Re: [Libreoffice-commits] .: sw/Module_sw.mk

On Friday 27 of April 2012, Lubos Lunak wrote:
>  sw/Module_sw.mk |    1 -
>  1 file changed, 1 deletion(-)
>
> New commits:
> commit e8897be1e1883b327cd324ab45e84fbed3af41df
> Author: Luboš Luňák <l.lunak@suse.cz>
> Date:   Fri Apr 27 19:29:07 2012 +0200
>
>     temporarily disable a test with a build race condition
>
> diff --git a/sw/Module_sw.mk b/sw/Module_sw.mk
> index bc0679c..de2bd90 100644
> --- a/sw/Module_sw.mk
> +++ b/sw/Module_sw.mk
> @@ -55,7 +55,6 @@ $(eval $(call gb_Module_add_slowcheck_targets,sw,\
>      CppunitTest_sw_filters_test \
>      CppunitTest_sw_macros_test \
>      CppunitTest_sw_subsequent_ooxmltok \
> -    CppunitTest_sw_subsequent_ww8tok \
>      CppunitTest_sw_subsequent_rtfexport \
>      CppunitTest_sw_subsequent_rtftok \
>  ))

 Miklos, as I said, this had nothing to do with the Clang compiler. I've set 
up another tindebox with exactly the same setup except for using GCC, and it 
has the problem as well. The Clang tinderbox has a setup that is actually 
unique or rare in several aspects besides the compiler (32bit, icecream, make 
check), so I'll keep it running, so that people do not go looking for Clang 
problems that are not there. 

 In this case I assume the problem is because this tinderbox uses icecream and 
so the build is massively parallel and more likely to trigger build race 
conditions. Build finishes fine if I run make again. I don't know why exactly 
it breaks (wouldn't it be nice if the warning messages actually said why 
things went wrong?), but I suspect the reference to 
$(OUTDIR)/unittest/registry in the *.mk file.

-- 
 Lubos Lunak
 l.lunak@suse.cz