Can two ATK toolkits cooperate in the same app?

[sbq]

Greetings,

 

I am working on making a Python application accessible by creating ATK objects corresponding to the GUI objects of the application.  The Python app is under development and I recently fetched a new version of the source and it caused my ATK interface to stop working.  I've tracked it down to a single line in the Python app:

 

import gtk

 

As I mentioned in a previous posting, my implementation of add_global_event_listener() in the AtkUtilClass prints the name of the event for each listener added.  If I initialize the ATK system:

 

before "import gtk":

1. The usual list of events is printed

2. My AT-SPI clients sees the ATK object with role ATK_ROLE_APPLICATION that my ATK code defines.

3. The toolkit name (obtained by my AT-SPI client from AccessibleApplication_getToolkitName) is "PAPI" - the ATK tookit I'm using

 

after "import gtk":

1. no list of events is printed

2. the application object is not mine, it has name "atkapplication.py"

3. the toolkit name is "GAIL"

 

Marcus von Appen has documented "Wrapping ATK" here:  http://www.sysfault.org/atk.html.  It looks to me like the first toolkit to perform this wrapping is the one the app uses.  Is this limitation documented somewhere?    When my AT-SPI client starts exploring the accessible object tree rooted at the application object, it will see either the tree created by GTK or the tree create by PAPI, but not both.

 

I forsee a problem when I start trying to interface to Cally, Clutter's accessibility implementation.  Sure enough, this is a problem that Alejendro Iglesias, the author of Cally, has identified, see http://bugzilla.o-hand.com/show_bug.cgi?id=1738).  Cally is a toolkit just like GAIL (see page 12 of this PDF: http://personales.igalia.com/apinheiro/files/cally_guadec_2009_slides.pdf.gz).  If I have a GTK app that is accessible (GAIL, ATK) and I add Clutter along with its accessibility (Cally, ATK) to my app, only one of Cally or GAIL will successfully connect to ATK.  I do not see a way for the two toolkits to cooperate.

 

-Sam

_______________________________________________
gnome-accessibility-list mailing list
gnome-accessibility-list@...
http://mail.gnome.org/mailman/listinfo/gnome-accessibility-list

[Piñeiro]

From: "Quiring, Sam" <Sam.Quiring@...>

> I am working on making a Python application accessible by creating ATK
> objects corresponding to the GUI objects of the application.  The Python
> app is under development and I recently fetched a new version of the
> source and it caused my ATK interface to stop working.  I've tracked it
> down to a single line in the Python app:
>  
> import gtk
>  
> As I mentioned in a previous posting, my implementation of
> add_global_event_listener() in the AtkUtilClass prints the name of the
> event for each listener added.  If I initialize the ATK system:
>  
> before "import gtk":
> 1. The usual list of events is printed
> 2. My AT-SPI clients sees the ATK object with role ATK_ROLE_APPLICATION
> that my ATK code defines.
> 3. The toolkit name (obtained by my AT-SPI client from
> AccessibleApplication_getToolkitName) is "PAPI" - the ATK tookit I'm
> using
>  
> after "import gtk":
> 1. no list of events is printed
> 2. the application object is not mine, it has name "atkapplication.py"
> 3. the toolkit name is "GAIL"

This means that you loaded GAIL. I'm don't know anything about this
new PAPI tookit, but if you want to use it as the root toolkit, you
need to be sure to be loaded after GAIL. So you'll need something
like:

GTK_MODULES=gail:papi:atk-bridge

or just define GTK_MODULES as empty

GTK_MODULES=
and then load the modules by hand.

BTW: if you are just adding the a11y support for a specific
application, with custom widgets, you'll probably don't need to create
a full new module for that. If this application is a GTK application,
you can just define the a11y object, and then redefine the method
gtk_widget_get_accessible in this widget.

> Marcus von Appen has documented "Wrapping ATK" here:
> http://www.sysfault.org/atk.html.  It looks to me like the first toolkit
> to perform this wrapping is the one the app uses.  Is this limitation
> documented somewhere?    When my AT-SPI client starts exploring the
> accessible object tree rooted at the application object, it will see
> either the tree created by GTK or the tree create by PAPI, but not both.

