Minh Nguyen wrote: > Hi David, > > On Wed, Jul 22, 2009 at 10:53 PM, Dr. David > Kirkby<david.kir...@onetel.net> wrote: >> Until I downloaded 4.1.1.alpha0, every time there has been a failure of >> Sage on Solaris, there has been a useful error message. >> >> Since downloading 4.1.1.alph0, I've found the install just dies for no >> known reason some times. (I've also seen the usual thing about running >> sage sh etc). >> >> The last time the install died was on t2, and the install.log looks like >> this: >> >> **************************************************** >> Starting build... >> Removing old PolyBoRi install... >> Done removing old PolyBoRi install. >> Running build_polybori... >> >> >> >> And there the install.log has stopped. Nothing. The make process has died. >> >> It's not only polybori this has happened to, but I think mpir and flint. >> Every time it's restarted, so it works ok. >> >> This has happened on both my Sun Blade 2000 and the Sun T5240 't2'. >> >> Is there anything which might have changed in this release that could be >> responsible for this? Once I could understand, but several times on two >> different machines seems a bit too much of a coincidence to me. > > I don't understand why either. The only new packages are > > #6380 [ntl-5.4.2.p9.spkg] > #6443 [zn_poly-0.9.p1.spkg] > #6445 [pari-2.3.3.p1.spkg] > #6451 [flint-1.3.0.p2.spkg] > #6493 [numpy-1.3.0.p1.spkg]
It's possible I've made more use of make -k j 50 or something like, but I've certainly built packages in the past in parallel. It is possible some of these might have race conditions which cause random failures like this. On my home machine (2 x 1200 MHz), I've tended to not build in parallel so much, as I want to leave some CPU power for other tasks. But I did run with 8 threads the other day. I just started make again, and polybori built without incident. > I have started building Sage 4.1.1.alpha0 on t2 to confirm. Great. Perhaps you could add http://sage.math.washington.edu/home/kirkby/Solaris-fixes/atlas-3.8.3.p6/atlas-3.8.3.p6.spkg which is the package which is a very trivial review http://sagetrac.org/sage_trac/ticket/6558 There is only one line of code changed, from: if os.uname()[0] == 'SunOS': to if os.uname()[0] == 'SunOS' and os.uname()[4] == 'sun4v': Also, perhaps you can review the polybori fix http://sagetrac.org/sage_trac/ticket/6528 http://sage.math.washington.edu/home/kirkby/Solaris-fixes/polybori-0.5rc.p9-Second-try/polybori-0.5rc.p9.spkg The polybori fix now works on t2 - assuming it does not died for some random reason. Also the MPFR fix. I am now firmly of the opinion this is a Solaris bug, not a gcc bug, and this fix will work all the time. It's a bit complicated by the fact it only installs by default on some platforms, and not on others. But it's all surrounded by an if [ `uname` = "SunOS" ]; then ... fi so will have no impact on other systems. The patch into in trac shows this only causes a problem on gcc later than 4.2.4, but in fact, it could happen on any gcc release, as it is a bug in Solaris, not in gcc. My singular patch http://sage.math.washington.edu/home/kirkby/Solaris-fixes/singular-3-1-0-2-20090620.p0/singular-3-1-0-2-20090620.p0.spkg http://sagetrac.org/sage_trac/ticket/6563 is very easy to review, but it does do something you might not like - it needs install-sh added to the top level directory of singular. It is probably a bug in singular or autoconf - probably the latter in fact. Why the singular developers have 15 or so copies of install-sh in the source and still fail to find it on one occasion is beyond me. I'd like to get another patch for ATLAS out some time, which resolves the issues of slow build on t2. But that path is somewhat more complex than the trivual one above, and will take more testing, so I'd like to get the simple one done first, then more onto the more complex one, which I doubt will make 4.1.1 There are still at least two Solaris issues remaining. 1) The 'modified sage library code' wont build without a bit of manual editing, to comment out a few lines in a header file 2) ecl wont build at all. Dave --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
