gnome-settings-daemon can't connect to D-Bus and eats CPU on ia64 due to -Wl,-z-defs

[Jonathan Nieder]

# sorry, Joss.  This needs more triage before the binutils maintainers
# will be able to work on it; please work with ia64 porters, and if no
# one is available, then discuss with the release team how to move
# forward.
#
# Putting it another way: unless other packages are affected, removing
# binutils/ia64 from the next stable release is not a possible avenue
# for fixing this.  But removing gnome-settings-daemon/ia64 could be.

reassign 537572 gnome-settings-daemon 2.24.1-1
affects 537572 = binutils
quit

Hi Émeric,

Émeric Maschino wrote:

> I've only archived files having differences in order to keep the tbz
> archives as small as possible.

Ok, so we still have some more steps before this is phrased as a binutils
bug.

Based on the files you sent, adding -Wl,-z,defs to the command line
changes the content of the following files:

	gnome-settings-daemon/.libs/libgsd-profile.a
	plugins/a11y-keyboard/.libs/liba11y-keyboard.{a,so}
	plugins/a11y-settings/.libs/liba11y-settings.{a,so}
	plugins/automount/.libs/libautomount.{a,so}
	plugins/background/.libs/libbackground.{a,so}
	plugins/clipboard/.libs/libclipboard.{a,so}
	plugins/common/.libs/libcommon.a
	plugins/dummy/.libs/libdummy.{a,so}
	plugins/gconf/.libs/libgconf.{a,so}
	plugins/housekeeping/.libs/libhousekeeping.{a,so}
	plugins/keybindings/.libs/libkeybindings.{a,so}
	plugins/keyboard/.libs/libkeyboard.{a,so}
	plugins/media-keys/cut-n-paste/.libs/libgvc.{a,so}
	plugins/media-keys/.libs/libmedia-keys.{a,so}
	plugins/mouse/.libs/libmouse.{a,so}
	plugins/print-notifications/.libs/libprint-notifications.{a,so}
	plugins/smartcard/.libs/libsmartcard.{a,so}
	plugins/sound/.libs/libsound.{a,so}
	plugins/updates/.libs/libupdates.{a,so}
	plugins/wacom/.libs/libwacom.{a,so}
	plugins/xrandr/.libs/libxrandr.{a,so}
	plugins/xsettings/.libs/libxsettings.{a,so}

However, the two versions of libgsd-profile.a, for example, are
actually the same after you extract them with "ar x" (of course).  It
is the timestamps and details like that that seem to be different.
(I don't think ld even entered into the picture.)

So I suspect it would be most fruitful to find out

 (1) what code is actually running when g-s-d gets stuck
 (2) which library that is in
 (3) whether re-linking _just_ that library without "-z defs"
     makes the symptoms go away

and if (3), to provide

 (i)  a complete set of input files and a single "ld" command that _I_
      _can_ _type_ to produce the _single_ library in question.  When
      I include "-z defs" it should be faulty, and when I do not
      include "-z defs" it should not be faulty.

 (ii) a description of how that library can be seen to be faulty ---
      for example, "when we call such-and-such function, it hangs, and
      swapping in the non-faulty version avoids that".

Until the bug has been so reduced, looking at what binutils does
is just not going to help much.

Thanks for your patience, and sorry for the trouble.