We follow the xdg spec and implement a fallback for XDG_RUNTIME_DIR (in ~/.cache) if it is unset.

Unfortunately, wayland logs an error in this case.  Our wayland log handler then does this:

static void
log_handler(const char *format, va_list args)
{
  g_logv (G_LOG_DOMAIN, G_LOG_LEVEL_ERROR, format, args);
}


which is fatal.


This happens any time someone tries to initialise wayland with XDG_RUNTIME_DIR unset.  That can happen if DISPLAY is unset, for example (which will cause the X11 init to fail, moving on to the next backend, which is wayland).

So we've now gone from a place where unset DISPLAY and XDG_RUNTIME_DIR used to cause gtk_init() to fail to a place where it causes the entire program to abort, simply because we happen to be trying wayland as a second effort.

We can fix this in some ways:

 1) only try the wayland backend if GDK_BACKEND=wayland explicitly

 2) fix libwayland not to throw an error in this case (and maybe fallback
    to ~/.cache/ in the way the spec suggests)

 3) avoid bringing up the wayland backend if XDG_RUNTIME_DIR is unset to
    avoid the certain abort that will follow

 4) log wayland messages in a less-fatal way

I'm in favour of 2.  If that's not on the table, I'd consider 3 to be the best option.