Bug#448158: more information
OK, so petsc seems to be ok now. I am going to try libmesh. Please don't close this issue yet - there could be other problems with g77/gfortran and maybe the proposed fix is not really a fix (i.e. causes other problems). Need to check libmesh out first. I was able to compile libmesh, with the help of upstream, the preliminary packages are here: http://debian.certik.cz/ When compiled in the upstream tree by hand, it seems to work, when used from the Debian package it fails, but the problem seems to be libmesh related, not petsc/openmpi related. I also built python-petsc4py, the package is also available at the link above. You can try it: $ apt-get install python-petsc4py $ python Python 2.4.4 (#2, Aug 16 2007, 02:03:40) [GCC 4.1.3 20070812 (prerelease) (Debian 4.1.2-15)] on linux2 Type help, copyright, credits or license for more information. from petsc4py.PETSc import Mat [fuji:06162] mca: base: component_find: unable to open osc pt2pt: file not found (ignored) libibverbs: Fatal: couldn't read uverbs ABI version. -- [0,0,0]: OpenIB on host fuji was unable to find any HCAs. Another transport will be used instead, although this may result in lower performance. -- A=Mat() A.createSeqAIJ(1000) Mat object at 0xb5ee98ec (0x82cac40) So it seems to work, but the ugly warning/error messages seems to be openmpi related. I am going to ask on the debian openmpi mailinglist about it. Let's leave this issue open until all these errors are either resolved, or we are sure they are not petsc related. Ondrej -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#448158: more information
Hello Ondrej, Thank you for your detailed inquiry into this issue. I'm glad you got it working. On Wed, 2007-11-07 at 23:54 +0100, Ondrej Certik wrote: On Nov 7, 2007 9:52 PM, Ondrej Certik [EMAIL PROTECTED] wrote: TESTING: checkCxxCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:541) TESTING: checkFortranCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:708) * UNABLE to CONFIGURE with GIVEN OPTIONS(see configure.log for details): --- bin/mpif90[f77] you provided with --with-mpi-dir=/usr does not work Run with --with-fc=0 if you wish to use this MPI and disable Fortran * make: *** [build-arch] Error 1 dpkg-buildpackage: failure: debian/rules build gave error exit status 2 pbuilder: Failed autobuilding of package - Aborting with an error - unmounting dev/pts filesystem - unmounting proc filesystem - cleaning the build env - removing directory /var/cache/pbuilder/build//9858 and its subdirectories Are you able to reproduce this problem? I found and reported this particular problem: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=450518 I provide a temporary fix in there, but this cannot be easily fixed in pbuilder, until a new version of openmpi is uploaded. Fixing it by hand and using debuild seems to make petsc build. Uff... It's funny, I had the same problem with mpich and openmpi: when I purged mpich, it still had mpich links to mpicc and friends, so I needed to update-alternatives --auto mpicc and mpirun. It's odd because mpich has its postinst and prerm, so I'm not sure why the prerm wasn't successfully removing the links... That problem is fixed. However, the pbuilder still fails - it should fail for you too. I discovered, that you need to change this in build-depends: gfortran | fortran-compiler to this: gfortran and when you do that, it builds in pbuilder. The problem is, that it probably calls g77, instead of gfortran. Okay, I will upload a new PETSc with this fix and the new watch file you sent me. Thanks! OK, so petsc seems to be ok now. I am going to try libmesh. Please don't close this issue yet - there could be other problems with g77/gfortran and maybe the proposed fix is not really a fix (i.e. causes other problems). Need to check libmesh out first. Great. I can't wait to try out the new libmesh with its new elements for 4th order Cahn-Hilliard simulations... -Adam -- GPG fingerprint: D54D 1AEE B11C CE9B A02B C5DD 526F 01E8 564E E4B6 Engineering consulting with open source tools http://www.opennovation.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#448158: more information
Okay, I will upload a new PETSc with this fix and the new watch file you sent me. Thanks! Nice. Check it in pbuilder first please - it's working for me, but it's better if it is tested twice. Great. I can't wait to try out the new libmesh with its new elements for 4th order Cahn-Hilliard simulations... If you could help me improve the libmesh package (by reporting bugs), it'd be awesome. I myself only use the libmesh*-pure packages, that don't include petsc, because it clashes with python-petsc4py, that I am using. I only use the FEM part of libmesh and do the rest myself. At the moment. Ondrej -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#448158: more information
