Hi Venkat, Thank you for all the details. I managed to set up a second Gfortran library on my Mac OS X test system and run Art Palmer's Modelfree 4.20 binary for Intel systems on it. I discovered that there is a problem in the Modelfree4 binary for Intel Mac systems in that the number of significant figures of the numbers in the mfout file are not treated correctly, as described in the fix: http://article.gmane.org/gmane.science.nmr.relax.scm/18350 This causes the numbers to be truncated compared to the Linux gcc and pgf binaries, and is likely to be a FORTRAN portability issue. The relax test suite now detects the Intel Mac Modelfree4 results and for now treats those as being correct, allowing the test suite to pass. As this does not affect how relax operates I will wait until releasing relax 3.0.1 with the test suite fix. This version will also correctly identify the GCC verses PGF Modelfree 4.20 results and handle those minor differences too. Cheers, Edward On 14 August 2013 15:43, Edward d'Auvergne <edward.dauvergne@xxxxxxxxx> wrote:
Hi Venkat, Thank you for the information. I have to ask a few more questions though as I unfortunately cannot get this to run on my Mac test system. Are you using the Mac OS X gfortran from http://r.research.att.com/tools/, MacPorts, HPC, or Fink? I am using the one from http://r.research.att.com/tools/, as this is the only one that can compile the Python Scipy package. Would you happen to know the exact version, as GCC has renamed the old libgfortan.3.dylib file to libgfortan.2.dylib in the current releases hence I might need to hunt down an old version to make this run. For the Linux systems, Art Palmer's binaries are compiled only for i386 using both the old g77 and pgf fortran compilers (in the gnu/ and pgf/ directories respectively). I cannot run the pgf version as it segfaults (it was compiled on Linux 2.2 and doesn't seem to run any more), but the g77 version compiled on Linux 2.6 runs fine. Do you know which of these two binaries you are running? Are they self compiled or Art's binaries? The commands 'file' or 'ldd' followed by the modelfree4 binary will give some useful information. On my system with the g77 binary, unfortunately I do not see the same problems as you do. This is a difficult problem because for every Modelfree4 binary variant which gives a different result, I have to manually add the Modelfree4 results to the test. Anyway, I really hope I can get down to the bottom of this problem. Cheers, Edward