...it's when you watch a television show and there's a second screen with it and you can play along with the game, answer questions. One of the second screens we made is this one.
Last month, we had elections here in the Netherlands, and you could play along via your phone or via an iPad. This was all developed in HTML5, of course. It was one version for both the iPad and the phone.
But one thing that's really cool and why I'm showing this is what my boss told me afterwards of the statistics. You always have these exit polls, and they tell you like, "This part is going to win or this party is going to win".
One of the questions was "Wie was het sterkst in de eerste ronde van het debat?" Now this is not the screen of the debate of the fourth of September, but at the end, they asked "Who do you think is the winner of the debate?" There were four party leaders among them, actually these four I believe, and the outcome of this poll was the most exact outcome, closest to the actual outcome of the elections. No poll actually got as close as we got with this second screen, which I think is pretty awesome.
So transitions, this is what I want to talk about. First a brief recap of transition and transformation properties. My guess is everybody has at least worked somewhat with transitions and transformations. Is there anyone who hasn't by a show of hands? I see some hands, but fortunately not too many.
Then what I actually propose to do is to animate everything with transitions, so not just using them a little bit but using them for everything you can possibly think of. Of course, there is a problem with that. That's what this talk is about, and I found the solution to it. I would like to show some demos and then some closing thoughts.
So to begin with, about vendor prefixes, I'm not going to talk about vendor prefixes. I get that you know that you have to use vendor prefixes for most transformation and transition properties, so I'm just not going to name them. Imagine these were there.
Another one, scale, probably known scale X, scale Y. You can use short hands. You can use dual scale, something like that.
Rotate is awesome too in degrees. I thought that it also worked in radians, but somehow it doesn't. Now as a coder, and I love math, I would love to rotate stuff in radians. I don't know, something happened or didn't happen. That's a simple rotation. Of course, there's skew and skews it. As Lea Verou
showed yesterday, you can skew and unskew and nest elements and make all kinds of cool stuff. Skew, skew, skew, skew.
And, of course, you can combine transformations into one property and they'll all be applied consecutively. So the order matters. If you do one thing before the other, the outcome changes. That's a transformation combined.
My personal favorite, of course, is the matrix. The matrix takes six properties. It's scale, skew, and translate. There's no rotation in there because rotation is basically a combination of two skews and two scales.
That might look something like this. Now these properties are mostly used internally by your browser. I don't think anyone does this kind of stuff. Does this work? Yeah, it works. Does anyone know if there are pointers with another color than red because I am color blind, and I cannot see that pointer. [laughter]
I have to look very closely, then I can see it.
There are green ones.
There are green ones. Green is not good either. [laughter]
I don't think anybody does these, writes these manually. So about a year ago I made a little tool for that. Probably some of you might have seen that. It's a lovely tool. It allows you to move around an element, and it shows you the matrix code that goes along with it.
So I can rotate this element, and it gives me the matrix. Rotate it back. I can move it around a little, skew, and scale. And it gives me the matrix that is the CSS matrix. It's a funny little tool. I use it from time to time.
Great! Those were 2D transformation properties. Of course, there is also the 3D variance. 3D basically adds a Z axis transformation along with all the X and Y transform properties except for skew.
Skew in 2D obviously does something like this, but in 3D, I'd expect a Z skew to do something like this. I don't know why it's omitted. I don't know if it's in the spec actually. Does anyone know if there's a skew Z in the spec? No? OK. But it's missing.
I'm not going to go over all the properties. 3D is basically like 2D but then it's really...
3D also has the 4x4 homogenous transformation matrix of 16 values in column major order. Have fun with that.
This is a quote from, I don't think it's readable, Webkit.org. It looks a bit like this. It does something like that.
Just like the 2D matrix, this matrix is mostly an internal value. You can do really cool stuff with that, and, actually, I'm going to do some real cool stuff with that. But, of course, you're not going to write this manually. You're going to make tools for this.
3D also has several other properties. Of course, you need perspective to show anything in 3D. Perspective origin allows you to kind of shift your perspective, and that is in respect to the original elements box.
So if you have 100x100 pixel box and you move it backwards, the perspective origin allows you to shift the perspective in regards to the original 100x100 box. You'd have to see it.
