Remove indices from fresh ty/const vars.

[nnethercote]

Each of InferTy::Fresh{Ty,IntTy,FloatTy} and InferConst::Fresh have a u32 index. It turns out the index is only used for some weak sanity checking during freshening and can be removed, making these values a bit like ReErased.

This change simplifies things quite a bit, e.g. TypeFreshener no longers needs maps, and we can eliminate a bunch of pre-interned fresh types.

r? @lcnr

[rustbot]

changes to the core type system

cc @compiler-errors, @lcnr

Some changes occurred to the CTFE machinery

cc @RalfJung, @oli-obk, @lcnr

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri

[nnethercote]

I'm not sure if this is legitimate. All the tests pass locally, but I'm getting an ICE when compiling nalgebra-0.33.0:

error: internal compiler error: compiler/rustc_trait_selection/src/traits/select/confirmation.rs:243:17: Where clause `Binder { value: <base::default_allocator::DefaultAllocator as base::allocator::Allocator<R, C>>, bound_vars: [] }` was applicable to `Obligation(predicate=Binder { value: TraitPredicate(<base::default_allocator::DefaultAllocator as base::allocator::Allocator<_, _>>, polarity:Positive), bound_vars: [] }, depth=1)` but now is not

[nnethercote]

@bors try

[rust-bors]

⌛ Trying commit db9c00f with merge 12882b8…

To cancel the try build, run the command @bors try cancel.

[lcnr]

it's not legitimate sadly

we use their index to make sure different infer vars result in different fresh variables. We use freshened infer vars for caching and need a differetn cache entry for T: Trait<?0, ?1> and T: Trait<?0,?1>

[rust-bors]

☀️ Try build successful (CI)
Build commit: 12882b8 (12882b84109d06697f8278cd473a9906315add23, parent: 65b6cdb6a6d33987b9d642a4882283c71fbe3957)

[nnethercote]

need a different cache entry for T: Trait<?0, ?1> and T: Trait<?0,?1>

Would that end up being T: Trait<Fresh(0), Fresh(1)> and T: Trait<Fresh(2), Fresh(3)>?

It's curious how close this is to working. When I first tried it I was expecting failures all over the place. When the ui tests and other tests passed I thought it was legitimate.

[lcnr]

aaaah, T: Trait<?0, ?0> and T: Trait<?0, ?1> 😅 we also need different ones for T: Trait<?0, ?1> and T: Trait<?1, ?2> for cycle detection in evaluate, cc #102713 (comment)