[Libreoffice] feature/gbuild_cppuhelper

[Stephan Bergmann]

Matúš,

I had intended to come back much earlier with feedback on your 
feature/gbuild_cppuhelper branch, but the situation turns out to be 
somewhat problematic:  I set out with a recent master (including Michael 
Stahl's fix for adding the ".3" at the end of libuno_*.dylib install 
names) on Mac OS X and cherry-picked the two commits from the 
feature/gbuild_cppuhelper branch on top of it.  Two problems I 
identified so far are:

- cppuhelper/CppunitTest_cppuhelper_qa_weak.mk cannot be included in the 
build, as it depends on module bridges (cppu::getCaughtException needs 
the C++ UNO bridge library), which already depends on module cppuhelper. 
  A short-term solution is probably to disable the test again (it looks 
like it was not enabled before gbuild'ification, right?).

- odk/pack/checkbin fails, complaining about missing HTML documentation 
files for various entities from cppuhelper.  It appears that autodoc 
cannot correctly parse things like

   /** ... */
   class CPPUHELPER_DLLPUBLIC AccessControl { ... };

to associate the documentation comment with class AccessControl, so HTML 
documentation for cppu::AccessControl is missing.  (The same autodoc 
problem already hits salhelper classes like salhelper::Condition, but 
went unnoticed because odk/util/check.pl fails to check for any 
salhelper content.)  The mid-term solution is certainly to get rid of 
autodoc (and hope that alternatives are mature enough to have no problem 
here).  A short-term solution might be to adapt odk/util/check.pl to not 
check for the missing HTML files, and get the mid-term solution in place 
until LO 3.6.

With those two obstacles out of the way, builds kept failing with 
strange crashes when running executables that themselves were created 
during the build, like the help compiler in module helpcontent2 or UNO 
bridge test code in module testtools.  Some of this I could track down 
to unclean rebuilds, but others remained rather mysterious.

I then went back to a clean recent master (as of this morning), without 
the feature/gbuild_cppuhelper changesets, and doing a clean build of it 
fails for me as well, in helpcontent2, sdext, and tail_build. 
(Curiously, current Mac OS X tinderbox builds apparently only fail in 
tail_build.)  So, it looks like master needs to be fixed first before 
verification of that branch can continue...

Have a nice Christmas time,
Stephan