trait_sel: resolve vars in host effects

In the standard library, the `Extend` impl for `Iterator` (specialised
with `TrustedLen`) has a parameter which is constrained by a projection
predicate. This projection predicate provides a value for an inference
variable but host effect evaluation wasn't resolving variables first.

Adding the extra resolve can the number of errors in some tests when they
gain host effect predicates, but this is not unexpected as calls to
`resolve_vars_if_possible` can cause more error tainting to happen.

Co-authored-by: Boxy <rust@boxyuwu.dev>

compiler/rustc_trait_selection/src/traits/effects.rs

@@ -31,6 +31,8 @@ pub fn evaluate_host_effect_obligation<'tcx>(
         );
     }
 
+    let ref obligation = selcx.infcx.resolve_vars_if_possible(obligation.clone());
+
     // Force ambiguity for infer self ty.
     if obligation.predicate.self_ty().is_ty_var() {
         return Err(EvaluationFailure::Ambiguous);

tests/ui/traits/const-traits/unconstrained-var-specialization.rs

@@ -0,0 +1,36 @@
+//@ check-pass
+//@ compile-flags: --crate-type=lib
+#![no_std]
+#![allow(internal_features)]
+#![feature(rustc_attrs, min_specialization, const_trait_impl)]
+
+// In the default impl below, `A` is constrained by the projection predicate, and if the host effect
+// predicate for `const Foo` doesn't resolve vars, then specialization will fail.
+
+#[const_trait]
+trait Foo {}
+
+pub trait Iterator {
+    type Item;
+}
+
+#[rustc_unsafe_specialization_marker]
+pub trait MoreSpecificThanIterator: Iterator {}
+
+pub trait Tr {
+    fn foo();
+}
+
+impl<A: const Foo, Iter> Tr for Iter
+    where
+        Iter: Iterator<Item = A>,
+{
+    default fn foo() {}
+}
+
+impl<A: const Foo, Iter> Tr for Iter
+    where
+        Iter: MoreSpecificThanIterator<Item = A>,
+{
+    fn foo() {}
+}