Transformation style is the flat versus preserve 3D variant. It's a really handy property when you need it, but I'm not going to really go into depth about that. Backface visibility obviously allows you to define whether backface is visible or not. OK?
Those were the transformations. Also I wanted to do a brief recap of transitions which we have also seen yesterday. Transition is defined over a property with a given duration, has easing or timing, whatever you'd like to call it, and a possible delay.
Single transition is just a single transition. If you want to define multiple transitions, you can just separate them with a comma. It's straightforward.
A typical transition would use a hover. This is a very simple hover transition over a beautiful div with a drop shadow. It scales it to 1.25, it's original size.
One funny thing to notice is that default easing is actually not linear, which I would have expected it to be because linear is, I don't know, linear is linear. But if you have an undefined easing, it actually does an ease in and an ease out.
So if you're doing anything code-wise with a transition and easing you really should not forget to specify easing. For instance, imagine you have a progress bar or a timer bar and it performs an animation via transition and you forgot to specify the easing, you're going to have a progress bar with ease in and ease out. That obviously makes no sense. So don't forget easing. It's a really cool effect.
Delays too are funny. We saw some examples using delays yesterday. This is a really simple one. It does a fade in, and it does a rotate. Then it does a rotate and a fade out. That's basically your trick, switching which transition goes first.
The confusing thing is that derp has initial transition on transform with no delay and a half second duration and opacity has the half a second delay, but it performs the fade in first.
That is because as soon as you hover, the new delay is applied first and then the transition is performed. So, again, just like easing, you have to take a little bit of care when you're using delays.
Good, animating things. Well, I really love to animate everything. If there is something that can be animated, I typically do it. I've done DHTML games, moving divs around, canvas implementations, playing with pixels, all kinds of stuff.
So when it comes to animating things, I basically think there are two kinds of animations. You have your static animations that go from A to B like a mouse over that does scale or a movement from some place to another. A and B are defined, are fixed, and those are easy because most of these transitions you can do with CSS, with a simple transition and a hover where a class will change.
Then aside from that there are dynamic animations. Dynamic animations mostly follow some kind of user interaction. It's like you're dragging, you drag and drop something around, you're doing something with inertia.
As you see, I think I'm the only speaker that has a Windows laptop. I hope I'm not the only one present with a Windows laptop. Two, three, four, oh back in there too. Thank God I'm not alone, upstairs too.
I also don't have an iPhone, but I really love the scroll thingy. When you scroll, you get this beautiful scroll bar. As you pull it out if you pull too far, the scroll bar becomes smaller. When you let go, it bounces back. It's a really cool effect.
They may use transformations. In fact I would say you should use transformations, of course, but they probably don't use transitions because you cannot enter variables.
Couldn't these animations too use transitions? That is what I ask myself. There are really great benefits to using transitions.
For one, the CSS animation interface, it is a complete animation engine. I really don't want to write easing equations. I have done so. I know a couple of them. But if I don't have to, I don't want to write easing equations.
So this is a complete animation engine, and if I could use the easing equations from CSS, that would be really awesome. So no per frame script calculations. When you start out animating stuff, you typically use a time out or an interval. You add several pixels to the distance in regard to the last step. That is your next step.
But when things evolve and it becomes more complex, you need time based animations. It starts at a certain point. It has a given velocity, and after two seconds or so it has to be somewhere or two seconds is your duration and you don't know how many frames you're going to have to render that thing. So that's pretty complex stuff. If CSS can do that for me, that would be awesome.
It's already optimized. Per frame calculations are, perhaps, on a desktop computer running Chrome on a JIT compiler or script engine.
That's all really awesome, but on a mobile phone if I'm going to do per frame calculations for all my animations, it's going to stutter like hell. It's going to drain the battery. That's really terrible stuff. Why would you want to use 3D transitions, 3D transitions in particular or transitions at all?
Transitions don't cause reflows of elements. If you animate a width or a height property, it's going to trigger a reflow, and the browser has to rerender the element. This goes for most properties in CSS. I believe, actually, all of them. I'm not sure.
Transitions, they don't actually cause these reflows. Aside from that, they are hardware accelerated so you get lots of frames per second. They are also hardware accelerated on mobiles.
I don't know how many of you have done animations on, for instance, an older iPad. But if you do a transition of a width or a height property on an old iPad, say an iPad 1, it's going to be stuttery and, well, you don't want that. Even the iPad 1 already supports these 3D CSS properties.
