Android platform engineer on application architecture

It already seems to be the goal to make all Java apps work on Native Client on Chrome OS. But once that's done, couldn't they just allow anyone to write apps in Go, Rust, Python, etc for Native Client directly? (and thus eliminating what will probably be some performance penalty of running Dalvik apps on Native Client).

That way Google gets to deprecate Java apps in the next 5-10 years and be done with Oracle forever, and developers would have a choice of some really good and safe languages as well.

Welcome to Worse is Better. This is literally what Unix went against. People used to write that way before; now it has to be easiest for platform developer/maintainer.

Android invested up front in more complex technology that would allow adaptation to future hardware devices. Most obviously, by using layout managers instead of pixel positioning like iOS did. That meant Android was able to adapt to a range of screen sizes way earlier than Apple could. Every time Apple tried to introduce a new screen size or resolution it resulted in major pain across their entire app ecosystem, something many Android apps avoided.

Phones are still memory constrained. Yes they have "shitloads" of RAM compared to previous devices but they still don't have a swap file of any kind. The activity model allows phones to have an apparently infinite number of apps open, without the UI ever stalling due to hitting a paging storm like desktop apps can. Simply increasing the amount of RAM doesn't solve the problems the activity model was designed to solve, and it's worth noting that when Apple introduced multi-tasking into iOS, they ended up with something quite similar to activities.

The choice of Java had complex effects on memory usage. Whilst Java heaps can be kind of piggy, a lot of memory usage comes from loaded code, and bytecode is a lot more compact than native code is. That's why they're reintroducing JIT compilation in Android N - having moved to fully AOT when ART was introduced they are now going back again to reduce memory pressure.

The decision to support SD cards made a lot of sense back in the days when internal storage was a lot more limited.

Layout managers are complex technology now ? The layout managers Android uses are pretty much identical to what Java used in AWT and Swing and is trivial to implement. I know because I implemented just such a thing for a UI toolkit I made for J2ME back in the day. Desktop apps have been doing it for decades. iOS by contrast uses a much more complicated system (not saying it's better, just more complex).

Also, I vastly prefer iOS's sytem with just a handful of resolutions so you can fine-tune your designs to the exact screen dimensions. In practice I've found it very hard to explain Android's layout mechanisms to designers and to get good designs from them that scale well. Most designers I know prefer being able to make a pixel-perfect design like you can on iOS. Sure it works fine for business-style apps but if you want to make something spiffy, you're fucked on Android.

> when Apple introduced multi-tasking into iOS, they ended up with something quite similar to activities.

iOS has nothing even remotely resembling activities. Background apps can be suspended and possibly killed, but that's as far as the parallelels go. Don't even get me started on the insane idea of Intents (sounds like a great concept but horrible in practice and badly implemented too).

> The activity model allows phones to have an apparently infinite number of apps open, without the UI ever stalling due to hitting a paging storm like desktop apps can.

Then why is this not an issue at all on iOS. In fact, my 6S pretty much never kills background apps and even when it does that does not affect foreground performance. And if they were so concerned about the UI stalling, they should have stayed away from Java. It has improved a little bit but especially in the early days the GC would regularly freeze the entire JVM for 100+ ms. Try hitting 60fps like that.

> The choice of Java had complex effects on memory usage. Whilst Java heaps can be kind of piggy, a lot of memory usage comes from loaded code, and bytecode is a lot more compact than native code is.

Wut ? Loaded code is a completely negligible part of an apps memory usage. And again, not an issue on iOS. In fact memory is far more constrained on Android than on iOS even when Android devices have more physical RAM. Android severely limits the amount of RAM an application can use (the heap size limit). I've never run into problems with RAM usage on iOS and regularly on Android. On a 1GB iPhone you can easily use 700-800MB RAM for your app if you wanted, while you're lucky if you can get 192MB on a 3GB Android phone.

Furthermore, garbage collected languages use about 3 times as much RAM as a non-GC language for similar performance (tradeoff between how often the GC runs vs performance) negating any and all gains of a smaller code size. There is also this thing called the ARM Thumb instruction set which results in smaller binaries if that really is an issue.

> The decision to support SD cards made a lot of sense back in the days when internal storage was a lot more limited.

And this goes to the heart of my criticism: yes internal storage was a lot more limited back then. Both Google and Apple knew this. They both also knew (or should have) that internal storage was going to improve a LOT and very quickly at that. This is the point I'm trying to make: Google develops for yesterdays and todays constraints while they should be developing for tomorrows. In fact, Google should look farther into the future than Apple does because it takes so long for their software to reach consumers, it's taking 2-3 years for any new Android release to gain any significant market share. Android N should be built with 2019's phone hardware in mind, instead it's being built for 2015 hardware.