On Wed, Oct 21, 2009 at 10:08 AM, William Stein <wst...@gmail.com> wrote: > On Wed, Oct 21, 2009 at 8:38 AM, Nils Bruin <nbr...@sfu.ca> wrote: >> >> Just a datapoint that might provide useful feedback for those who are >> trying to make binary installs a smooth experience: >> >> I tried to install 4.1.2 on a Fedora 10 (i686) laptop. I tried both >> the Fedora 9 and the Fedora 11 image. With either I got errors: >> >> ImportError: /usr/lib/libstdc++.so.6: version `GLIBCXX_3.4.11' not >> found (required by <lib>.so) >> >> doing a "sage -f ..." for the package that provides <lib>.so solved >> that problem, but then the next C++ library would play up. >> I ended up getting stranded on: >> >> local/lib/python2.6/site-packages/sage/combinat/partitions.so >> >> for which I was not readily able to determine which spkg to install. >> >> It may just be that F9/F10/F11 is a particularly active time of libstdc >> ++ development. However, it gave me the impression that binary >> distributions of sage are very fragile. It's good that there is always >> the source fall-back option, but the prospect of having my laptop >> churn for 2 hours to produce an upgrade actually put me off upgrading >> for now. >> >> If this is a more common problem, how difficult is it to have a list >> of c++ spkgs so that libstd++ problems can be solved by recompiling >> those? Or are there so many that you might as well do a source >> install? > > Would you be willing to test taking a fresh install of the fedora 11 > binary and dropping > libstdc++ from Fedora 11 (etc.) into your local/bin/? For a very > long time, with Sage we > *distributed* a bunch of libstdc++ files with the binary. Maybe a > combination of that > with some instructions like you mention bight be the best option. > > Linux is a such an "exciting challenge" when it comes to making binaries.
Check out the VirtualBox binary download page for Linux: http://www.virtualbox.org/wiki/Linux_Downloads There are 39 distinct binaries listed there: * Ubuntu 9.10 ("Karmic Koala") i386 | AMD64 * Ubuntu 9.04 ("Jaunty Jackalope") i386 | AMD64 * Ubuntu 8.10 ("Intrepid Ibex") i386 | AMD64 * Ubuntu 8.04 LTS ("Hardy Heron") i386 | AMD64 * Debian 5.0 ("Lenny") i386 | AMD64 * Debian 4.0 ("Etch") i386 | AMD64 * openSUSE 11.1 i386 | AMD64 * openSUSE 11.0 i386 | AMD64 * openSUSE 10.3 i386 | AMD64 * SUSE Linux Enterprise Server 10 (SLES10) i386 | AMD64 * Fedora 12 ("Constantine") i386 | AMD64 * Fedora 11 ("Leonidas") i386 | AMD64 * Fedora 9 ("Sulphur") / 10 ("Cambridge") i386 | AMD64 * Fedora 8 ("Werewolf") i386 | AMD64 * Mandriva 2009.1 i386 | AMD64 * Mandriva 2008.0 i386 * Red Hat Enterprise Linux 5 ("RHEL5") / CentOS 5 i386 | AMD64 * Red Hat Enterprise Linux 4 ("RHEL4") / CentOS 4 i386 * Turbolinux 11 i386 | AMD64 * PCLinuxOS 2007 i386 * All distributions i386 | AMD64 Opera is similar. If these guys don't have a better solution, then probably this is simply the sort of thing that is *demanded* by distributing nontrivial binary software for Linux. I am not opposed to trying to target making binaries for far more system, if we get organized and work together. It's just a matter of making VirtualBox machines that have g++, make, m4, gcc, installed into them. I like them to also have a /tmp that has at least 32GB free disk space on it. I think we could support building 30-40 binaries on our current rather powerful hardware resources. The main issue is the workload of creating all these VirtualBox virtual machines. I don't want to do it all myself! William --~--~---------~--~----~------------~-------~--~----~ To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
