seg fault with too large "number of bases"? (FIXED)

[psimon]

failed to make on Jaunty with all repo library builds, but now does compile on Karmic, however cant run any of the examples, i get seg fault;

tried different optimization levels, still same;

i noticed that other example outputs dont have such a large value here; [I] Number of Bases= 156479720

specifically on example/cylinder;

simon@hal3 ~/free-cfd-0.1.2/examples/cylinder $ mpirun -np 2 ../../src/fcfd cylinder
[I] Found input file cylinder.in
[I] Read input section: flamelet
[I] Read input section: turbulence
[I] Read input section: grid
[I] Read input section: reference
[I] Read input section: timeMarching
[I] Read input section: numericalOptions
[I] Read input section: linearSolver
[I] Read input section: jacobian
[I] Read input section: fluidProperties
[I] Read input section: writeOutput
[I] Read input section: probes
[I] Read input section: integrateBoundary
[I] Read input section: initialConditions
[I] Read input section: boundaryConditions
[I] Found grid file cylinder.cgns
[I] Number of Bases= 156479720
--------------------------------------------------------------------------
mpirun noticed that process rank 0 with PID 7477 on node hal3 exited on signal 11 (Segmentation fault).
--------------------------------------------------------------------------

here is some version info;

g++-4.4
libopenmpi1.3
libpetsc3.0.0
libparmetis3.1
libcgns2

Computer:
Processor 2x Intel(R) Core(TM)2 Duo CPU E4700 @ 2.60GHz
Memory 2060MB (508MB used)
Operating System Linux Mint 8 Helena - 32bit - Main Edition

[esozer]

This sounds like a cgns library problem. Do you have 2.5? Can you try to recompile and link?

[psimon]

very much thanks for this help.

i am using 2.5

the versions i gave were just the main, i thought if there was an incomparability it would be at that level, anyway the detailed sub-version etc are;

g++-4.4.1-4ubuntu8
libopenmpi1.3.2.3ubuntu1
libpetsc3.0.0.dfsg-5ubuntu2
libparmetis3.1.1-1
libcgns2.5.3-3

this is all from the Karmic repos. none of these are compiled, all that i compiled was freeCFD linking using the repos. header, *.dev, packages, and as i said it succeeded.

i thought a simple test would be to try other/newer cgns files, maybe some parameter has changed/obsoleted, so that the newer libraries are confused, do you have something i could try?

[esozer]

It is hard to really find a newer cgns file. We would have to know that the grid generator is actually using the newest cgns libraries. I suggest grabbing an older cgns library source from http://sourceforge.net/projects/cgns/files/ (I think what I had was 2.5.1 which is no longer there but release 2 could be OK). It should be straightforward to compile it from source. Please let me know if you try this and verify there is a new version issue so that I can find out what has changed.

[psimon]

it now works;

as a test i tried all the versions ( -2 ,-3 and -4 ), from http://sourceforge.net/projects/cgns/files/ and they ALL worked, so its only the original Karmic repo. libcgns that failed.

my build Notes:
1. by default these build a static lib *.a
to get dynamic lib *.so; ./configure --enable-shared

2. these make into /usr/local/lib where the repo lib was in /usr/lib
so CGNS_BASE needed to be changed to /usr/local from /usr
and to run;
export LD_LIBRARY_PATH=/usr/local/lib, was needed because /usr/local is not in the default lib search path