Reduce span to function name in unreachable calls

[ayuusweetfish]

As title suggests, this might close #64103. Refer to the updated tests for expected output.

There is potential to further improve usability. In particular, is it favourable that the exact diverging expression/statement be pointed out (not only in this case, but for all unreachable code)? Certainly that would deserve another issue, but I'm interested in the opinions.

[rust-highfive]

r? @petrochenkov

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

[petrochenkov]

r? @estebank for reassigning, can't spend time on digging into this right now

[petrochenkov]

(The PR as is doesn't seem to address the issue?)

[estebank]

The reduced span to the call itself makes it slightly easier to follow what is going on, but as you point out the only thing that will actually make this clear is to actually point at the diverging expression (panic!(), return, etc.) that is the underlying cause for the call itself to be unreachable. Would you be open to investigate what would be required to accomplish that?

Beyond that r=me on this PR once it's green.

[ayuusweetfish]

Thanks for the prompt reply!

I'd love to dig into this! For now, I think a viable solution is to store the span of the diverging Expr/Stmt/Block inside the Diverges enum, which can aid the lint for all types of unreachable code. All changes are still in crate librustc_typeck.

[Centril]

I'd like to not complicate Diverges further at this point and instead go for the simpler fix in this PR.

[ayuusweetfish]

Would you mind explaining the reason? And will it be useful if the diverging expression is pointed out for every unreachable code warning? The only downside I can think of is becoming too verbose, but there is always #[allow(unreachable_code)].

[Centril]

Would you mind explaining the reason?

Mainly internal compiler complexity reasons, but if it turns out it does not substantially add to the complexity of understanding how Diverges works then go for it.

[estebank]

@Centril I can see it being as "cheap" as adding a single span: Cell<Option<Span>> field in FnCtxt pointing at the last divergent path...

[estebank]

Lets do this, lets merge this PR and @kawa-yoiko can work on the enhancement in a follow up that @Centril and I can review and have a longer convo if needed there. This as is is already a clear, if smaller, improvement.

@bors r+

[bors]

📌 Commit e1d27eb has been approved by estebank