Apple Watch is encoding each frame as PNG(lionet.livejournal.com)
lionet.livejournal.com
Apple Watch is encoding each frame as PNG
http://lionet.livejournal.com/137350.html
8 comments
Hi, 3rd party watch dev here!
WatchKit is really limited. It would be impossible, I believe, to make an app anywhere near as good as the native Stopwatch using WatchKit.
So this comment is pretty frustrating, actually: "you have to understand at least the basics of the underlying architecture if you want to write efficient code". What basic architecture? WatchKit is all we've got as a 3rd party devs! And it sucks.
It seems insane and very ill-advised to me that Apple isn't using the same SDKs for their own apps as what's available to 3rd parties.
WatchKit is really limited. It would be impossible, I believe, to make an app anywhere near as good as the native Stopwatch using WatchKit.
So this comment is pretty frustrating, actually: "you have to understand at least the basics of the underlying architecture if you want to write efficient code". What basic architecture? WatchKit is all we've got as a 3rd party devs! And it sucks.
It seems insane and very ill-advised to me that Apple isn't using the same SDKs for their own apps as what's available to 3rd parties.
Your point is actually orthogonal to the OP's.
The OP was claiming that Apple's architecture is flawed and that once they realize the error of their ways, battery life will improve dramatically. That's just not true.
Your point is that the public SDK limits 3rd party developers from doing some stuff that Apple can do with the private SDK. That is, of course, absolutely true.
Letting the public SDK (or at least parts of it) lag behind the private SDK, while certainly frustrating to 3rd party developers, is the right thing to do. The situation is exactly the same on the iPhone, iPad, and Apple TV. Apple has employed that strategy from the beginning with iOS, and they believe it to be the correct strategy, so they will almost certainly stick with it.
I don't work at Apple any more (pushing toward launch on a startup!), but based on over 8 years of iOS history, I think it's safe to say that the public watch SDK will get more powerful over time, as private-only features are deemed ready to release to 3rd parties.
The OP was claiming that Apple's architecture is flawed and that once they realize the error of their ways, battery life will improve dramatically. That's just not true.
Your point is that the public SDK limits 3rd party developers from doing some stuff that Apple can do with the private SDK. That is, of course, absolutely true.
Letting the public SDK (or at least parts of it) lag behind the private SDK, while certainly frustrating to 3rd party developers, is the right thing to do. The situation is exactly the same on the iPhone, iPad, and Apple TV. Apple has employed that strategy from the beginning with iOS, and they believe it to be the correct strategy, so they will almost certainly stick with it.
I don't work at Apple any more (pushing toward launch on a startup!), but based on over 8 years of iOS history, I think it's safe to say that the public watch SDK will get more powerful over time, as private-only features are deemed ready to release to 3rd parties.
Yeah, OP's claims about battery life are manifestly silly. The HN headline makes it sound like the article is making claims about the Watch in general, but it isn't, just that one dev's app.
I understand and take your point about public APIs lagging behind private ones, but in this case it's not that the public SDK is lagging, it's a totally different beast.
The fundamental execution model is the same as watchos 1 -- you don't get to write code that runs in the app process, you write code that runs in a separate extension process and get to twiddle ui elements in the main process using a very limited API.
Which is why if you're trying to do anything dynamic, you have to resort to generating and beaming over images.
Besides this fundamental limitation, the management of app process vs extension process vs network daemons is still buggy (contributed to, probably, by the fact that apple isn't dogfooding), which means it's hard to impossible to make your app stable. I think this has been a huge contributor the general perception that 3rd party Watch apps are slow and buggy.
Obviously, I can only speculate about Apple's motives for going with WatchKit vs exposing native development, as on the iPhone.
My guess is they were very worried about an initial public perception of the watch being "battery life sucks," so they didn't want to open up 3rd party code to run on the Watch. Hence WatchKit and OS1.
Then there was a huge public negative perception about not having native apps, so they rushed out OS2, which has 'native' apps, but still following the (now nonsensical) 2-process model of WatchKit.
