Really bizarre linking error!

[bajichuan]

Hello! I have the world's strangest linking error, and I'm hoping that
someone can help me sort it out. I recently installed and compiled a
library called LinBox without a problem. I have an object-oriented
software application, and I want it to call the library. When I add
the following 3 lines (copied directly from the library tutorial),

#include <linbox/field/modular.h>
using namespace LinBox;
typedef Modular<shortFi eld;

to a .cpp file, my code compiles perfectly. However, when I add the
same 3 lines to a .h file (where they *must* ultimately go), I get
(among others) the following error:

/tmp/cckvjpY5.o: In function `~edge':/usr/include/gmp.h:1590: multiple
definition of
`LinBox::Modula rBase<Integer>: :write(std::bas ic_ostream<char ,
std::char_trait s<char&) const'

Comments? This is incredibly strange to me. It looks like some sort of
linking error. Namespace conflict? I don't have root access on the
school machines, and so I have compiled and am pointing to a local
copy of the LinBox library. This is so strange... why does the code
work from
the .cpp file but not the .h file?

I hope this is an error in my code and not in the library itself.
Thoughts and suggestions on how to track this down would be carefully
considered and gratefully appreciated!

Very best regards,
Susan

PSHere is a copy of my very simple make file:

grob_heur: gc_heur.cpp Graph.cpp MonExpEnum.cpp SetCombEnum.cpp
GraphLinearSyst e
m.cpp
g++ -Wall -g \
-I/LINBOX_PATH/linbox-1.1.3/include \
-I/usr/include \
-o gc_heur gc_heur.cpp \
Graph.cpp MonExpEnum.cpp SetCombEnum.cpp GraphLinearSyst em.cpp
\
/LINBOX_PATH/linbox-1.1.3/linbox/util/gmp++/.libs/libgmpxx.a \
/usr/lib/libgmp.a \
/LINBOX_PATH/linbox-1.1.3/lib/liblinbox.a

[bajichuan]

Because you have managed to include the header file in the same cpp file

twice, or you have included the header file in the cpp file where you
manually placed the typedef.

Yes, certainly that is the obvious first guess. Unfortuneatly, all of
my header files do have the traditional guards

#ifndef MY_HEADER
#define MY_HEADER

Also, all of the library files have the traditional guards as
expected. Also, as Arun suggested, I am not double declaring the
typedef, although I certainly double checked just to be sure, because
that is another obvious error.

It looks like you're including the header in (at least) two different
source files (not surprising -- that's the usual reason for putting
something into a header). From the looks of things, in modular.h, you
have a _definition_ (not just a declaration) of a function named
'write' that's part of your ModularBase class. That would be all right
IF it was an inline function, but it's apparently not.

I find this comment very interesting. Here is the function itself, as
declared within the library header file (which has guards).

template <>
std::ostream& ModularBase<Int eger>::write (std::ostream &os)
const
{ return os << "GMP integers mod " << _modulus; }

Obviously, if I comment out this function within the library header
file, the code compiles fine, but of course that is not the right
solution AT ALL. Does this function declaration look okay? I think
this is a function definition, not a declaration. Isn't it also
inline? A shouldn't the guards around the header file in the library
prevent any double declaration?

Thanks for your comments!

Susan

[Ian Collins]

ba*******@hotma il.com wrote:

Please don't snip attributions, it is considered rude.

>It looks like you're including the header in (at least) two different
source files (not surprising -- that's the usual reason for putting
something into a header). From the looks of things, in modular.h, you
have a _definition_ (not just a declaration) of a function named
'write' that's part of your ModularBase class. That would be all right
IF it was an inline function, but it's apparently not.

I find this comment very interesting. Here is the function itself, as
declared within the library header file (which has guards).

template <>
std::ostream& ModularBase<Int eger>::write (std::ostream &os)
const
{ return os << "GMP integers mod " << _modulus; }

Obviously, if I comment out this function within the library header
file, the code compiles fine, but of course that is not the right
solution AT ALL. Does this function declaration look okay? I think
this is a function definition, not a declaration.

It is.

Isn't it also inline?

No, only class member functions defined inline have inline linkage, not
normal functions.

Your original linker error shows that this function is not declared as
inline and is should be.

A shouldn't the guards around the header file in the library
prevent any double declaration?

They can't.

--
Ian Collins.

