[Q] Template function - linking problem

[Dmitry Epstein]

Here is an example of a problem, which I tried to reduce to its
bare essentials:

// begin test1.cpp
class E {
public:
template<class T> static void f();
};

template<class T> void E::f() {}

class C {};

int main()
{
E::f<C>();
return 0;
}
// end test1.cpp


This compiles and links fine. Now I split the same code into three
different files:

// begin e.h
class E {
public:
template<class T> static void f();
};
// end e.h

// begin e.cpp
#include "e.h"

template<class T> void E::f() {}
// end e.cpp

// begin test2.cpp
#include "e.h"

class C {};

int main()
{
E::f<C>();
return 0;
}
// end test2.cpp


This compiles fine, but when I try to link (in Visual Studio 2003),
I get this error:

test error LNK2019: unresolved external symbol "public: static void
__cdecl E::f<class C>(void)" (??$f@VC@@@E@@S AXXZ) referenced in
function _main

What is wrong here?

--
Dmitry Epstein
Northwestern University, Evanston, IL. USA
mitia(at)northw estern(dot)edu

[Leor Zolman]

Re: [Q] Template function - linking problem

On 04 Apr 2004 00:48:47 GMT, Dmitry Epstein
<mitia.nospam@n orthwestern.edu .invalid> wrote:
[color=blue]
>alfps@start. no (Alf P. Steinbach) wrote in
>news:406f5750. 158635171@news. individual.net:[color=green]
>> * Dmitry Epstein <mitia.nospam@n orthwestern.edu .invalid>
>> schriebt:[color=darkred]
>>> // begin e.h
>>> class E {
>>> public:
>>> template<class T> static void f();
>>> };
>>> // end e.h
>>>
>>> What is wrong here?[/color]
>>
>> The compiler cannot instantiate the template without having
>> access to the full template definition, which you should
>> simply put in the header file.[/color]
>
>Hm, I know that I can compile a template function all by itself,
>and some sort of code is generated at that point. How does the
>compiler deal with the absence of a concrete instantiation, and why
>does it become a problem elsewhere, like in my example?[/color]

If a compiler just sees a function /template/ in a TU, with no implicit or
explicit specializations of it, then no code is generated whatsoever. If
there /are/ specializations , only the code for those is generated. (In a
class template, in fact, only member functions actually used somehow
[called, or have their addresses taken] get instantiated, unless you use
explicit instantiation.)
[color=blue]
>[color=green]
>> Another possibility is to avoid instantiation at the point of
>> usage by instantiating the template somewhere else for the
>> relevant types.
>>
>> But that restricts usage of the template to those types.[/color]
>
>This doesn't sound like what I want to do, but how would I
>instantiate the template "somewhere else"? Can you give an
>example?[/color]

For a class template C defined beginning with:

template<typena me T>
class C {
...
};

explicit instantiation would be accomplished with:

template<>
class C<int>;


You can do the same for function templates, or selected member functions of
a class template.
-leor
[color=blue]
>
>Thanks.[/color]

--
Leor Zolman --- BD Software --- www.bdsoft.com
On-Site Training in C/C++, Java, Perl and Unix
C++ users: Download BD Software's free STL Error Message Decryptor at:

https://www.bdsoft.com/tools/stlfilt.html

[Alf P. Steinbach]

Re: [Q] Template function - linking problem

* Dmitry Epstein <mitia.nospam@n orthwestern.edu .invalid> schriebt:[color=blue]
> // begin e.h
> class E {
> public:
> template<class T> static void f();
> };
> // end e.h
>
> What is wrong here?[/color]

The compiler cannot instantiate the template without having access to
the full template definition, which you should simply put in the header
file.

Another possibility is to avoid instantiation at the point of usage by
instantiating the template somewhere else for the relevant types.

But that restricts usage of the template to those types.

--
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Dmitry Epstein]

Re: [Q] Template function - linking problem

alfps@start.no (Alf P. Steinbach) wrote in
news:406f5750.1 58635171@news.i ndividual.net:[color=blue]
> * Dmitry Epstein <mitia.nospam@n orthwestern.edu .invalid>
> schriebt:[color=green]
>> // begin e.h
>> class E {
>> public:
>> template<class T> static void f();
>> };
>> // end e.h
>>
>> What is wrong here?[/color]
>
> The compiler cannot instantiate the template without having
> access to the full template definition, which you should
> simply put in the header file.[/color]

Hm, I know that I can compile a template function all by itself,
and some sort of code is generated at that point. How does the
compiler deal with the absence of a concrete instantiation, and why
does it become a problem elsewhere, like in my example?
[color=blue]
> Another possibility is to avoid instantiation at the point of
> usage by instantiating the template somewhere else for the
> relevant types.
>
> But that restricts usage of the template to those types.[/color]

This doesn't sound like what I want to do, but how would I
instantiate the template "somewhere else"? Can you give an
example?

Thanks.

--
Dmitry Epstein
Northwestern University, Evanston, IL. USA
mitia(at)northw estern(dot)edu

[Leor Zolman]

Re: [Q] Template function - linking problem

On 04 Apr 2004 00:48:47 GMT, Dmitry Epstein
<mitia.nospam@n orthwestern.edu .invalid> wrote:
[color=blue]
>alfps@start. no (Alf P. Steinbach) wrote in
>news:406f5750. 158635171@news. individual.net:[color=green]
>> * Dmitry Epstein <mitia.nospam@n orthwestern.edu .invalid>
>> schriebt:[color=darkred]
>>> // begin e.h
>>> class E {
>>> public:
>>> template<class T> static void f();
>>> };
>>> // end e.h
>>>
>>> What is wrong here?[/color]
>>
>> The compiler cannot instantiate the template without having
>> access to the full template definition, which you should
>> simply put in the header file.[/color]
>
>Hm, I know that I can compile a template function all by itself,
>and some sort of code is generated at that point. How does the
>compiler deal with the absence of a concrete instantiation, and why
>does it become a problem elsewhere, like in my example?[/color]

If a compiler just sees a function /template/ in a TU, with no implicit or
explicit specializations of it, then no code is generated whatsoever. If
there /are/ specializations , only the code for those is generated. (In a
class template, in fact, only member functions actually used somehow
[called, or have their addresses taken] get instantiated, unless you use
explicit instantiation.)
[color=blue]
>[color=green]
>> Another possibility is to avoid instantiation at the point of
>> usage by instantiating the template somewhere else for the
>> relevant types.
>>
>> But that restricts usage of the template to those types.[/color]
>
>This doesn't sound like what I want to do, but how would I
>instantiate the template "somewhere else"? Can you give an
>example?[/color]

For a class template C defined beginning with:

template<typena me T>
class C {
...
};

explicit instantiation would be accomplished with:

template<>
class C<int>;


You can do the same for function templates, or selected member functions of
a class template.
-leor
[color=blue]
>
>Thanks.[/color]

--
Leor Zolman --- BD Software --- www.bdsoft.com
On-Site Training in C/C++, Java, Perl and Unix
C++ users: Download BD Software's free STL Error Message Decryptor at:

https://www.bdsoft.com/tools/stlfilt.html