CTFE mir is sufficient for statics and consts

[jsgf]

The issue isn't that the MIR is missing, but that it's for CTFE, so extend the check for that as well.

Closes issue #85401

[rust-highfive]

r? @davidtwco

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

[tmiasko]

With proposed change statics from another crate will be codegened locally, i.e., duplicated each time they are referenced from another crate.

Under normal circumstances the issue never arise because is_reachable_non_generic(def_id) is true for statics, and should_codegen_locally returns false earlier.

[jsgf]

@tmiasko I see. Would doing something like

if tcx.is_ctfe_mir_available(def_id) {
    return false;
}

be better then?

My goal here is to make running separate rustc --emit metadata -Zalways-encode-mir=yes/rustc --emit link commands equivalent to the cargo pipelined build invocation of rustc --emit link,metadata - that is, so that the .rmetas can be used as inputs to downstream dependent builds.

[tmiasko]

I suspect that anything short of computing reachable_non_generics when building the metadata crate will run into trouble. Without this information all functions will be generated locally.

[jsgf]

@tmiasko That must be what --emit metadata,link must doing then, yes?

[tmiasko]

That must be what --emit metadata,link must doing then, yes?

There are two implementation of reachable_non_generics. One for the local crate that actually computes it and the second one for non-local crates that merely decodes what had been computed by the other. Consequently the content for non-local crates will be reflection of compilation flags used when the metadata had been emitted.

The --emit metadata,link will include reachable_non_generics for that crate, --emit metadata will omit it.

[RalfJung]

Cc @oli-obk who wrote the code that is being modified here.

[jsgf]

Closing in favour of #85793