[cfe-commits] r84140 - in /cfe/trunk: include/clang/AST/Decl.h lib/AST/Decl.cpp lib/CodeGen/CodeGenModule.cpp lib/Sema/SemaTemplateInstantiateDecl.cpp test/CXX/temp/temp.spec/temp.explicit/p1-emit.cpp

[Douglas Gregor]

On Oct 25, 2009, at 3:18 PM, Daniel Dunbar wrote:

> Hi Doug,
>
> On Wed, Oct 14, 2009 at 2:29 PM, Douglas Gregor <dgregor at apple.com>  
> wrote:
>> Author: dgregor
>> Date: Wed Oct 14 16:29:40 2009
>> New Revision: 84140
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=84140&view=rev
>> Log:
>> Give explicit and implicit instantiations of static data members of
>> class templates the proper linkage.
>>
>> Daniel, please look over the CodeGenModule bits.
>
> Comments below.
>
>> --- cfe/trunk/lib/CodeGen/CodeGenModule.cpp (original)
>> +++ cfe/trunk/lib/CodeGen/CodeGenModule.cpp Wed Oct 14 16:29:40 2009
>> @@ -541,7 +541,12 @@
>>     }
>>   }
>>
>> -  return VD->getStorageClass() == VarDecl::Static;
>> +  // Static data may be deferred, but out-of-line static data  
>> members
>> +  // cannot be.
>> +  // FIXME: What if the initializer has side effects?
>
> For C++ we shouldn't be deferring anything that could have side
> effects. Can you fix or file this FIXME?

Filed at http://llvm.org/bugs/show_bug.cgi?id=5308

>> +  return VD->isInAnonymousNamespace() ||
>> +         (VD->getStorageClass() == VarDecl::Static &&
>> +          !(VD->isStaticDataMember() && VD->isOutOfLine()));
>>  }
>
>> +static CodeGenModule::GVALinkage
>> +GetLinkageForVariable(ASTContext &Context, const VarDecl *VD) {
>> +  // Everything located semantically within an anonymous namespace  
>> is
>> +  // always internal.
>> +  if (VD->isInAnonymousNamespace())
>> +    return CodeGenModule::GVA_Internal;
>> +
>> +  // Handle linkage for static data members.
>> +  if (VD->isStaticDataMember()) {
>> +    switch (VD->getTemplateSpecializationKind()) {
>> +    case TSK_Undeclared:
>> +    case TSK_ExplicitSpecialization:
>> +    case TSK_ExplicitInstantiationDefinition:
>> +      return CodeGenModule::GVA_StrongExternal;
>> +
>> +    case TSK_ExplicitInstantiationDeclaration:
>> +      assert(false && "Variable should not be instantiated");
>> +      // Fall through to treat this like any other instantiation.
>
> I think we are moving towards using llvm_unreachable for these.

I didn't even realize that existed :) Thanks!

	- Doug