[GTK3] JVM crash when copying data (when there are no seats)

[jonahgraham]

While working on #2768 to get tests wayland tests running on I consistently see a crash like this:

#  SIGSEGV (0xb) at pc=0x0000774d6acc7741, pid=2214417, tid=2214423
#
# JRE version: OpenJDK Runtime Environment Temurin-21.0.5+11 (21.0.5+11) (build 21.0.5+11-LTS)
# Java VM: OpenJDK 64-Bit Server VM Temurin-21.0.5+11 (21.0.5+11-LTS, mixed mode, sharing, tiered, compressed oops, compressed class ptrs, g1 gc, linux-amd64)
# Problematic frame:
# C  [libgdk-3.so.0+0x83741]  gdk_wayland_selection_add_targets_libgtk_only+0x1f1

Stack: [0x0000774dfc300000,0x0000774dfc400000],  sp=0x0000774dfc3f9f30,  free space=999k
Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
C  [libgdk-3.so.0+0x83741]  gdk_wayland_selection_add_targets_libgtk_only+0x1f1
C  [libgtk-3.so.0+0x2ab8c8]  gtk_selection_add_targets+0x118
C  [libgtk-3.so.0+0x3b919b]
C  [libgtk-3.so.0+0x3b3554]  gtk_clipboard_set_with_owner+0x74
C  [libswt-pi3-gtk-4972r1.so+0x569de]  Java_org_eclipse_swt_internal_gtk3_GTK3_gtk_1clipboard_1set_1with_1owner+0x1f
j  org.eclipse.swt.dnd.ClipboardProxy.setData(Lorg/eclipse/swt/dnd/Clipboard;[Ljava/lang/Object;[Lorg/eclipse/swt/dnd/Transfer;I)Z+263
j  org.eclipse.swt.dnd.Clipboard.setContents([Ljava/lang/Object;[Lorg/eclipse/swt/dnd/Transfer;I)V+127
j  org.eclipse.swt.tests.junit.Test_org_eclipse_swt_dnd_Clipboard.test_LocalClipboard(I)V+33

I think I have identified why - no seat when running weston headless.

[jonahgraham]

One of the things that made debugging this confusing is that the stack trace Java reported is not correct - the real stack trace looks like:

bt right before the SIGSEGV

and right after:

So the gdb + java unwinder are providing slightly different results.

[jonahgraham]

The issue is running without any "seat". Since we are trying to run fully headless, there is no input devices (and hence no seats). I filed a GTK bug report here #7915.

[jonahgraham]

When I reproduce this locally I can run with a seat, provided by x11!

So instead of running:

wlheadless-run -c weston --width 1920 --height 1080 -- mvn ...

This may work, by wrapping in xvfb-run:

xvfb-run wlheadless-run -c weston --backend=x11  --width 1920 --height 1080 -- mvn ... 

There may be more options in https://gitlab.freedesktop.org/ofourdan/xwayland-run to consider that seem a little more sane, such as using xwfb-run intead of xvfb-run wlheadless-run.

---

In reality what I am doing, so I can connect Eclipse's launches to that weston instance, is this:

$ cat hold.sh 
#!/bin/bash
env | sort | grep DISPLAY
sleep 1d

$ xvfb-run --auth-file=/tmp/my_xvfb_auth_file.Xauthority  --server-num=55 wlheadless-run -c weston --backend=x11  --width 1920 --height 1080 -- ./hold.sh 
--snip--
DISPLAY=:55
WAYLAND_DISPLAY=wayland-2291814

And using those settings in my launch config like this:

[jonahgraham]

I am closing this as a "not SWT" bug.

[jonahgraham]

The issue is running without any "seat". Since we are trying to run fully headless, there is no input devices (and hence no seats). I filed a GTK bug report here #7915.

While this GTK3 bug report is closed with no plan to fix, the attention of this bug report and another one has triggered some new improvements to weston to allow a seat when testing headlessly.

See https://gitlab.freedesktop.org/ofourdan/xwayland-run/-/issues/12 and associated PRs in weston and xwayland run