Hi Christoph, thanks for the info. We had disabled the optional postgresql on hurd-i386 in the previous pspp release.
Is it maybe an idea to include this test case in the regression in postgresql? That would stop the release of the non-functional postgresql on hurd-i386, no? Friedrich > Am 12.04.2016 um 17:39 schrieb Christoph Berg <m...@debian.org>: > > Re: Friedrich Beckmann 2016-04-12 > <3ad966ac-2984-4449-94ba-addfe00d9...@gmx.de> >> Package: postgresql-9.5 >> Version: 9.5.2-1 >> >> During test installation on buildd for pspp, the pspp build fails on >> hurd-i386 during the setup >> of the postgresql. When I disable postgresql, then pspp builds and works. >> >> This is specific to the hurd-i386 architecture. The pspp package builds on >> other architectures. >> >> Friedrich >> >> The full log: >> https://buildd.debian.org/status/fetch.php?pkg=pspp&arch=hurd-i386&ver=0.10.1-2&stamp=1460387447 > > Hi Friedrich, > > the problem is in the hurd kernel that doesn't implement semaphores. > The interesting initdb error is this: > > creating template1 database in > /«PKGBUILDDIR»/build/src/test/regress/./tmp_check/data/base/1 ... FATAL: > could not create semaphores: Function not implemented > DETAIL: Failed system call was semget(1, 17, 03600). > > We have the weird situation where the server compiles successfully > including semget(), but isn't able to get executed because the > the kernel doesn't implement the feature. > > I've talked to the hurd people, and Richard Braun was confirming that > inter-process semaphores are not implemented yet. > > What we can do from the PostgreSQL side is to replace the current sysv > semaphores (semget) by POSIX semaphores (sem_init) which don't work > either, but have a much greater chance of getting implemented in the > future. So some day, it will just work, but until then, PostgreSQL > will unfortunately not really be available on that platform. > > Christoph
