[cfe-users] compiler bug or language feature?

[Daniel Schwalbe]

Hmm, if I do this it gives me another error:

test.cpp:15:42: error: 'SmartPtrType' following the 'template' keyword 
does not refer to a template
  ~SmartUnion() {m_sptr.SmartPtrType<T>::~SmartPtrType<T>(); }

Again, g++ takes it without complaining.

Since I noticed that my original attachment got scrubbed I am pasting 
the modified test program directly. Perhaps the context will help to 
sort things out:

#include <memory>
#include <string>
using namespace std;

//comment this in to see the compiler error
#define show_compiler_error 1

template<typename T, template<class> class SmartPtrType = shared_ptr>
struct SmartUnion
{
     SmartUnion() {new(&m_sptr) SmartPtrType<T>();}


#ifdef show_compiler_error
  ~SmartUnion() {m_sptr.SmartPtrType<T>::~SmartPtrType<T>(); }
#else
  typedef SmartPtrType<T> smartptr_type;
   ~SmartUnion() {m_sptr.smartptr_type::~smartptr_type(); }
#endif

  union {
      SmartPtrType<T> m_sptr;
      T * m_ptr;
  };

};

using namespace std;
int main()
{
     SmartUnion<string,shared_ptr> smu;

     return 0;
}


Am 08.10.2013 20:08, schrieb Richard Smith:
> You need to write
>
>   m_sptr.SmartPtrType<T>::~SmartPtrType*<T>*();
>
> Per 3.4.3/6, "in a qualified-id of the form nested-name-specifier[opt] 
> class-name::~class-name, the second class-name is looked up in the 
> same scope as the first". So you can't use the injected-class-name 
> after the ~, because it's not visible.
>
> On Tue, Oct 8, 2013 at 10:05 AM, Daniel Schwalbe <dansch491 at gmail.com 
> <mailto:dansch491 at gmail.com>> wrote:
>
>     When I add the 'template' keyword as suggested clang gives me the
>     very same error. g++ still eats it without problem.
>
>     Am 08.10.2013 18:23, schrieb David Blaikie:
>>
>>
>>
>>     On Tue, Oct 8, 2013 at 8:36 AM, Daniel Schwalbe
>>     <dansch491 at gmail.com <mailto:dansch491 at gmail.com>> wrote:
>>
>>         Hi,
>>
>>         I don't know if I encountered a compiler bug in clang 3.3 or
>>         if this is one of the many subtle c++ exceptional cases:
>>
>>         When I try to compile the attached file (using: clang++
>>         -std=c++11 test.cpp) I receive a compiler error:
>>
>>         "test.cpp:15:42: error: identifier 'SmartPtrType' in object
>>         destruction expression does not name a type
>>          ~SmartUnion() {m_sptr.SmartPtrType<T>::~SmartPtrType(); }"
>>
>>         g++ compiles the file without problem.
>>
>>         clang compiles it, too, if I explicitly introduce a typedef
>>         for SmartPtrType<T> and call this typedef name instead:
>>
>>         typedef SmartPtrType<T> smartptr_type;
>>         ~SmartUnion() {m_sptr.smartptr_type::~smartptr_type(); }
>>
>>         Is this a subtle language feature or a compiler bug?
>>
>>
>>     I believe this is a language feature because the expression is
>>     ambiguous if m_sptr is type dependent (which I assume it is)
>>
>>     The other solution, rather than introducing a typedef, is to use
>>     the 'template' keyword:
>>
>>     ~SmartUnion() {m_sptr.template SmartPtrType<T>::~SmartPtrType(); }
>>
>>
>>
>>         Daniel
>>
>>
>>
>>
>>         _______________________________________________
>>         cfe-users mailing list
>>         cfe-users at cs.uiuc.edu <mailto:cfe-users at cs.uiuc.edu>
>>         http://lists.cs.uiuc.edu/mailman/listinfo/cfe-users
>>
>>
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-users/attachments/20131009/27893248/attachment.html>