I do hope Apple opens up true native dev on the watch at some point.
It won't be "the public SDK getting more powerful over time," though, it will be making a qualitative change to the SDK itself.
P.S. If you haven't and are curious, I encourage you to take a look at WatchKit. I think you will be surprised at how limited it is.
P.P.S. I think Apple may have been better served by doing something akin to the original iPhone model -- just saying, "no 3rd party apps yet, but it's coming," rather than having a half-assed SDK on day 1.
I understand and take your point about public APIs lagging behind private ones, but in this case it's not that the public SDK is lagging, it's a totally different beast.
The fundamental execution model is the same as watchos 1 -- you don't get to write code that runs in the app process, you write code that runs in a separate extension process and get to twiddle ui elements in the main process using a very limited API.
Which is why if you're trying to do anything dynamic, you have to resort to generating and beaming over images.
Besides this fundamental limitation, the management of app process vs extension process vs network daemons is still buggy (contributed to, probably, by the fact that apple isn't dogfooding), which means it's hard to impossible to make your app stable. I think this has been a huge contributor the general perception that 3rd party Watch apps are slow and buggy.
Obviously, I can only speculate about Apple's motives for going with WatchKit vs exposing native development, as on the iPhone.
My guess is they were very worried about an initial public perception of the watch being "battery life sucks," so they didn't want to open up 3rd party code to run on the Watch. Hence WatchKit and OS1.
Then there was a huge public negative perception about not having native apps, so they rushed out OS2, which has 'native' apps, but still following the (now nonsensical) 2-process model of WatchKit.
I do hope Apple opens up true native dev on the watch at some point.
It won't be "the public SDK getting more powerful over time," though, it will be making a qualitative change to the SDK itself.
P.S. If you haven't and are curious, I encourage you to take a look at WatchKit. I think you will be surprised at how limited it is.
P.P.S. I think Apple may have been better served by doing something akin to the original iPhone model -- just saying, "no 3rd party apps yet, but it's coming," rather than having a half-assed SDK on day 1.
Actually, thinking about it, they are doing something akin to the iPhone model -- initially iPhone development was supposed to be done as html+js web apps, and devs revolted, and then we got a real SDK.
WatchKit is the html+js of WatchOS development. OK, not that bad, but the situation is analogous.
WatchKit is the html+js of WatchOS development. OK, not that bad, but the situation is analogous.
I was going to reply to you and say the same thing: it's extremely similar to the initial iPhone 3rd party developer story, and for the same reasons. Apple is pretty predictable, really.
I know how WatchKit works, I was in on some of those meetings... Trust me, everyone at Apple had the same concerns you expressed and they still do. It will get better over time.
Also note that some built-in Apple apps do use WatchKit.
And I would still argue that allowing 3rd party code to run on the watch instead of in an extension on the phone would be just another example of an internal-only feature being opened up to the public. It's the same pattern. I realize that's a significant "feature", but so was the switch from html+js to native apps on the phone. And those significant advances go through the same decision process as smaller ones: do we feel comfortable opening up this feature to 3rd parties? what are the risks? have we used it enough internally to feel confident? has it iterated enough internally that we consider it stable enough to release (because it's hard to change after release)? etc.
I know how WatchKit works, I was in on some of those meetings... Trust me, everyone at Apple had the same concerns you expressed and they still do. It will get better over time.
Also note that some built-in Apple apps do use WatchKit.
And I would still argue that allowing 3rd party code to run on the watch instead of in an extension on the phone would be just another example of an internal-only feature being opened up to the public. It's the same pattern. I realize that's a significant "feature", but so was the switch from html+js to native apps on the phone. And those significant advances go through the same decision process as smaller ones: do we feel comfortable opening up this feature to 3rd parties? what are the risks? have we used it enough internally to feel confident? has it iterated enough internally that we consider it stable enough to release (because it's hard to change after release)? etc.
