Don't codegen impossible to satisfy impls

[compiler-errors]

Fixes #109098

[rustbot]

r? @jackh726

(rustbot has picked a reviewer for you, use r? to override)

[jackh726]

This is certainly a way to do this. But just want to clarify.

When we monomorphize an impl with type parameters, we check that the where clauses hold, right? But this is after we've already type checked the call sites and errored on use. I guess we can't eagerly monomorphize those impls?

Might be worth expanding the comment a little bit. r=me after that

[compiler-errors]

Left a more descriptive error message.

I guess we can't eagerly monomorphize those impls?

Yeah, we only eagerly monomorphize impls with no ty/ct params.

When we monomorphize an impl with type parameters, we check that the where clauses hold, right?

We don't actually have to re-check where-clauses during monomorphization in general. If something is reachable by fn main, then it must "inductively" have its where-clauses satisfied after monomorphization, we can never call something impossible from main since it has no where-clauses. Probably not the best explanation, not sure how to put it into other words.

This breaks down for eager monomorphization since we monomorphize impls with no type params that are not necessarily called by anything.

[compiler-errors]

@bors r=jackh726

[bors]

📌 Commit b36bbb0 has been approved by jackh726

It is now in the queue for this repository.