Please add these sparse matrix packages that are needed for octave 2.9.10 to have sparse support. Reproducible: Always Steps to Reproduce:
Created attachment 115802 [details] sci-libs/camd initial ebuild
Created attachment 115804 [details, diff] camd-2.1.2-autotools.patch
Created attachment 115805 [details] sci-libs/ccolamd initial ebuild
Created attachment 115806 [details, diff] ccolamd-2.5.2-autotools.patch
Created attachment 115808 [details] sci-libs/cholmod initial ebuild
Created attachment 115810 [details, diff] cholmod-1.4.0-autotools.patch
Created attachment 115812 [details] sci-libs/colamd initial ebuild
Created attachment 115814 [details, diff] colamd-2.6.0-autotools.patch
Created attachment 115815 [details] sci-libs/cxsparse initial ebuild
Created attachment 115816 [details, diff] cxsparse-2.0.7-autotools.patch
the cholmod package also require metis, which has an ebuild here in bugzilla: http://bugs.gentoo.org/show_bug.cgi?id=53394
cxsparse doesn't build because cs.h is not found. First I thought it was a problem with the patch, 'cause there's +AM_CFLAGS = -I ../Source +include_HEADERS = Source/cs.h whereas cs.h is placed in Include. So I replaced Source with Include, but that didn't work, too. I'm no autotools-expert, so I re-replaced Include with Source again and simply changed the ebuild: src_unpack() { unpack ${A} cd "${WORKDIR}" epatch "${FILESDIR}"/${P}-autotools.patch cd "${S}" >>> cp Include/cs.h Source/ <<< rm -f Makefile */Makefile eautoreconf } Works, but it's probably not the most elegant solution :-) Greetings, Simon
I have some version bumps for these, added src_test, and tweaked the autotools patches a little adding header checks and support for building tests mainly. Cholmod doesn't actually require ccolamd/camd, unless it's being built with metis support (which isn't required), so I added that to use autotools as well and the corresponding USE flag. Cholmod also comes with the stress suite, located in Tcov. I was able to get this to build, but it fails half-way through the tests. This is because they copy all the source files that build libcholmod into the current directory and parse them using gcov. If anyone is actually interested, they could probably fix this. Also, the gcov scripts are writting in csh, so there's an added depend on tcsh.
Created attachment 131415 [details] camd-2.2.0.ebuild
Created attachment 131417 [details, diff] camd-2.2.0-autotools.patch
Created attachment 131419 [details] ccolamd-2.7.0.ebuild
Created attachment 131421 [details, diff] ccolamd-2.7.0-autotools.patch
Created attachment 131422 [details] cholmod-1.5.0.ebuild
Created attachment 131423 [details, diff] cholmod-1.5.0-autotools.patch
Created attachment 131424 [details] colamd-2.7.0.ebuild
Created attachment 131425 [details, diff] colamd-2.7.0-autotools.patch
I forgot to mention above, but as Johan pointed out earlier, there is already a bug for metis, #53394. I've also commited an updated ebuild there as well that I used to test cholmod with.
Very nice work done indeed.
Folks, any reason why can't you move this to science overlay and work on it there? This bug is getting messy and noisy with all those tons of attachments. :/
Hi Thanks a lot for all the great work. All these packages are now in the science overlay. Please test them and report here any problems. I would like to push them to the main tree soon. I made quite a lot of changes with respect to the one submitted here. For the ebuild writers, please review these changes and the autotools patches. Now we are almost done to the full suitesparse (any candidate?) Sebastien
(In reply to comment #25) Everything looks good to me. I pushed your changes to our local overlay and have updated the developer's machines. No one has reported any issues. Thanks,
Now in cvs. Thanks all for your hard work.