Activating a view from cairo-dock crashes wayfire

[dkondor]

Steps to reproduce:

1. Build and run cairo-dock from this branch: https://github.com/dkondor/cairo-dock-core/tree/wayland_egl
2. Start wayfire and cairo-dock (cairo-dock -o -L)
3. Start any app (gnome-calculator in my example)
4. Minimize it
5. Right click on the icon
6. From the right click menu, click on the magnifying glass icon (see screenshot)

Expected result: app is restored and activated
Actual result: wayfire crashes (see below for stacktrace)

In this case, cairo-dock is using the activate request of wlr-foreign-toplevel-management.

The crash does not happen if left clicking on the application icon (app is restored and activated) and also does not happen with wf-dock -- both of these methods use the same interface. The crash happens on master without my changes to the foreign toplevel protocol.

Output from wayfire:

wayfire: ../types/xdg_shell/wlr_xdg_surface.c:364: void handle_xdg_surface_commit(struct wlr_surface *): Assertion `false' failed.
EE 08-10-20 12:38:37.817 - [src/main.cpp:244] Fatal error: Fatal error(SIGABRT)
EE 08-10-20 12:38:37.865 - #1  signal_handler(int) ../src/main.cpp:245
EE 08-10-20 12:38:38.040 - #2  killpg ??:?
EE 08-10-20 12:38:38.178 - #3  __GI_raise ../sysdeps/unix/sysv/linux/raise.c:51 (discriminator 3)
EE 08-10-20 12:38:38.317 - #4  __GI_abort /build/glibc-2ORdQG/glibc-2.27/stdlib/abort.c:81
EE 08-10-20 12:38:38.454 - #5  __assert_fail_base /build/glibc-2ORdQG/glibc-2.27/assert/assert.c:89
EE 08-10-20 12:38:38.629 - #6  __assert_fail ??:?
EE 08-10-20 12:38:38.641 - #7  handle_xdg_surface_commit ../types/xdg_shell/wlr_xdg_surface.c:366
EE 08-10-20 12:38:38.655 - #8  surface_commit_pending ../types/wlr_surface.c:430
EE 08-10-20 12:38:38.669 - #9  surface_commit ../types/wlr_surface.c:502
EE 08-10-20 12:38:38.677 - #10 ffi_call_unix64 ??:?
EE 08-10-20 12:38:38.684 - #11 ffi_call ??:?
EE 08-10-20 12:38:38.694 - #12 wl_closure_invoke /home/dkondor/program/wayland/src/connection.c:1020
EE 08-10-20 12:38:38.703 - #13 wl_client_connection_data /home/dkondor/program/wayland/src/wayland-server.c:432
EE 08-10-20 12:38:38.711 - #14 wl_event_loop_dispatch /home/dkondor/program/wayland/src/event-loop.c:1027
EE 08-10-20 12:38:38.721 - #15 wl_display_run /home/dkondor/program/wayland/src/wayland-server.c:1349
EE 08-10-20 12:38:38.750 - #16 main ../src/main.cpp:406
EE 08-10-20 12:38:38.885 - #17 __libc_start_main ../csu/libc-start.c:344
EE 08-10-20 12:38:38.924 - #18 ?? ??:0
EE 08-10-20 12:38:38.927 - [EGL] command: eglMakeCurrent, error: EGL_BAD_DISPLAY (0x3008), message: "eglMakeCurrent"
EE 08-10-20 12:38:38.927 - [render/egl.c:431] eglMakeCurrent failed
wayfire: ../render/gles2/renderer.c:35: struct wlr_gles2_renderer *gles2_get_renderer_in_context(struct wlr_renderer *): Assertion `wlr_egl_is_current(renderer->egl)' failed.
EE 08-10-20 12:38:38.927 - [src/main.cpp:244] Fatal error: Fatal error(SIGABRT)

[soreau]

What call is cairo-dock making in this case?

[dkondor]

zwlr_foreign_toplevel_handle_v1_activate()

zwlr_foreign_toplevel_handle_v1_activate()

But the crash does not happen instantly. Actually, it does not happen at all if I set a breakpoint at this line and single step or even if I just continue after the break. It still happens if I run it in a debugger without setting a breakpoint.

[ammen99]

This seems to be a race condition, and hitting this assert would usually mean a bug in wlroots. Have you tried running with ASAN to rule out that Wayfire corrupts the memory?

[dkondor]

Just tried it, ASAN does not report any errors related to this.

I get the following error, but I think it is unrelated:

=================================================================
==24012==ERROR: AddressSanitizer: odr-violation (0x000000d58720):
  [1] size=56 'vtable for wf::touch::gesture_action_t' ../subprojects/wf-touch/src/touch.cpp
  [2] size=56 'vtable for wf::touch::gesture_action_t' ../subprojects/wf-touch/src/touch.cpp
These globals were registered at these points:
  [1]:
    #0 0x5c5810  (/usr/local/bin/wayfire+0x5c5810)
    #1 0x7fc10aada68d  (/usr/local/lib/x86_64-linux-gnu/wayfire/libmove.so+0x7568d)

  [2]:
    #0 0x5c5810  (/usr/local/bin/wayfire+0x5c5810)
    #1 0xa25a9d  (/usr/local/bin/wayfire+0xa25a9d)

==24012==HINT: if you don't care about these errors you may set ASAN_OPTIONS=detect_odr_violation=0
SUMMARY: AddressSanitizer: odr-violation: global 'vtable for wf::touch::gesture_action_t' at ../subprojects/wf-touch/src/touch.cpp
==24012==ABORTING

Telling it to ignore this, i only see memory leaks reported if terminated normally, but not anything if the crash is triggered.

[dkondor]

I looked into this a bit more:

• the process sending the signal that triggers the crash (client in surface_commit()) seems to be cairo-dock
• I can also reproduce this with opening multiple apps and trying to switch between them with the magnifying glass icon
• I cannot reproduce this on sway when switching between apps

[dkondor]

Fixed in wlroots: swaywm/wlroots#2454