On Thu, Feb 25, 2010 at 04:46:43PM +0100, Edward d'Auvergne wrote:
Hi, This failure is because it takes 201 iterations to reach the minimum compared to 204 with x86_64 hardware. I have now removed this iteration check as testing the values and chi-squared number is sufficient to see that the minimum has been reached. Could you update the repository copy (svn up) and check again? For the Fink specific changes to the scons/install.py file, is there a way of signalling that this is a Fink install rather than a normal install? Maybe I could add an scons target called install_fink so that you type: $ scons install_fink That way we can bring in these changes into the relax repository, something which might be useful for installing relax on Macs in the distant future. Cheers, Edward
Edward, The current relax 1.3 svn no longer shows any failures in the testsuite when built under fink on powerpc-apple-darwin9. I'll check the other arches in the next day or so. As for adding the patches from fink, those are problematic. The fink approach is to support python variants (hence the relax-py package name that can build against the fink python 2.5 or 2.6. The relax-py.patch basically... 1) patches the relax script so that python is called as python2.X with the X replaced according which python fink is building relax-py against. 2) the generation of pyc files are suppressed in the scons/install.py script because fink guidelines are not to package .pyc or .pyo files but generate then from a postinstall packaging script. 3) the sconstruct script is patched to allow the relax path to be renamed relax-py2X to indicate which python relax was built against and to install the binaries in the bin subdirectory of the fink installation. The only way to handle this would be to have separate .fink versions of these three files with the patches applied. This would be problematic for sconstruct. Jack