From maintainers-request at octave dot org Fri Mar 31 02:19:17 2006 Subject: Re: Test failures in building the Debian packages for octave-2.9.5 From: David Bateman To: Rafael Laboissiere Cc: maintainers at octave dot org, Debian Octave Group Date: Fri, 31 Mar 2006 10:14:28 +0200 Rafael Laboissiere wrote: >In building octave-2.9.5 for Debian, several tests failed, most of them >related to sparse matrices. The log is available at: > > http://pkg-octave.alioth.debian.org/tmp/debian-octave-2.9.5_fntests.log > >Is there anything I can do to improve the situation? I am building >against libumfpack4-dev version 4.4-2. > >Thanks, > > > As far as I can tell these are all due to missing dependencies on UFSparse v1.2 or more recent, which supercedes umfpack in the sense that it includes umfpack v4.6. I suppose I should have made the tests of the sparse stuff conditional in the same way as the test of named tokens in regexp.cc is conditional on pcre being installed. However, there is so many tests involved and the combinations of what you might or might not have installed and the results of the test is so large that I'd rather just support the testing with all dependencies. I'd convinced Quentin Spencer to package UFsparse as a single package for Fedora rather than UMFPACK to reduce the number of dependencies as octave 2.9.5 is dependent on the AMD, CCOLAMD, COLAMD, CHOLMOD, UMFPACK and CXSPARSE subpackages of UFSparse. I produced a sample RPM available at http://www.dbateman.org/?page=octave of UFSparse, which Quentin took and improved, and pointed Christophe Prud'homme the maintainer of umfpack under debian to the same package. Maybe a debian package of UFSparse v1.2 or more recent (v1.3 is supposed to be release with minor changes to CXSPARSE) could be created based on Quentin's or my RPM. Note that a difference between Fedora and debian is that Quentin decided not to include the dependency on Metis for CHOLMOD, which results in reduced speed by not capability. However, as debian has PARMETIS in nonfree, the debian package of UFSparse might prefer to come in two flavours, one with a dependency on parmetis and one without. There is a slight difference in the build flags of CHOLMOD in this case, which you can find in the SRPMS on my site above.. Regards David -- David Bateman David dot Bateman at motorola dot com Motorola Labs - Paris +33 1 69 35 48 04 (Ph) Parc Les Algorithmes, Commune de St Aubin +33 6 72 01 06 33 (Mob) 91193 Gif-Sur-Yvette FRANCE +33 1 69 35 77 01 (Fax) The information contained in this communication has been classified as: [x] General Business Information [ ] Motorola Internal Use Only [ ] Motorola Confidential Proprietary