AFAIK, not the first is the last, as each time a module defining this
methods is loaded, it tries to use their own implementation of this
AtkUtil methods. At least, this should work.

> I forsee a problem when I start trying to interface to Cally, Clutter's
> accessibility implementation.  Sure enough, this is a problem that
> Alejendro Iglesias, the author of Cally, has identified, see
> http://bugzilla.o-hand.com/show_bug.cgi?id=1738).  Cally is a toolkit
> just like GAIL (see page 12 of this PDF:
> http://personales.igalia.com/apinheiro/files/cally_guadec_2009_slides.pd
> f.gz).  If I have a GTK app that is accessible (GAIL, ATK) and I add
> Clutter along with its accessibility (Cally, ATK) to my app, only one of
> Cally or GAIL will successfully connect to ATK.  I do not see a way for
> the two toolkits to cooperate.

One of the top things on my TODO list is just try to check the
cooperation between GAIL and Cally. In the past I implemented the a11y
support for the GtkClutterEmbed widget. This allows me to have a11y
support for a mixed Gtk+Clutter application.

In this case I load Cally, and *then* I loaded GAIL, so was GAIL the
one redefining (wrapping) AtkUtil methods. We can say that in this
case, GAIL is the "chief".

But as I said, and pointed in the presentation that you showed, this
is just a specific environment, and we need to check how to resolve
the other issues.

I still need to push this GtkClutterEmbed a11y support somewhere. But
I prefer to take a look to the whole gtk-clutter stack, now that it is
updatd, to see how to do that.

I tried to explain my concerns about all this gtk vs cally, and the
problems with only a root toolkit in a past mail [1]

[1] http://mail.gnome.org/archives/gnome-accessibility-list/2009-August/msg00028.html

Best regards

===
API (apinheiro@...)
_______________________________________________
gnome-accessibility-list mailing list
gnome-accessibility-list@...
http://mail.gnome.org/mailman/listinfo/gnome-accessibility-list

[sbq]

> GTK_MODULES=gail:papi:atk-bridge

This is a new one on me.  Where is the effect of this
environment variable documented?  What form would papi
have to be (.so file?) and where would it have to be
located in the file system for this to work?

I just did a couple of google's for this and finally found
http://library.gnome.org/devel/gtk/unstable/gtk-running.html
which says GTK_MODULES defines a list of modules to load.
It does not say which order they are initialized.  Is there
more definitive documentation?

= > Marcus von Appen has documented "Wrapping ATK" here:
= > http://www.sysfault.org/atk.html.  It looks to me like the first
= > toolkit to perform this wrapping is the one the app uses.  Is this
= > limitation documented somewhere?    When my AT-SPI client starts
= > exploring the accessible object tree rooted at the application object, it
= > will see either the tree created by GTK or the tree create by PAPI, but
= > not both.
=
= > AFAIK, not the first is the last, as each time a module defining this
= > methods is loaded, it tries to use their own implementation of this
= > AtkUtil methods. At least, this should work.

We may be talking about two different strategies.  I think you
are referring to the GTK_MODULES= order of loading (and initializing?).
I am describing the effects of PAPI executing the code describe in
"Wrapping ATK".  If PAPI wraps ATK after "import GTK" is executed,
GAIL will be the toolkit, not PAPI.

I've tracked through some of the code in the ATK Bridge, and I am pretty
sure (but could be wrong) the first toolkit to execute
gnome_accessibility_module_init() is the only toolkit that will work in
the app.  This is consistent with the evidence I reported in my email.

-Sam


-----Original Message-----
From: Piñeiro [mailto:apinheiro@...]
Sent: Monday, October 05, 2009 3:30 AM
To: Quiring, Sam
Cc: gnome-accessibility-list@...
Subject: Re: Can two ATK toolkits cooperate in the same app?

