Simplify eager normalization of constants

[oli-obk]

r? @nikomatsakis

[nikomatsakis]

Looks good but I have one nlggling concern -- though maybe think it's pre-existing. Can we have type inference variables in the type of the constant (or some other place?). In that case we might want to invoke infcx.fully_resolve or some such thing before invoking eval?

[oli-obk]

It's not possible to end up with an unevaluated ty::Const whose ty or substs field has inference variables. We may get there in the future once we have generic constants, but for now that can't happen.

I can add a debug assert assert here to make sure of that though.

[nikomatsakis]

@bors r+

[bors]

📌 Commit 181ed55 has been approved by nikomatsakis

[nikomatsakis]

@oli-obk

I can add a debug assert assert here to make sure of that though.

oh, I thought you meant you had added such an assert -- in any case, I don't think it'd be catastrophic if an inference variable leaked through. We might ICE or be overly conservative, but I think not worse than that.

[bors]

⌛ Testing commit 181ed55 with merge ac21131...

[bors]

☀️ Test successful - checks-azure
Approved by: nikomatsakis
Pushing ac21131 to master...