explicit instantiation of methods bug.

explicit instantiation of methods bug.

[Andrew Fitzgibbon]

Hi,

I occasionally explicitly instantiate single methods of a template class,
e.g. Given
   
   template<class Type>
   struct A {
     ....
     bool f (A<Type>& u, A<Type>& w, A<Type>& v) const { }
   };

   template bool A<double>::f (A< double >&, A< double >&, A< double >&) const;

I want to instantiate f and not anything else in A.

egcs gives

/users33/awf/etest.cc:7: no matching template for `A<double>::f(A<double> &, A<double> &, A<double> &) const' found

But CD2 (the latest std I have access to), section 14.7.2.8 contains the phrase
   
   and the template may be a [member template or] 
   member function [which would not normally be accessible]
   
which would seem to imply that it ought ot be allowed, as in earlier gcc's.

Re: explicit instantiation of methods bug.

[Mark Schaefer]

	Andrew,

	I'm not sure about the state of egcs, but what you are doing could
easily be considered a partial specialization.  The SGI compiler, for instance,
would try to interpret your function as "Here's how I want to define A::f for
type double"

	Also, from the snippet of the standard you gave, what you are doing is
neither a member template definition, nor a member function definition.  You
should know that you can't declare a member function outside of the class
declaration so what does it fall into?

	I'm not sure where I stand on template instantiation.  According to
Stroustrup, the compiler should be smart enough to instantiate the proper
template code without operator intervention.  I consider it cheesy to insist
(like gcc 7.2 did, at least) that the user explicitly tell the compiler when he
planned to instantiate a template.

	I think a lot of compilers (SGI included) have a very hard time with
template dependencies, especially when the instantiation comes from a library
function call.

--
Mark Schaefer