Re: scalapack tests missing on m68k
Hi Drew! On 08/15/2017 11:11 AM, Drew Parsons wrote: > The standard build pushes the BLACS test executables into TESTING > instead of the local location BLACS/TESTING. When ctest runs (i.e. > when nocheck is not activated) it rebuilds the tests, leaving test > executables in BLACS/TESTING (where testing for BLACS is performed). > My scripts were pulling those executables (xCbtest and xFbtest) out of > BLACS/TESTING to install in scalapac-mpi-test. > > When nocheck is active and ctest is not run, then the executables never > get created in BLACS/TESTING, so dh_install fails. But they have been > built already in the toplevel TESTING dir. > > The solution is to not install from BLACS/TESTING, but instead link > from there to the corresponding executable installed in TESTING. > > So the m68k build should be successful for scalapack 2.0.2-1exp7. Great! > Thanks for your help debugging. Thanks for fixing your software on m68k ;-). Adrian -- .''`. John Paul Adrian Glaubitz : :' : Debian Developer - glaub...@debian.org `. `' Freie Universitaet Berlin - glaub...@physik.fu-berlin.de `-GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
Re: scalapack tests missing on m68k
On Wed, 2017-08-09 at 08:07 -0400, John Paul Adrian Glaubitz wrote: > On 08/09/2017 01:17 AM, Drew Parsons wrote: > > > I can run the testsuite manually if you like. > > > > That would be handy, especially to know if the tests have in fact > > been > > built. The log doesn't show any complaints about the test > > executables, > > but on the other hand the empty test results together with the > > dh_install error suggests they're missing. > > The builds are all performed with "nocheck nobench" which is > intentional. I can reproduce the scalapack build failure on amd64 using nocheck. The standard build pushes the BLACS test executables into TESTING instead of the local location BLACS/TESTING. When ctest runs (i.e. when nocheck is not activated) it rebuilds the tests, leaving test executables in BLACS/TESTING (where testing for BLACS is performed). My scripts were pulling those executables (xCbtest and xFbtest) out of BLACS/TESTING to install in scalapac-mpi-test. When nocheck is active and ctest is not run, then the executables never get created in BLACS/TESTING, so dh_install fails. But they have been built already in the toplevel TESTING dir. The solution is to not install from BLACS/TESTING, but instead link from there to the corresponding executable installed in TESTING. So the m68k build should be successful for scalapack 2.0.2-1exp7. Thanks for your help debugging. Drew
Re: scalapack tests missing on m68k
On Wed, 2017-08-09 at 08:07 -0400, John Paul Adrian Glaubitz wrote: > On 08/09/2017 01:17 AM, Drew Parsons wrote: > > > I can run the testsuite manually if you like. > > > > That would be handy, especially to know if the tests have in fact > > been > > built. The log doesn't show any complaints about the test > > executables, > > but on the other hand the empty test results together with the > > dh_install error suggests they're missing. > > The builds are all performed with "nocheck nobench" which is > intentional. > I see. That explains why the buildd didn't run the tests, which is fine. In that case it's just dh_install that we need to fix, where it reports missing files build-openmpi/BLACS/TESTING/x* build-mpich/BLACS/TESTING/x* The logs indicate that xCbtest and xFbtest should have been built there. (I presume that error means that build-openmpi/TESTING/x*, etc. have been found by dh_install) Drew
Re: scalapack tests missing on m68k
On 08/09/2017 01:17 AM, Drew Parsons wrote: >> I can run the testsuite manually if you like. > > That would be handy, especially to know if the tests have in fact been > built. The log doesn't show any complaints about the test executables, > but on the other hand the empty test results together with the > dh_install error suggests they're missing. The builds are all performed with "nocheck nobench" which is intentional. Adrian -- .''`. John Paul Adrian Glaubitz : :' : Debian Developer - glaub...@debian.org `. `' Freie Universitaet Berlin - glaub...@physik.fu-berlin.de `-GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
Re: scalapack tests missing on m68k
On Tue, 2017-08-08 at 23:49 -0400, John Paul Adrian Glaubitz wrote: > On 08/08/2017 02:30 AM, Drew Parsons wrote: > > scalapack 2.0.2 is in experimental. It appears to build well > > enough on > > m68k [1], but something is going wrong with the test executables. > > > > dh_auto_test isn't running any tests (the executables are in build- > > openmpi/TESTING, and elsewhere in .../BLACS, .../PBLAS, build- > > mpich). > > I can run the testsuite manually if you like. That would be handy, especially to know if the tests have in fact been built. The log doesn't show any complaints about the test executables, but on the other hand the empty test results together with the dh_install error suggests they're missing. > > > dh_install reports: > > Unsupported ioctl: cmd=0x40049409 > > Unsupported ioctl: cmd=0x40049409 > > Unsupported ioctl: cmd=0x40049409 > > dh_install: Cannot find (any matches for) "build- > > openmpi/BLACS/TESTING/x*" (tried in ., debian/tmp) > > dh_install: scalapack-mpi-test missing files: build- > > openmpi/BLACS/TESTING/x* > > > > Is an "Unsupported ioctl" message like this normal, when dh_install > > has trouble finding files? > > This is just coming from the fact that the builds are performed on > QEMU. The messages aren't from dh_install but QEMU and can be > ignored. > > > What is m68k doing differently from the other arches? > > We're not building on real hardware at the moment. Ah, that explains the difference. Thanks for your help, Drew
Re: scalapack tests missing on m68k
On 08/08/2017 02:30 AM, Drew Parsons wrote: > scalapack 2.0.2 is in experimental. It appears to build well enough on > m68k [1], but something is going wrong with the test executables. > > dh_auto_test isn't running any tests (the executables are in build- > openmpi/TESTING, and elsewhere in .../BLACS, .../PBLAS, build-mpich). I can run the testsuite manually if you like. > dh_install reports: > Unsupported ioctl: cmd=0x40049409 > Unsupported ioctl: cmd=0x40049409 > Unsupported ioctl: cmd=0x40049409 > dh_install: Cannot find (any matches for) > "build-openmpi/BLACS/TESTING/x*" (tried in ., debian/tmp) > dh_install: scalapack-mpi-test missing files: > build-openmpi/BLACS/TESTING/x* > > Is an "Unsupported ioctl" message like this normal, when dh_install has > trouble finding files? This is just coming from the fact that the builds are performed on QEMU. The messages aren't from dh_install but QEMU and can be ignored. > What is m68k doing differently from the other arches? We're not building on real hardware at the moment. -- .''`. John Paul Adrian Glaubitz : :' : Debian Developer - glaub...@debian.org `. `' Freie Universitaet Berlin - glaub...@physik.fu-berlin.de `-GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
