Package building on xen- or kvm-based buildds?
As was mentioned on IRC last night, there seems to be a difference of
build success on xen versus kvm based buildds. Some packages are
building OK for me on KVM and failing at the Xen-based buildds. One
major problem is the fakeroot package, which on other systems is
fakeroot-sysv, but on GNU/Hurd it is fakeroot-tcp. According to Olaf B.
there seems to also exist a fakeroot-hurd package, however not yet
complete. The fakerot-tcp problem is a race condition, see bug #534879
According to Pino T. a sysv-package can be built when shared memory is
implemented (I thought it was already available?)
Package building OK for me on kvm currently are clisp, aptitude and
rootskel (Samuel T. , which patch is needed?). One difference between a
buildd and a user environment is the chroot environment on a buildd.
However, aptitude was built on the kvm-based buildd ironforge last night
and the build succeeded. So maybe problematic packages should be tried
on ironforge if they fail on the other buildds.
Another problematic package id postgresql-8.4, which builds OK on my
box, but fails when initializing the database. The fist try gave the
common fakeroot output:
libfakeroot: connect: Connection refused
but the second try revealed another problem:
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).
child process exited with exit code 1
Looks like semget() is a stub. In that case these tests should be
disabled, or there is no hope to get this package built before the
functionality is available. The semget problems seems to be the same
with postgresql-9.1 when looking at the buildd logs as well.