Template compilation

Template compilation

[Alexander Sclearuc]

Hi,

I try to compile a simple template example (see attachments)
and I get the next error:
    Undefined			first referenced
      symbol  			    in file
    ostream & operator<<<int>(ostream &, array1<int> const &)template.o
    a.out: fatal error: Symbol referencing errors. No output written to a.out
    collect2: ld returned 1 exit status

My system is SCO 5.0.2 and I use egcs-1.0.2 (build by me too:(
with no flags on command line

The standard C++ compiler for SCO (CC) compiles this example with
no problems.

What is the problem? 

PS. I haven't programmed in C++ for more than 2 years, so, may
be I really forgot C++;) But not the basics;)

-- 
Alexander Sclearuc                      Phone: (373) 2 21 20 96
NetInfo S.R.L.				Fax:   (373) 2 21 20 96
Chisinau 				       (373) 2 24 00 83
Moldova, Republic of			mailto:avs@netinfo.moldova.net

Re: Template compilation

[JC Loredo-Osti]

On Thu, 7 May 1998, Alexander Sclearuc wrote:
> I try to compile a simple template example (see attachments)
> and I get the next error:
>     Undefined			first referenced
>       symbol  			    in file
>     ostream & operator<<<int>(ostream &, array1<int> const &)template.o
>     a.out: fatal error: Symbol referencing errors. No output written to a.out
> 

If you put your template1.C file in template1.h, you would compile without
problems. (g++ needs to see everything, I guess). Some thing like,

template1.h
------------
ifndef __template1_h
#define __template1_h


#include <iostream.h>

template<class T>
class array1 {
...
};

template<class T>
inline ostream& operator<<(ostream& os, const array1<T>& a) {
  for (int i = 0; i < 4; i++)
    os << a[i] << " ";
  return os;
}

#endif

--------

-j

email: josti@mscs.dal.ca

Re: Template compilation

[Alexandre Oliva]

Alexander Sclearuc <avs@netinfo.moldova.net> writes:

> [1  <text/plain; US-ASCII (7bit)>]
> Hi,

> I try to compile a simple template example (see attachments)
> and I get the next error:
>     Undefined			first referenced
>       symbol  			    in file
>     ostream & operator<<<int>(ostream &, array1<int> const &)template.o

Every template definition must be available whenever it is used for it
to be implicitly instantiated (unless the template is defined with the
`export' keyword, which is not implemented yet, anyway).  So, move the
contents of template1.C into template1.h and it should work.

> The standard C++ compiler for SCO (CC) compiles this example with
> no problems.

It probably uses some non-standard template repository mechanism.

-- 
Alexandre Oliva
mailto:oliva@dcc.unicamp.br mailto:aoliva@acm.org
http://www.dcc.unicamp.br/~oliva
Universidade Estadual de Campinas, SP, Brasil

Re: Re: Template compilation

[Alexander Sclearuc]

Alexandre Oliva wrote:
> > I try to compile a simple template example (see attachments)
> > and I get the next error:
> >     Undefined			first referenced
> >       symbol  			    in file
> >     ostream & operator<<<int>(ostream &, array1<int> const &)template.o
> 
> Every template definition must be available whenever it is used for it
> to be implicitly instantiated (unless the template is defined with the
> `export' keyword, which is not implemented yet, anyway).  So, move the
> contents of template1.C into template1.h and it should work.

Thank you for your answer, this variant works.

But I still have a question.
I compiled recently with MSVC one library and all was OK.
The README file says that the library could be compiled with
EGCS-1.0.2, so I tried to compile the library with it.
And I get a lot of errors like one pointed above.
It took me almost a day to understand what the problem is;)

I don't want to change all library sources and
I would like to know what the C++ standard says 
about templates (from this point of view)

> > The standard C++ compiler for SCO (CC) compiles this example with
> > no problems.
> It probably uses some non-standard template repository mechanism.
Yes, it is a Cfront and it creates a repository directory 

-- 
Alexander Sclearuc                      Phone: (373) 2 21 20 96
NetInfo S.R.L.				Fax:   (373) 2 21 20 96
Chisinau 				       (373) 2 24 00 83
Moldova, Republic of			mailto:avs@netinfo.moldova.net

Re: Template compilation

[Martin von Loewis]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1367 bytes --]

> I don't want to change all library sources and I would like to know
> what the C++ standard says about templates (from this point of view)

[temp]/6 says

>> A namespace­scope declaration or definition of a non­inline
>> function template, a non­inline member function template, a
>> non­inline member function of a class template or a static data
>> member of a class template may be preceded by the export
>> keyword. If such a template is defined in the same translation unit
>> in which it is declared as exported, the definition is considered
>> to be *exported*.

Since you do not use the export keyword, your templates are not
exported.

With this definition, [temp]/8 says

>> A non­exported template that is neither explicitly specialized nor
>> explicitly instantiated must be defined in every translation unit
>> in which it is implicitly instantiated (14.7.1) or explicitly
>> instantiated (14.7.2); no diagnostic is required.

Since you instantiate templates which you didn't export, you must
define them in the translation unit were they are implicitly
instantiated.

So, your library is wrong, and egcs is right. MSVC is also right,
since it does not need to diagnose this violation.

As Alexandre points out, the export keyword is currently not supported
by egcs, so you'll have to define everything you instantiate.

Regards,
Martin

Re: Template compilation

[Robert Lipe]

> I try to compile a simple template example (see attachments)
> and I get the next error:

Others that know C++ and g++ addressed this question.   I'll just
throw in some SCO-specific trivia for free.

> My system is SCO 5.0.2 and I use egcs-1.0.2 (build by me too:(
[ munch ] 
> The standard C++ compiler for SCO (CC) compiles this example with
> no problems.

Assuming you're using the CC that ships as part of the SCO OpenServer
Development System and not the CC that ships as part of the (new) U/ODK,
you're talking about a Cfront mutant that is mostly of early 1990's 
vintage.

I don't think it's at all safe to consider this compiler anything 
resembling authoritative regarding C++ as it exists in the late '90's.
There are surely reasons why everyone, including SCO defected from 
Cfront.