MingW and GNU MP?

I've been trying to get GMP and MingW to play ball for the last couple of hours, but it's been a struggle.

I finally got GMP installed under Windows XP, but now i've hit another problem.

I'm trying to compile something simple like this:

Code:

---

#include <gmp.h>

int main()
{
    mpz_t sum;
    return 0;
}

---

Basically a program that will try out one of GMPs data types, but the program only works if i put the entire path between the '<'s, like <C:\GMP\include\gmp.h>.

How do i avoid this, the gmp documentation doesn't do this, and i don't wan't to either?

Btw, the documentation tells me to do some linking, (-lgmpxx and -lgmp), but when i try this, nothing really happens, i suspect this to be the problem. What do i need to do to get it right?

Quote:

---

nothing really happens

---

The segment of code you posted doesn't do anything. "nothing really happens" would be the correct behavior.

As for having to use the full path, gmp.h isn't in your include path. You can pass -IC:\GMP\include *dash - uppercase 'eye' - path" to the compiler, and then use <gmp.h> in your program.

Personally, I have a directory structure like:

Code:

---

+ addons
  + include
  + lib
  + bin

  + libgmp
  + SDL
  + any_other_3rd_party_library

---

I then copy any headers to include, any libraries (*.a files) to lib, and any dlls/useful tools to bin. Then I have the environment variable CPATH and CPLUS_INCLUDE_PATH set to the include directory. This keeps things organized, for one, and allows me to upgrade gcc without losing/having to move the libraries. (Although C++ libraries need to be rebuilt - at least, wxWidgets did.)

See this page for environment variables affecting gcc/g++.

Quote:

---

Originally Posted by Cactus_Hugger

The segment of code you posted doesn't do anything. "nothing really happens" would be the correct behavior.

---

That's not what i meant, when i try to compile the program with the -lgmp flag, it gives me the same errors as it would if i hadn't included the -lgmp (Undeclared reference, blablabla).

Thanks for the "-I" tip though, i tried passing "-IC:\GMPB\include" to the compiler, but it just comes up with some "undefined reference" when i try to run some functions from the GMP library.

Probably should have added this in the first post.
"undefined reference" occurs when you link, so the -I flag is either working, or it wasn't needed. To link, you need to specify where the library files are, so the compiler can find them. The flag is -Ldirectory. (Just like -I, but with -L) You'll need use the directory to the library files (*.a files), instead of the include folder.

Your compiler should also say something like "couldn't find library -lgmp", or something along those lines, when you pass -lgmp. (Since it's not there, it ought to be complaining...)

(There's also an environment variable for libraries, but use the flags if you're not comfortable with environment variables.)

Quote:

---

Originally Posted by Cactus_Hugger

Probably should have added this in the first post.
"undefined reference" occurs when you link, so the -I flag is either working, or it wasn't needed. To link, you need to specify where the library files are, so the compiler can find them. The flag is -Ldirectory. (Just like -I, but with -L) You'll need use the directory to the library files (*.a files), instead of the include folder.

Your compiler should also say something like "couldn't find library -lgmp", or something along those lines, when you pass -lgmp. (Since it's not there, it ought to be complaining...)

(There's also an environment variable for libraries, but use the flags if you're not comfortable with environment variables.)

---

Awesome, it's working now :) I did "g++ main.cpp -IC:\GMPB\include -LC:\GMPB\lib -lgmp" and it compiled right away.

Last problem:
It still wont work if i just do "g++ main.cpp", that gives me a ton of errors. Please don't tell me i have to show it the way to the header and to the library and do "-lgmp" each time i want to compile?

At very best, you'll have to type "g++ main.cpp -lgmp"

g++ can't (unfortunately) magically determine the location to the header files. (And it's not a good idea to use absolute locations... who says I even have a C: ?) You can supply the location in a couple of ways:
1) Using the -I / -L flags as you already have. If you use a makefile, you can simply put CFLAGS = -IC:\etc..., and add the -L flag to wherever you link. (If you're compiling everything on the command line, I would highly suggest looking into either makefiles or an IDE - either helps simplify large projects with many files.)

2) You can copy the headers and *.a files into their respective directories wherever you have g++, alongside the standard headers/libraries.

