Make some invalid codegen attr errors structured/translatable

[estebank]

No description provided.

[rustbot]

r? @nnethercote

rustbot has assigned @nnethercote.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

Some changes occurred in compiler/rustc_codegen_ssa/src/codegen_attrs.rs

cc @jdonszelmann

[compiler-errors]

I thought we were in limbo w.r.t. translatable/structured diagnostics. https://rust-lang.zulipchat.com/#narrow/channel/147480-t-compiler.2Fwg-diagnostics/topic/.E2.9C.94.20translation.20effort

I kinda don't want to land more PRs to make things into translatable diagnostics for no reason if it's not clear that this effort has a primary owner in the compiler and it's possible that we may just want to rip it all out?

cc @davidtwco @jieyouxu

[jieyouxu]

(idk what's the plan for the translation infra, I recorded what I found out about the current status over at #132181)

AFAICT, the current status is that:

• You don't need to use the translatable diagnostics infra if it's overly restrictive/annoying (e.g. in const-eval usually)
• ... but if you want (or if it makes the code clear or sth), then by all means use the infra.

But yeah, probably doesn't really make sense to migrate existing diagnostics over to the translatable one unless it makes the code nicer or if we have a path forward for the translatable infra, I guess.

FWIW, ripping the translatable infra out is equally churny/disruptive so...

[estebank]

I wasn't aware of the policy change discussions, only vaguely remembered a different thread from davidtwco about some of the issues he saw with the current approach.

I'm ok with not landing these.

But I think there's still value in the current infrastructure even independently of the (demise of the?) translation effort. One case is #137201, where I made it so that instead of having to manually shorten types and set the message for where the full type name was written on every error, the translation machinery handles that on every appropriate type without manual intervention.

[nnethercote]

I will say r=me because it looks like conversion is valid, code-wise, and it gets rid of a few calls to struct_span_code_err! which is a macro I don't like, and because it marginally improves the error output in one test. I will leave it up to @estebank whether or not to merge this.

[bors]

☔ The latest upstream changes (presumably #137497) made this pull request unmergeable. Please resolve the merge conflicts.