The WebKit team inside Igalia is looking for some fresh blood. On the technical side we work on all things WebKit, from rendering, networking and accessibility to JavaScript, multimedia and the GNOME web browser. We have a strong commitment to the GTK+ port, but we are by no means restricted to it. On the social side Igalia practices workplace democracy, which in practical terms means that you’ll have, shortly after joining, a voice and vote in how the company is run, from the short term tactical considerations (should we do this project or hire this guy?) to the long term strategic investments (I tell you, this “Web” thing is totally the future).

If this sounds like something you’d like to do drop me a line to any of my multiple mail addresses (I’m sure you’ll manage to find at least one of them) with any background information that you consider relevant.

PS: We are extremely flexible in both location (ask me about my last 9 months travelling around the world) and how you distribute your working hours, so we should be able to accommodate pretty much anyone that is both human and living on planet Earth.

Like last year around these dates, last week some of the WebKitGTK+ hackers gathered in the Igalia offices to spend a few days hacking the good hack, eating tortilla and playing Street Fighter II.

Others have already blogged about the event (Mario, Gustavo, Alex, Diego in multiple occasions … eh…) in some detail, so I’ll just try to give some extra information about some of the things that I did.

JIT + Oprofile

One of the few negative side effects of the not-so-recently acquired JIT superpowers of JavaScriptCore is that JIT-generated code does not play very well with the tools we use everyday to debug and improve our code: gdb will have no idea of the name of the chunks of generated code it passes through (since they have none), and will print something unhelpful like “#9 ??”, profile tools like oprofile and sysprof will have exactly the same problem, so you’ll be unable to know which, if any, of the generated code is the culprit of excessive CPU usage, valgrind does not expect executable code to modify itself at runtime, and will freak out and crash unless instructed to not do so, etc. Fortunately some of these problems have solutions, so I spent the beginning of the hackfest reworking (to please the Reviewer gods) a beautiful patch started some time ago by Holger Freyther and the wizards of the University of Szeged to instruct oprofile about our JIT maneuvers. With this in place the tools is able to know the context of the JIT memory chunks it goes through, and is able to go from some useless complains about anonymous memory ranges to printing something like:

141       0.1910  8581.jo   <jit-func>:fannkuch[tests/sunspider-0.9/access-fannkuch.js:5-62]

Nice!

about:plugins

Another topic where I spent a few days is the long struggle to resurrect “about:plugins” in Epiphany. While Dan was busy kicking libsoup into shape so that we can implement about: URLs in a non-terrible way, I worked on adding the APIs in WebKitGTK+ that the browser will eventually use to fetch the plugin data. A couple of patches (here and here) have already landed, and we can now query WebKit for all the plugins it has loaded in the session, ask information about them, and even disable and enable them at runtime. Because seeing is believing you can see a hacky implementation of the about page that I have implemented locally:

Rest assured, this will look nicer (and will have more data and features) when it finally lands upstream!

Tradition

It was of course great to see everyone again, both colleagues from Igalia and all the other hackers, and I hope we’ll be able to repeat the event again next year. If we manage to do it a third time this would pretty much become an ancient tradition of the GNOME community as far as these things go. Thanks to all the contributors and sponsors (Igalia, Collabora and the GNOME foundation) and until next time.

It’s time for a brief (and late!) recap of some of the most notorious things we did in Epiphany for 2.30, and a short update on what’s already happening in the road to 3.0.

2.30

For 2.30 we focused on fixing all the regressions introduced by the switch to WebKit in 2.28. Overall we did a pretty good job, and there’s only a few things missing to reach the no-regressions goal. Here’s everything that used to work in 2.26, didn’t work in 2.28 and that we fixed for 2.30: middle-click to open clipboard contents, bring back our old context menus, up arrow functionality, do a web search with the URL entry contents on Ctrl+Enter too, EphyEmbedSingle::handle-content signal for external download managers, favicon support, send HTTP ‘Accept-Language’ headers with the user language preferences, custom User-Agent support, javascript: and mailto: URI support, close and movement DOM methods for window, shift+click in text areas, and, yes, form authentication auto-completion.

Wew.

OK, so that’s a lot of stuff, but we were not happy with only bringing back old functionality, we also made some improvement here and there:

GNOME Keyring storage for form auths + Infobar goodness:

In 2.30 not only all your form authentication data is stored in the keyring, but we walked the extra mile to migrate the data in the gecko profile to the new format (not a particularly funny thing to do) and we now show a shiny infobar each time you are about to submit never-seen-before data:

Page cache

After much painful debugging we managed to make the Page Cache support in WebKitGTK+ stable enough for widespread usage. These means that when a page passes some preconditions its whole in-memory representation will be saved for some time, making going back to it with the Back/Forward button blindingly fast. Fast is Good.

Favicons in the Tab Menu:

Back in 2007 I opened a bug suggesting to put favicons next to the page titles in our Tabs menu, arguing that this would make it much much easier to identify a certain page when you have lots of tabs opened; having said that, I promised to attach a patch with the fix “soonish”.

3 years and one week later Olivier Tilloy probably got fed up of waiting and decided to just send a patch himself, making our lives (well, at least mine) much better:

Hey, can you see the bug in that screenshot? An opportunity to contribute!

