[PATCH] Sema: Implement DR244

[Richard Smith]

================
Comment at: lib/Sema/SemaExprCXX.cpp:136-137
@@ -135,4 +135,4 @@
     //
     //   the class-names are looked up as types in the scope designated by
     //   the nested-name-specifier.
     //
----------------
Please update this to match the standard (in particular, "the second class-name is looked up in the same scope as the first").

================
Comment at: lib/Sema/SemaExprCXX.cpp:161-167
@@ -158,11 +160,9 @@
       isDependent = isDependentScopeSpecifier(PrefixSS);
     } else if (ObjectTypePtr) {
       LookupCtx = computeDeclContext(SearchType);
       isDependent = SearchType->isDependentType();
     } else {
       LookupCtx = computeDeclContext(SS, EnteringContext);
       isDependent = LookupCtx && LookupCtx->isDependentContext();
     }
   } else if (ObjectTypePtr) {
----------------
By a strict reading of the standard, I think:

  namespace N {
    struct A {
      struct B {};
    };
    A *a;
    A::B *b;
  }
  void f() {
    a->N::~A(); // ok
    b->N::A::~B(); // ill-formed, B is *only* looked up in N
  }

... but see also core issue 399. I think our behavior here (with your patch) is appropriate for now.

================
Comment at: test/CXX/drs/dr2xx.cpp:1016
@@ -1015,3 +1015,2 @@
   B::A() {} // ok
-  C::~C() {} // expected-error {{expected the class name}}
   C::~A() {} // ok
----------------
Why was this deleted? If we no longer reject this, that's a regression, and we can no longer claim to implement DR298...

http://reviews.llvm.org/D3583