Implement support for GeneratorWitnessMIR in new solver

[compiler-errors]

r? @cjgillot

I mostly want this to cut down the number of failing UI tests when running the UI test suite with --compare-mode=next-solver, but there doesn't seem like much reason to block implementing this since it adds minimal complexity to the existing structural traits impl in the new solver.

If others are against adding this for some reason, then maybe we should just make GeneratorWitnessMIR return NoSolution for these traits. Anything but an ICE please 😸 🧊

[rustbot]

Some changes occurred to the core trait solver

cc @rust-lang/initiative-trait-system-refactor

[lcnr]

think this is fine, still not completely sure about the current approach wrt to GeneratorWitnessMIR but this feels simple enough to not cause any issue.

[cjgillot]

When I implemented it in the current solver, the subtlety was the handling of lifetimes, since generator_hidden_types only has erased lifetimes.
Do those functions need to account for lifetimes somehow, or are erased lifetimes ok?

[compiler-errors]

Oh, I guess we could instantiate fresh placeholder lifetimes for all the types in here when we encounter ReErased...

I didn't check but the solver probably handles erased lifetimes (I think...), but the region constraints returned by the solver are probably not sound or correct -- so if we had some auto trait bound that holds with region constraints, they're probably not enforced.

[lcnr]

Oh, I guess we could instantiate fresh placeholder lifetimes for all the types in here when we encounter ReErased...

I didn't check but the solver probably handles erased lifetimes (I think...), but the region constraints returned by the solver are probably not sound or correct -- so if we had some auto trait bound that holds with region constraints, they're probably not enforced.

We currently ICE when encountering ReErased in a query response. If these lifetimes are existentials instantiating them with infer vars would work? 🤔

[compiler-errors]

I thought they would be universal vars, since they're acting more like for-all over the generator interior's choice of lifetimes, but idk

[lcnr]

🤷 universals is also fine using placeholders. But it feels weird to have universal regions be represented using ReErased instead of using a Binder?

[cjgillot]

In the current solver, they are modeled as universals. The idea is: if x: MyType<'erased> appears in MIR, 'a could be the lifetime of a local (so arbitrary small) or of a global (arbitrary large). See bind_generator_hidden_types_above.

If we want lifetime knowledge in generators, we'd need to compute those hidden types in borrowck instead of a separate query. I don't have such plans yet.

[cjgillot]

Great!
@bors r+

[bors]

📌 Commit 4a4fc3b has been approved by cjgillot

It is now in the queue for this repository.