eagerly prove WF when resolving fully qualified paths

Author: lcnr

compiler/rustc_hir_typeck/src/fn_ctxt/_impl.rs

@@ -798,13 +798,16 @@ impl<'a, 'tcx> FnCtxt<'a, 'tcx> {
                 bug!("`resolve_ty_and_res_fully_qualified_call` called on `LangItem`")
             }
         };
+
+        self.register_wf_obligation(
+            ty.raw.into(),
+            qself.span,
+            ObligationCauseCode::WellFormed(None),
+        );
+        self.select_obligations_where_possible(|_| {});
+
         if let Some(&cached_result) = self.typeck_results.borrow().type_dependent_defs().get(hir_id)
         {
-            self.register_wf_obligation(
-                ty.raw.into(),
-                qself.span,
-                ObligationCauseCode::WellFormed(None),
-            );
             // Return directly on cache hit. This is useful to avoid doubly reporting
             // errors with default match binding modes. See #44614.
             let def = cached_result.map_or(Res::Err, |(kind, def_id)| Res::Def(kind, def_id));
@@ -849,14 +852,6 @@ impl<'a, 'tcx> FnCtxt<'a, 'tcx> {
                 result
             });
 
-        if result.is_ok() {
-            self.register_wf_obligation(
-                ty.raw.into(),
-                qself.span,
-                ObligationCauseCode::WellFormed(None),
-            );
-        }
-
         // Write back the new resolution.
         self.write_resolution(hir_id, result);
         (

tests/ui/traits/next-solver/method/path_lookup_wf_constraints.rs

@@ -0,0 +1,39 @@
+//@ compile-flags: -Znext-solver
+//@ check-pass
+
+// A regression test for trait-system-refactor-initiative#161
+
+trait Constrain<T> {
+    type Assoc;
+}
+impl<T> Constrain<T> for () {
+    type Assoc = ();
+}
+struct Foo<T, U = <() as Constrain<T>>::Assoc>(T, U);
+
+impl<T: Copy> Foo<T> {
+    fn foo() {}
+}
+struct B;
+impl Foo<B> {
+    fn foo() {}
+}
+
+type Alias<T> = Foo<T>;
+fn via_guidance<T: Copy>()
+where
+    (): Constrain<T>,
+{
+    // Method selection on `Foo<?t, <() as Constrain<?t>>::Assoc>` is ambiguous.
+    // only by unnecessarily constraining `?t` to `T` when proving `(): Constrain<?t>`
+    // are we able to select the first impl.
+    //
+    // This happens in the old solver when normalizing `Alias<?t>`. The new solver doesn't try
+    // to eagerly normalize `<() as Constrain<?t>>::Assoc` so we instead always prove that the
+    // self type is well-formed before method lookup.
+    Alias::foo();
+}
+
+fn main() {
+    via_guidance::<()>();
+}