Hi Andreas, On Tue, 4 Feb 2014 21:15:41 +0100, Andreas Tille <andreas@an3as.eu> wrote: > recently I uploaded seqan and I admit I fail to build the package on two > of three machines I control and use to build. Only one of these three > amd64 machines has a big enough memory and does not die from swapping. > When looking at the build log of i386: > > https://buildd.debian.org/status/fetch.php?pkg=seqan&arch=i386&ver=1.4.1-3&stamp=1390975155 > > I see [...] > cc1plus: out of memory allocating 7372636 bytes after a total of 45391872 > > When looking at > > https://buildd.debian.org/status/package.php?p=seqan > > it seems that only the "more powerful" architectures are able to build > this package but also kfreebsd-amd64 fails (with a different error but I > suspect the same problem). I would consider excluding some > architectures but the package is really used on i386 and I wonder whether > there are some options to get it builded at least for this architecture. The memory allocation isn't caused by memory exhaustion but by address space exhaustion - pair_align.cpp needs 5GB of RAM to build with the current CXXFLAGS. All the failed architectures apart from kfreebsd-amd64 are 32-bit architectures. One solution is to build with no optimisations (-O0 instead of -O2, perhaps using noopt); that allows seqan to build on my i386 buildd. I didn't try with -O1. Regards, Stephen
Attachment:
signature.asc
Description: PGP signature