On 22/07/12 11:25, Kristian Rietveld wrote: > (Please keep me on CC as I am not subscribed to the list)
> During the build, however, the "sw_filters_test" unit test crashes. The > crash happens in rtl_string_release from the FieldEntry (the one in > sw/source/file/ww8/ww8par.hxx) destructor. The FieldEntry destructor is > called through a dyld stub. > > After reading assembly code I figured that the FieldEntry destructor > being executed does not match the FieldEntry's fields. What was already > suspicious above is that rtl_string_release is being called even though > this FieldEntry uses OUStrings. > > From the assembly code I guess the FieldEntry destructor being executed > is actually the one in registry/source/reflwrit.cxx. So my impression > is that the dynamic linker on OS X is simply calling the wrong > destructor (the symbols have the same name). that is really not supposed to happen, we really should build everything with -fvisibility=hidden on all platforms where unnecessarily exported symbols cause such problems (guess that means everything but Windows). on master and libreoffice-3-6, in solenv/gbuild/platform/macosx.mk the -fvisibility=hidden is inside: > ifeq ($(HAVE_GCC_VISIBILITY_FEATURE),TRUE) can you check that that is actually detected properly for your compiler? and if so, that it's actually on the command line in a "make registry VERBOSE=t"? perhaps the check for it is broken somehow. ah, there is another one, i wonder what that does: > ifneq ($(HAVE_GCC_VISIBILITY_BROKEN),TRUE) possibly in this day and age we should strongly consider hard-requiring visibility support in configure, i.e. reject compilers that can't do it. _______________________________________________ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice
