Rollup merge of rust-lang#127574 - lcnr:coherence-check-supertrait, r=compiler-errors

elaborate unknowable goals

A reimplemented version of rust-lang#124532 affecting only the new solver. Always trying to prove super traits ends up causing a fatal overflow error in diesel, so we cannot land this in the old solver.

The following test currently does not pass coherence:
```rust
trait Super {}
trait Sub<T>: Super {}

trait Overlap<T> {}
impl<T, U: Sub<T>> Overlap<T> for U {}
impl<T> Overlap<T> for () {}

fn main() {}
```

We check whether `(): Sub<?t>` holds. This stalls with ambiguity as downstream crates may add an impl for `(): Sub<Local>`. However, its super trait bound `(): Super` cannot be implemented downstream, so this one is known not to hold.

By trying to prove that all the super bounds of a trait before adding a coherence unknowable candidate, this compiles. This is necessary to prevent breakage from enabling `-Znext-solver=coherence` (rust-lang#121848), see tests/ui/coherence/super-traits/super-trait-knowable-2.rs for more details. The idea is that while there may be an impl of the trait itself we don't know about, if we're able to prove that a super trait is definitely not implemented, then that impl would also never apply/not be well-formed.

This approach is different from rust-lang#124532 as it allows tests/ui/coherence/super-traits/super-trait-knowable-3.rs to compile. The approach in rust-lang#124532 only elaborating the root obligations while this approach tries it for all unknowable trait goals.

r? `@compiler-errors`

Author: matthiaskrgr

compiler/rustc_next_trait_solver/src/solve/assembly/mod.rs

@@ -698,6 +698,18 @@ where
                 if ecx.trait_ref_is_knowable(goal.param_env, trait_ref)? {
                     Err(NoSolution)
                 } else {
+                    // While the trait bound itself may be unknowable, we may be able to
+                    // prove that a super trait is not implemented. For this, we recursively
+                    // prove the super trait bounds of the current goal.
+                    //
+                    // We skip the goal itself as that one would cycle.
+                    let predicate: I::Predicate = trait_ref.upcast(cx);
+                    ecx.add_goals(
+                        GoalSource::Misc,
+                        elaborate::elaborate(cx, [predicate])
+                            .skip(1)
+                            .map(|predicate| goal.with(cx, predicate)),
+                    );
                     ecx.evaluate_added_goals_and_make_canonical_response(Certainty::AMBIGUOUS)
                 }
             },

compiler/rustc_type_ir/src/inherent.rs

@@ -432,6 +432,7 @@ pub trait Predicate<I: Interner<Predicate = Self>>:
     + UpcastFrom<I, ty::OutlivesPredicate<I, I::Ty>>
     + UpcastFrom<I, ty::OutlivesPredicate<I, I::Region>>
     + IntoKind<Kind = ty::Binder<I, ty::PredicateKind<I>>>
+    + Elaboratable<I>
 {
     fn as_clause(self) -> Option<I::Clause>;
 

tests/ui/coherence/normalize-for-errors.next.stderr

@@ -7,6 +7,7 @@ LL |
 LL | impl<S: Iterator> MyTrait<S> for (Box<<(MyType,) as Mirror>::Assoc>, S::Item) {}
    | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ conflicting implementation for `(Box<(MyType,)>, <_ as Iterator>::Item)`
    |
+   = note: upstream crates may add a new impl of trait `std::clone::Clone` for type `(MyType,)` in future versions
    = note: upstream crates may add a new impl of trait `std::marker::Copy` for type `std::boxed::Box<(MyType,)>` in future versions
 
 error: aborting due to 1 previous error

tests/ui/coherence/normalize-for-errors.rs

@@ -18,5 +18,6 @@ impl<S: Iterator> MyTrait<S> for (Box<<(MyType,) as Mirror>::Assoc>, S::Item) {}
 //~^ ERROR conflicting implementations of trait `MyTrait<_>` for type `(Box<(MyType,)>,
 //~| NOTE conflicting implementation for `(Box<(MyType,)>,
 //~| NOTE upstream crates may add a new impl of trait `std::marker::Copy` for type `std::boxed::Box<(MyType,)>` in future versions
+//[next]~| NOTE upstream crates may add a new impl of trait `std::clone::Clone` for type `(MyType,)` in future versions
 
 fn main() {}

tests/ui/coherence/super-traits/super-trait-knowable-1.current.stderr

@@ -0,0 +1,13 @@
+error[E0119]: conflicting implementations of trait `Overlap<_>` for type `()`
+  --> $DIR/super-trait-knowable-1.rs:16:1
+   |
+LL | impl<T, U: Sub<T>> Overlap<T> for U {}
+   | ----------------------------------- first implementation here
+LL | impl<T> Overlap<T> for () {}
+   | ^^^^^^^^^^^^^^^^^^^^^^^^^ conflicting implementation for `()`
+   |
+   = note: downstream crates may implement trait `Sub<_>` for type `()`
+
+error: aborting due to 1 previous error
+
+For more information about this error, try `rustc --explain E0119`.