3) You can setup the environment variables as mentioned above. I do this in my autoexec.bat file:

Code:

---

SET CPATH=N:\MinGW\addons\include
SET LIBRARY_PATH=N:\MinGW\addons\lib

---

Substitute your paths as needed, of course. You can also run those SET commands in just a command prompt (cmd or command) and they'll work, but only for that prompt until that prompt is closed, which is a great way to make sure they work first. (Of course, they work in batch files too, if you carry gcc on a usb drive like I do. ;) )

Actually, I find it easier to just do a make install to MinGW's own include and lib directories.

Quote:

---

Originally Posted by laserlight

Actually, I find it easier to just do a make install to MinGW's own include and lib directories.

---

That's fine as long as you don't mind updating all your installed tools when you get the next newer version of the compiler.

It also tends make "interesting" job of updating your next tools library version when there is a change in the name of a header-file, and you include the old one, as you don't know which to remove... :-(

Any project larger than one C-file requires some sort of "build-tool" anyways - either a makefile or a batch-script to build, so why not use this to specify your paths to include/library files...

--
Mats

Quote:

---

That's fine as long as you don't mind updating all your installed tools when you get the next newer version of the compiler.

---

For GMP I think you pretty much should re-compile if you upgrade compiler, considering that they keep on warning of the dangers of GMP being miscompiled.

Quote:

---

Originally Posted by Cactus_Hugger

At very best, you'll have to type "g++ main.cpp -lgmp"

g++ can't (unfortunately) magically determine the location to the header files. (And it's not a good idea to use absolute locations... who says I even have a C: ?) You can supply the location in a couple of ways:
1) Using the -I / -L flags as you already have. If you use a makefile, you can simply put CFLAGS = -IC:\etc..., and add the -L flag to wherever you link. (If you're compiling everything on the command line, I would highly suggest looking into either makefiles or an IDE - either helps simplify large projects with many files.)

2) You can copy the headers and *.a files into their respective directories wherever you have g++, alongside the standard headers/libraries.

3) You can setup the environment variables as mentioned above. I do this in my autoexec.bat file:

Code:

---

SET CPATH=N:\MinGW\addons\include
SET LIBRARY_PATH=N:\MinGW\addons\lib

---

Substitute your paths as needed, of course. You can also run those SET commands in just a command prompt (cmd or command) and they'll work, but only for that prompt until that prompt is closed, which is a great way to make sure they work first. (Of course, they work in batch files too, if you carry gcc on a usb drive like I do. ;) )

---

Thanks alot, you've been a great help! :)

Sorry for reviving this old thread but i've hit into more problems and i don't want to fill up the boards with threads on the same subject.

As i said, i finally got it working with the C part of GMP, but really, what i wanted was the class interface of gmp so i can do stuff like:

Code:

---

mpz_class a, b;
a = 89;
b = a * 3;

---

After reading the documentation i realized that i had only gotten the C part of the library so i'm forced to use functions for everything. The doc for GMP reads:

Quote:

---

The C++ support consists of a library 'libgmpxx.la' and header filer 'gmpxx.h'

---

Well, i've only got the C files for GMP, and the header file, but i'm lacking the 'libgmpxx.la' file, even though i downloaded the newest version from the GMP project site (Ver 4.2.1), and the FAQ clearly states that the C++ wrapper files are included in the gmp files on the site.

Where is the file im lacking?

To get gmpxx, GMP needs to be compiled with --enable-cxx. Unfortunately this is disabled by default.

There are good reasons for not enabling --enable-cxx by default.

Quote:

---

It is a pity that --enable-cxx does not default to `yes' when an adequate C++ compiler is available. This means several other "bug reports" for those of us relying on the C++ interface. What are the reasons behind the current default? The C++ interface should somehow be out of its experimental phase. Or not? The language itself and the GMP C++ code has probably matured by now, but the compilers still cause major headaches. If we made --enable-cxx the default, the result would be that GMP didn't build for most users. We might be able to design tests that reject many of the compilers/include files/libraries that cause headaches. But that'd be a lot of work, and since C++ compiler writers are very quick to break new things, wouldn't be successful unless we made very frequent gmp releases. :-)

---

Or at least there were when that was written.

Anyway, this means that you'll need to find a special binary distribution or compile GMP yourself. After extensive googling I couldn't find any pre-compiled Windows binaries with gmpxx enabled, but maybe you'll have better luck. Or maybe you could try compiling GMP. It probably wouldn't be that difficult, but I wouldn't know as I've never tried it.