As I said, it's battery friendly, and it's actually pretty widely supported. The only browser really not supporting these transitions is Internet Explorer, but it's going to catch up with version 10.
This talk is not specifically about what browsers support or what they don't support, so I'm not really considering this. If I want to do dynamic animations using CSS transformations, there is a huge problem. The transitions cannot be stopped at a current position.
So I made this beautiful image. Imagine I have some kind of div element. I'm dragging it around, and when I let go, I want it to have a moment of inertia, that it comes to a halt over a given time, say one or two seconds.
This is fine, and I could write some script that finds out what the velocity is of when I was moving my mouse, but during that moment of inertia, if that were to be a CSS transition, there would be no way to actually pick it up halfway and move it somewhere else, because it's currently running a transition, and it has only a starting point and an ending point.
I don't know where it is, I don't know how to stop it, so I need some kind of way to interrupt a running transition.
Computed Style gives you the computed style of an element, and you can request the current property value of an element. If I have an element, and this element has two transitions, it can either move to the right with an ease out, or it can move to the left with an ease out.
Behind this is a little script, and the script outputs the current value of the transformation using the requestAnimationFrame callback.
Wow. This is funny, because I defined the transition on transformation, and there is nowhere, no place that I defined the matrix to be used, and still it returns a matrix.
This is actually funny, because if it returns a matrix, then in theory, I could request the property, and at a given time, just apply that current value to the element. And then in theory, it should actually stop. That's the plan of attack, query the current transform value and assign it.
These are the first lines from before, querySelector ("#derp"), window.getComputedStyle (derp), and the current transform is (computed.getPropertyValue('transform')). Obviously it should say WebKit here if you want it to run in WebKit. Then at the point where I want to stop the transition, I should remove the transition property.
If I would do nothing else than remove the transition property, the element would just to its final destination, to the B point. But if I immediately set the current property transform to that property that I requested, then it should actually stop at the point it currently is.
That's all theory, of course. We have to try this. One example is a draggable element, and when I let go, it takes the current velocity and comes to a halt with an ease out. I didn't do any animation coding for this. It's just some script tracking my mouse. When I mouse down, I can pick it up, move it. It finds out the velocity at which I was moving, and when I let go, it constructs a CSS transition which is applied immediately to moving element.
Not only that, when I'm moving it, I can stop it by just picking it up again. We've all seen this before, it's just a movable element with inertia and some dragging, but it's completely animated via CSS. There's no more animation script in this. If you do this on a mobile phone, it's going to be battery friendly, it's going to be hardware accelerated, and it's fully interactive.
Yeah. [laughs] So, I figured, if querying the current 2D transformation value and applying it works in 2D space, would that also work in 3D space?
That would allow me to make fully interactive 3D whatever. It doesn't matter. They could be completely run using CSS animations or transitions. I made one too. What this does...this is a beautiful cube and this is the view from the front. There is an entire cube. It works the same way as the little element from before. I can pick it up and move it. When I stop moving, it has a moment of inertia. It keeps rotating.
At any given time that I want, I can pick it up again and remove it. This goes around. If I give it a little bit of a push, it moves. It doesn't matter when I let it go or when I pick it up. It just is a beautiful, moving cube. [applause]
This is all running on CSS transitions. There is not a single line of animation script used in this. The only script that I used is for interaction. Behind this, just like the 2D one, is a 4x4 matrix implementation, allowing 3D transformations. So, I can pick up...it's in the next slide, actually.
Should I get a Mac? Yeah? No? Who said that? [laughs] Marcin is selling his.
Who is selling whose? The one that fell and [inaudible].
Hmm. One other thing and I don't know if it's visible actually, because there is a bit of a reflection here...is it visible?
No. We trust you.
OK. That is -webkit-box-reflect. I don't know if you know that property. -webkit-box-reflect lets you do iOS style reflections. Of course, they are beautiful. It just turns out that it also works for 3D transformations. Below this cube is a reflected, rotating cube via -webkit-box-reflect.
It should be visible. There is a bit of a thing here. But maybe only I can see that.
Great! There are some things to keep in mind. Querying CSS properties doesn't always work. We had some issues on, for instance, Android, with inherited properties like fonts and colors. What it returns is an empty string. I didn't really look into that.