I too was expecting a technical article from an expert, so I closed it after reading the first sentence. While reading the first sentence, a Flash plugin on the page crashed my browser, making the author even less credible of an authority on performance benchmarking.
It is a hosted blogging service and author has no power over what they do with the flash
> Constants are defined via let, and variable are allowed to vary through a var. Though constants are lame, as only the binding itself is constant and you can change everything inside them as usual.
That is at best misleading, as it is not true for collections. From the Swift docs:
> If you create an array, a set, or a dictionary, and assign it to a variable, the collection that is created will be mutable. This means that you can change (or mutate) the collection after it is created by adding, removing, or changing items in the collection. If you assign an array, a set, or a dictionary to a constant, that collection is immutable, and its size and contents cannot be changed.
Swift is the only language that I've seen that actually does make a distinction depending on the keyword used.
That is at best misleading, as it is not true for collections. From the Swift docs:
> If you create an array, a set, or a dictionary, and assign it to a variable, the collection that is created will be mutable. This means that you can change (or mutate) the collection after it is created by adding, removing, or changing items in the collection. If you assign an array, a set, or a dictionary to a constant, that collection is immutable, and its size and contents cannot be changed.
Swift is the only language that I've seen that actually does make a distinction depending on the keyword used.
It also is one of the only languages where collections are value types.
And it has to, to make 'let' create truly immutable objects. Without them, this:
And it has to, to make 'let' create truly immutable objects. Without them, this:
var a = [1,2,3]
let b = a
a += [4,5,6]
You are correct. But in FP, we understand immutability differently, so it becomes very painful to watch structural equivalence of the following variables:
class Foo { var x = 0 }
let a = [Foo(), Foo(), Foo()]
var b = a
a[0].x = 1
var c = b // [0].x == 1 ;(
let d = a // [0].x == 1 ;(
class Foo { var x = 0 }
let a = [Foo(), Foo(), Foo()]
var b = a
a[0].x = 1
var c = b // [0].x == 1 ;(
let d = a // [0].x == 1 ;(
Rust does this as well, just FYI.
> When Apple realizes how to avoid that, the watch will last for a week instead of hours.
In order for this statement to be true, the CPU would have to be using 85% of the entire watch's power consumption just to shuffle PNGs. I bet the entire CPU doesn't use nearly that much power, once you include the display.
In order for this statement to be true, the CPU would have to be using 85% of the entire watch's power consumption just to shuffle PNGs. I bet the entire CPU doesn't use nearly that much power, once you include the display.
This is for 3rd party apps -- despite the "native" 2.0 SDK, where all your code runs on the watch, it's still split clumsily into app (which you can only control by providing interface templates) and extension, which is where your code runs. The extension can control the app only through a very limited SDK that has to do IPC to effect changes in the main app.
Apple's own apps don't have this restriction, and I can't see a good one for 3rd party apps to have it, either -- hopefully they'll open up a full API in the future.
Apple's own apps don't have this restriction, and I can't see a good one for 3rd party apps to have it, either -- hopefully they'll open up a full API in the future.
Would you rather they use uncompressed bitmaps instead?
I remember the little MP3/MP4 players which were very popular around a decade ago(https://en.wikipedia.org/wiki/S1_MP3_player ) used a similar strategy of making all the UIs bitmaps. The LCD had the windowing functionality built into its controller so it was easy to define a rectangular window and then just stream the appropriate pixels into it. They only had a 24MHz Z80 and 384KB of RAM, but hundreds of MB or more of secondary storage, so it made sense to tradeoff size for speed that way.
You can see an example of that here:
http://www.mympxplayer.org/pafiledb/images/screenshots/12088...
I remember the little MP3/MP4 players which were very popular around a decade ago(https://en.wikipedia.org/wiki/S1_MP3_player ) used a similar strategy of making all the UIs bitmaps. The LCD had the windowing functionality built into its controller so it was easy to define a rectangular window and then just stream the appropriate pixels into it. They only had a 24MHz Z80 and 384KB of RAM, but hundreds of MB or more of secondary storage, so it made sense to tradeoff size for speed that way.