From: "Quiring, Sam" <Sam.Quiring@...>

> I am working on making a Python application accessible by creating ATK
> objects corresponding to the GUI objects of the application.  The Python
> app is under development and I recently fetched a new version of the
> source and it caused my ATK interface to stop working.  I've tracked it
> down to a single line in the Python app:
>  
> import gtk
>  
> As I mentioned in a previous posting, my implementation of
> add_global_event_listener() in the AtkUtilClass prints the name of the
> event for each listener added.  If I initialize the ATK system:
>  
> before "import gtk":
> 1. The usual list of events is printed
> 2. My AT-SPI clients sees the ATK object with role ATK_ROLE_APPLICATION
> that my ATK code defines.
> 3. The toolkit name (obtained by my AT-SPI client from
> AccessibleApplication_getToolkitName) is "PAPI" - the ATK tookit I'm
> using
>  
> after "import gtk":
> 1. no list of events is printed
> 2. the application object is not mine, it has name "atkapplication.py"
> 3. the toolkit name is "GAIL"

This means that you loaded GAIL. I'm don't know anything about this
new PAPI tookit, but if you want to use it as the root toolkit, you
need to be sure to be loaded after GAIL. So you'll need something
like:

GTK_MODULES=gail:papi:atk-bridge

or just define GTK_MODULES as empty

GTK_MODULES=
and then load the modules by hand.

BTW: if you are just adding the a11y support for a specific
application, with custom widgets, you'll probably don't need to create
a full new module for that. If this application is a GTK application,
you can just define the a11y object, and then redefine the method
gtk_widget_get_accessible in this widget.

> Marcus von Appen has documented "Wrapping ATK" here:
> http://www.sysfault.org/atk.html.  It looks to me like the first toolkit
> to perform this wrapping is the one the app uses.  Is this limitation
> documented somewhere?    When my AT-SPI client starts exploring the
> accessible object tree rooted at the application object, it will see
> either the tree created by GTK or the tree create by PAPI, but not both.

AFAIK, not the first is the last, as each time a module defining this
methods is loaded, it tries to use their own implementation of this
AtkUtil methods. At least, this should work.

> I forsee a problem when I start trying to interface to Cally, Clutter's
> accessibility implementation.  Sure enough, this is a problem that
> Alejendro Iglesias, the author of Cally, has identified, see
> http://bugzilla.o-hand.com/show_bug.cgi?id=1738).  Cally is a toolkit
> just like GAIL (see page 12 of this PDF:
> http://personales.igalia.com/apinheiro/files/cally_guadec_2009_slides.pd
> f.gz).  If I have a GTK app that is accessible (GAIL, ATK) and I add
> Clutter along with its accessibility (Cally, ATK) to my app, only one of
> Cally or GAIL will successfully connect to ATK.  I do not see a way for
> the two toolkits to cooperate.

One of the top things on my TODO list is just try to check the
cooperation between GAIL and Cally. In the past I implemented the a11y
support for the GtkClutterEmbed widget. This allows me to have a11y
support for a mixed Gtk+Clutter application.

In this case I load Cally, and *then* I loaded GAIL, so was GAIL the
one redefining (wrapping) AtkUtil methods. We can say that in this
case, GAIL is the "chief".

But as I said, and pointed in the presentation that you showed, this
is just a specific environment, and we need to check how to resolve
the other issues.

I still need to push this GtkClutterEmbed a11y support somewhere. But
I prefer to take a look to the whole gtk-clutter stack, now that it is
updatd, to see how to do that.

I tried to explain my concerns about all this gtk vs cally, and the
problems with only a root toolkit in a past mail [1]

[1] http://mail.gnome.org/archives/gnome-accessibility-list/2009-August/msg00028.html

Best regards

===
API (apinheiro@...)
_______________________________________________
gnome-accessibility-list mailing list
gnome-accessibility-list@...
http://mail.gnome.org/mailman/listinfo/gnome-accessibility-list

[Piñeiro]

From: "Quiring, Sam" <Sam.Quiring@...>

> > GTK_MODULES=gail:papi:atk-bridge
>
> This is a new one on me.  Where is the effect of this
> environment variable documented?  What form would papi
> have to be (.so file?) and where would it have to be
> located in the file system for this to work?

You can found this on the accessibility section in library.gnome [1].

