Enforce alphabetical sorting with tidy

[Noratrieb]

We have many places where things are supposed to be sorted alphabetically. For the smaller and more recent size assertions, this is mostly upheld, but in other more... alive places it's very messy.

This introduces a new tidy directive to check that a section of code is sorted alphabetically and fixes all places where sorting has gone wrong.

[rustbot]

This PR changes MIR

cc @oli-obk, @RalfJung, @JakobDegen, @davidtwco, @celinval, @vakaras

Some changes occurred to the CTFE / Miri engine

cc @rust-lang/miri

[rust-highfive]

r? @jackh726

(rust-highfive has picked a reviewer for you, use r? to override)

[jackh726]

This is nice. My only thought here is that, because it's a very manual annotation, it will be forgotten. Is there any way we could maybe even get rid of the "end" annotation?

[saethlin]

because it's a very manual annotation, it will be forgotten

I agree, but seeing how this has tidied up some sorting errors already, it's hard to argue this isn't a step forward.

Is there any way we could maybe even get rid of the "end" annotation?

Don't know if we want to get this advanced, but there are places in ui tests where we use regular expressions. The begin comment could contain a regular expression, then sort all nonempty lines that match that, perhaps sorting by a capture group. The downside would be that any change which causes one line to not match (like a comment in the middle of an argument list) would cause the elements below to not be sorted. So maybe we want a way to diagnose that... But now we're talking about tooling for the tools. Maybe a simple end delimiter is better.

[jackh726]

I mean, we could be a bit stricter and put all the items that we want to be in alphabetical order in their own block, then annotate that block.

[faptc]

What about making rustfmt works with feature attributes?

[davidtwco]

Could you add this annotatation here as well?

rust/compiler/rustc_error_messages/src/lib.rs

Lines 37 to 66 in 4bd3078

	fluent_messages! {
	ast_lowering => "../locales/en-US/ast_lowering.ftl",
	ast_passes => "../locales/en-US/ast_passes.ftl",
	attr => "../locales/en-US/attr.ftl",
	borrowck => "../locales/en-US/borrowck.ftl",
	builtin_macros => "../locales/en-US/builtin_macros.ftl",
	compiletest => "../locales/en-US/compiletest.ftl",
	const_eval => "../locales/en-US/const_eval.ftl",
	codegen_gcc => "../locales/en-US/codegen_gcc.ftl",
	driver => "../locales/en-US/driver.ftl",
	expand => "../locales/en-US/expand.ftl",
	hir_analysis => "../locales/en-US/hir_analysis.ftl",
	infer => "../locales/en-US/infer.ftl",
	interface => "../locales/en-US/interface.ftl",
	lint => "../locales/en-US/lint.ftl",
	metadata => "../locales/en-US/metadata.ftl",
	middle => "../locales/en-US/middle.ftl",
	mir_dataflow => "../locales/en-US/mir_dataflow.ftl",
	monomorphize => "../locales/en-US/monomorphize.ftl",
	parser => "../locales/en-US/parser.ftl",
	passes => "../locales/en-US/passes.ftl",
	plugin_impl => "../locales/en-US/plugin_impl.ftl",
	privacy => "../locales/en-US/privacy.ftl",
	query_system => "../locales/en-US/query_system.ftl",
	save_analysis => "../locales/en-US/save_analysis.ftl",
	session => "../locales/en-US/session.ftl",
	symbol_mangling => "../locales/en-US/symbol_mangling.ftl",
	trait_selection => "../locales/en-US/trait_selection.ftl",
	ty_utils => "../locales/en-US/ty_utils.ftl",
	}

[Noratrieb]

Seeing hir_analysis being moved in a follow-up PR in this macro was actually what originally gave me the idea to do this, so it's funny that I forgot about it :D

[rustbot]

rustc_error_messages was changed

cc @davidtwco, @compiler-errors, @JohnTitor, @estebank, @TaKO8Ki

[Noratrieb]

This is nice. My only thought here is that, because it's a very manual annotation, it will be forgotten. Is there any way we could maybe even get rid of the "end" annotation?

I think we could get rid of the end annotation with more heuristics, but I think it's quite complex already and I don't think the end annotation is particularly annoying, so I'd rather not do this.

Our best chance for not forgetting to add it to new places is to use it in enough places so that many people become aware of it, and then the assigned reviewer or people that randomly come across can remind the author of this. I'll also ping @nnethercote here since you've been adding a lot of size assertions which are particularly impacted by this.

[Noratrieb]

What about making rustfmt works with feature attributes?
#[rustfmt::sorted] might be an interesting attribute, but adding that is gonna be harder and require more effort than this simple tidy check, which is quite nice already.

[nnethercote]

This is great. A+, 10/10, would recommend. Can you use it for dependencies in Cargo.toml files? 😀

I agree the tidy-alphabetical-end is a little clumsy, but certainly not a showstopper. If you leave it off will tidy complain?

[Noratrieb]

It will complain about lines after the section not being sorted (unless you do this at the end of the file, in which case it will probably work)

[nnethercote]

Could it be made to complain if it reaches either the end of file or another start annotation, without hitting an intervening end annotation?

[Noratrieb]

I can implement this tomorrow, yes.

I could also add it to all the Cargo.toml but I'd rather not do this right now. I think it should check those implicitly, which takes some more code, which I'd rather do in a follow up PR. The big diff in the unstable options is making me feel very uneasy and I'd like to see this merged rather soon :D

[Noratrieb]

Oh, and I just noticed that I have a bad typo in the commit message, it should be "directive"

[Noratrieb]

I implemented that error message and CI is green now

[bors]

☔ The latest upstream changes (presumably #102926) made this pull request unmergeable. Please resolve the merge conflicts.

[Noratrieb]

Thank you @Dylan-DPC for doing the rebase as I requested, as I can't do it right now and would like to get this merged quickly

[jackh726]

@bors r+

[bors]

📌 Commit ce35609 has been approved by jackh726

It is now in the queue for this repository.

[joshtriplett]

Great to have this available and checked in CI!

I would love to see this available in rustfmt, as an attribute on a block or declaration. #[rustfmt::sort] could keep the block or declaration in sorted order.