[Jerry Coffin]

In article <11************ **********@x40g 2000prg.googleg roups.com>,
ba*******@hotma il.com says...

[ ... ]

It looks like you're including the header in (at least) two different
source files (not surprising -- that's the usual reason for putting
something into a header). From the looks of things, in modular.h, you
have a _definition_ (not just a declaration) of a function named
'write' that's part of your ModularBase class. That would be all right
IF it was an inline function, but it's apparently not.

I find this comment very interesting. Here is the function itself, as
declared within the library header file (which has guards).

template <>
std::ostream& ModularBase<Int eger>::write (std::ostream &os)
const
{ return os << "GMP integers mod " << _modulus; }

No -- to be inline, it has to either be written inside of the class
definition, or else explicitly defined as 'inline'. You haven't shown us
the remainder of the class definition to be absolutely certain, but
qualifying the name (i.e. using "ModularBase<In teger>::write" instead of
just "write") suggests that this function is being defined outside of
the class definition.

Obviously, if I comment out this function within the library header
file, the code compiles fine, but of course that is not the right
solution AT ALL. Does this function declaration look okay? I think
this is a function definition, not a declaration. Isn't it also
inline? A shouldn't the guards around the header file in the library
prevent any double declaration?

No -- they prevent it from being declared twice in the same translation
unit (pre-processed source file). It still attempts to define the
function once in each source file where you include the header -- but
unless it's inline, you can only define it once across all the source
files that make up the complete program.

--
Later,
Jerry.

The universe is a figment of its own imagination.

[bajichuan]

Making the function inline did the trick!!!! Thank you so much. I will
let the people who put out the library know about the error.

Thanks again for your help and comments!

Best,
Susan

[Jack Klein]

On Sun, 05 Aug 2007 18:58:19 -0000, "ar********@gma il.com"
<ar********@gma il.comwrote in comp.lang.c++:

Why are you top posting? Did you not read the signature block in
Alf's post that explains why it is wrong?

Yup this is a multiple decleration problem and is completly to do with
ur program and nothing to do with the library.

Why do you refer to ur's program, when the original poster's name was
"arun.darra ", and the post you replied to was from "Alf P. Steinbach"?
Did you reply to the wrong post by mistake? Actually, since you seem
to be implying that Alf's code is incorrect, even though he is not the
one who posted it, you are replying to the wrong post.

there are 2 reasons for this:

1. because ur programs .h files dont have header guards
and the same header is being included multiple times.
use heade guards in ur header file, some thing like:

# ifndef _FILE_
# define _FILE_

This, of course, is completely wrong because it violates a namespace
reserved for the implementation. All identifiers beginning with an
underscore followed by an upper case letter, and all identifiers
containing two consecutive underscores anywhere within them, are
reserved in all contexts.

--
Jack Klein
Home: http://JK-Technology.Com
FAQs for
comp.lang.c http://c-faq.com/
comp.lang.c++ http://www.parashift.com/c++-faq-lite/
alt.comp.lang.l earn.c-c++
http://www.club.cc.cmu.edu/~ajo/docs/FAQ-acllc.html

[arun.darra]

This, of course, is completely wrong because it violates a namespace

reserved for the implementation. All identifiers beginning with an
underscore followed by an upper case letter, and all identifiers
containing two consecutive underscores anywhere within them, are
reserved in all contexts.

Well thanks for the info Jack but the idea was show the usage of
header guards... But the point u put across was well taken.. thanks.

[bhadorn]

On 5 Aug., 18:47, bajich...@hotma il.com wrote:

Hello! I have the world's strangest linking error, and I'm hoping that
someone can help me sort it out. I recently installed and compiled a
library called LinBox without a problem. I have an object-oriented
software application, and I want it to call the library. When I add
the following 3 lines (copied directly from the library tutorial),

#include <linbox/field/modular.h>
using namespace LinBox;
typedef Modular<shortFi eld;

to a .cpp file, my code compiles perfectly. However, when I add the
same 3 lines to a .h file (where they *must* ultimately go), I get
(among others) the following error:

/tmp/cckvjpY5.o: In function `~edge':/usr/include/gmp.h:1590: multiple
definition of
`LinBox::Modula rBase<Integer>: :write(std::bas ic_ostream<char ,
std::char_trait s<char&) const'

Hi,
I've got the same linking error not using "inline". See

http://www.xatlantis.ch/cpp_blog/compiler.html

wbr

bhadorn