scipy 0.4.4 is released (see http://new.scipy.org/Wiki). This is a major release, it uses numpy (see bug #119305) as its array module. I have tested the attached ebuild on my system and it works with no problems.
Created attachment 77448 [details] scipy-0.4.4 ebuild
I forgot to mention: SRC_URI is incorrect in the ebuild. The actual URI is http://new.scipy.org/Wiki/Download?action=AttachFile&do=get&target=scipy-0.4.4.tar.gz, but when I put that in the ebuild, I couldn't get it to download the file and save it as scipy-0.4.4.tar.gz instead of "Download?action=Attac...". I am not sure if this is possible from within an ebuild to specify an output filename, or if we should put the tarball on gentoo mirrors and then use that in SRC_URI.
I have successfully tested this ebuild and made a small modification for the SRC_URI.
Created attachment 78599 [details] scipy-0.4.4.ebuild updated ebuild
scipy-0.4.4 in the Gentoo scientific overlay (http://gentooscience.org) Please test and report here.
(In reply to comment #5) > scipy-0.4.4 in the Gentoo scientific overlay (http://gentooscience.org) > Please test and report here. > Looks like they made some changes in numpy (see http://www.scipy.net/pipermail/scipy-user/2006-January/006613.html), so scipy.test() fails on many cases, with errors like: AttributeError: 'numpy.ndarray' object has no attribute 'dtypechar'. When I run scipy.test, I get: Ran 1208 tests in 1.866s FAILED (failures=1, errors=329) I am using numpy 0.9.4 and scipy 0.4.4. From the above mail: SciPy SVN has been updated and fixed with the changes.
Er... Do we need a requirement on f2py ? As far as I understand, numpy comes with f2py, and having an external f2py installed caused me some troubles...
In reply to comment #7: ebuild in overlay (see comment #5) does not require f2py. In reply to comment #6: yes it does the same for me. from the scipy svn, we understand scipy-0.4.6 is not far from release, and correct the dtype errors. The ebuild in the overlay now uses lapack-atlas. It seems that scipy only works with some non f2c-ed clapack implementation, which only (so far in portage) lapack-atlas provides. But it can use more blas implementation (I tested with mkl).
0.4.6 was released on 2006-02-16
(In reply to comment #9) > 0.4.6 was released on 2006-02-16 And is in the overlay (as well as the required dep. numpy-0.9.5). btw, I tested with python -c "import scipy; scipy.test(level=10)", and all tests were OK.
scipy-0.4.8 on the overlay also with better treatment of atlas and fftw. (note now atlas is not an option anymore).
Version 0.4.8 now in Portage.