But, basically, you will probably have to bubble up into the DOM and try requesting the property, until you get something.
I don't know why I put this here actually. I believe, if you remove a property from the style member from an element directly, it becomes empty and it reverts back to the style sheet property. So there is a way to work around specificity issue. It also works if you empty the property, so not remove property but set value to empty.
So what about fall backs? Of course, this does not work in any version of Internet Explorer. It's going to work in IE10 presumably. If this is really important to you, you could use a fall back using left and top properties. Of course, you cannot do that for 3D, so 3D is going to revert to 2D. But this is really all fine.
I think if you're making a website or an application and it runs on either desktop and it's also available on your iPad and on your iPhone, the environment in which it runs defines which technologies you can use.
It's fine to use 3D transforms on an iPad. In fact, I would want to encourage you to do so. It doesn't matter that it doesn't work in Internet Explorer because you can use other effects in a desktop version of your app. So otherwise, if you don't have to rely on fall backs, just use them as progressive enhancements. Actually, they're a lot of fun to play with. That's the main reason for me doing these kinds of things. It's just fun. Mathias
likes to read specs, and I too read specs, but I also like math. So in my spare time, I do stuff with math. I don't know.
Finally, convince your client or project manager that this is OK. We all have the knowledge that we should use progressive enhancements. We know the technologies, but we are the ones that need to convince the people deciding things or selling things or creative people making beautiful designs. We have to tell them what we can do for them. That's our job. That was it. [applause]
I think next time we'll give you a bottle of water. I think there were like three breaks going over there. Interesting, I found that the end transition not firing, I think that's by design, isn't it? I wouldn't call it a bug at all.
I think so, yeah, yeah.
Did you find, then, in general, cause transitionEnd and transitionStart also gives you a time stamp. Is that something that could be used for that as well without having to interrupt the transition at all?
How do you mean? TransitionStart and transitionEnd has a property in between which is a time stamp of the animation, so on click you could say "where would it be in between beginning and end" and things like that. Let's do it on Github, don't worry about it. I like events, but I've found that transition events are not firing reliably.
I don't even use them. OK, the transitionStart even also coincides with interaction in some way, so you already have a click or a mouse over or something. You know when the transition starts because you trigger it. The only thing you don't know is when it ends.
The point is not waiting until ends but interrupting it in the middle, and there is no event for that.
Cool. So in terms of syntax, you said you're a person that likes math. I always find that CSS transition and CSS animation syntax, especially the order of it, that it can be arbitrary...
Yeah, it's like the demo that Lea Verou showed yesterday of the smiley face that didn't rotate but it rotated because it rotated against the other rotation.
If you rotate something and then translate it, that is something different from first translating it and then rotating. You're changing the point from where you're transforming it. Maybe it's hard to understand or it's difficult or it's in some unpredictable or something, but more than that, it's actually a powerful tool.
I really like it that the order in which I define transformations gives a different result, and that's the same thing that matrixes do anyway.
It just feels odd the first few times that you have to remember it.
But at the same time, it supposedly doesn't matter which order it's in.
There are things like we don't exactly know how it works so we shun away from it or it's somehow scary. I think that is the wrong way to look at things. It may work differently based on how you write it, and there's a reason for that. There's a very good reason for that. What we need to do is find out what that reason is and just understand it.
Yeah, I like the matrix transformation tool because I did this CSS 3D transformation animation tool as well for myself to understand what's going on.
It is really the best way of learning something, isn't it? Yeah.
Because you know the syntax is annoying. You might as write a code editor for you to do it.
Yeah, that's how tools come to be.
I found one thing that really annoys me about Chrome especially, and I think it's just...I don't know why it's on design or not. Like when I do a transition on something or an animation on something and I load the page, it starts from where the normal flow is to the first stage. It does a transformation before I actually say do one.
Is there any way to avoid this? To get the transition? To stop it from animating from undefined to the first defined state when I just put a transition all on something.
I should see that issue. I don't...do you have an example.
Not here, on my desktop. [laughs] We can do that later as well.
Yeah, well, it's the other way around actually. Sometimes when there's non defined, it still animates to the first state of the page as it is when you do a transition in it.
That's really weird.