I just tried building petsc in my unstable chroot, and it worked fine. Perhaps I have a package installed which is not in the Build-Depends? Here's the post-configure Fortran build info: Fortran Compiler version: - Using C/C++ linker: Using Fortran linker: /usr/bin/mpif90 - Using libraries: -L/home/hazelsct/petsc-2.3.3/lib/linux-gnu-c-opt -L/home/hazelsct/petsc-2.3.3/lib/linux-gnu-c-opt -lpetscts -lpetscsnes -lpetscksp -lpetscdm -lpetscmat -lpetscvec -lpetsc-lX11 -L/usr/lib -L/usr/lib -lsuperlu -L/usr/lib -L/usr/lib -lumfpack -L/usr/lib -L/usr/lib -lamd -llapack-3 -lblas-3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -lmpi_f90 -lmpi_f77 -lgfortranbegin -lgfortran -lm -lm -lm -lm -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -lmpi_f90 -lmpi_f77 -lgfortranbegin -lgfortran -lm -lm -lm -lm -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -lmpi_f90 -lmpi_f77 -lgfortranbegin -lgfortran -lm -lm -lm -lm -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -ldl Can you do me a favor and check to see whether all of those libraries are installed in your chroot? If not, then we've found the missing lib and can add that build-dep. It seems I have all of them. Let's use pdebuild to get a reproducible results. Please use these commands: sudo pbuilder --create #create your pbuilder environment sudo pbuilder --update sudo apt-get source petsc cd petsc-2.3.3 pdebuild and you will get this (it's a different error than above, don't know why): [...] TESTING: checkCxxCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:541) TESTING: checkFortranCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:708) * UNABLE to CONFIGURE with GIVEN OPTIONS(see configure.log for details): --- bin/mpif90[f77] you provided with --with-mpi-dir=/usr does not work Run with --with-fc=0 if you wish to use this MPI and disable Fortran * make: *** [build-arch] Error 1 dpkg-buildpackage: failure: debian/rules build gave error exit status 2 pbuilder: Failed autobuilding of package - Aborting with an error - unmounting dev/pts filesystem - unmounting proc filesystem - cleaning the build env - removing directory /var/cache/pbuilder/build//9858 and its subdirectories Are you able to reproduce this problem? I fear we will have to switch to lam, to satisfy both petsc4py and libmesh. I am going to try to build everything myself using lam and if I succeed, I'll send you a patch. Dealt with this in the previous message. If you need lam for now to get your work done, petsc should build with it; please let me know if it doesn't. If openmpi is the implementation that people from lam moved to, let's get everything run using openmpi, once and for all. I am not so much interested in a hot fix, but rather for a systematic way of doing things. Right now, I am trying to discover and fix the problem with petsc. Then I'll continue with petsc4py and libmesh. Ondrej
Bug#448158: more information
TESTING: checkCxxCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:541) TESTING: checkFortranCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:708) * UNABLE to CONFIGURE with GIVEN OPTIONS(see configure.log for details): --- bin/mpif90[f77] you provided with --with-mpi-dir=/usr does not work Run with --with-fc=0 if you wish to use this MPI and disable Fortran * make: *** [build-arch] Error 1 dpkg-buildpackage: failure: debian/rules build gave error exit status 2 pbuilder: Failed autobuilding of package - Aborting with an error - unmounting dev/pts filesystem - unmounting proc filesystem - cleaning the build env - removing directory /var/cache/pbuilder/build//9858 and its subdirectories Are you able to reproduce this problem? I found and reported this particular problem: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=450518 I provide a temporary fix in there, but this cannot be easily fixed in pbuilder, until a new version of openmpi is uploaded. Fixing it by hand and using debuild seems to make petsc build. Uff... Ondrej -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#448158: more information
