New tests: GTK3 platform accessibility

[Michael Weghorn]

On 2023-09-12 10:49, Colomban Wendling wrote:
> All that is a bit confusing to me, but let me try ton understand:
> - On a system using dbus-broker, when logged in under GNOME, test fails?
> - On *the same system*, but logging-in through SSH works!??

Yes, that's what I've observed in a Fedora 38 VM.

> That's worrying, as it suggests the running environment has too much 
> influence on the test run, which is not good.
> 
> However, IIUC you suggest it might be a dbus/dbus-borken conflict 
> somehow?

That's what it looked like to me during first analysis.

What might be happening is that at-spi-bus-launcher detects a systemd 
environment that makes it think running dbus-broker should work, but 
something about the systemd/D-Bus environment in the virtual X session 
provided by xfvb-run and dbus-launch doesn't match what's expected/required.

See also this comment in the at-spi-bus-launcher source code [1]:

   /* This detects whether we are running under systemd. We only try to
    * use dbus-broker if we are running under systemd because D-Bus
    * service activation won't work otherwise.
    */

The following sample program, based on the at-spi-bus-launcher-code, 
could explain why the SSH case works: The "systemd, use dbus-broker" 
case is triggered for me on Debian testing in a KDE Plasma X11 session 
(and probably the same way for a Fedora 38 GNOME session), but not in an 
SSH session ("no systemd, don't use dbus-broker").


---------------------------------------------------

// build: g++ -o test test.cxx -lsystemd

#include <iostream>
#include <unistd.h>
#include <systemd/sd-login.h>

int main()
{

     // s. 
https://gitlab.gnome.org/GNOME/at-spi2-core/-/blob/e7310f8a57cf9d9744f68059fad451cb3acd116a/bus/at-spi-bus-launcher.c#L453-460
     char* unit;
     if (sd_pid_get_user_unit (getpid (), &unit) >= 0)
         std::cout << "systemd, use dbus-broker" << std::endl;
     else
         std::cout << "no systemd, don't use dbus-broker" << std::endl;
}

---------------------------------------------------

> What if we try `dbus-broker-launch` (assuming it's compatible) 
> instead of `dbus-launch` if the former is available? 
> Would that help?


As Stephan mentioned already, it's not compatible as a drop-in-replacement.

Maybe this more complicated way mentioned in the Github discussion I 
mentioned earlier could work:

Question [2]:

"What is the recommended way to run a nested dbus session, if not by 
using dbus-run-session? I mostly use it to run debug instances of GNOME 
Shell, as well as the for the CI pipeline when running inside Docker."
Answer (next comment):

"We don't have a recommended way to do that. If that functionality is
needed, it should be straightforward to achieve by just invoking your
private instance of `dbus-broker-launch`. Socket-activation for the
listener-socket is mandatory, though. So I would imagine creating
`dbus-broker at .socket` and `dbus-broker at .service` as systemd template
units, which take a path as template-argument where to place the socket.
It should then be as simple as `systemctl start
dbus-broker at foobar.service` to get it up and running.

If, however, you intend to integrate it into a test-suite, I would
rather expect the test-suite to create the listener manually and pass it
through `execve(2)` to `dbus-broker-launch`."

> I don't have any system to test this with yet, and it might take some 
> time before I can have a F38 VM up and running with a LO build setup, 
> but I'll try and see to this soon if it's a concern.

I'm on vacation this week, but could give this another try in my Fedora 
VM sometime afterwards. Or, if you're also attending LibreOffice 
conference next week in person, we could also take a look at it together 
then for example.


[1] 
https://gitlab.gnome.org/GNOME/at-spi2-core/-/blob/e7310f8a57cf9d9744f68059fad451cb3acd116a/bus/at-spi-bus-launcher.c#L453-460
[2] https://github.com/bus1/dbus-broker/issues/145#issuecomment-437874214
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/libreoffice/attachments/20230913/fe1cd40d/attachment.sig>