", binding, ident)
  	} else {
  	ident.to_string()
  	};

[estebank]

I think the idea of changing the api surface to always defer evaluation of the message would be a relatively painless big win.

[Centril]

Filed #67927.

[bors]

☀️ Try build successful - checks-azure
Build commit: 228ed11 (228ed1118a30b6bfb1f725f66259681e8d2c2cf8)

[rust-timer]

Queued 228ed11 with parent 71bb0ff, future comparison URL.

[rust-timer]

Finished benchmarking try commit 228ed11, comparison URL.

[Centril]

@bors rollup=never

cc @oli-obk

[oli-obk]

cc @rust-lang/wg-diagnostics

r? @oli-obk

[Dylan-DPC-zz]

@Marwes can you rebase this? Thanks

[Marwes]

It looks like there is in progress work to further improve this ( #67927 ) so maybe this should just be closed?

[Dylan-DPC-zz]

Closing it then. Thanks :)

[estebank]

I feel like merging this PR is a reasonable stop gap measure, if you wish to put in the work to clean it up @Marwes.

[Marwes]

Looks like things have changed quite a bit (files deleted/moved) so I will leave this alone. Got other performance PRs that needs investigation anyways.