From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sven Neumann Date: Wed, 19 Dec 2012 23:03:58 +0100 Subject: [Buildroot] [PATCH 1/4] libglib2: bumped version to 2.32.4 In-Reply-To: <1898585649.1424283.1355934492400.JavaMail.root@advansee.com> References: <1355936324-9585-1-git-send-email-bogdan@nimblex.org> <1898585649.1424283.1355934492400.JavaMail.root@advansee.com> Message-ID: <1355954638.17510.17.camel@bender> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: buildroot@busybox.net Hi, On Wed, 2012-12-19 at 17:28 +0100, Beno?t Th?baudeau wrote: > > - ac_cv_func_posix_getgrgid_r=no glib_cv_long_long_format=ll \ > > - ac_cv_func_printf_unix98=yes ac_cv_func_vsnprintf_c99=yes \ > > Can you explain in detail why you removed these? Did you test that gst-inspect > still works fine after that? No, actually you added this to buildroot way after we've done our changes to update libglib2. So the removal of these two lines is probably just an artefact from rebasing our old patch on top of current buildroot. I don't think we've removed them on purpose. It's been a while since we've done the switch to libglib2. And as I said in the patch that Bogdan submitted, not all the changes are ready for inclusion in buildroot. But let me try to give some background on the changes that we've done. I've collected this from our buildroot git log, so bear with me if I don't remember all the details. + ac_cv_alignof_guint32=4 ac_cv_alignof_guint64=4 ac_cv_alignof_unsigned_long=4 \ These are definitely not right for all platforms. But I found that the alignment checks in the glib configure script don't get the correct result when cross-compiling. So I've added this to force the correct values for our platforms (ARM and Geode). + CPPFLAGS=-D__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 \ This turned out to be necessary on our ARM platform. On ARMv5 atomic operations are not actually available as inlines but libgcc does still provide them as functions. Without tricking glib into using the functions it would use mutex-based fallbacks. + ac_cv_func_newlocale=no ac_cv_func_uselocale=no \ + ac_cv_func_strtod_l=no ac_cv_func_strtoll_l=no ac_cv_func_strtoull_l=no \ The xlocale based implementations of g_ascii_strto{d,ll,ull} and g_ascii_formatd crash with uClibC. At least they did this with the version of uCLibC that we used back then (that was version 0.9.32 as far as I can see). So this change tricks the configure script into believing that this functionality was not available so that the good old home-grown code is used again as it was the case before 2.32. So the submitted patch for libglib2 should not be taken as final. But I hope that this will nevertheless help to get libglib2 updated in buildroot for the next release. Regards, Sven