You can see an example of that here:
http://www.mympxplayer.org/pafiledb/images/screenshots/12088...
PNG is not cheap on the compression side, due to the lossless coder. It's a little better to decompress:
Some benchmarks on iPhone 6 are here: https://browser.primatelabs.com/geekbench3/4423501
e.g., JPEG is 15x faster to compress and 2.5x faster to decompress.
Some benchmarks on iPhone 6 are here: https://browser.primatelabs.com/geekbench3/4423501
e.g., JPEG is 15x faster to compress and 2.5x faster to decompress.
You can always adjust the compression settings on a PNG to make compression faster.
I think I'm more impressed that LiveJournal is still around 10 years after it first came out.
LiveJournal is older than that; it was launched in 1999.
10 years after it died, you mean.
It's the small accomplishments I appreciate.
The big take-away:
In the end of the day, the watch battery is
there just to shuffle these .PNGs around. When
Apple realizes how to avoid that, the watch
will last for a week instead of hours.Except he provides no evidence this is true, and in fact, it's almost certainly 100% not true ;-)
Hint: the power draw on these watches is not from drawing.
On phones, with powerful CPU's, the display is usually 10x the draw of the CPU (depends on the CPU).
On a watch, what do you think that ratio is?
Hint: the power draw on these watches is not from drawing.
On phones, with powerful CPU's, the display is usually 10x the draw of the CPU (depends on the CPU).
On a watch, what do you think that ratio is?
I have this weird feeling his statement might have been a joke, rather than to be taken literally as an engineering calculation.
Nawh, I'm probably wrong.
Nawh, I'm probably wrong.
As a joke, it's stupid.
Because it is basically "look at these idiots drawing pngs".
Except they aren't idiots, because if it worked, helped them develop faster, and doesn't actually matter, it's actually probably the right thing to do.
So it just makes him look ignorant.
Because it is basically "look at these idiots drawing pngs".
Except they aren't idiots, because if it worked, helped them develop faster, and doesn't actually matter, it's actually probably the right thing to do.
So it just makes him look ignorant.
Today I learned that even HN trolls may not always understand a concept of trolling.
It was indeed a joke.
(author).
It was indeed a joke.
(author).
Jokes are funny when they contain an element of truth. There's a difference between hyperbole and being wrong.
Except Android Wear watches are built essentially identically, and only get about a days worth of life, and don't do any of this crazy shit.
Yeah. Pebble does get a week of battery life, however. The distinction between them and AppleWatch/AndroidWear is that the Pebble uses an E-ink screen that consumes no power when it doesn't change. That makes me think that the display is the biggest power drain on a smartwatch.
It's not E-ink, it's a transflextive LCD
Android Wear watches typically have the screen always on, and in the case of the OLED one (LG Urbane) I've got that means bright enough to see my way through a dark room. I usually wear mine overnight (with the screen off) for sleep tracking, and after 23h it's usually around 50%. I'd be quite happy suggesting you'd get two full days' use out of it, especially if you don't keep it awake all night.
This is ridiculous. Even if you don't use the screen the watch still only lasts less than 2 days.
> In the end of the day, the watch battery is there just to shuffle these .PNGs around. When Apple realizes how to avoid that, the watch will last for a week instead of hours.
Drawing every frame using CoreGraphics is a terrible idea, on any iOS device. Just like any platform, you have to understand at least the basics of the underlying architecture if you want to write efficient code.
I happened to make the built-in Stopwatch app on the watch, which displays an animated graph very similar to his. Suffice it to say, I did not use a CADisplayLink and CoreGraphics at 60fps. Especially considering my graph is horizontally scrollable by the user.
Even if WatchKit still doesn't provide a more efficient way to do this (I don't know; I've never used WatchKit), rest assured that Apple's apps certainly don't use this method, so the week-long battery life claim is just a dream, unfortunately.