Hello folks!

I'm a very happy eigen3 user! Recently, I bumped into a very strange problem, which seems to be related to the compiler I use. Below is a contained c++ file that demonstrates the crash, in what looks like very simple code at first sight. The crash occurs when using a GCC 4.7.0 compiler. (when I use GCC 4.9.1, the problem does not appear).

Code: Select all

#include <stdio.h>
#include <complex>
#define dcomplex std::complex<double>

#include "../../contrib/eigen32/eigen/Eigen/Core"
#include "../../contrib/eigen32/eigen/Eigen/LU"
#include "../../contrib/eigen32/eigen/Eigen/Sparse"

int main() {
    printf( "Will try out some things\n");
    dcomplex cnum(1, 1.0);
   
    Eigen::SparseMatrix<dcomplex, Eigen::RowMajor, int> data(6, 6);

    std::stringstream s;
    s << data;
    printf("data: %s", s.str().data());
    s.str("");
   
   
    // Test: inserting / using coeffRef (both crash)
    data.insert(0, 0) = cnum;
    //data.coeffRef(0, 0) += cnum;
   
   
    // Test: using makeCompressed: doesn't change the behavior
    //data.makeCompressed();
   
    // Test: finalizing the matrix: doesn't change the behavior
    //data.finalize();
   
    // It will allow performing operations on the matrix:
    printf("number of nonzeros now: %d\n", data.nonZeros());
    printf("norm: %f\n", data.norm());
   
    // But it will not allow printing the matrix for a second time. Crashes:
    s << data;
    printf("data: %s", s.str().data());
    s.str("");
   
};


Unfortunately, switching to gcc 4.9 is not an option, because other third-party tools / modules that are loaded at runtime, use gcc 4.7. I am using Windows. I tried looking through the issue tracker but could not find anything that could be related.

Could you paste the backtrace. Thanks.

And what about using std::cout << "data:" << std::endl << data << std::endl; instead of C code?

Hello,

Thank you for the response!
I just tested it on Linux, using gcc 4.7.0 (I was hoping to then use gdb / valgrind to look for something suspicious). Unfortunately I cannot reproduce it there: the program does not crash. So it's only on Windows, gcc 4.7.0. Here's the full specs of the compiler on Windows:

Code: Select all

> gcc -v                                           
Using built-in specs.                                                                                   
COLLECT_GCC=c:\Users\mfiers\Anaconda\envs\myenv\Scripts\gcc.bat\..\..\MinGW\bin\gcc.exe             
COLLECT_LTO_WRAPPER=c:/users/mfiers/anaconda/envs/myenv/mingw/bin/../libexec/gcc/x86_64-w64-mingw32/4
.7.0/lto-wrapper.exe                                                                                     
Target: x86_64-w64-mingw32                                                                               
Configured with: ../../../build/gcc/src/configure --target=x86_64-w64-mingw32 --prefix=/c/bb/vista64-mingw32/mingw-x86-x86_64/build/build/root --with-sysroot=/c/bb/vista64-mingw32/mingw-x86-x86_64/build/build/root --enable-languages=all,obj-c++ --enable-fully-dynamic-string --disable-multilib                     
Thread model: win32                                                                                     
gcc version 4.7.0 20111220 (experimental) (GCC)                                                         


Additional things I tried:
- I've been looking on how to create a backtrace on Windows but it seems to be extremely difficult.
- I tried compiling with -g -O0, but the same problem occurs.
- I tried the code you suggested, but it will crash, too.

Are there compiler options from eigen which I could use that could give me some more information?

I don't think it's really related to the printing: I think the matrix is somehow corrupted. This is because in my production code I don't explicitly print the matrix, and I also get crashes.

Without a backtrace we cannot do anything.

Cannot you upgrade to a non experimental GCC, like gcc-4.7.2-1

Upgrading to 4.7.2 seems to be solving the problem. Furthermore, it seems the runtime libstdc++-6.dll is 'compatible' between 4.7.0 and 4.7.2-compiled modules.

Thank you very much for the assistance! I'll accept the answer, but I still have to do some more testing. if anything weird pops up, I'll re-open this topic.

Thanks again!
Martin