Bug#958025: gnome-shell: Wayland session never starts

[Sam Morris]

On Fri, 2020-04-17 at 16:19 +0100, Simon McVittie wrote:
> I notice you're using the latest GNOME from testing/unstable, on a
> system where many relevant packages have not been upgraded from 
> stable.

Yes, I usually roll with only the upgrades apt installs itself, and on
the (fairly rare) occasions where problems are caused by version skew
accross a source package I upgrade packages that apt missed by hand.

> On Fri, 17 Apr 2020 at 15:48:51 +0100, Sam Morris wrote:
> >     Apr 17 15:16:09 gnome-session-binary[26371]: WARNING: Error
> > creating FIFO: File exists
> 
> From the gnome-session source code, this is something to do with
> gnome-session running sessions as systemd user services, which is new in
> GNOME 3.36. It wouldn't surprise me if this relies on relatively recent
> features or fixes in systemd or a related component.

Just as soon as I bet a grip on how GNOME sessions are started
everytihng changes again! :)

> > ... it seems suspicious that trying to connect to (presumably) Xwayland
> > on the login screen
> 
> AT-SPI :'-(

Indeed, installing some debug symbols reveals:

   #0  0x00007f6a55f79b4f in __GI___poll (fds=fds at entry=0x7ffc6c2e5ae0, nfds=nfds at entry=1, timeout=timeout at entry=-1) at ../sysdeps/unix/sysv/linux/poll.c:29
   #1  0x00007f6a52c41d97 in poll (__timeout=-1, __nfds=1, __fds=0x7ffc6c2e5ae0) at /usr/include/x86_64-linux-gnu/bits/poll2.h:46
   #2  read_block (len=8, buf=0x556e9049b9b0, fd=44) at ../../src/xcb_in.c:388
   #3  _xcb_in_read_block (c=c at entry=0x556e904b85c0, buf=0x556e9049b9b0, len=len at entry=8) at ../../src/xcb_in.c:1075
   #4  0x00007f6a52c3fb31 in read_setup (c=0x556e904b85c0) at ../../src/xcb_conn.c:177
   #5  xcb_connect_to_fd (fd=fd at entry=44, auth_info=auth_info at entry=0x7ffc6c2e5c20) at ../../src/xcb_conn.c:359
   #6  0x00007f6a52c43ac2 in xcb_connect_to_display_with_auth_info (displayname=displayname at entry=0x556e9048f260 ":1024", auth=auth at entry=0x0, screenp=screenp at entry=0x0)
       at ../../src/xcb_util.c:532
   #7  0x00007f6a52c43c3a in xcb_connect (displayname=displayname at entry=0x556e9048f260 ":1024", screenp=screenp at entry=0x0) at ../../src/xcb_util.c:489
   #8  0x00007f6a55a13db2 in _XConnectXCB (dpy=dpy at entry=0x556e904b7370, display=display at entry=0x556e9048f260 ":1024", screenp=screenp at entry=0x7ffc6c2e5d6c)
       at ../../src/xcb_disp.c:78
   #9  0x00007f6a55a04972 in XOpenDisplay (display=0x556e9048f260 ":1024") at ../../src/OpenDis.c:129
   #10 0x00007f6a54751074 in atspi_get_a11y_bus () from /usr/lib/x86_64-linux-gnu/libatspi.so.0
   #11 0x00007f6a5648a780 in atk_bridge_adaptor_init (argc=<optimized out>, argv=<optimized out>) at ../atk-adaptor/bridge.c:1044
   #12 0x0000556e8f5986c7 in shell_a11y_init () at ../src/main.c:313
   #13 main (argc=<optimized out>, argv=<optimized out>) at ../src/main.c:525

And 'ss' reveals that the socket gnome-shell is trying to connect to is
owned by...

   u_str              LISTEN              0                    1                                                                              /tmp/.X11-unix/X1024 1037591                                              * 0                         users:(("gnome-shell",pid=70011,fd=38))                                        
   u_str              LISTEN              1                    1                                                                             @/tmp/.X11-unix/X1024 1037590                                              * 0                         users:(("gnome-shell",pid=70011,fd=37))                                        

gnome-shell!?

> ii  libatk-bridge2.0-0                           2.30.0-5
> ii  libatk1.0-0                                  2.36.0-2
> 
> One of these relatively-closely-related libraries is 18 months older than
> the other. Perhaps the missing versioned dependency involves ATK/AT-SPI?

No luck so far having updated libatk1.0-0 to match...

> > ii  libegl-mesa0 [libegl-vendor]  18.3.6-2+deb10u1
> > ii  libgl1-mesa-dri               18.3.6-2+deb10u1
> > ii  libglx-mesa0 [libglx-vendor]  20.0.4-2
> 
> This also looks suspicious to me: you're mixing up modules from versions
> of Mesa released 2 years apart.
> 
>     smcv

I've updated these ones too for good measure :)

Anyway... I'll keep upgrading things and follow up if I figure out
which package fixes things...


-- 
Sam Morris <sam at robots.org.uk>