Avoid indeterminism in kcolorschemeeditor
Without this patch,
build/kcms/colors/CMakeFiles/kcolorschemeeditor.dir/kcolorschemeeditor_autogen/mocs_compilation.cpp.o
and the resulting /usr/bin/kcolorschemeeditor varied from parallelism.
See https://reproducible-builds.org/ for why this matters.
See also https://bugzilla.opensuse.org/show_bug.cgi?id=1109420
| Automatic diff as part of commit; lint not applicable. |
| Automatic diff as part of commit; unit tests not applicable. |
Well it's not really a dependency, is it? They only use the same sources to build up two different views of the same widget. This patch is hacky as the original cmakefile... I understand the need but can't the patch fix the root cause instead?
If you look at http://rb.zq1.de/compare.factory-20181121/plasma5-desktop-compare.out
You can see that the asm diff between a -j1 and a -j4 build of openSUSE's plasma5-desktop package contains (among many others)
-_ZN18SchemeEditorDialog18qt_static_metacallEP7QObjectN11QMetaObject4CallEiPPv.part.1:
+_ZN19SchemeEditorEffects18qt_static_metacallEP7QObjectN11QMetaObject4CallEiPPv.part.1:
and I found that it goes away when you tell cmake or make to build one after the other (either order works) like with this patch.
But what is the root cause then? And how do you fix it better?
The root cause is that we build twice the same elements instead of having one share lib used on the two contexts.
I guess this is more involved to do the proper fix, so at first yours can be OK... Push it!
Can you please push it?
I am contributing so rarely, that it is not worth having commit access.