Re: how to track down parallel build issues?

On Saturday 22 May 2010, Andreas Radke wrote:
> Am Mon, 12 Apr 2010 14:33:17 +0200
>
> schrieb Petr Mladek <[email protected]>:
> > What do you mean with "I can build the package with -j1"? Do you mean
> > that the error does not happen when you build with -j1?
>
> Yes. Build finished when built non-smp.

Just to be sure. Did you do both build on the same system, same machine?

> I don't run into the lucene issue right now but something that looks
> similar:

grr, I feel that it is a random bug.

> This happened with ooo-build 3.2.1.1/m17 here. Maybe that's something
> different.

It is hard to guess.

> Should I try the patching again to get a strace.log?

Yes, please. The following steps worked for me (just tried):

# set paths
ooo_build_dir="/tmp/go-openoffice/trunk/src/ooo-build-3.2.1.1/"
solver_dir="$ooo_build_dir/build/ooo320-m17/solver/320/unxlngx6.pro/"

# copy the attached regcomp wrapper instead of the current hardlinked binary
# IMPORTANT: it is important to move the file before you copy the new one;
# otherwise you would rewrite regcomp.bin via hardlink!!
mv $solver_dir/bin/regcomp $solver_dir/bin/regcomp.saved
cp /saved_attachements/regcomp $solver_dir/bin/regcomp
chmod 755 $solver_dir/bin/regcomp

# run installation
cd $ooo_build_dir/bin
./ooinstall /opt/ooo-test 2>&1 | tee /install.1.log

Please, send me the /tmp/strace.1.log and the problematic 
LuceneHelpWrapper.jar.

Please, bzip2 the strace log if it is too huge.

BTW: The strace from my system is available at
http://download.go-oo.org/tstnvl/3.2.1.1-lucene-component-registration/strace.pmladek.1.log.bz2

> And yes, I'm building official packages for ArchLinux. Several branches:
> Usually vanilla stable released, latest testing (m18 now) and development
> version (m78) + latest stable bugfix release from go-openoffice/ooo-build.
>
> The smp install trouble doesn't happen in the vanilla versions. Is there a
> way to prevent smp usage for the install process?

Do you use the same configure flags with the vanilla version and ooo-build? 
Especially, do you use the same system libraries, java, ...? Do you do all 
builds on the same machine?

> make -j1 DESTDIR=${pkgdir} install || return 1  - that doesn't help here.

IMHO, the installation is never parallelized. I think that the root of the 
problem is in the build phase.

> The build is configured --with-max-jobs=${MAKEFLAGS/-j/} where I usually
> build -j6.

I use even more parallelized build and do not see such problems here.

-- 
Best Regards,

Petr Mladek
software developer
---------------------------------------------------------------------  
SUSE LINUX, s. r. o.                        e-mail: [email protected]
Lihovarská 1060/12                          tel: +420 284 028 952
190 00 Prague 9                             fax: +420 284 028 951
Czech Republic                              http://www.suse.cz/