investigate autopkgtest regressions in 2.24-7ubuntu1

[Martin Pitt]

I went through the current http://people.canonical.com/~ubuntu-archive/proposed-migration/update_excuses.html#glibc regressions, and these packages cannot trivially be excluded to be unrelated, so they need a closer look:

http://autopkgtest.ubuntu.com/packages/a/apt/zesty/i386
http://autopkgtest.ubuntu.com/packages/g/gnome-color-manager/zesty/armhf
http://autopkgtest.ubuntu.com/packages/g/gspell/zesty/i386
http://autopkgtest.ubuntu.com/packages/g/gtk+3.0/zesty/i386 (could also have been introduced by new mesa)
http://autopkgtest.ubuntu.com/packages/m/mercurial/zesty/i386
http://autopkgtest.ubuntu.com/packages/w/why/zesty/armhf

These fail on all architectures, I just picked a random one. Note that I killed the majority of the amd64 tests, as we already have enough data from the other architectures, and the queue lag became too painful.

[dobey (dobey)]

The GTK+ thing definitely looks like mesa:

Thread 1 (Thread 0x7ffff7f11ac0 (LWP 13309)):
#0 rawmemchr () at ../sysdeps/x86_64/rawmemchr.S:37
#1 0x00007ffff49f9702 in _IO_str_init_static_internal (
    sf=sf@entry=0x7fffffffdef0, ptr=ptr@entry=0x0, size=size@entry=0,
    pstart=pstart@entry=0x0) at strops.c:41
#2 0x00007ffff49e6b97 in __GI___isoc99_vsscanf (string=0x0,
    format=0x7ffff30592f2 "%d.%d", args=args@entry=0x7fffffffe018)
    at isoc99_vsscanf.c:41
#3 0x00007ffff49e6b37 in __isoc99_sscanf (s=<optimized out>,
    format=<optimized out>) at isoc99_sscanf.c:31
#4 0x00007ffff303bd22 in epoxy_glx_version ()
   from /usr/lib/x86_64-linux-gnu/libepoxy.so.0
#5 0x00007ffff6b2bb49 in ?? () from /usr/lib/x86_64-linux-gnu/libgdk-3.so.0
#6 0x00007ffff6b2befa in ?? () from /usr/lib/x86_64-linux-gnu/libgdk-3.so.0
#7 0x00007ffff6b34ac6 in ?? () from /usr/lib/x86_64-linux-gnu/libgdk-3.so.0
#8 0x00007ffff6b31b36 in ?? () from /usr/lib/x86_64-linux-gnu/libgdk-3.so.0
#9 0x00007ffff6b21a48 in ?? () from /usr/lib/x86_64-linux-gnu/libgdk-3.so.0

This is deep inside the gdk_init() path.

[Dimitri John Ledkov (xnox)]

I do not believe glibc is at fault here.

I've been test building ubiquity. It builds in zesty, and it builds in zesty-proposed, with all packages from zesty proposed but: libgl1-mesa-glx libglapi-mesa.

Let me try with libgl1-mesa-glx libglapi-mesa from zesty-proposed, but glibc from zesty-release, to conclusively exclude glibc. And that fails, downgrading those two mesa packages still resolves everything. Therefore I am certain that current glibc in zesty-proposed is good, and it is the mesa that is now borked / at fault with Xvfb. See bug #1647600 for mesa. https://bugs.launchpad.net/ubuntu/+source/mesa/+bug/1647600

[Martin Pitt (pitti)]

This still leaves apt, mercurial, and why to be investigated.