Disable all plugins at runtime:

Back in the day all you could do in Epiphany was to disable the Java plugin, which nowadays is, to say the least, a feature of questionable usefulness. What we all want to do, surely, is to disable Flash unless we actually need to use it, right? Actually, make it all plugins in general, just in case:

Now

And what about the next release? We haven’t stopped working. In fact at Igalia we are increasing our commitment with the platform, and we are growing the team: Alejandro García is working on rendering performance and in reminding everyone how many years it takes to get anything done in software, always; Sergio Villar is working on libsoup, aiming to finish the disk-cache for 3.0; Mario Sánchez is focused on our accessibility support, fixing bugs left and right (with the help of two of our interns, Diego Escalante and José Millán); Philippe keeps rocking on the media front, and our most recent hiring, Martin Robinson is rewriting and improving our DnD support so fast that it’s hard to keep track of it! Oh, and yeah, myself I’m still working on the GObject DOM bindings, a new post about all the new features I’ve added these past weeks is way overdue.

And what about Epiphany? For the next release (due on Wednesday) you can expect a bunch of bugfixes and some UI improvements. One that I’m enjoying a lot is something I implemented last week; I finally got rid of our statusbar for good, and now by default we’ll show contextual messages in a Chrome-like embedded statusbar (which we already had, but that I have reimplemented and made visible by default):

More vertical space for web content, less code!

My goodness was that cheese good

After finishing some loose ends in a remarkable place I found in Bairro Alto the GObject DOM bindings are in good enough shape to start doing some damage in this world (queue video that probably won’t appear in Planet GNOME, click into my blog to see it):

What you just (hopefully) saw is a GTK+ button that, when clicked, gets all the links from its cousin WebKitWebView and makes them do a small bounce, all with the DOM APIs you know and hate^Wlove exported through GObject.

For instance, to get all links simply call webkit_dom_document_get_links on the WebKitDOMDocument associated with the WebView, which will return a WebKitDOMHTMLCollection. Now we just iterate through the elements, get the WebKitDOMCSSStyleDeclaration associated to each one, and set the CSS properties we want on each one. Simple as that! No “evaluate this string as JavaScript”, and no using the JavaScript bindings through JavaScriptCore. In fact, since we are generating our own bindings instead of using the JavaScript ones, we can get the proper semantics for GObject toolkits instead of having to live with the decisions other people made for other languages and contexts. You can see the code snippet here; I use a two-step animation instead of using the full-blown css transitions because of a small technical issue that I didn’t want to fix before pushing this post, but of course we should be doing it right in the future. In fact, we should create proper APIs for CSS animations in general, not too different from what you have in, say, Clutter.

Most of the code for this is already in Webkit trunk, and now that the initial phase of the work is done I keep improving the code incrementally and pushing the patches to bugzilla, where my compadre Gustavo promptly reviews them. Now if you’ll excuse me I’ll go to expose all events as glib signals on the DOM objects, watch this space (or the upstream bug) for more updates. Happy hacking!

I hadn’t blogged about it yet, but the last weekend I crossed 10 time zones, one ocean and some insane border inspections to attend the first ever WebKit Contributors Meeting. I just returned to my hotel in San Francisco from two hectic days at the Apple campus in Cupertino, putting faces to the names I see daily in my WebKit work and attending some really interesting working sessions and hackatons. Of course a lot of productive discussions and coding happened, but I also value some incorporeal sense of unity and direction that you can get when you put a bunch of people that work together in a project physically in the same place for a couple of days; I’m already looking forward to next year’s meeting, and I thank Apple for organizing the event.

4 of us from Igalia have come to the US (Álex, Philippe, Juanjo and myself), and we’ll stay until Saturday to attend the Linux Collaboration Summit and do some fast-paced sight-seeing around the city (although we already enjoyed a great day off on Sunday with Martin around the city, including the awesome shoe-garden in Alamo Square).

To finish it off, I have to mention that I used the ever productive airplane time and some dead hours these days to advance quite a bit in the GObject DOM Bindings for WebKitGTK+, and that I can already correctly generate large enough portions of it to do actual applications and meaningful unit testing (basically, I cover Document and most of its dependencies!). More about this soon!

Fresh from the ovens of Igalia‘s Industrial Web Hackery division I bring you what all of you were waiting for: support for the DOM methods window.{moveTo, resizeTo, moveBy, resizeBy} in WebKitGTK+, and the corresponding fix in Epiphany. What does this mean? It means that Browser Pong now works in Epiphany!

What? Browser Pong? Yes: Browser Pong.

Go nuts, this is better than World of Warcraft.

Arrived yesterday night to Coruña for the WebKitGTK+ hackfest, a couple of hours before Gustavo did. Today he and I kicked off the day zero of the hackfest, before everybody arrives starting tomorrow.

We spent the whole day hacking on form login/password saving, and despite some issues with GNOME keyring being unhappy and dying on us, I can say we made good progress for one day of work:

This is epiphany/webkit master auto-filling my twitter.com login/password after launch, which as some people know is one of our last nasty regressions. There’s still a few things to do, but I’m confident about landing this before we leave Spain. Also, for those of you not following our development closely, the screenshot also shows the twitter favicon, since Gustavo recently fixed our favicon support in master.

Later today, Álex and Philippe joined us. Álex continued working in a tough accessibility bug in WebKitGTK+ he’s been fighting with, and Philippe arrived just in time for a nice dinner downtown. Not bad for one day, considering we were even not supposed to be here today!

Gustavo has written a nice sumary of all the new APIs in 1.1.14 (yes, all that stuff was only for 1.1.14!), so go and check it out if you are interested.