For reference, here's some information about gmpxx. http://gmplib.org/manual/C_002b_002b...lass-Interface

Quote:

---

Originally Posted by dwks

To get gmpxx, GMP needs to be compiled with --enable-cxx. Unfortunately this is disabled by default.

There are good reasons for not enabling --enable-cxx by default.

Or at least there were when that was written.

Anyway, this means that you'll need to find a special binary distribution or compile GMP yourself. After extensive googling I couldn't find any pre-compiled Windows binaries with gmpxx enabled, but maybe you'll have better luck. Or maybe you could try compiling GMP. It probably wouldn't be that difficult, but I wouldn't know as I've never tried it.

For reference, here's some information about gmpxx. http://gmplib.org/manual/C_002b_002b...lass-Interface

---

Well, i already did compile it once using Msys so why shouldn't i be able to compile it again, it's not really hard at all, it just takes forever. But the "enable-cxx" flag, should i put that after "./configure" or after "make install"?

Quote:

---

Originally Posted by Neo1

Well, i already did compile it once using Msys so why shouldn't i be able to compile it again, it's not really hard at all, it just takes forever. But the "enable-cxx" flag, should i put that after "./configure" or after "make install"?

---

Code:

---

./configure --enable-cxx

---

(At least I presume so, as that's how other tools, such as gcc take parameters of such type).

--
Mats

This is quite a struggle! I've recompiled the GMP files with the --enable-cxx flags, and i have now got the correct headers and library files, but it still won't work, very frustrating!

First of all: Cactus_hugger mentioned copying the library files and headers to my compiler directory to make compiling easier, i put the "gmp.h" in there, but there is also a subdirectory called "C++". This contains all the c++ headers, like string and iostream, should i put the "gmpxx.h" file in there?

Also, if i try to compile a simple program using the c++ gmp classes, g++ gives me an error in the header file gmpxx.h - "Calling fdopen: No such file or directory". This is the program i am trying to compile:

Code:

---

#include <gmp.h>
#include <gmpxx.h>

int main()
{
        mpz_class a, b, c;
       
        a = 1423354;
        return 0;
}

---

This is getting to me, i've shown the compiler the way to both library files and header files, and i've added "-lgmp -lgmpxx" just as the GMP documentation states, so why on earth isn't it working?

Quote:

---

Originally Posted by Neo1

First of all: Cactus_hugger mentioned copying the library files and headers to my compiler directory to make compiling easier, i put the "gmp.h" in there, but there is also a subdirectory called "C++". This contains all the c++ headers, like string and iostream, should i put the "gmpxx.h" file in there?

---

Say your standard include path is /usr/include. That's where all of the standard C header files go, among other things. Standard C++ header files tend to go in /usr/include/c++ (or some other variation). Header files for libraries go in their own folder typically; for example, SDL includes go in /usr/include/SDL; ncurses includes go in /usr/include/curses; and some OpenGL includes go in /usr/include/GL. I would think that gmp header files should go in, say, /usr/include/gmp. You could put the C++ header files in the same place.

If you did this, rather than going

Code:

---

#include <gmp/gmpxx.h>

---

in your code, consider adding /usr/include/gmp to the include path and just using

Code:

---

#include "gmpxx.h"

---

You can add a path to the search path for header files with the gcc/g++ option

Code:

---

g++ -I /usr/include/gmp

---

The space between -I and the rest of the path is optional.

Quote:

---

Also, if i try to compile a simple program using the c++ gmp classes, g++ gives me an error in the header file gmpxx.h - "Calling fdopen: No such file or directory". This is the program i am trying to compile:

Code:

---

#include <gmp.h>
#include <gmpxx.h>

int main()
{
        mpz_class a, b, c;
       
        a = 1423354;
        return 0;
}

---

This is getting to me, i've shown the compiler the way to both library files and header files, and i've added "-lgmp -lgmpxx" just as the GMP documentation states, so why on earth isn't it working?

---

Sticking "fdopen" into google gives you this page:

Code:

---

NAME
    fdopen - associate a stream with a file descriptor

SYNOPSIS
    #include <stdio.h>

    FILE *fdopen(int fildes, const char *mode);

---

Apparently you need <stdio.h>. gmpxx.h should include it, but perhaps they forgot to . . . .

