debuginfo: Extraneous descriptions of enum discriminator are emitted

[vadimcn]

This probably isn't a huge deal, but... if one were to examine the output of llvm-dwarfdump for, say, libstd (if compiled with -g flag), you'd find that rustc emits a description of the type of enum discriminator field for each monomorphization of generic enums. For Option alone, there's probably a couple hundred of them.
I think this is unnecessary, because discriminator never depends on enum's type parameters, so all monomorphizations could share the same discriminator description.

Edit: This doesn't apply, of course, to optimized forms of Option, such as Option<&T>, which don't have the discriminator field to begin with.

[thestinger]

It does depend on the type parameters in some cases. The compiler has logic for specialized enum representations.

[vadimcn]

You are right, I forgot about the pointer Option's. The point still stands, though, for unoptimized forms.

[michaelwoerister]

I'll look into it.

[michaelwoerister]

This shouldn't be hard to optimize. Just cache discriminator types with the enums DefId as key.

[michaelwoerister]

This should have been closed automatically with #13441, but wasn't because of PR rollup #13443.