Add short error message-format

[GuillaumeGomez]

Fixes#42653.

[rust-highfive]

r? @eddyb

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

[eddyb]

r? @nrc

[nrc]

Rather than adding a new emitter, could you have the human-readable emitter take an argument for detail level - either short or verbose? It seems that we'd save a lot of code dup (and a macro) that way.

Could you put this behind a feature gate?

[GuillaumeGomez]

Yes and yes. I was just thinking that we'd want to fully split both of the features but if not...

[GuillaumeGomez]

Ok, now just remain to put it behind a feature flag.

[kennytm]

Is it possible to make each error only occupy one line, similar to clang -fno-caret-diagnostics?

Mock up:

$DIR/short-error-format.rs:16:9: error[E0308]: mismatched types
$DIR/short-error-format.rs:18:7: error[E0599]: no method named `salut` found for type `u32` in the current scope
error: aborting due to 2 previous errors 

[eddyb]

@kennymismatched types is the worst error in the entire compiler, if you can't see the detail lines. I know it looks pretty but it has screwed over pretty much everyone who doesn't look at the console, including the official RLS extension for VS Code.
I only recently pulled off a hack to have the information available to kate in a manageable way.

[kennytm]

@eddyb That sample is referring to the UI test in this PR https://github.com/GuillaumeGomez/rust/blob/a58becd60d7d7a965ed246cf8e290e2bd94043b9/src/test/ui/short-error-format.stderr:

error[E0308]: mismatched types
--> $DIR/short-error-format.rs:16:9
error[E0599]: no method named `salut` found for type `u32` in the current scope
--> $DIR/short-error-format.rs:18:7
error: aborting due to 2 previous errors

Improving E0308 could happen independent of this PR.

[eddyb]

Oh yeah, in that case, I agree, the example output in #44636 (comment) is a subset of the lines I have from my errfilt conversion, but I also have labels/notes (within some limitations), e.g.:

src/librustc_trans/mir/lvalue.rs:97:62: error[E0308]: mismatched types
src/librustc_trans/mir/lvalue.rs:97:62: expected struct `rustc::ty::layout::FullLayout`, found enum `rustc::mir::tcx::LvalueTy`
src/librustc_trans/mir/lvalue.rs:97:62: note: expected type `rustc::ty::layout::FullLayout<'_>`
src/librustc_trans/mir/lvalue.rs:97:62: found type `rustc::mir::tcx::LvalueTy<'_>`

[GuillaumeGomez]

So which output do you want? So I can do it once and for all. :)

[estebank]

I feel if the idea is to use as little vertical space the output should be closer to:

$DIR/short-error-format.rs:16:9 - error[E0308]: mismatched types
$DIR/short-error-format.rs:18:7 - error[E0599]: no method named `salut` found for type `u32` in the current scope
error: aborting due to 2 previous errors

Beyond that, I wonder wether it'd be worth it to try to keep in the diagnostic machinery space for custom single line error messages that would work better for short output (and for text editors). May be I'll just go ahead and introduce it for while and see how much of a pain it becomes. If it is too much, we can rip it out. I do feel strongly against improving things for text editors in detriment of cli output.

[GuillaumeGomez]

New output available!

[estebank]

Some nitpicks, but I'm ok with the PR. Leaving it to @nrc / @eddyb to give final r+.

[estebank]

human, json or short

[estebank]

Fits in single line?

[estebank]

Fits in single line?

[estebank]

Fits in single line?

[GuillaumeGomez]

The branch has been created on rustfmt so I point to it instead.

[GuillaumeGomez]

We now have the following error:

[00:00:36] Submodule 'src/tools/rustfmt' (https://github.com/rust-lang-nursery/rustfmt.git) unregistered for path 'src/tools/rustfmt'
[00:00:36] Submodule 'src/tools/rustfmt' (https://github.com/rust-lang-nursery/rustfmt.git) registered for path 'src/tools/rustfmt'
[00:00:37] error: Server does not allow request for unadvertised object fc2c85d423a34086243f019c15e2d9cc2afcbb8f
[00:00:37] Fetched in submodule path 'src/tools/rustfmt', but it did not contain fc2c85d423a34086243f019c15e2d9cc2afcbb8f. Direct fetching of that commit failed.

Anyone has an idea?

[topecongiro]

@GuillaumeGomez Could you please use e35e27659dec7a70d1e0e45be2c5f3a02b2c742c? I think I have erased fc2c85d423a34086243f019c15e2d9cc2afcbb8f while I pushed the update :/ I am sorry.

[GuillaumeGomez]

Ok, let's hope it'll be the good one this time. :)

[GuillaumeGomez]

New error and it doesn't get any better:

[00:02:06] extracting /checkout/obj/build/cache/2017-08-29/cargo-beta-x86_64-unknown-linux-gnu.tar.gz
[00:02:06] error: failed to resolve patches for `https://github.com/rust-lang/crates.io-index`
[00:02:06]
[00:02:06] Caused by:
[00:02:06] patch for `rustfmt-nightly` in `https://github.com/rust-lang/crates.io-index` did not resolve to any crates
[00:02:06] failed to run: /checkout/obj/build/x86_64-unknown-linux-gnu/stage0/bin/cargo build --manifest-path /checkout/src/bootstrap/Cargo.toml --locked
[00:02:06] Build completed unsuccessfully in 0:00:46

Anyone has an idea?

[GuillaumeGomez]

Fixed.

[michaelwoerister]

@bors r+

[bors]

📌 Commit b46c014 has been approved by michaelwoerister

[bors]

⌛ Testing commit b46c014 with merge f9d2416...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: michaelwoerister
Pushing f9d2416 to master...

[alexcrichton]

@nrc@GuillaumeGomez this landed without a feature flag, was that intentional? I assume not given #44636 (comment)?

[nrc]

I suppose this should be behind -Zunstable-features given that comment. I filed #45995

Thanks for noticing @alexcrichton !

[GuillaumeGomez]

Hum indeed, it should had have a feature flag. I'll add it as soon as possible.

[vitiral]

❤️ Thanks for fixing the bug I reported, looking forward to using this when it becomes available!

[vitiral]

On this comment my preference would be something like:

src/librustc_trans/mir/lvalue.rs:97:62: error[E0308]: mismatched types. Expected struct `rustc::ty::layout::FullLayout`, found enum `rustc::mir::tcx::LvalueTy`

It's all in one line but that's probably okay. That's just, like, my opinion though man.