Gtk3 on a HTML5 backend | Hacker News Reader

Gtk3 on a HTML5 backend

blogs.gnome.org

255 pointsbiehl36 comments

Does anyone else remember the X11 browser plugin? It was an X server that ran as a plugin in Netscape and IE, with a little bit of magic ("xrx") to handle authentication to the web server. So you could visit a web page and start running, say, FrameMaker in your browser.

There's a guide to getting it set up still alive at http://www.csn.ul.ie/~caolan/TechTexts/Broadway.html

This was in 1996, in the X11R6.3 release ("Broadway"), the same release they came out with LBX, the low-bandwidth extension to X, so that you could run your X11 apps over a modem with reasonable throughput.

It never got much adoption. I don't really know why. My guess, though, is that it didn't address the real reasons people were moving from X11 to browser apps: server load, security, ease of administration, latency-tolerance, and ease of development.

Today, server load and latency tolerance are a lot less important than they were in 1996, because computers are a hundred times faster and hundreds of millions of people have ping times under 20 milliseconds, ten times faster than you'd get on a modem.

But security, ease of administration, and ease of development are a lot more important now than they were in 1996.

So, I suspect this won't get widely adopted, but it could go either way. I hope it does!

Notice the dropdown menu when he types in the address: it says "Broadway 2.0".

I just read the post; I didn't watch the video.

Why would someone implement such an awful protocol as X11 when they could just directly render. They could make application-level X11 clients for legacy X apps running on some server somewhere if they wish, there's no reason to cripple the whole experience with X11.

It's hard to understand your question because the quality of its writing is as poor as its punctuation. Can you clarify? What do you mean by "directly render", "application-level X11 clients", "they", and "server"?

Perhaps you were suggesting that remotely displaying a window of, say, FrameMaker in your browser would be better done by rasterizing the application window on the same machine where FrameMaker was running, as with Xvnc, only as a browser plugin.

That is indeed a workable and fairly simple solution, but it usually uses a lot more bandwidth on the network and memory on the machine running FrameMaker than just running X11, particularly with LBX or ssh compressing the X11 protocol.

The question is pretty straight forward, but where are the punctuation issues? Are expecting more commas?

When asking a question, you should use a question mark. Your last comma should be a colon or semicolon, or even better, you could split the sentence in two.

I find it perfectly readable anyway, but I'm quite sure the comment is technically right.

Using a period instead of a question mark is a writing technique to imply that the question isn't really a question (most commonly used to form polite but firm requests). Touchè on the [semi]colon.

Minimizing latency is still very important in mobile, times of network congestion, and the many places around the world that still have slow connections. HTML also has the benefits that it's indexable and semi-open-source. So I hope this is not widely adopted, but it would be awesome for porting existing apps and maybe replacing Flash in some instances.

I think this raises the question: how much longer will the web still be indexable in the classical HTML sense?

(re-posting my comment from the blog because I want to hear people's opinion)

This is extremely useful for getting apps written for the desktop to run on the browser for remote access and virtualization.

I don’t know if there are any companies with GTK3 desktop apps that need this right now, but for the past six months I’ve been working on doing exactly the same thing for JazzScheme (http://jazzscheme.org/). JazzScheme has its own widget set and UI library that runs on top of Cairo. Instead of sending bitmap data across I’m providing a thin indirection layer on top of Cairo surface and HTML 5 canvas (they’re almost identical in terms of drawing commands).

We’re using it for a pretty involved business app that needs a desktop version. Doing a web version any other way would have been a complete nightmare.

The technical description:

Each toplevel window is mapped to a canvas element, and the content in the windows is updated by streaming commands over a multipart/x-mixed-replace XMLHttpRequest that uses gzip Content-Encoding to compress the data. Window data is pushed as region copies (for scrolling) and image diffs. Images are sent as data: uris of uncompressed png data.

Input is gathered via dom events and sent to the server using websockets.

