HTML 5 canvas graphics library | Hacker News Reader

HTML 5 canvas graphics library

exsprite.com

55 pointsdjd16 comments

I'm always glad to see more canvas libraries, but the performance of this one is pretty horrendous. I'm writing a comprehensive canvas diagramming library that is now up to 70K lines and its no where near this slow.

In case the author reads these comments I'll mention a few I see at first glance:

They are doing things like setting the font every on single draw unnecessarily (which takes a surprising amount of time, performance-wise). Actually they are resetting nearly all of the canvas state by hand on each draw, which can just as easily be done in the much faster way by doing "canvas.width = canvas.width". But even that they shouldn't do if they can avoid it.

Similarly the author should not set the globalCompositeOperation all the time unless he or she really needs to.

I'm sure there are a lot of other low-hanging fruit that can be fixed. I'll try send the author an email later tonight. I hope they keep at it.

I agree, I watched the demos and thought it had horrible performance.

Did you say your library had 70K lines!?!!?

Yes, but there's an awful lot of stuff that needs to run once at the start (premade geometric figures and the like). Once compressed of course it is much smaller.

I'm also kind of sick of seeing stuff done with canvases that could probably be done as efficiently or more efficiently without. (It would be particularly cool for someone to implement a library that abstracted out the back-end, so you could run in DHTML4 or excanvas or a real canvas.)

limejs does that, but it uses google closure

I've seriously thought about surgically removing Closure from limejs. It uses a lot of Closure classes (i.e. Vector2) so it wouldn't be clean, but improving the build process would be worth it.

How does the performance of this compare to other libraries like PaperJS (http://paperjs.org) or EaselJS (http://easeljs.com) or ProcessingJS (http://processingjs.org/ though I understand Processing doesn't really have a display list)?

From my experience EaselJS is pretty fast. You gotta be aware of the demos on Firefox, because it gets pretty slow due to multipe error messages unless you set the following flag:

    DisplayObject.suppressCrossDomainErrors = true

I made something similar a ways back (a little rough, but some fun features like defining sprites using ASCII characters):

https://github.com/mcantelon/grout

Here's a game I made with it:

http://mikecantelon.com/demo/demo_blood_funnel.html

Interesting. But I wonder why I'd use this instead of EaselJS: http://easeljs.com/.

well, Grant's a joker that ripped off plenty of my ideas and code; but it's not horrible if you have to use somethin.

This is great, it's similar to Flash's display list. I actually think a display list should be part of the canvas spec. There's really no reason to have a bunch of competing display list libs and it could be optimized if it was part of the browser.

Cool...other than the parts that need click or keyboard input, all the demos seem to work, more or less, on my iPhone 4S (Canvas Creatures is pretty sluggish, though) and Kindle Fire (both Canvas Creatures and Snake are sluggish there, and Canvas Creatures looks a little weird, like the maybe the alpha channel isn't quite working. This may be a limitation of the Kindle display or browser; I haven't had it long enough to judge).

How about adding support for touch events and gestures? :-)

Something like this with an alternative WebGL renderer would be cool! It can make it a lot faster - see a blog post I wrote on it: http://www.scirra.com/blog/58/html5-2d-gaming-performance-an...

I miss Flash. I know it had big issues but this stuff makes for some hard livin'.

Okay. I wrote StrikeDisplay.blogspot.com which does what this is trying to do. I haven't had much time to work on it for awhile, but I try to keep an eye on developments in this area.

I reviewed the code. I think it's a good effort and it's on the right track, as far as developing a seamless parent/child display chain coupled with event chain for the kind of things we take for granted in Flash. It gets some things right; like apparently re-scoping mouse events with target and currentTarget, which is good if you're expecting methods listening for the events to be scoped in some other way than window or document.

The two issues I see with this library are serious deficiencies in performance -- which could be improved upon, of course, although some of the architecture requiring redraws on every frame is just inefficient as has been pointed out -- and also issues with the way bounding boxes are implemented and how collision/rollover checks are implemented. First of all, it useless to have rollovers always based on rectangles, especially if there aren't careful z-ordering methods. Secondly, having an inner and outer bounding box is a very slow method for traversing the redraws and checking events (so are a lot of other bb methods in canvas. everything's slower in canvas than in Flash). The closer to native browser bone you can cut on this, the better; my solution to click collision checks in StrikeDisplay was to implement a hidden canvas on top of the visible one, that duplicated everything drawn in white, everything else in black, and check the pixel color value of that canvas when the mouse clicked or went over anything inside a bounding box. This turned out to be a hell of a lot faster.

I'm happy to see these kinds of libs come out, because if we do have to give up all the comforts of flash or unity or javafx to develop in this nightmare DOM with nothing but a raw canvas, the sooner we can bridge the gap between visual-oriented coders and everyone else, the better for everybody. This one needs work, but I'd rather see something along these lines than most of the other stuff I've seen come out that's DOM-centric rather than based on a display chain.

Reply on news.ycombinator.com