On Nov 7, 2007 9:52 PM, Ondrej Certik [EMAIL PROTECTED] wrote: TESTING: checkCxxCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:541) TESTING: checkFortranCompiler from config.setCompilers(/tmp/buildd/petsc-2.3.3/python/BuildSystem/config/setCompilers.py:708) * UNABLE to CONFIGURE with GIVEN OPTIONS(see configure.log for details): --- bin/mpif90[f77] you provided with --with-mpi-dir=/usr does not work Run with --with-fc=0 if you wish to use this MPI and disable Fortran * make: *** [build-arch] Error 1 dpkg-buildpackage: failure: debian/rules build gave error exit status 2 pbuilder: Failed autobuilding of package - Aborting with an error - unmounting dev/pts filesystem - unmounting proc filesystem - cleaning the build env - removing directory /var/cache/pbuilder/build//9858 and its subdirectories Are you able to reproduce this problem? I found and reported this particular problem: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=450518 I provide a temporary fix in there, but this cannot be easily fixed in pbuilder, until a new version of openmpi is uploaded. Fixing it by hand and using debuild seems to make petsc build. Uff... That problem is fixed. However, the pbuilder still fails - it should fail for you too. I discovered, that you need to change this in build-depends: gfortran | fortran-compiler to this: gfortran and when you do that, it builds in pbuilder. The problem is, that it probably calls g77, instead of gfortran. OK, so petsc seems to be ok now. I am going to try libmesh. Please don't close this issue yet - there could be other problems with g77/gfortran and maybe the proposed fix is not really a fix (i.e. causes other problems). Need to check libmesh out first. Ondrej -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#448158: more information
Hello Ondrej, On Fri, 2007-11-02 at 11:22 -0400, Adam C Powell IV wrote: I've retitled this bug to your issue, which is an FTBFS, and will try to get to it sometime next week. I'll also see if I can build and test petsc4py and libmesh using OpenMPI. On Fri, 2007-11-02 at 02:48 +0100, Ondrej Certik wrote: Package: petsc --- Please enter the report below this line. --- Hi Adam, I have some bad news - I am not able to build the package from source. How to reproduce: $ sudo apt-get update $ apt-get source petsc $ cd petsc-2.3.3 $ debuild [...] TESTING: SGIMPICheck from config.packages.MPI(/home/ondra/debian/test/petsc-2.3.3/python/BuildSystem/config/packages/MPI.py:515) TESTING: FortranMPICheck from config.packages.MPI(/home/ondra/debian/test/petsc-2.3.3/python/BuildSystem/config/packages/MPI.py:525) * UNABLE to CONFIGURE with GIVEN OPTIONS(see configure.log for details): --- Fortran error! mpi_init() could not be located! * make: *** [build-arch] Error 1 debuild: fatal error at line 1237: debian/rules build failed I just tried building petsc in my unstable chroot, and it worked fine. Perhaps I have a package installed which is not in the Build-Depends? Here's the post-configure Fortran build info: Fortran Compiler version: - Using C/C++ linker: Using Fortran linker: /usr/bin/mpif90 - Using libraries: -L/home/hazelsct/petsc-2.3.3/lib/linux-gnu-c-opt -L/home/hazelsct/petsc-2.3.3/lib/linux-gnu-c-opt -lpetscts -lpetscsnes -lpetscksp -lpetscdm -lpetscmat -lpetscvec -lpetsc-lX11 -L/usr/lib -L/usr/lib -lsuperlu -L/usr/lib -L/usr/lib -lumfpack -L/usr/lib -L/usr/lib -lamd -llapack-3 -lblas-3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -lmpi_f90 -lmpi_f77 -lgfortranbegin -lgfortran -lm -lm -lm -lm -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -lmpi_f90 -lmpi_f77 -lgfortranbegin -lgfortran -lm -lm -lm -lm -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -lmpi_f90 -lmpi_f77 -lgfortranbegin -lgfortran -lm -lm -lm -lm -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3 -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../../../lib -L/lib/../lib -L/usr/lib/../lib -L/usr/lib/gcc/x86_64-linux-gnu/4.2.3/../../.. -ldl -lmpi -lopen-rte -lopen-pal -lnsl -lutil -lgcc_s -lpthread -ldl Can you do me a favor and check to see whether all of those libraries are installed in your chroot? If not, then we've found the missing lib and can add that build-dep. But there are more serious problems: the libmesh doesn't build when petsc is compiled against openmpi and I am not able to get it run, not even the old version. Sorry I didn't find this earlier - I just tested my program with the binary libmesh package and that worked. Maybe also it was the case that it built, but something with fortran has changed, since petsc also doesn't built anymore. That's odd, I just built libmesh just fine too... Also petsc prints some very ugly messages, see for example: http://code.google.com/p/petsc4py/issues/detail?id=9 I fear we will have to switch to lam, to satisfy both petsc4py and libmesh. I am going to try to build everything myself using lam and if I succeed, I'll send you a patch. Dealt with this in the previous message. If you need lam for now to get your work done, petsc should build with it; please let me know if it doesn't. Cheers, -Adam -- GPG fingerprint: D54D 1AEE B11C CE9B A02B C5DD 526F 01E8 564E E4B6 Engineering consulting with open source tools http://www.opennovation.com/
Processed: Re: Bug#448158: more information
