Fix crash when an incorrect redeclaration only differs in __unaligned type-qualifier

[rogfer01]

Summary

This patch fixes an assertion that is hit when a redeclaration with differing types only differs in the unaligned type-qualifier.

[rjmccall]

LGTM.

[rogfer01]

Ping?

[rjmccall]

In D29986#684811, @rogfer01 wrote:

Ping?

Are you expecting something even more conclusive than "looks good to me"? Because I sent you that feedback a week ago. :)

[rogfer01]

Sorry for the noise. I was expecting some more formal approval in the usual form of "Ready to land" but if you think it is OK I will commit it shortly.

Regards.

cfe/trunk/lib/AST/ASTContext.cpp

if (getLangOpts().OpenCL) {
         return LHS;
       if (RQuals.isAddressSpaceSupersetOf(LQuals))
         return RHS;
     }
     // If any of these qualifiers are different, we have a type
     // mismatch.
     if (LQuals.getCVRQualifiers() != RQuals.getCVRQualifiers() ||
         LQuals.getAddressSpace() != RQuals.getAddressSpace() ||
-        LQuals.getObjCLifetime() != RQuals.getObjCLifetime())
+        LQuals.getObjCLifetime() != RQuals.getObjCLifetime() ||
+        LQuals.hasUnaligned() != RQuals.hasUnaligned())
       return QualType();
 
     // Exactly one GC qualifier difference is allowed: __strong is
     // okay if the other type has no GC qualifier but is an Objective
     // C object pointer (i.e. implicitly strong by default).  We fix
     // this by pretending that the unqualified type was actually
     // qualified __strong.
     Qualifiers::GC GC_L = LQuals.getObjCGCAttr();

cfe/trunk/test/Sema/unaligned-qualifier.c

+// RUN: %clang_cc1 %s -verify -fsyntax-only -fms-extensions
+
+int __unaligned * p1; // expected-note {{previous definition is here}}
+int * p1; // expected-error {{redefinition of 'p1' with a different type: 'int *' vs '__unaligned int *'}}