GSL library

Hi everyone!

I am using a code on a machine, and it works fine.

I compile with

gcc -c -g KANT.c
g77 -c -g tn.f
g77 -c -g blas.f
gcc -g KANT.o tn.o blas.o -lgsl -lgslcblas -lg2c


(Thanks to this forum I found out how to mix fortran and c in the same code).

Now I want to work on another machine that doesn't have gsl libraries installed and I don't have administration rights.

So I downloaded gsl localy.

and I type

gcc -c -I/data/mtsamado/KANT_8-8-2007/gsl/include/ KANT.c
gcc -L/data/mtsamado/KANT_8-8-2007/gsl/lib KANT.o -lgsl -lgslcblas

the executable a.out is generated but when I try to run it I get the error message :

./a.out: error while loading shared libraries: libgsl.so.0: cannot open shared object file: No such file or directory

Any ideas ??
Thanks a lot.
Michel

I expect you haven't actually got the library itself of libgsl - only the devel variant of it. You would need to install the libgsl as well.

--
Mats

In the working directory I have the directory gsl/ that contains

include/
lib/

where lib/ contains the following
libgsl.a libgslcblas.la libgslcblas.so.0 libgsl.la libgsl.so.0 pkgconfig
libgslcblas.a libgslcblas.so libgslcblas.so.0.0.0 libgsl.so libgsl.so.0.9.0


Is it not enough ??

Thanks

Aha, ok. So the quick fix would be to add the path to that local directory to LD_LIBRARY_PATH. However, this is also a possible source of security holes, as someone may add a libc.so that contains some extra code to compromize your account into that directory, so the correct option would be to actually install libgsl into the /usr/lib or some other path that is already in LD_LIBRARY_PATH.

--
Mats

Thanks I am not expert with linux and all these path things

I found on the net that I should do something like

export LD_LIBRARY_PATH=/data/mtsamado/KANT_8-8-2007/gsl/lib

but when I do that I get

export: Commande introuvable. (which in english the computer cannot find the command export)

1) Is it the correct syntax to ling my code to GSL?
2) Is there another way to link directly the code to the library I need ?

By the way I have no administrator rights, so I don't think I could instal GSL in /usr/lib ?? Is that right ?

thanks a lot for the help.

You probably want to ADD to the existing LD_LIBRARY_PATH, something like:

Code:

---

export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:mypath

---

You could also try to make a static link - add "-static" at the beginning of your final gcc command (the one that builds the final executable).

--
Mats

You're a genious !!

I type

gcc -c -I/data/mtsamado/KANT_8-8-2007/gsl/include/ KANT.c
gcc -static -L/data/mtsamado/KANT_8-8-2007/gsl/lib KANT.o -lgsl -lgslcblas -lm

As you suggested and it works !!
Thanks

Something more occured !

I run the code everything is fine until a message segmentation fault occured (I run with gdb)


Program received signal SIGSEGV, Segmentation fault.
gsl_fit_wlinear (x=0x7fffd06210a0, xstride=1, w=0x7fffd0621000, wstride=1, y=0x7fffd0621050, ystride=1, n=13783292549630263302,
c0=0x7fffd0621000, c1=0x7fffd0621000, cov_00=0x7fffd0621000, cov_01=0x7fffcfe65420, cov_11=0x7fffd0621000, chisq=0x7fffd0621000)
at linear.c:131
131 linear.c: Aucun fichier ou répertoire de ce type.
in linear.c

What does this mean ?
I guess gsl_fit_wlinear is not found ???


Is it still a library issue or something else ?

Thanks

Any ideas, anyone ??

Thanks
Michel

Quote:

---

Originally Posted by mimi1981

n=13783292549630263302,

---

I find that line highly odd. looks like that function was called with an enormous value for *n*. Now, I don't know what that library does or what n is used for, but that just doesn't look right to me. What's that function supposed to do?

QuantumPete

Pete is right. According to the the GSL docs (http://www.csse.uwa.edu.au/programmi...sl-ref_34.html) the 7th argument of that function is the number of elements in the array you are passing to the function. You are getting a seg. fault because GSL is running over the buffer and eating you stack (until it hit projected memory -- or something like that).

Go to the line in you code where you call the gsl_fit_wlinear function and make sure you have got the function arguments correct. Post you code if you want I have used GSL before.

Cheers,

Dan.

Actually, looking at the value of N as a hex value gives: BF48178C00000006. BF48178C seems like a stack address, 00000006 is a sensible small number.

My guess would be that the arguments are messed up for this function - something that should be a stack address is taken as a long long int.

Another alternative is that n is actually a double floating point that is negative.

--
Mats

Quote:

---

Originally Posted by matsp

My guess would be that the arguments are messed up for this function - something that should be a stack address is taken as a long long int.

---

Maybe passing a pointer instead of an integer?

QuantumPete

Quote:

---

Originally Posted by QuantumPete

Maybe passing a pointer instead of an integer?

QuantumPete

---

Surely a possibility.

--
Mats

Quote:

---

Originally Posted by matsp

Surely a possibility.

--
Mats

---

Then it pays to take notice of your sig and compile with -Wall :)

QuantumPete

Quote:

---

Originally Posted by QuantumPete

Then it pays to take notice of your sig and compile with -Wall :)

QuantumPete

---

Yes, but in this case we're dealing with cross-language interfaces, which usually means that there's no way to actually let the compiler verify that the header file / prototype matches the actually generated code, since the Fortran compiler will not accept C style header files.

Care and attention needed.

--
Mats

If I'm not mistaken, you'd still need an extern declaration for the FORTRAN function. Hopefully that declaration is in some header file shipped with the FORTRAN, but not compiled into it.
So you include that header in C and the compiler will perform argument checking. It would, of course, suck if you didn't get the header with the FORTRAN, but had to make it up yourself. FORTRAN variable translations are always tricky, especial logical.

QuantumPete