Re: Problem building in Gerrit, Windows only.

On 24.10.2016 14:20, Michael Stahl wrote:
> On 20.10.2016 10:52, Stephan Bergmann wrote:
> > On 10/20/2016 09:13 AM, Giuseppe Castagno wrote:
> > > a patch of mine:
> > > <https://gerrit.libreoffice.org/#/c/30036/4>
> > > always fails in Gerrit Windows build machine in test
> > > PythonTest_solenv_python:
> > > <http://ci.libreoffice.org/job/lo_gerrit/1657/Config=windows_msc_dbgutil_32/console>
> > >
> > >
> > > Unfortunately I'm not able to reproduce it on my Window 10 build machine
> > > (Win10 64bit, VS2013 64bit, LO 32bit) , where that same test doesn't fail.
> > >
> > > I need help, because I have no idea what's going on.
> > > May be make internal error, as stated in these build log lines:
> > >
> > > "
> > > make[3]: Entering directory
> > > '/home/tdf/lode/jenkins/workspace/lo_gerrit/Config/windows_msc_dbgutil_32/solenv/qa/python/selftest'
> > >
> > > make[3]: *** internal error: invalid --jobserver-fds string
> > > 'gmake_semaphore_5488'.  Stop.
> > > make[3]: Leaving directory
> > > '/home/tdf/lode/jenkins/workspace/lo_gerrit/Config/windows_msc_dbgutil_32/solenv/qa/python/selftest'
> > >
> > > "
> > > can give some hint?
> >
> > Looking at 
> > <http://ci.libreoffice.org/job/lo_gerrit/Config=windows_msc_dbgutil_32/>, 
> > quite a number of the recent master Gerrit/Jenkins Windows builds are 
> > red, and, scanning through their console output, quite a number fail 
> > with the same "ERROR: test_gbuildtoide (gbuildtoide.CheckGbuildToIde)" 
> > as your 
> > <http://ci.libreoffice.org/job/lo_gerrit/1657/Config=windows_msc_dbgutil_32/console>.
> >
> > Looks like there is still problems there after 
> > <https://cgit.freedesktop.org/libreoffice/core/commit/?id=f5c54089b50718abf7c35aa81b150c509809d5c4>
> >  
> > "prep WinResTarget for WNT in testdir".  Björn?
>
> if i insert a "subprocess.check_call(['/bin/sh', '-c', 'set'])" in that
> test i see these variables:
>
> MAKEFLAGS='rs -j --jobserver-fds=3,4'
> MAKELEVEL=2
> MAKE_TERMERR=/dev/pts/4
> MAKE_TERMOUT=/dev/pts/4
>
> but not MAKE itself so the test just invokes the first make in the PATH.
>
> let's try if this helps:
>  https://gerrit.libreoffice.org/30226

this succeeded on the same "tb72" that produced the above failed build
log, so i've pushed it to master now.