On Jan 24, 2008 3:26 PM, Tobias Schwinger <
tschwinger@isonews2.com> wrote:
[...]
The wording is a bit ambiguous: A pointer to an /incomplete type/ is in
fact OK (using standard terminology), but 'TreeImplBase' is a /template
id/ which is not a type at all.
As Ovanes pointed out that construct is illegal, however (except inside
a definition of 'TreeImplBase' where it's the /injected class name/).
/Specializing/ a template with template arguments yields a type. Using
(whatever) member of this type's interface causes the template to be
implicitly (possibly partially) /instantiated/. It's most important to
realize that specializing a template does not cause the template to be
instantiated automatically.
Note that "template specialization" has a context-dependent meaning, as
it can either refer to a type or of a class (template) definition of a
variant implementation of the template:
// /primary template/
template< typename T > struct A { ... };
// definition of the (full) specialization A<int>
template< > struct A<int> { ... };
// definition of a partial specialization
template< typename T > struct A< B<T> >
{
typedef A self; // /injected class name/ is a type
};
// X and Y are specializations (not instantiations)!
typedef A< B<int> > X;
typedef A< long > Y;
int main()
{
X x; // instantiation of the 'B<int>'-specialization of A
// at this point
// ...
OK, that's about template terminology in five minutes :-).
Regards,
Tobias