Regression for well known trait impls for builtin types

[jackh726]

So, this regression happened sometime between 0.14 and 0.20.

Basically, if we have a goal like

Canonical {
  value: InEnvironment {
    environment: Env([]),
    goal: Implemented(Ty(^0.0): TraitId(DefId(2:1942 ~ core[d282]::marker[0]::Sized[0]))),
  },
  binders: [U0 with kind integer type],
}

then we don't generate the clause for Sized impl (at least with SLG).

We pass Implemented(Ty(?0i): TraitId(DefId(2:1942 ~ core[d282]::marker[0]::Sized[0]))) to program_clauses_for_goal; the ?0i self type is not covered in add_sized_program_clauses. I haven't tested Copy/Clone, but I'm almost certain the same problem occurs. Edit: forgot that these are in libcore, so aren't builtin for ints/floats.

[flodiebold]

I think we would have floundered on this before #555, but now it's actually hitting the built-in clauses code which doesn't handle it.

I.e. we just need to actually handle this case in the built-in clauses code.

[jackh726]

Well, we used to be able to solve that goal fine in SLG (I found this when updating the rustc integration, because lots of tests there fail now)

Yeah, this should be a fairly simple fix, but its actually a pretty big issue.

[flodiebold]

My guess is it worked because rustc tried again later when the variable was resolved. Now Chalk will say the goal can't be solved.