There is literally no such thing as a compiler bug in a language defined by "whatever the compiler does." The compiler is always, tautologically, correct.

By that reasoning specifications can't have bugs either. Sure, by some extreme prescriptive understanding of what a compiler or specification is, there are no bugs, but if a cryptography specification allows attacks in the threat model or rustc segfaults, all but the most asinine parties would agree there's a bug -- what's the point of asserting otherwise?

You are being pedantic here; this stance doesn't survive contact with the real world. The fact that Rust's language reference is not complete/up to date, doesn't mean that Rust is defined _just_ by "whatever the compiler does." There are community-wide accepted documents (the reference, Rustonomicon, RFCs, and more informally, Language/Compiler team meeting notes/comments that reflect their consensus about the semantics) how Rust _ought_ to work.