On Tuesday, January 31, 2012 1:01:41 PM UTC-8, R. Andrew Ohana wrote: > > On Mon, Jan 30, 2012 at 10:04, John H Palmieri <jhpalm...@gmail.com> > wrote: > > > > On Tuesday, January 17, 2012 1:59:45 PM UTC-8, William wrote: > >> > >> On Tue, Jan 17, 2012 at 1:39 PM, William Stein <wst...@gmail.com> > wrote: > >> > >> > On Mon, Jan 16, 2012 at 3:43 PM, William Stein <wst...@gmail.com> > wrote: > >> >> On Mon, Jan 16, 2012 at 9:26 AM, John H Palmieri <jhpa...@gmail.com> > >> >> wrote: > >> >>> > >> >>> > >> >>> On Monday, January 16, 2012 7:42:49 AM UTC-8, William wrote: > >> >>>> > >> >>>> Hi, > >> >>>> > >> >>>> A major blocker for Sage-5.0 is supporting OS X (version 10.7 -- > the > >> >>>> version that has been out for months now). > >> >>>> > >> >>>> Fortunately, it is now "relatively easy" to build sage-5.0.beta1 on > >> >>>> OS > >> >>>> X 10.7 with XCode 4.x, and have it start up. > >> >> > >> >> After building Sage on OS X 10.7, "make test" did this: > >> > > >> > For the record, running tests with "sage -t devel/sage/sage" yields > >> > hundred(s) of failing files: > >> > > >> > http://wstein.org/home/wstein/tmp/test-sage-5.0.beta1-osx10.7.txt > >> > > >> > It could be that most of these boil down to some code at the core of > >> > PARI (the bezout function) being miscompiled. > >> > > >> > There is a discussion about this from August 2011 here: > >> > > >> > http://pari.math.u-bordeaux.fr/archives/pari-dev-1108/msg00000.html > >> > > >> > It unfortunately appears based on the mailing list -- and searching > >> > through the ** 5 months ** since then -- that nothing at all was done > >> > to try to fix the problem, even though I had setup access to 10.7 for > >> > the Pari developers. Probably the main problem is that the machine I > >> > setup -- sqrt5.math.washington.edu -- is on the CS network, and for > >> > some reason it keeps getting kicked off. > >> > > >> > This really sucks. Maybe I can rewrite their bezout to get around > >> > the problem. > >> > >> Another option is to turn of optimization (-O0) when building part of > >> PARI under XCode 4.x, since the problem is bad compiler optimization. > >> I just tried, export SAGE_DEBUG="yes", then building PARI with > >> SAGE_CHECK="yes", and I get: "The PARI self-tests all passed". > >> > >> So something based on optimization flags is the workaround I'll pursue > >> for now. Something that is slightly slower than optimal is way > >> better than infinitely slower. > > > > > > I've been trying to build Sage with OS X Lion over the past few days, and > > things are going pretty well: there are a few doctest failures, but I > think > > most are numerical noise. Most spkgs build just fine, but because of > bugs > > in Apple's compiler, a few -- pari, gsl, symmetrica -- have problems > unless > > we modify them. So I propose the following modifications in the > > spkg-install scripts for those packages: > > > > - check to see if running Lion, and if so > > - check to see if /usr/bin/gcc-4.2 exists. this might be available > from an > > older installation of XCode, or it can be installed using the gcc package > > Georg posted a link to: http://r.research.att.com/tools/. If it's > there, > > set CC=gcc-4.2. > > - otherwise, turn off optimization (compile with -O0) for those > packages. > > This seems to avoid the bugs, although it will slow down those specific > > pieces of code. > > > > Opinions? > Do we know which files build incorrectly, we should only set the flag > for those (and not the whole library).
I tried that and still got self-test failures. I didn't try all possible combinations of compiling some files with -O3 and some with -O0 until tests passed. > Also, we should see if -O1 > works too. > No, it doesn't. I tried that. [snip] I've started looking into the difficulties of getting sage to build > with clang (on lion), and have made some progress on the sage > libraries. > > Using CC=clang, CXX=clang++, and SAGE_CHECK=yes for all spkgs except > I thought I had tried this with pari and had failures, but now it seems to work. In an existing Sage build, installing cvxopt this way doesn't work for me -- I get self test failures -- but maybe setting CC=clang (etc.) and building the whole thing from scratch will work better. Please keep working on this; it sounds like you're getting somewhere. -- John -- 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
