Report "missing lifetime specifier" when an anonymous lifetime is used in bounds

[cjgillot]

Both diagnostics pretty much do the same thing: report erroneous lifetime elision.

The "missing lifetime specifier" diagnostic has better diagnostics, as it suggests adding a named lifetime parameter.

This change does not accept more code, it just changes the diagnostic.

r? @petrochenkov

[rust-highfive]

Some changes occurred in diagnostic error codes

cc @GuillaumeGomez

[GuillaumeGomez]

Please don't remove error code explanations! There are examples where they are not emitted anymore. Please do the same for this error code as well.

[cjgillot]

I don't understand your comment. The md file has been removed for other errors that are not emitted any more, hasn't it?

[GuillaumeGomez]

It wasn't supposed to. I realized that a few months ago (maybe more than a year now, not sure...). The point is: if people with older rustc version look at newer rust error code page, they will not see their error anymore. Just update the md file like this one.

[GuillaumeGomez]

Ah bingo, my CI check worked. :)

[GuillaumeGomez]

Thanks! It's possible that the test will fail because of the compile_fail,E0106,E0637. If so, just update it to compile_fail,E0106 and it should work.

[petrochenkov]

@cjgillot

This change does not accept more code, it just changes the diagnostic.

How it happens exactly? What serves as a replacement for AnonymousReportError?

[cjgillot]

This was used for where predicates, and generic parameters. In these cases, there is no resolution for the elided lifetime, so we correctly print "missing lifetime specifier".
However, I forgot to account for universal impl-trait, where we need to explicitly forbid anonymous lifetimes, because purely syntactic rules would do the wrong thing.

[bors]

☔ The latest upstream changes (presumably #96833) made this pull request unmergeable. Please resolve the merge conflicts.