Remove ParamEnv::reveal_all from codegen

[davidtwco]

Instead of using ParamEnv::reveal_all, the ParamEnv should be tracked in the mono item - this is necessary to enable more complex forms of polymorphization in future.

something simple like trying layout_of(ty) on various types in the MIR body and ignoring generics in ty if it succeeds, should work, but the lack of a correct ParamEnv would make it fail later (especially if e.g. it's something like &T which only has a constant layout if T: Sized)

(source: eddyb in Zulip)

[RalfJung]

Likely Miri would need similar changes? Or does it not matter because Miri doesn't actually monomorphize, we just run polymorphic MIR and track the current subst?

[davidtwco]

Likely Miri would need similar changes? Or does it not matter because Miri doesn't actually monomorphize, we just run polymorphic MIR and track the current subst?

It's possible - I'm not that sure.

[fmease]

Should we close this now that -Zpolymorphize has been removed or is this something future designs/implementations are bound to be affected by, too? Rephrased, is it useful to keep open "in limbo"?

[RalfJung]

@lcnr @BoxyUwU is there still something to track here or has this been resolved with the removal of -Zpolymorphize?

[BoxyUwU]

I think the only purpose this serves is that if -Zpolymorphize is re-added this might remind folks of this pitfall 🤔 I think its safe to close this; its not actionable until someone picks up polymorphization again which I haven't seen any movement on