tests/ui/coherence/super-traits/super-trait-knowable-1.rs

@@ -0,0 +1,19 @@
+// Added in #124532. While `(): Super` is knowable, `(): Sub<?t>` is not.
+//
+// We therefore elaborate super trait bounds in the implicit negative
+// overlap check.
+
+//@ revisions: current next
+//@ ignore-compare-mode-next-solver (explicit revisions)
+//@[next] compile-flags: -Znext-solver
+//@[next] check-pass
+
+trait Super {}
+trait Sub<T>: Super {}
+
+trait Overlap<T> {}
+impl<T, U: Sub<T>> Overlap<T> for U {}
+impl<T> Overlap<T> for () {}
+//[current]~^ ERROR conflicting implementations
+
+fn main() {}

tests/ui/coherence/super-traits/super-trait-knowable-2.rs

@@ -0,0 +1,33 @@
+// A regression test for pyella-0.1.5 which broke when
+// enabling the new solver in coherence.
+//
+// `Tensor: TensorValue` is knowable while `Tensor: TensorOp<?t2>`
+// may be implemented downstream. We previously didn't check the
+// super trait bound in coherence, causing these impls to overlap.
+//
+// However, we did fail to normalize `<Tensor as TensorValue::Unmasked`
+// which caused the old solver to emit a `Tensor: TensorValue` goal in
+// `fn normalize_to_error` which then failed, causing this test to pass.
+
+//@ revisions: current next
+//@ ignore-compare-mode-next-solver (explicit revisions)
+//@[next] compile-flags: -Znext-solver
+//@ check-pass
+
+pub trait TensorValue {
+    type Unmasked;
+}
+
+trait TensorCompare<T> {}
+pub trait TensorOp<T>: TensorValue {}
+
+pub struct Tensor;
+impl<T2> TensorCompare<T2> for Tensor {}
+impl<T1, T2> TensorCompare<T2> for T1
+where
+    T1: TensorOp<T2>,
+    T1::Unmasked: Sized,
+{}
+
+
+fn main() {}

tests/ui/coherence/super-traits/super-trait-knowable-3.current.stderr

@@ -0,0 +1,13 @@
+error[E0119]: conflicting implementations of trait `Overlap<_>` for type `()`
+  --> $DIR/super-trait-knowable-3.rs:19:1
+   |
+LL | impl<T, U: Bound<W<T>>> Overlap<T> for U {}
+   | ---------------------------------------- first implementation here
+LL | impl<T> Overlap<T> for () {}
+   | ^^^^^^^^^^^^^^^^^^^^^^^^^ conflicting implementation for `()`
+   |
+   = note: downstream crates may implement trait `Sub<_>` for type `()`
+
+error: aborting due to 1 previous error
+
+For more information about this error, try `rustc --explain E0119`.

tests/ui/coherence/super-traits/super-trait-knowable-3.rs

@@ -0,0 +1,22 @@
+// Unlike in `super-trait-knowable-1.rs`, the knowable
+// super trait bound is in a nested goal so this would not
+// compile if we were to only elaborate root goals.
+
+//@ revisions: current next
+//@ ignore-compare-mode-next-solver (explicit revisions)
+//@[next] compile-flags: -Znext-solver
+//@[next] check-pass
+
+trait Super {}
+trait Sub<T>: Super {}
+
+struct W<T>(T);
+trait Bound<T> {}
+impl<T: Sub<U>, U> Bound<W<U>> for T {}
+
+trait Overlap<T> {}
+impl<T, U: Bound<W<T>>> Overlap<T> for U {}
+impl<T> Overlap<T> for () {}
+//[current]~^ ERROR conflicting implementations of trait `Overlap<_>` for type `()`
+
+fn main() {}

tests/ui/issues/issue-48728.stderr renamed to tests/ui/issues/issue-48728.current.stderr

@@ -1,5 +1,5 @@
 error[E0119]: conflicting implementations of trait `Clone` for type `Node<[_]>`
-  --> $DIR/issue-48728.rs:4:10
+  --> $DIR/issue-48728.rs:9:10
    |
 LL | #[derive(Clone)]
    |          ^^^^^ conflicting implementation for `Node<[_]>`