This doesn't really sound that impressive. It's kind of cool and you could also do the same and port some RDP or VNC client over which would be even cooler. But I don't really feel that this is the way to go. It would be much nicer if the native HTML widgets would be used.

I personally don't give much of a crap about native HTML. That's just a transport, and let's face it, for doing apps it's shit. What I would want it to have is deep linking, back button behaviour and the other nice bits of HTTP.

There is not a 1-to-1 mapping between all Gtk+ widgets and all HTML widgets.

HTML5 allows making any element focusable and supports ARIA that exposes custom elements to accessibility API, so you can make pretty good custom widgets using DHTML.

There doesn't have to be an exact one-to-one mapping. I doubt there are many Gtk+ widgets that cannot be reasonably emulated in the latest browsers.

If this turns into a complete and mature implementation, I won't miss anything when Ubuntu drops X11 next year, quite the opposite!

Now this is getting serious. I was considering porting xul for web apps (with the default and obvious coffeescript dsl, so that no xml verbosity would be needed), but gtk3 is even more outstanding. Any idea when it will be out?

It's not a product, it's a demo he hacked up to see if it was worth pursuing.

How come is the UI so responsive (much more than what I'm used to with remote X sessions)? Is it because both the app and the HTTP server run on the same machine?

The Qt guys have a similar thing. It however needs the Google native client plugin.

http://labs.qt.nokia.com/2010/06/25/qt-for-google-native-cli...

There's also the QWebClient which does not need a plugin, but also does not look very beautiful or native.

http://labs.qt.nokia.com/2009/09/18/qt-in-the-cloud-with-qwe...

The biggest gotcha with these kinds of demos is, and always will be, keyboard input. Unless browser vendors are willing to relinquish all keyboard controls for a special kind of popup window, HTML5 VNC type solutions are dead in the water.

The GTK ports on Mac are pretty bad looking. This would make them awesomer again!

Does it just use the default GTK+ theme, or are there rendering issues?

This actually makes some sense (if I understand it correctly), vs. a VNC client solution in a browser.

You can, in theory, use the higher-level hints given at the toolkit level to optimize the bits sent for each toolkit API invocation.

It looks cool, but (and please forgive my ignorance in this subject)...

Isn't this the sort of thing X was built for? Client rendering, but the underlying app in a remote server? How is rendering this in the browser more useful/cool?

(edit: typo)

Browsers exist in more places than X clients.

Getting X tunneled over HTTP is actually pretty complicated:

* The client/server thing is reversed * X expects everything to be fully asynchronous (WebSockets get a lot of hype but they're pretty useless at this stage IMO) * XRender is a big mismatch to canvas drawing API (the old X drawing protocol is in some ways better, in many ways (esp. color) much worse)

I can think of potential use cases similar to what 280 North are doing with Atlas and Cappuccino.

Check out http://280atlas.com/

Yep, in the sense of cocoa vs gtk. But rather than a port to js, it's a port to canvas.

"Each toplevel window is mapped to a canvas element" Does this means it's not crawl-able, or is it?

I'm not an expert, but I believe this to be correct - the contents of the canvas elements are created/managed by javascript, so they would likely appear empty to most spiders.

Additionally, I don't believe there is any good way to crawl the contents of a canvas element - the best you can get from it is an image representation as a data uri, so no text or html elements with semantic meaning.

Didn't some people do this with VNC a few months back? Either way, this is damn cool.

Related work: the Rich Ajax Platform, an implementation of Eclipse's SWT using the Qooxdoo Javascript framework.

Thats awesome, I'm more than a little jealous I didn't think of that.

I was just lying on the couch thinking about this idea. Well, I have much larger, grander idea that is a mix of Palm/HP's new Mocha platform and some play at ubiquitous interfaces and data powering apps on many devices simultaneously. I was thinking though, that some of the frontends for rtorrent look neat, but still fail to look native.

What if the browser exposed the GTK resources and allowed web applications to render using the same visual components as native applications on the platform. That'd be sweeet.

Reply on news.ycombinator.com