Koji build results fail to match local rawhide installation's

Tom Lane tgl at redhat.com
Thu Feb 21 20:29:01 UTC 2008


mysql is currently failing to build in rawhide on ppc, with symptoms
that indicate a problem in its code for detecting how much stack space
remains available.  The latest example is
http://koji.fedoraproject.org/koji/taskinfo?taskID=446073

I've been fighting this for more than a month with no success in
isolating the problem.  In order to have enough access to poke at it
with gdb, I recently decided to resort to installing rawhide on an old
Mac laptop I wasn't doing anything with ... and behold, mysql builds
and passes its regression tests just fine!  So now I don't know what
to think.  It seems that either

(1) the koji environment doesn't match a rawhide install, or
(2) there's a relevant hardware difference between my poor old
    Powerbook G3 and the PPC machines in the koji build farm.

Since the problem has been present in Koji since at least early January,
while my Mac worked both just after an F9-Alpha install and after a
subsequent "yum update" to yesterday's rawhide, it's not just a matter
of being slightly out of sync on package versions.  Whatever's busted
has been so for awhile.

I'd appreciate any hints on where to look or what to try next.  In
particular, are there any real details available on the machines in
the build farm, such as exact CPU types?

			regards, tom lane




More information about the fedora-devel-list mailing list