Should we be eagerly evaluating consts in the "old"-style normalizer?

[compiler-errors]

Right now, the AssocTypeFolder will eagerly evaluate consts unless GCE is enabled. We still have usages of this folder (the "old"-style normalizer) all over the place, and tearing that out is a lot of work.

Eventually, I'd like to move away from normalizing types or consts unless there's a clear reason for the normalization (i.e. we structurally normalize types in HIR typeck for various reasons like field acceses).

Right now we continue to eagerly normalize types in HIR typeck. Doing this for consts and not for types does introduce an incongruency that may make subtle alias-relate bugs appear just for consts, but it does get us one step closer to always-lazy-norm...

rust-lang/rust#112183 (comment)

[lcnr]

We normalize constants in all places which normalizes types and expect it to stay that way. I feel confident in our current normalization approach and don't expect us to cause const and ty normalization to diverge.

I don't think this issue is useful to track anything because of that. Even if normalizing consts will cause issues, I don't think keeping this issue open is valuable for that.