dbus at-spi splatter

[totaam]

We already try our best to silence this stuff:

xpra/xpra/platform/posix/features.py

Lines 14 to 16 in c6a040f

	"#silence some AT-SPI and atk-bridge warnings:",
	"NO_AT_BRIDGE=1",
	)

But unfortunately, it seems to be happening somewhere else now:

** (Xpra:636376): WARNING **: 11:50:42.677: AT-SPI: Could not obtain desktop path or name


** (Xpra:636376): WARNING **: 11:50:42.679: atk-bridge: GetRegisteredEvents returned message with unknown signature

** (Xpra:636376): WARNING **: 11:50:42.679: atk-bridge: get_device_events_reply: unknown signature

This happens with:

xpra start --minimal --dbus=yes --dbus-launch="dbus-launch --sh-syntax --close-stderr"

But not with:

xpra start --minimal --dbus=yes --dbus-launch="no"

So something dbus-launch does triggers it.
It seems to create a bunch of /usr/bin/dbus-broker-launch --config-file=/usr/share/defaults/at-spi2/accessibility.conf --scope user processes we never asked for.
And looking at this config file is not pleasant - basically anyone is allowed to do anything!?:

<policy context="default">
    <!-- Allow root to connect -->
    <allow user="root"/>
    <!-- Allow everything to be sent -->
    <allow send_destination="*" eavesdrop="true"/>
    <!-- Allow everything to be received -->
    <allow eavesdrop="true"/>
    <!-- Allow anyone to own anything -->
    <allow own="*"/>
  </policy>

Is this what does it?

$ cat /lib/systemd/user/at-spi-dbus-bus.service
[Unit]
Description=Accessibility services bus
PartOf=graphical-session.target

[Service]
Type=dbus
BusName=org.a11y.Bus
ExecStart=/usr/libexec/at-spi-bus-launcher
Slice=session.slice
TimeoutStopSec=5

Who knows.

RedHat bug AT-SPI: Could not obtain desktop path or name
Gnome bug atk-bridge: get_device_events_reply: unknown signature

[totaam]

Does not occur with xpra 5.x! So this is a regression.
git bisect landed on the wrong commits, twice! Perhaps I was not cleaning the source tree well enough before rebuilding. So I did by hand..

e3b595a bad
72b4f11 good
ef665e2 good
4831c63 good
c7b12a6 bad
a5ac22a good

So c7b12a6 is to blame!? How!?

[totaam]

What a very strange bug!
It took a lot of effort to narrow it down, moving files around in both directions, and rebuilding fully every time to prevent cythonized files from squeezing in the way, I found that this one line introduces the bug:

toggle_packages(gtk_x11_ENABLED, "xpra.x11.gtk")

Importing Gtk and Gdk from the module init file causes the bug!?
What on earth does this have to do with dbus and ATK!?