diff options
Diffstat (limited to 'tests/ui/higher-ranked/trait-bounds/hrtb-exists-forall-trait-contravariant.rs')
-rw-r--r-- | tests/ui/higher-ranked/trait-bounds/hrtb-exists-forall-trait-contravariant.rs | 36 |
1 files changed, 36 insertions, 0 deletions
diff --git a/tests/ui/higher-ranked/trait-bounds/hrtb-exists-forall-trait-contravariant.rs b/tests/ui/higher-ranked/trait-bounds/hrtb-exists-forall-trait-contravariant.rs new file mode 100644 index 000000000..921061916 --- /dev/null +++ b/tests/ui/higher-ranked/trait-bounds/hrtb-exists-forall-trait-contravariant.rs @@ -0,0 +1,36 @@ +// Test a case where variance and higher-ranked types interact in surprising ways. +// +// In particular, we test this pattern in trait solving, where it is not connected +// to any part of the source code. + +trait Trait<T> {} + +fn foo<T>() +where + T: Trait<for<'b> fn(&'b u32)>, +{ +} + +impl<'a> Trait<fn(&'a u32)> for () {} + +fn main() { + // Here, proving that `(): Trait<for<'b> fn(&'b u32)>` uses the impl: + // + // - The impl provides the clause `forall<'a> { (): Trait<fn(&'a u32)> }` + // - We instantiate `'a` existentially to get `(): Trait<fn(&?a u32)>` + // - We unify `fn(&?a u32)` with `for<'b> fn(&'b u32)` -- this does a + // "bidirectional" subtyping check, so we wind up with: + // - `fn(&?a u32) <: for<'b> fn(&'b u32)` :- + // - `&'!b u32 <: &?a u32` + // - `!'b: ?a` -- solveable if `?a` is inferred to `'empty` + // - `for<'b> fn(&'b u32) <: fn(&?a u32)` :- + // - `&?a u32 u32 <: &?b u32` + // - `?a: ?b` -- solveable if `?b` is also inferred to `'empty` + // - So the subtyping check succeeds, somewhat surprisingly. + // This is because we can use `'empty`. + // + // NB. *However*, the reinstated leak-check gives an error here. + + foo::<()>(); + //~^ ERROR implementation of `Trait` is not general enough +} |