[Buildroot] autobuilder improvements and fixes

[Buildroot] autobuilder improvements and fixes

[Thomas Petazzoni]

Hello,

Two quick informations about the autobuilder tests running on the Free
Electrons server:

 * I have updated the Buildroot external toolchains. They are now all
   built with Buildroot 2014.02-rc1, which should solve a certain
   number of static linking issues against libstdc++.a. However, the
   SH4 toolchain is still an old one generated by Buildroot 2013.05,
   due to the impossibility of generating a SH4 toolchain since the
   internal toolchain backend rework.

 * I have found the reason for the various "TIMEOUT" builds that we saw
   from time to time. Then occurred when a (e)glibc toolchain was
   chosen, and when the test script was deciding to do a static library
   build. In this case, the script was detecting that this
   configuration should not be tested (because glibc + static is not
   supported by glibc), but it was doing that in a completely stupid
   way which was making the script regenerate a random selection of
   packages (make randpackageconfig) during 4 hours in a loop, until
   the build was killed. The result was not only that we were having
   these stupid erroneous results, but also that one of the three build
   instances running on the server was consuming 4 hours of CPU time
   for nothing. I have worked around the problem for now, we will see
   if it works correctly within the next hours/days. It is still on my
   todo-list to rewrite the autobuilder script, to fix this kind of
   ugly things :)

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Buildroot] autobuilder improvements and fixes

[Peter Korsgaard]

>>>>> "Thomas" == Thomas Petazzoni <thomas.petazzoni@free-electrons.com> writes:

 > Hello,
 > Two quick informations about the autobuilder tests running on the Free
 > Electrons server:

 >  * I have updated the Buildroot external toolchains. They are now all
 >    built with Buildroot 2014.02-rc1, which should solve a certain
 >    number of static linking issues against libstdc++.a. However, the
 >    SH4 toolchain is still an old one generated by Buildroot 2013.05,
 >    due to the impossibility of generating a SH4 toolchain since the
 >    internal toolchain backend rework.

Great!

 >  * I have found the reason for the various "TIMEOUT" builds that we saw
 >    from time to time. Then occurred when a (e)glibc toolchain was
 >    chosen, and when the test script was deciding to do a static library
 >    build. In this case, the script was detecting that this
 >    configuration should not be tested (because glibc + static is not
 >    supported by glibc), but it was doing that in a completely stupid
 >    way which was making the script regenerate a random selection of
 >    packages (make randpackageconfig) during 4 hours in a loop, until
 >    the build was killed. The result was not only that we were having
 >    these stupid erroneous results, but also that one of the three build
 >    instances running on the server was consuming 4 hours of CPU time
 >    for nothing. I have worked around the problem for now, we will see
 >    if it works correctly within the next hours/days. It is still on my
 >    todo-list to rewrite the autobuilder script, to fix this kind of
 >    ugly things :)

Heh, sounds funky ;)

Thanks.

-- 
Bye, Peter Korsgaard