Victor Bazarov schrieb:
Arne Petersen wrote: Hy,
Hy yourself.
I've got a problem with member function templates compiled into
libraries.
I'm trying to get a library collection (coded for GNU gcc, where its
compiled completly) being compiled on Visual Studio .NET. The problem is
that for gcc the template functions (members of a class) are explicitly
instantiated in a cpp file that includes the classes .h and .cpp file.
Let me understand, the template functions are instantiated in a cpp file
that includes (among other things) the cpp file, right?
The VS compiler does not correctly bring together the function template
declaration and its instanciation.
What's "incorrect" about it? Could you be a bit more verbose here?
An other problem is that in VS it's not possible to declare just the
generic function template in the header and define the specialization in
the .cpp file. At least a declaration of the specialized template has to
be written in the header.
It's not possible in any compiler that exists today that doesn't implement
"export" keyword (and I'm yet to see one that does).
Now how can I get GNU version libraries be compiled under VS?
If you need a compiler-specific solution, you need to ask in a newsgroup
dedicated to that compiler. However, it is usually enough to instantiate
templates explicitly to place them into the library:
-------------------- some.h
template<class T> void foo(T t);
-------------------- some.cpp
#include <some.h>
template<class T> void foo(T t) {
42;
}
template void foo(int);
template void foo(double);
template void foo(const char*);
#include <myspecialclass.h>
template void foo(MySpecialClass&);
------------------------------------------
All those declarations that begin with 'template' without the argument
specification after it are the explicit instantiations. When the compiler
sees an explicit instantiation, it generates code for that function. Next
time there is a call to that function, the compiler doesn't need to see
the body of the function, it should be resolved at link time.
The advantage of this is that you can put the implementation in a separate
C++ translation unit, thus hiding it from the user of the template. Also,
the compiler doesn't waste time compiling the template definition every
time it sees that it's used.
The disadvantage if it is that you have to be able to predict for which
template arguments your template is going to be used (and make that list
exhaustive). In the example above, if I ever try using foo(100L), then
I got a problem: unresolved external "void foo<long>(long);" because there
is no explicit instantiation for it in my library, and the compiler cannot
instantiate it at the time of use because I didn't provide the definition.
Victor
Thanks for reply!
The problem I have is that the function temlpate is a class member. The
template specialization (for int) is instanciatet correctly but the
explicit instanciation (for double) for the generic template goes wrong.
----------------------------hello1.h
class klasse {
public:
klasse(){};
template<class T> int blah(T foo);
template <> int blah<int>(int foo);
};
----------------------------hello1.h
----------------------------hello1.cpp
#include "hello1.h"
#include <iostream>
template<class T> int klasse::blah(T foo){
std::cout<<"blah"<<std::endl;
return (int)0;
}
//correct implicite instanciation
template<> int klasse::blah<int>(int foo){
std::cout<<"integer"<<std::endl;
return (int)1;
}
//incorrect explicit instanciation
template int klasse::blah(double foo);
----------------------------hello1.cpp
Maybe this problem depends on the used compiler. So if you know a good
ngroup for the Visual Studio .NET compiler I would be thankfull.
Arne
