On 07/12/2014 01:53 PM, Pádraig Brady wrote: > On 07/12/2014 12:39 PM, Pádraig Brady wrote: >> On 07/11/2014 03:43 PM, Daiki Ueno wrote: >>> Bernhard Voelker <m...@bernhard-voelker.de> writes: >>> >>>> On 07/11/2014 03:13 PM, Daiki Ueno wrote: >>>>> Could you provide an example of the build failures with 0.19.1? >>>> >>>> https://build.opensuse.org/package/live_build_log/Base:System/coreutils/openSUSE_Factory/i586 >>>> >>>> [ 172s] *** Error in `/usr/bin/xgettext': realloc(): invalid pointer: >>>> 0x098f4ea8 *** >>> >>> Thanks, it should be fixed with: >>> http://git.savannah.gnu.org/cgit/gettext.git/commit/?h=maint&id=7b2d8d61 >>> It's a shame that it wasn't found before the 0.19.1 release. Will make >>> a new bug-fix release this weekend. >>> >>> Regards, >>> >> >> FWIW I hit this just now when rebuilding coreutils from scratch. >> I rebuilt my gettext rpm with the above patch and all is well again. > > I'm having another issue with gettext-0.19.1 though. > With that installed, coreutils ./bootstrap is causing an older > config.rpath file to be added to the current directory. > I.E. a version with copyright 2010, compared to the > build-aux/config.rpath file which is 2014? > > Reverting back to gettext-0.18.3 avoids the issue.
I tracked this down to autoconf 2.68 being used on this older system. The sequence is: ./bootstrap if gettext --version == 0.18.3 apply autopoint m4 include workaround so compat with 2.68 else rely on autopoint avoiding m4 include errors This autopoint code though relies on autoconf >= 2.69 fi > Unless there is a simple fix for this, > I suggest we revert the gnulib patch inducing the 0.19 dependency > until all these issues are resolved. So the fix would be to have coreutils AC_PREREQ([2.69]) in configure.ac which would be incorporated in the coreutils gettext update patch at the start of this thread. This is feasible since 2.69 was released two years ago, and I've confirmed this works fine. So with a forward looking view, I'm 55:45 in favor of doing that, rather than reverting the gnulib gettext 0.19 update. However it would be great to decouple things on the gettext side. I.E. could func_trace in autopoint perhaps fall back to the older sed scheme if it gets empty output from autom4te? At least a warning about older autoconf should be issued, if we really do need to depend on autoconf 2.69 functionaltity. BTW the fact that func_trace currently sends all errors to /dev/null complicated the debugging here. thanks, Pádraig.