Challenge for real experts: why does cmake 3.1.3 not use automoc for a certain file?

[Friedrich W. H. Kossebau]

Am Dienstag, 16. Juni 2015, 16:00:04 schrieb Friedrich W. H. Kossebau:
> Am Montag, 8. Juni 2015, 14:52:12 schrieb Ben Cooksley:
> > On Sat, Jun 6, 2015 at 10:07 PM, Alex Merry <alex.merry at kde.org> wrote:
> > > On Friday 05 June 2015 00:58:30 Friedrich W. H. Kossebau wrote:
> > >> you are a cmake and KDE buildsystem expert?
> > >> Prove that yourself and improve the world for others ;)
> > > 
> > > Or, indeed, work towards becoming such an expert...
> > > 
> > >> Here is the challenge:
> > >> 
> > >> https://build.kde.org/job/calligra%20frameworks%20kf5-qt5/PLATFORM=Linu
> > >> x,
> > >> Var iation=All,compiler=gcc/28/consoleText
> > >> 
> > >> CMakeFiles/KisAsyncMergerTest.dir/kis_async_merger_test.cpp.o: In
> > >> function
> > >> `KisAsyncMergerTest::KisAsyncMergerTest()':
> > >> /home/jenkins/builds/calligra/kf5-
> > >> qt5/krita/image/tests/kis_async_merger_test.h:24: undefined reference
> > >> to
> > >> `vtable for KisAsyncMergerTest'
> > >> CMakeFiles/KisAsyncMergerTest.dir/kis_async_merger_test.cpp.o: In
> > >> function
> > >> `KisAsyncMergerTest::~KisAsyncMergerTest()':
> > >> /home/jenkins/builds/calligra/kf5-
> > >> qt5/krita/image/tests/kis_async_merger_test.h:24: undefined reference
> > >> to
> > >> `vtable for KisAsyncMergerTest'
> > >> collect2: error: ld returned 1 exit status
> > >> krita/image/tests/CMakeFiles/KisAsyncMergerTest.dir/build.make:89:
> > >> recipe
> > >> for target 'krita/image/tests/KisAsyncMergerTest' failed
> > >> 
> > >> Right, you say, smells like automoc not run. And indeed it is not run.
> > >> So
> > >> far you are an expert.
> > >> 
> > >> But why?
> > >> Why does cmake on CI not want to run automoc on
> > >> kis_async_merger_test.{h,cpp}?
> > >> 
> > >> Locally it works fine for anyone (cmake 3.0.1/3.0.2), noone yet was
> > >> able
> > >> to
> > > 
> > >> reproduce the problem, instead everyone sees:
> > > So my first question is: what version of cmake is being run on the build
> > > servers? If it's not 3.0.1 or 3.0.2, then that could explain why you
> > > can't
> > > reproduce it locally.
> > 
> > The CI system is using "cmake version 3.1.3" on Ubuntu Vivid (15.04).
> > Not sure if that makes a difference.
> 
> Seems it does make a difference 8\ ...
> 
> With cmake version 3.0.1 (from OpenSUSE Tumbleweed package):
> $ cd build/calligra.moment/krita/image/tests
> $ make help  | grep sync
> ... KisAsyncMergerTest
> ... KisAsyncMergerTest_automoc
> ... KisAsyncMergerTest_automoc.o
> ... KisAsyncMergerTest_automoc.i
> ... KisAsyncMergerTest_automoc.s
> ... kis_async_merger_test.o
> ... kis_async_merger_test.i
> ... kis_async_merger_test.s
> 
> With cmake version 3.1.3 (manually compiled on my OpenSUSE Tumbleweed):
> $ cd build/calligra.moment/krita/image/tests
> $ make help  | grep sync
> ... KisAsyncMergerTest
> ... kis_async_merger_test.o
> ... kis_async_merger_test.i
> ... kis_async_merger_test.s
> 
> Uh? Anyone seen something similar? Any idea what could make that newer cmake
> decide that kis_async_merger_test.* does not need automoc?
> Putting in some #include "moc_kis_async_merger_test.cpp" also did not have
> any effect.
> 
> Anyone an idea for a workaround?

Interesting... seems renaming the target "KisAsyncMergerTest" to something 
else, e.g. "KisAsyncMergerTest_" unbreaks this.
Any explanation? There is no other target with that name "KisAsyncMergerTest".

Having done that workaround, now another target is hit with the same problem, 
KisBookmarkedConfigurationManagerTest. Got the same treatment with _ appended, 
proper automoc targets now created for me locally.

Currently waiting if CI finds the next problem now (not enough disk space left 
right now locally for giving a full test myself with the other cmake version, 
Calligra/Krita is too large :), so partially relying on build.kde.org).

Could there be some hash collision or something else? There are a lot of 
targets in that CMakeLists.txt, perhaps more than someone expected...
http://quickgit.kde.org/?p=calligra.git&a=blob&h=082747066eb6910d29a7fd3a59b6b531af529e95&f=krita%2Fimage%2Ftests%2FCMakeLists.txt&o=plain

For the Qt4/kdelibs4 version of Calligra, which has almost the same code, no 
such issues could be seen...

What else could I try to get more insight?

Cheers
Friedrich