Processing commands for [EMAIL PROTECTED]: retitle 448158 FTBFS: Fortran MPI check fails Bug#448158: petsc - build-depends against libopenmpi-dev Changed Bug title to `FTBFS: Fortran MPI check fails' from `petsc - build-depends against libopenmpi-dev'. thanks Stopping processing here. Please contact me if you need assistance. Debian bug tracking system administrator (administrator, Debian Bugs database) -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#448158: more information
retitle 448158 FTBFS: Fortran MPI check fails thanks Hello Ondrej, This is a separate issue, and should have been a new bug. The limited platform availability of OpenMPI is a non-issue, and I was about to close it after a week of inactivity. I've retitled this bug to your issue, which is an FTBFS, and will try to get to it sometime next week. I'll also see if I can build and test petsc4py and libmesh using OpenMPI. LAM development has stopped, and its developers have moved to OpenMPI. Please see item 9 in http://www.open-mpi.org/faq/?category=general and the LAM homepage at http://www.lam-mpi.org/ for more info. I therefore feel very strongly that it is worth trying to make OpenMPI the primary base for MPI software in the lenny release, rather than reverting to the deprecated lam distribution. On the other hand, the lam build option for PETSc should still work using fakeroot debian/rules PETSC_MPI=lam binary -- or if it doesn't work, then please file a separate bug against petsc. Next time please check the bug more carefully to see if it is the same thing you're reporting, and if it's not, then open a new one. Thanks, Adam On Fri, 2007-11-02 at 02:48 +0100, Ondrej Certik wrote: Package: petsc --- Please enter the report below this line. --- Hi Adam, I have some bad news - I am not able to build the package from source. How to reproduce: $ sudo apt-get update $ apt-get source petsc $ cd petsc-2.3.3 $ debuild [...] TESTING: SGIMPICheck from config.packages.MPI(/home/ondra/debian/test/petsc-2.3.3/python/BuildSystem/config/packages/MPI.py:515) TESTING: FortranMPICheck from config.packages.MPI(/home/ondra/debian/test/petsc-2.3.3/python/BuildSystem/config/packages/MPI.py:525) * UNABLE to CONFIGURE with GIVEN OPTIONS(see configure.log for details): --- Fortran error! mpi_init() could not be located! * make: *** [build-arch] Error 1 debuild: fatal error at line 1237: debian/rules build failed But there are more serious problems: the libmesh doesn't build when petsc is compiled against openmpi and I am not able to get it run, not even the old version. Sorry I didn't find this earlier - I just tested my program with the binary libmesh package and that worked. Maybe also it was the case that it built, but something with fortran has changed, since petsc also doesn't built anymore. Also petsc prints some very ugly messages, see for example: http://code.google.com/p/petsc4py/issues/detail?id=9 I fear we will have to switch to lam, to satisfy both petsc4py and libmesh. I am going to try to build everything myself using lam and if I succeed, I'll send you a patch. Ondrej -- GPG fingerprint: D54D 1AEE B11C CE9B A02B C5DD 526F 01E8 564E E4B6 Engineering consulting with open source tools http://www.opennovation.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#448158: more information
Package: petsc --- Please enter the report below this line. --- Hi Adam, I have some bad news - I am not able to build the package from source. How to reproduce: $ sudo apt-get update $ apt-get source petsc $ cd petsc-2.3.3 $ debuild [...] TESTING: SGIMPICheck from config.packages.MPI(/home/ondra/debian/test/petsc-2.3.3/python/BuildSystem/config/packages/MPI.py:515) TESTING: FortranMPICheck from config.packages.MPI(/home/ondra/debian/test/petsc-2.3.3/python/BuildSystem/config/packages/MPI.py:525) * UNABLE to CONFIGURE with GIVEN OPTIONS(see configure.log for details): --- Fortran error! mpi_init() could not be located! * make: *** [build-arch] Error 1 debuild: fatal error at line 1237: debian/rules build failed But there are more serious problems: the libmesh doesn't build when petsc is compiled against openmpi and I am not able to get it run, not even the old version. Sorry I didn't find this earlier - I just tested my program with the binary libmesh package and that worked. Maybe also it was the case that it built, but something with fortran has changed, since petsc also doesn't built anymore. Also petsc prints some very ugly messages, see for example: http://code.google.com/p/petsc4py/issues/detail?id=9 I fear we will have to switch to lam, to satisfy both petsc4py and libmesh. I am going to try to build everything myself using lam and if I succeed, I'll send you a patch. Ondrej --- System information. --- Architecture: i386 Kernel: Linux 2.6.21-1-686 Debian Release: lenny/sid 500 unstableftp.cz.debian.org --- Package information. --- Depends (Version) | Installed ===-+-=== | -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]