Non-primitive-types static assertions result in an LLVM failure

[LeoTestard]

The code:

struct S;

#[static_assert]
static FOO: S = S;

fn main() {}

results in:

issue-13951.rs:1:1: 1:10 warning: struct is never used: `S`, #[warn(dead_code)] on by default
issue-13951.rs:1 struct S;
                 ^~~~~~~~~
issue-13951.rs:4:1: 4:19 warning: static item is never used: `FOO`, #[warn(dead_code)] on by default
issue-13951.rs:4 static FOO: S = S;
                 ^~~~~~~~~~~~~~~~~~
rustc: /home/rustbuild/src/rust-buildbot/slave/nightly-dist-rustc-linux/build/src/llvm/include/llvm/Support/Casting.h:237: typename llvm::cast_retty<X, Y*>::ret_type llvm::cast(Y*) [with X = llvm::ConstantInt; Y = llvm::Value; typename llvm::cast_retty<X, Y*>::ret_type = llvm::ConstantInt*]: Assertion `isa<X>(Val) && "cast<Ty>() argument of incompatible type!"' failed.

This has no meaning and should be at least warned about. Putting #[static_assert] on non-static items (like struct definitions... ) is silently ignored.
I think that we should restrict #[static_assert] to boolean statics, and emit a warning or an error in the other cases.

[LeoTestard]

(I think I can fix this myself, actually I found this while trying to fix #13951 and it's a bit blocking since I don't know which #[static_assert]-flagged statics considered as used or not. But I just need someone from the team to agree or not that we should disallow such statics.)

[dotdash]

IMO, it's not so much about allowing such statics but about allowing #[static_assert] on statics that aren't plain integers. You get the same assertion with struct S { x: i32 }, and we surely want to allow such statics.

As far as I'm concerned, I'd forbid #![static_assert] on such statics.