dogtail-devel Dogtail - headless mode on a build server? (using mock)

[Stepan Kasal <skasal redhat com>]

Hello,

Len DiMaggio asked:
> team wanting to run Dogtail tests as part of a build verification 
> process. One of the issues in doing this is the need to install an X 
> server on the system that Dogtail will use to run headless.
> 
>   Stepan - I've heard that you have some ideas to get around this - can 
> you elaborate?

Here is what I gathered from my experiments:

The X serever is definitely needed for headless testing of X
applications.  I heard xvnc is used sometimes, but dogtail uses Xvfb,
which is more natural.  That is why the dogtail rpm in Fedora Devel
requires xorg-x11-server-Xvfb.

Another requirement is xorg-x11-xinit, which is missing from
dogtail.spec.  frysk.spec requires it.  (I mentioned it in a mail,
but have not filed a bug until now: fedora bug #203189.)
(Zack, is it OK with you if I add the requirement and rebuild dogtail
in rawhide?)

Then you need an environment/window manager which supports SPI, the
interface to the accessibility description of the application.
One of the option is metacity.  frysk.spec requires that.  I do not
know whether this should be made a requirement.
Perhaps dogtail should require a `spi-capable-wm', which would be
provided by metacity, or by certain Gnome package, or (in the future)
by a KDE package.  (I heard that KDE does not support SPI yet, but
plans to.)

After all of that, the dogtail-run-frysk script failed, because Xvfb
was not able to connect to font server (xfs) through the usual socket
(/tmp/.font-unix/fs7100).  (An X client asked for "fixed10" and the
server was not able to satisfy the requiest.) This socket is not
present in the chrooted environment used by mock, where we build the
frysk rpm.

This is where I get stuck last time I worked on this.
Then I stopped and started thinking about it and asking coleagues.
After that, I formed an opinion that the above direction is generally
right and the next step is to tell Xvfb not to use font server.
(Using server config options passed on the command line of xinit.)

One reason why the above should be successful is that we are speaking
about Gtk2 applications, which do not use server fonts.  (So it is
perfectly possible that the "fixed10" was the only one font
requested.)

So the next step is obvious: get to where I once was, and then craft
the right set of options to tell the Xvfb to use the font files
directly.

I will try this tomorrow, and post the results to this list.

(I apologize for any inaccuracies in the above text.)

Stepan Kasal