The summary is that GAIL, HAIL, Cally and any other modules that add
a11y support for any UI toolkit (Gtk+, Hildon, Clutter) are supposed
to be loaded optionally, due the overhead. In GNOME they tried to make
it as general as possible, so this was the option to solve that.

GTK_MODULES [2] is just a variable where you put the dynamic modules
that you want to load using GModule, when you call gtk_init. So if you
are using a "pure-gtk" application, use that its enough. And this is
the reason atk-bridge is placed on the gtk directory for modules,
although technically, is not a gtk module [3]. In a mixed environment
it would be more complex [4].

In fact, right now use GTK_MODULES to load this modules in a general
way, has some problems related (see bugs [5][6][7] as some examples).

GTK_MODULES has a set of default directories to search for.

> I just did a couple of google's for this and finally found
> http://library.gnome.org/devel/gtk/unstable/gtk-running.html
> which says GTK_MODULES defines a list of modules to load.
> It does not say which order they are initialized.  Is there
> more definitive documentation?

Ups, just my [2] link ;). The order in which is loaded is left-right,
so with:

  GTK_MODULES=gail:atk-bridge

You are asking to load GAIL, and then load atk-bridge. And this is the
correct form. If you define it on the reverse, you'll get a crash, as
atk-bridge could call atk_util_get_root, that is not defined until
gail is loaded.

> = > Marcus von Appen has documented "Wrapping ATK" here:
> = > http://www.sysfault.org/atk.html.  It looks to me like the first
> = > toolkit to perform this wrapping is the one the app uses.  Is this
> = > limitation documented somewhere?    When my AT-SPI client starts
> = > exploring the accessible object tree rooted at the application object, it
> = > will see either the tree created by GTK or the tree create by PAPI, but
> = > not both.
> =
> = > AFAIK, not the first is the last, as each time a module defining this
> = > methods is loaded, it tries to use their own implementation of this
> = > AtkUtil methods. At least, this should work.
>
> We may be talking about two different strategies.  I think you
> are referring to the GTK_MODULES= order of loading (and initializing?).
> I am describing the effects of PAPI executing the code describe in
> "Wrapping ATK".  If PAPI wraps ATK after "import GTK" is executed,
> GAIL will be the toolkit, not PAPI.

Hmm, yes I think that you are right. This html shows you how to wrap
the current toolkit. As far as I understand, this means that you can
keep with the normal GTK_MODULES:

GTK_MODULES=gail:atk-bridge.

I suppose that you import gtk will call gtk_init (sorry I'm not really
used to python apps), so gail will be the one in use.

Then, you need to execute the methods explained in your link, and if
it works, you are providing your own AtkUtil method implementation, so
the toolkit should be PAPI.

> I've tracked through some of the code in the ATK Bridge, and I am pretty
> sure (but could be wrong) the first toolkit to execute
> gnome_accessibility_module_init() is the only toolkit that will work in
> the app.  This is consistent with the evidence I reported in my email.

The first toolkit will redefine the AtkUtil methods. This is a
standalone class. When you load the GailUtil class, it define the
AtkUtil methods, if you call gnome_accessibility_module_init with
other toolkit, this one would redefine this methods, as cally does. In
my attempts with mixed gail-cally it works at least. Anyway is
something that I need to check with detail.

Take a look to the implementation of gail_util_class_init or
cally_util_class_init for more information.

[1] http://library.gnome.org/devel/accessibility-devel-guide/nightly/gad-how-it-works.html.en
[2] http://library.gnome.org/devel/gtk/unstable/gtk-running.html
[3] http://mail.gnome.org/archives/gnome-accessibility-list/2009-August/msg00019.html
[4] http://bugzilla.o-hand.com/show_bug.cgi?id=1738#c3
[5] https://bugzilla.gnome.org/show_bug.cgi?id=95280
[6] https://bugzilla.gnome.org/show_bug.cgi?id=564751
[7] https://bugzilla.gnome.org/show_bug.cgi?id=563876

===
API (apinheiro@...)
_______________________________________________
gnome-accessibility-list mailing list
gnome-accessibility-list@...
http://mail.gnome.org/mailman/listinfo/gnome-accessibility-list