Auto merge of #27258 - nikomatsakis:issue-26952, r=eddyb

Correct regression in type-inference caused by failing to reconfirm that
the object trait matches the required trait during trait selection.  The
existing code was checking that the object trait WOULD match (in a
probe), but never executing the match outside of a probe.

This corrects various regressions observed in the wild, including
issue #26952. Fixes #26952.

r? @eddyb 
cc @frankmcsherry

src/test/run-pass/infer-from-object-trait-issue-26952.rs

@@ -0,0 +1,33 @@
+// Copyright 2012 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// Test that when we match a trait reference like `Foo<A>: Foo<_#0t>`,
+// we unify with `_#0t` with `A`. In this code, if we failed to do
+// that, then you get an unconstrained type-variable in `call`.
+//
+// Also serves as a regression test for issue #26952, though the test
+// was derived from another reported regression with the same cause.
+
+use std::marker::PhantomData;
+
+trait Trait<A> { fn foo(&self); }
+
+struct Type<A> { a: PhantomData<A> }
+
+fn as_trait<A>(t: &Type<A>) -> &Trait<A> { loop {  } }
+
+fn want<A,T:Trait<A>+?Sized>(t: &T) { }
+
+fn call<A>(p: Type<A>) {
+    let q = as_trait(&p);
+    want(q); // parameter A to `want` *would* be unconstrained
+}
+
+fn main() { }