tags 625756 patch user ubuntu-de...@lists.ubuntu.com usertags 625756 ubuntu-patch oneiric
On Thu, May 05, 2011 at 05:54:35PM +0000, brian m. carlson wrote: > When I build electric-fence from source on amd64/sid with gcc-4.5 or > gcc-4.6, the ./tstheap 3072 call gets to iteration 100 and then starts > allocating massive amounts of memory. Within less than a minute it goes > through 3.5GB of physical memory and 5GB of swap. It is then killed by > the OOM killer. As a consequence, the package fails to build from > source. > > I've looked at the code and haven't been able to track down the problem > yet. The code does work correctly without optimization, but still goes > out of control at -O1. I suspect this is a consequence of optimizer > improvements. I've reported this as a GCC bug: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50337 -fno-tree-dse works around it. * Build with -fno-tree-dse, since otherwise GCC >= 4.5 misoptimises allocateMoreSlots() (closes: #625756, LP: #749139). diff -Nru electric-fence-2.1.16/debian/rules electric-fence-2.1.16ubuntu1/debian/rules --- electric-fence-2.1.16/debian/rules 2010-05-14 05:04:58.000000000 +0100 +++ electric-fence-2.1.16ubuntu1/debian/rules 2011-09-08 18:08:12.000000000 +0100 @@ -19,10 +19,10 @@ build-stamp: dh_testdir #if [ -x configure ]; then ./configure --prefix=/usr; fi - make CFLAGS="-O2 -g -Wall -fPIC" + make CFLAGS="-O2 -fno-tree-dse -g -Wall -fPIC" gcc -g -shared -Wl,-soname,libefence.so.0 -o libefence.so.0.0 efence.o page.o print.o -lc -lpthread rm *.o - make CFLAGS="-O2 -g -Wall" + make CFLAGS="-O2 -fno-tree-dse -g -Wall" touch build-stamp clean: Thanks, -- Colin Watson [cjwat...@ubuntu.com] -- To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
