IMPORTANT/WARNING buildsystem is not stable / unreliable

[Hans de Goede]

Dan Williams wrote:
> On Fri, 2005-11-25 at 13:59 +0100, Hans de Goede wrote:
>> Hi all,
>>
>> Sorry for shouting but I hope I've got some attention this time around. 
>> I've reported this before, see:
>> https://www.redhat.com/archives/fedora-extras-list/2005-November/msg00804.html
>>
>> Today I again got the same inconsistent results again on x86_64 but this 
>> with a FC-4 build (again of directfb). The firsttime it failed (crashed) 
>> with the (interesting?) message below.
>>
>> I did a requeue within minutes and this time it build fine. This does 
>> not sound like a consistent build environment to me!
>>
>> I've saved all the logs off the failed attempt before the requeue and 
>> the succeeded attempt logs are at:
>> http://buildsys.fedoraproject.org/logs/fedora-4-extras/1292-directfb-0.9.24-4.fc4/
>>
>> Maybe its an idea to run memtest for a couple of hours on hammer? and/or 
>> to run the old burn-it kernel compile script once used to detect the 
>> notorious hard to find K6 bug.
> 
> I track all the failed builds pretty closely, I usually comb through
> them every day or two.  And directfb seems to be the only one with these
> sorts of problems.  There are, of course, two explanations, one of which
> you've pointed out here:
> 
> 1) The build machines are screwed up or have a hardware fault
> 2) The directfb package has build issues either with gcc, with the build
> process, or both
> 
> Given that directfb is the only package in Extras that I've seen that
> actually _segfaults_ gcc, I'm tending to blame directfb at the moment.
> What I'll try to do is build with make -j4 to see if there are parallel
> make issues in directfb.  Also, does it always seem to fail on x86_64?
> Has it ever failed oddly on i386 or ppc32?
> 

Nope, it always fails on x86_64. I'm happy to hear that the 
buildmachines are "closely" monitored and that directfb is the only 
problem package. I agree with you that in this case it is most likely 
that directfb is tripping over a gcc or buildprocess bug.

Also notice that I've removed your CFLAGS patch, and that it does build 
without, but not always. Sofar 3 out of 4 build attempts without your 
cflags patch succeeded, 2 devel builds and 1 FC-4 build, the failed 
build was an FC-4 build.

Thanks for looking into this.

Regards,

Hans