gcc-3.4 requires 'typename' to use map in a template?

gcc-3.4 requires 'typename' to use map in a template?

[Dan Kegel]

The following code is rejected by gcc-3.4-20031001, but accepted
by gcc-2.95 and gcc-3.2.3:

#include <map>

template <typename T>
class myclass {
         typedef std::map<T, unsigned int> mymap;
public:
         void mymethod() {
             mymap::iterator myvar;
             myvar = 0;
         }
};

Command to reproduce:
$ /usr/local/gcc-3.4-20031001/bin/g++ foo2.cc
foo2.cc: In member function `void myclass<T>::mymethod()':
foo2.cc:8: error: expected `;'
foo2.cc:9: error: `myvar' undeclared (first use this function)

Using the typename keyword seems to be a workaround, as
gcc-3.4-20031001 accepts the following code without complaint:

#include <map>

template <typename T>
class myclass {
         typedef std::map<T, unsigned int> mymap;
public:
         void mymethod() {
             typename mymap::iterator myvar;
             myvar = 0;
         }
};

So is gcc-3.4-20031001 right to require typename here, or is this a bug?
- Dan

Re: gcc-3.4 requires 'typename' to use map in a template?

[Dan Kegel]

Dan Kegel wrote:
> #include <map>
> 
> template <typename T>
> class myclass {
>         typedef std::map<T, unsigned int> mymap;
> public:
>         void mymethod() {
>             typename mymap::iterator myvar;
>             myvar = 0;
>         }
> };
> 
> So is gcc-3.4-20031001 right to require typename here, or is this a bug?

The local language lawyer says yep, gcc-3.4 could be right;
since mymap depends on T, it needs the 'typename'.
But shoot, you'd *think* a typedef would be enough of a clue!
Man, is this nonintuitive...

I guess I'll run out and buy that book _C++ Templates: The Complete Guide_
mentioned in the previous post,
http://gcc.gnu.org/ml/gcc/2003-10/msg00082.html
- Dan

Re: gcc-3.4 requires 'typename' to use map in a template?

[David Carlton]

On Thu, 02 Oct 2003 17:29:40 -0700, Dan Kegel <dkegel@ixiacom.com> said:
> Dan Kegel wrote:

>> #include <map>
>> template <typename T>
>> class myclass {
>> typedef std::map<T, unsigned int> mymap;
>> public:
>> void mymethod() {
>> typename mymap::iterator myvar;
>> myvar = 0;
>> }
>> };
>> So is gcc-3.4-20031001 right to require typename here, or is this a
>> bug?

> The local language lawyer says yep, gcc-3.4 could be right;
> since mymap depends on T, it needs the 'typename'.
> But shoot, you'd *think* a typedef would be enough of a clue!

My understanding is that it's not that it needs help figuring out that
mymap is a type, it needs help figuring out that mymap::iterator is a
type.  (Which it might or might not be for different values of T, if
you inserted some particularly perverse specializations of std::map.)

David Carlton
carlton@kealia.com

Re: gcc-3.4 requires 'typename' to use map in a template?

[Dan Kegel]

David Carlton wrote:
> My understanding is that it's not that it needs help figuring out that
> mymap is a type, it needs help figuring out that mymap::iterator is a
> type.  (Which it might or might not be for different values of T, if
> you inserted some particularly perverse specializations of std::map.)

Ahh.  It makes sense.  Thanks!
- Dan