Yeah, it is. How do you find about performance? You said, "This is better for performance. This is better for battery life". Do you do some excessive testing on that? Do you have a test harness for it? Do you try it out by hand?"
We just try it out by hand. The cube also works on an iPad, so you can just drag it around and it rotates and stuff. The demo is not completely finished, but I'm going to finish it and put it online so everybody can play with it.
That's a good idea. Flickr, their iPad HTML5 version, they basically found that dragging images when you do a navigation left and right, that image elements were much, much slower than divs with background image and then translation was must faster than changing the X and Y.
That's normally the case, that the 3D acceleration kicks in with translate? Yeah. I try to do everything with these 3D translations because the performance is so, so amazing.
So you would say it's better than going with the old top and left style which is much easier to understand? Yeah, don't use left and top. [laughs]
You mean like shader style? Like putting something on an environment? That is just it, you've got a kind of curved perspective. It's fish eye effects.
Yeah. In canvas, I've seen some of them, but they were hard coded.
In canvas, yeah. Canvas can do anything. [laughs]
Why do you like Canvas? I see some math geek would be appealing to you.
Yeah, I don't know. I just love Canvas. It's a 2D...it's an empty piece of paper, and you can draw something on it. I just love it. There are people complaining about Canvas not having a DM and not being able to click on stuff, but that is exactly what makes it so fast. I don't know, I just love it.
I found it more too...I always said Canvas is like an Etch-a-Sketch on your page. You basically can paint on it, and you can shake it to actually delete it.
Canvas is like a mini Photoshop in your browser, a programmatically accessible Photoshop.
Yeah, except it has the wonderful thing of a state in terms of the coordinate system.
I like that you can paint a shape and then rotate it and scale it and reuse it that way.
So basically that would allow for things like Easel library and things like that to fall back to a CSS transformation in case Canvas is not supported.
For Internet Explorer. It's hard to actually test...
The browser I didn't want to name.
The browser I didn't want to name. [laughs] After saying that you like Windows much. One question here, there's actually not much coming in interestingly enough. "What do you mean by 'these things are available for browsers internally.' Are matrixes internally used for everything?"
Well, apparently so. Because if you apply a translate, if you do a combined transformation using the translate, rotate, and skew properties, what you get in return if you query the value is a matrix. So your browser uses this matrix anyway.
I don't know, I'm glad this matrix is exposed because it allows me to do more than just the more human readable transformation properties. But I only assume that that is what it internally does.
It's the same with 3D transformations. We simulated them with X and Y and depth of field before.
And now we've got the real XYZ axes to play with.
Yeah, yeah. You said you love playing with these things and doing this, and I find myself...How do you get around the issue of building something like the cube and then you find yourself sitting in front of it for half an hour and playing with it rather than doing something else?
Yeah, that happens, yeah.
How do you stop yourself? I know that feeling, so I just have to kick myself and continue working.
Yeah. The side bar on my blog has just z-index tilt, and I get so many emails about how is this done. I wrote a blog post about it. It's not hard to do, but it seems to be very engaging for people to play with things like that.
Yeah. It's extremely funny. You keep doing that, and that's also what keeps you going, what keeps you making these kinds of things.
So do you use any of that in your second screen application, as Easter eggs, for example, or...
Well, this particular technology in stopping a transition and programmatically creating a new one is not yet integrated into our system, but it's going to be.
So you do it for like expanding and bouncing transitions or...
Well, no, not really. One thing, for instance, which we remade, and that's an older discussion about should we recreate the native app field in web apps. Maybe we could agree or disagree that it is either right or wrong, but we did it anyway.
One thing we made is just that scrollable thing that you can scroll content like on iOS.
Like the bounce back...
With the bounce back and the scroll bar thingy.
Canvas in the back? Yeah, your native field.
Nice. But then you always try this and you put it online. You put it on Github and tell people about it, and then you get 6000 requests "oh, it doesn't perform the same as iOS."
Are we trying to simulate too much and not having the same access to the hardware? In my opinion, you can get really, really, really close, but there is always a tiny delay. If you scroll on your iPhone, it's just direct. Your page moves immediately, and there is a slight delay.
But it can be progressively enhanced as well. So if it doesn't work it's just a normal flow.
Cool. Well, I guess that's all we have for now.
OK. Thank you very much. [applause]