Last week some of the members of the Igalia WebKit team, attended to the second edition of the WebKit Contributors Meeting in Cupertino, California, in order to gather round with other WebKit contributors to discuss and work around our favorite web engine, away from IRC and with a more personal and “human” touch.
As for me, it was the first time I attended this unconference and I have to say it was a great experience overall, even if accessibility (the field I usually work on through the GTK port of WebKit) was not precisely a hot topic there. But in the other hand, I managed to put some faces to people I just knew from IRC, attended to several interesting discussions and sessions, did some actual work ™ and met Chris Fleizach, the guy from Apple involved in accessibility for the Mac port, who attended the meeting on Tuesday morning and discussed with me some interesting topics, mainly about the implementation of the accessibility support in the Mac port of WebKit2.
For those of you that are not up to date with what WebKit2 means and what’s the current state of the whole thing, I’d recommend you to take a look to the WebKit2 wiki, although I can already advance you that one of the main features (if not the main one) of WebKit2 is the new multi-process architecture, which will go bundled right along with the engine, instead of doing the split in the final application, like Google Chrome does now with WebKit1 (see the wiki for more details, and some nice diagrams too, like the following one).
The problem however, from the accessibility point of view, is how to expose such a new architecture to Assistive Technologies (ATs) while, at the same time, having those ATs seeing just one process (the UI process), regardless of the different processes that would be running on each of the tabs (the Web proceses). In other words, how to “hide” to the ATs that they are extracting accessibility related information from a multi-process based browser, so they still keep the illusion of having only one application exposing accessibility objects, as it used to be so far.
Talking to Chris about how they implemented this in the Mac port, he told me they basically needed to make up a way to transparently connect the UI process and the Web Process, so navigation between the root accessibility object in the Web Process and the “leaf” accessibility object in the UI process were done in a seamless way, through a mechanism that would basically allow bidirectional communication in the same way it used to be when there was just one process running. This makes a lot of sense, if you ask me, since you “just” (saying it as if it were not a complex task, even though it actually is) add this mechanism to WebKit2 accessibility code and you’ll automagically get your ATs working as they used to work, without any other changes needed from their side.
However, when it comes to ATK/AT-SPI, which is what we currently use in GNOME to expose accessible objects from applications to ATs, it looks like such a mechanism (or something pretty similar) is already available by means of the AtkSocket and the AtkPlug classes, both subclasses of AtkObject, which basically act as a bridge that allows to connect two AtkObject‘s so the children of one of those (the Plug) are exposed as children of the other one (the Socket), no matter they’re in the same process or in different ones. And this, unless I’m missing something, is exactly what we’d need in WebKit2, probably along with some other things and tweaks that I just can’t think of at this moment. But at least is definitely a very good start point, IMHO.
But… how to communicate those sockets and plugs if they are in different processes? You might be wondering… Well, if you are using the D-Bus implementation of AT-SPI, also known as AT-SPI2, the needed bits for that (the implementation for the atk_socket_embed() and atk_plug_get_id() functions) are already implemented by the ATK bridge (at-spi2-atk), so the only thing you’ll need to do is to provide the remaining implementation of the AtkSocket and AtkPlug classes, register those new AtkObject‘s as the right accessible objects for your widgets (or whatever your “normal” objects are) and to use the ID that at-spi2-atk provides for the AtkPlug object to connect from the AtkSocket object… and you’re done.
However, I have to say that the documentation I’ve found so far about AtkSocket and AtkPlug is not precisely very detailed, so I basically ended up looking directly at the code and trying to write myself a small example to better understand things. And at this point, I asked in the #a11y channel in GIMPNet and it was Mike Gorse who kindly handed me out a nice tarball with an example they had written in C#, which was exactly what I was looking for. Yay! Thanks, Mike!
However, I’d be more interested in a plain C, GObject based, implementation of that example which would serve me both to better understand how it works and to use it as the base for further tests in WebKitGTK, so I went ahead and wrote it, and this is how the results look now, as seen through Accerciser‘s eyes:
As you can see, there are two processes running, ta-socket and ta-plug, and in ta-socket we have an AtkObject named “The Socket” which is exposing, as his only child, the whole subtree present in the other process ta-plug, starting in another AtkObject named “The Plug”… which is exactly, although of course at a much smaller dimension, what I think we would like to see happening in the accessibility tree of any WebKitGTK based application in the future, once that WebKit2 is mature enough to replace WebKit1.
Obviously, this shouldn’t be taken as a “we’re done with WebKitGTK/WebKit2 when it comes to a11y” comment or the like. Actually this is just the beginning of the whole thing… But I think, or at least I would like to think, that is at least a good start point :-).
And before you shoot… “yes”, I’m already planning to help improve the ATK documentation in this regard with the knowledge I acquired while working on this examples, even perhaps next week during the ATK/AT-SPI hackfest (it’s actually one of the proposed tasks) I’ll be attending here in Coruña, at the Igalia offices.
By the way, feel free to grab the source for this example from its git respository at gitorious:
git clone http://git.gitorious.org/atksocket-atkplug-example/mainline.git