r? @michaelwoerister

(rust-highfive has picked a reviewer for you, use r? to override)

Question for reviewers: in attempting to clean up the error checking for link_ordinal in rustc_typeck/src/collect.rs beginning at about line 3156, I discovered that it's recording the ordinal, if present, to the function's CodegenFnAttrs struct. Should I be getting this information from there, rather than re-analyzing the attributes, in native_libs.rs? If so, how do I get a reference to the relevant CodegenFnAttrs object? Or should we drop this field from CodegenFnAttrs instead? As far as I can determine, it's only ever read by check_link_name_xor_ordinal in rustc_typeck/src/collect.rs.

(I refer to cleaning up the error checking because the messages rustc issues in certain circumstances are, at best, misleading: #[link_ordinal(3, 4)] is indeed invalid input, but complaining about an illegal argument format (unsuffixed integer value expected) in response is confusing.)

@ricobbe I'm just giving my personal views here, am no reviewer in any official capacity, but I think that duplication should be avoided and ideally it should be re-used. I think it should be possible to obtain the attrs with something like self.tcx.codegen_fn_attrs(item.id.def_id), not entirely sure though.

Also, IMO the error message doesn't have to be perfect, but of course improvements are always nice.

@est31 Ok, I've pushed changes that use the info from CodegenFnAttrs. As another benefit, this also streamlines the error-detection code as well!

r? @joshtriplett

@joshtriplett do you have a sense of when you'll be able to take a look at this?

LGTM from a purely language perspective. Deferring to compiler folks on the implementation strategy.

LGTM from a purely language perspective. Deferring to compiler folks on the implementation strategy.

@joshtriplett Thanks for the review! I'm happy for the compiler folks to take a look at it, though I will say that this PR is a straightforward extension of the basic import-by-name raw-dylib work from PR #84171.

r? @michaelwoerister
for compiler team review of the implementation. No issues from a language semantics perspective.

I'll take a look tomorrow.

Looks great, @ricobbe! Thanks for adding so many tests.

@bors r

@michaelwoerister: 🔑 Insufficient privileges: Not in reviewers

@rust-lang/infra, any idea what that is about?

Let's give it another try: @bors r

📌 Commit 142f6c0 has been approved by michaelwoerister