Or maybe it's a runtime error. In which case, it's probably not something you've done wrong, because all of fdopen()'s (likely) errors seem to be programmer-related:

Code:

---

ERRORS
    The fdopen() function may fail if:

    [EBADF]
        The fildes argument is not a valid file descriptor.
    [EINVAL]
        The mode argument is not a valid mode.
    [EMFILE]
        {FOPEN_MAX} streams are currently open in the calling process.
    [EMFILE]
        {STREAM_MAX} streams are currently open in the calling process.
    [ENOMEM]
        Insufficient space to allocate a buffer.

---

Unless you're using a lot of memory or opening a lot of streams, it's going to be either EBADF, in which case whatever file descriptor being passed to fdopen() is invalid, or the mode is invalid (which is really unlikely if they tested gmpxx even once!).

So . . . maybe some file or another can't be opened. Does gmpxx have a runtime? Perhaps putting your header and library files in a standard place would help, although I don't see how it could . . .

Quote:

---

Originally Posted by dwks

So . . . maybe some file or another can't be opened. Does gmpxx have a runtime? Perhaps putting your header and library files in a standard place would help, although I don't see how it could . . .

---

Thanks alot for the lengthy reply - i tried putting the gmp header into its own folder and then compiling it with the -i flag. This got me a million errors, so i tried specifying the new location of the header files in the #include in my program, but still:

"calling fdopen: no such file or directory"

So what do i do?

I'm not quite sure i understand what you mean by gmpxx having a runtime? Can you clarify :)?

Quote:

---

Originally Posted by Neo1

Thanks alot for the lengthy reply - i tried putting the gmp header into its own folder and then compiling it with the -i flag. This got me a million errors, so i tried specifying the new location of the header files in the #include in my program, but still:

"calling fdopen: no such file or directory"

So what do i do?

---

After reading this page, I think I know what is happening. Perhaps g++ is spitting out that error because one of the paths in its search paths (for include files, or for libraries) doesn't exist.

You could run "g++ -print-search-dirs" and make sure each of those directories exists, but beware that the list is going to be long. Besides, if your compiler compiles other programs than that's not going to be the problem.

What is the exact -I option that you were passing to g++? If you put the GMP files in /usr/include/gmp (or c:\cygwin\usr\include\gmp) like I suggested, try

Code:

---

g++ -I /usr/include/gmp ...

---

or

Code:

---

g++ -I c:/cygwin/usr/include/gmp ...

---

Do you get that error whilst compiling, linking, or running your program?

Quote:

---

I'm not quite sure i understand what you mean by gmpxx having a runtime? Can you clarify :)?

---

Some libraries are dynamically linked. This means that much of the code in the library is in an external file, rather than becoming part of your executable, which is the case with static libraries (such as GMP). The nice thing about dynamic libraries is that you can load them into memory once, and use them with many different programs; whereas statically linked libraries have to be loaded into memory for each program along with the rest of the executable. A "runtime" is usually a dynamically linked library.

But I'm pretty sure that GMP is statically linked.

Quote:

---

Originally Posted by dwks

After reading this page, I think I know what is happening. Perhaps g++ is spitting out that error because one of the paths in its search paths (for include files, or for libraries) doesn't exist.

You could run "g++ -print-search-dirs" and make sure each of those directories exists, but beware that the list is going to be long. Besides, if your compiler compiles other programs than that's not going to be the problem.

What is the exact -I option that you were passing to g++? If you put the GMP files in /usr/include/gmp (or c:\cygwin\usr\include\gmp) like I suggested, try

Code:

---

g++ -I /usr/include/gmp ...

---

or

Code:

---

g++ -I c:/cygwin/usr/include/gmp ...

---

Do you get that error whilst compiling, linking, or running your program?


Some libraries are dynamically linked. This means that much of the code in the library is in an external file, rather than becoming part of your executable, which is the case with static libraries (such as GMP). The nice thing about dynamic libraries is that you can load them into memory once, and use them with many different programs; whereas statically linked libraries have to be loaded into memory for each program along with the rest of the executable. A "runtime" is usually a dynamically linked library.

But I'm pretty sure that GMP is statically linked.

---

I tried compiling it with the following path: "-IC:/MINGW/include/gmp" - the gmp folder is where the 2 headers are located, but it just got me rather long "undefined reference" errors?

Btw, the program i'm compiling now looks like this, just to clarify:

Code:

---

#include "gmpxx.h"
#include "gmp.h"
#include <stdio.h>

int main()
{
    mpz_class a, b, c;

    a = 14324234;
    return 0;
}

---

Oh and i get the errors while compiling...