Re: [Libreoffice] Need help with failing cppunit tests

On Wed, Feb 23, 2011 at 12:36:14PM +0000, Caolán McNamara wrote:
> On Wed, 2011-02-23 at 09:46 +0100, Francois Tigeot wrote:
> > For now, I'm focussing on sc/qa/unit.
>
> Because your log shows "OK (10)" before it falls over and dies, I
> believe that the tests themselves are passing, and you are dying on
> exit.

This means the other errors such as
  UniversalContentBroker::configureUcb(): No configuration
  ContentBroker::initialize - already initialized!
  PropertySetRegistry::getConfigProvider - No config provider!
  etc...
are only caused by the first issue ?

> Could be a few reasons for that, I suppose the first one to check
> is what's your autogen/configure line ?

I use these values:

export CFLAGS="-O -g -pipe"
export CPPFLAGS=-I/usr/pkg/include
export LD_LIBRARY_PATH=/usr/pkg/lib

./autogen.sh \
        --with-max-jobs=2 --with-num-cpus=2 \
        --prefix=/usr/local/LibreOffice \
        --enable-symbols --enable-debug \
        --x-includes=/usr/pkg/include --x-libraries=/usr/pkg/lib \
        --with-lang="en" \
        --without-java \
        --disable-mozilla --disable-nss-module \
        --disable-neon --disable-Xaw \
        --with-system-headers --with-system-libs \
        --with-system-python --with-system-libtextcat \
        --without-system-jars \
        --without-system-db \
        --without-system-mdds \
        --without-system-vigra \
        --without-system-altlinuxhyph \
        --without-system-mythes \
        --without-system-odbc \
        --without-system-cppunit \

The environment variables are needed, otherwise some components do not find
the correct path for include files or libraries.

> To find out if you are using "system" cppunit or the internal one, the
> internal one has a fix for when building with debug-mode stl

I tried both. I ported locally one of the LibreOffice patches which fixed
a similar crash to the system package but it had no effect.
Currently I'm using the internal fork.
 
> Another possibility is some nasty atexit/__cxa_atexit problem. A hack to
> experiment for that possibility can be found at
> http://lists.freedesktop.org/archives/libreoffice/2011-January/005372.html

Hmm. This could be a bit ugly.
I have applied the patch and I'm doing a partial build right now to see if it
changes anything.

-- 
Francois Tigeot