Normalize lazilly for sized #32844
Normalize lazilly for sized #32844
Conversation
It should never have been public. Call `skip_binder` instead and make your intentions explicit. Where possible, I adopted `map_bound_ref` and `map_bound` instead.
This avoids normalizing for structural types, which kind of gives us a poor man's lazy normalization. It's enough to fix issue rust-lang#31299, anyway, though you can still get false failures if you try hard enough. Fixes rust-lang#31299.
Leverage the fact that wfcheck requires that all struct fields are of sized type except the last, so we don't need to check that again.
|
This does not handle newtyped pointers ( |
| ty::BoundCopy => ok_if(vec![element_ty]), | ||
| ty::BoundSized => ok_if(Vec::new()), | ||
| ty::BoundCopy => self.builtin_bound_ok_if(bound, vec![element_ty]), | ||
| ty::BoundSized => self.builtin_bound_ok_if(bound, vec![]), |
|
How does last-field-only not break our self-recursive struct checking? struct Recursive {
this: Recursive,
last: u32
} |
We have separate code that checks for this. It so happens that if we get a recursive When I run that test I get: /home/nmatsakis/tmp/x.rs:1:1: 4:2 error: recursive type `Recursive` has infinite size [E0072]
/home/nmatsakis/tmp/x.rs:1 struct Recursive {
/home/nmatsakis/tmp/x.rs:2 this: Recursive,
/home/nmatsakis/tmp/x.rs:3 last: u32
/home/nmatsakis/tmp/x.rs:4 }
/home/nmatsakis/tmp/x.rs:1:1: 4:2 help: run `rustc --explain E0072` to see a detailed explanation
/home/nmatsakis/tmp/x.rs:1:1: 4:2 help: insert indirection (e.g., a `Box`, `Rc`, or `&`) at some point to make `Recursive` representable
error: aborting due to previous errorI can add a test case if you like. |
No, it doesn't, no structs. Why not land it? Just because it's hacky? |
|
Withdrawing this PR in favor of #33138, which is way better. |
This is a bit of a hack, but if we eagerly expand structural type obligations, we can resolve #31299.
r? @arielb1