Tweak auto trait errors

[estebank]

Make suggestions to remove params and super traits verbose and make spans more accurate.

error[E0567]: auto traits cannot have generic parameters
  --> $DIR/auto-trait-validation.rs:6:19
   |
LL | auto trait Generic<T> {}
   |            -------^^^
   |            |
   |            auto trait cannot have generic parameters

error[E0568]: auto traits cannot have super traits or lifetime bounds
  --> $DIR/auto-trait-validation.rs:8:20
   |
LL | auto trait Bound : Copy {}
   |            -----   ^^^^
   |            |
   |            auto traits cannot have super traits or lifetime bounds

error[E0380]: auto traits cannot have associated items
  --> $DIR/issue-23080.rs:5:8
   |
LL | unsafe auto trait Trait {
   |                   ----- auto traits cannot have associated items
LL |     fn method(&self) {
   |        ^^^^^^

[compiler-errors]

I think that all of these removals are clear enough that we should probably make all of them hidden. A suggestion that says "remove the parameters" doesn't need to take 4 extra lines to render the removal of those parameters; it's pretty clear by the name and the span that's being highlighted what the user needs to do here, and really the only thing we want to make sure is that they're still suggestions so they can be applicable.

In general, I think instead of making what I'll call "unambiguous removal" suggestions verbose, it's probably best to make them hidden.

[estebank]

The only caveat that I have is that not everyone is well versed on the jargon we use. For generic parameters, that'd be weird, but for super traits, that can conceivably be true (although what would a person be doing adding super traits without knowing what they are called, but I digress). The other consideration that I have is the likelihood of an error being seen in the first place. A really uncommon diagnostic doesn't need to be as short as necessary. Having said that, I agree: suggestions that are clear from the message only for the vast majority of people should not be rendering the code to make the diagnostic shorter. The only thing we should be super careful about is to make sure that all "hidden" suggestions have a run-rustfix annotation (and maybe add a flag to render suggestions regardless in stderr files so that we can notice mistakes?).

[compiler-errors]

The other consideration that I have is the likelihood of an error being seen in the first place. A really uncommon diagnostic doesn't need to be as short as necessary.

I'm not totally sure if I agree with this part. For diagnostics that we expect users to never see (auto traits are almost never likely to be stabilized any time soon), I think we should be optimizing these diagnostics for the people who we do expect to see them (e.g. people working on the type system, or experimenting with the unstable feature).

So I'd still prefer shorter presentations for these, since anyone who's going to be messing with auto traits either knows what a supertrait is, or can and should find out what they are (if anything because without knowing what it is, they'll be in the dark by blindly applying the suggestion).

[compiler-errors]

I believe this is still waiting on adding a tracking issue for majorly refactoring diagnostic suggestion styles that I asked for in #t-compiler/wg-diagnostics > Why are we making suggestions verbose? @ 💬

@rustbot author

[compiler-errors]

r=me

[estebank]

@bors r=compiler-errors

[bors]

📌 Commit 91e606b has been approved by compiler-errors

It is now in the queue for this repository.

[Zalathar]

Still trying to track down #144995 (comment)

@bors try jobs=dist-ohos-x86_64

[rust-bors]

⌛ Trying commit 91e606b with merge 26a9d8f…

To cancel the try build, run the command @bors try cancel.

[rust-bors]

☀️ Try build successful (CI)
Build commit: 26a9d8f (26a9d8f3d1c06eba6fb06b2a34c7507abba9952c, parent: 6bcdcc73bd11568fd85f5a38b58e1eda054ad1cd)