Jake Wharton
Android’s Java 9, 10, 11, and 12 Support
27 November 2018
Note: This post is part of a series on D8 and R8, Android’s new dexer and optimizer, respectively. For an intro to D8 read “Android’s Java 8 support”.
The first post in this series explored Android’s Java 8 support. Having support for the language features and APIs of Java 8 is table stakes at this point. We’re not quite there with the APIs yet, sadly, but D8 has us covered with the language features. There’s a future promise for the APIs which is essential for the health of the ecosystem.
A lot of the reaction to the previous post echoed that Java 8 is quite old. The rest of the Java ecosystem is starting to move to Java 11 (being the first long-term supported release after 8) after having toyed with Java 9 and 10. I was hoping for that reaction because I mostly wrote that post so that I could set up this one.
With Java releases happening more frequently, Android’s yearly release schedule and delayed uptake of newer language features and APIs feels more painful. But is it actually the case that we’re stuck with those of Java 8? Let’s take a look at the Java releases beyond 8 and see how the Android toolchain fares.
Java 9
The last release on the 2 — 3 year schedule, Java 9 contains a few new language features. None of them are major like lambdas were. Instead, this release focused on cleaning up some of the sharp edges on existing features.
Concise Try With Resources
Prior to this release the try-with-resources construct required that you define a local variable (such as try (Closeable bar = foo.bar()) ). But if you already have a Closeable , defining a new variable is redundant. As such, this release allows you to omit declaring a new variable if you already have an effectively-final reference.
This feature is implemented entirely in the Java compiler so D8 is able to dex it for Android.
Unlike the lambdas or static interface methods of Java 8 which required special desugaring, this Java 9 feature becomes available to all API levels for free.
Anonymous Diamond
Java 7 introduced the diamond operator which allowed omitting a generic type from the initializer if it could be inferred from the variable type.
This cut down on redundant declarations, but it wasn’t available for use on anonymous classes. With Java 9 that is now supported.
Once again this is entirely implemented in the Java compiler so the resulting bytecode is as if String was explicitly specified.
Because there is nothing interesting in the bytecode, D8 handles this without issue.
Yet another language feature available to all API levels for free.
Private Interface Methods
Interfaces with multiple static or default methods can often lead to duplicated code in their bodies. If these methods were part of a class and not an interface private helper functions could be extracted. Java 9 adds the ability for interfaces to contain private methods which are only accessible to its static and default methods.
This is the first language feature that requires some kind of support. Prior to this release, the private modifier was not allowed on an interface member. Since D8 is already responsible for desugaring default and static methods, private methods were straightforward to include using the same technique.
Static and default methods are supported natively in ART as of API 24. When you pass —min-api 24 for this example, the static method is not desugared. Curiously, though, the private static method is also not desugared.
We can see that the getHey() method’s access flags still contain both PRIVATE and STATIC . If you add a main method which calls hey() and push this to a device it will actually work. Despite being a feature added in Java 9, ART allows private interface members since API 24!
Those are all the language features of Java 9 and they all already work on Android. How about that.
The APIs of Java 9, though, are not yet included in the Android SDK. A new process API, var handles, a version of the Reactive Streams interfaces, and collection factories are just some of those which were added. Since libcore (which contains implementation of java.* ) and ART are developed in AOSP, we can peek and see that work is already underway towards supporting Java 9. Once included included in the SDK, some of its APIs will be candidates for desugaring to all API levels.
String Concat
The new language features and APIs of a Java release tend to be what we talk about most. But each release is also an opportunity to optimize the bytecode which is used to implement a feature. Java 9 brought an optimization to a ubiquitous language feature: string concatenation.
If we take this fairly innocuous piece of code and compile it with Java 8 the resulting bytecode will use a StringBuilder .
The bytecode contains the code we otherwise would have written if the language didn’t allow simple concatenation.
If we change the compiler to Java 9, however, the result is very different.
The entire StringBuilder usage has been replaced with a single invokedynamic bytecode! The behavior here is similar to how native lambdas work on the JVM which was discussed in the last post.
At runtime, on the JVM, the JDK class StringConcatFactory is responsible for returning a block of code which can efficiently concatenate the arguments and constants together. This allows the implementation to change over time without the code having to be recompiled. It also means that the StringBuilder can be pre-sized more accurately since the argument’s lengths can be queried.
If you want to learn more about why this change was made, Aleksey ShipilГ«v gave a great presentation on the motivations, implementation, and resulting benchmarks of the change.
Since the Android APIs don’t yet include anything from Java 9, there is no StringConcatFactory available at runtime. Thankfully, just like it did for LambdaMetafactory and lambdas, D8 is able to desugar StringConcatFactory for concatenations.
This means that all of the language features of Java 9 can be used on all API levels of Android despite changes in the bytecode that the Java compiler emits.
But Java is now on a six-month release schedule making Java 9 actually two versions old. Can we keep it going with newer versions?
Java 10
The only language feature of Java 10 was called local-variable type inference. This allows you to omit the type of local variable by replacing it with var when that type can be inferred.
This is another feature implemented entirely in the Java compiler.
No new bytecodes or runtime APIs are required for this feature to work and so it can be used for Android just fine.
Of course, like the versions of Java before it, there are new APIs in this release such as Optional.orElseThrow , List.copyOf , and Collectors.toUnmodifiableList . Once added to the Android SDK in a future API level, these APIs can be trivially desugared to run on all API levels.
Java 11
Local-variable type inference was enhanced in Java 11 to support its use on lambda variables. You don’t see types used in lambda parameters often so a lot of people don’t even know this syntax exists. This is useful when you need to provide an explicit type to help type inference or when you want to use a type-annotation on the parameter.
Just like Java 10’s local-variable type inference this feature is implemented entirely in the Java compiler allowing it to work on Android.
New APIs in Java 11 include a bunch of new helpers on String , Predicate.not , and null factories for Reader , Writer , InputSteam , and OutputStream . Nearly all of the API additions in this release could be trivially desugared once available.
A major API addition to Java 11 is the new HTTP client, java.net.http . This client was previously available experimentally in the jdk.incubator.http package since Java 9. This is a very large API surface and implementation which leverages CompletableFuture extensively. It will be interesting to see whether or not this even lands in the Android SDK let alone is available via desugaring.
Nestmates
Like Java 9 and its string concatenation bytecode optimization, Java 11 took the opportunity to fix a long-standing disparity between Java’s source code and its class files and the JVM: nested classes.
In Java 1.1, nested classes were added to the language but not the class specification or JVM. In order to work around the lack of support in class file, nesting classes in a source file instead creates sibling classes which use a naming convention to convey nesting.
Compiling this with Java 10 or earlier will produce two class files from a single source file.
As far as the JVM is concerned, these classes have no relationship except that they exist in the same package.
This illusion mostly works. Where it starts to break down is when one of the classes needs to access something that is private in the other.
When these classes are made siblings, Outer$Inner.sayHi() is unable to access Outer.name because it is private to another class.
In order to work around this problem and maintain the nesting illusion, the Java compiler adds a package-private synthetic accessor method for any member accessed across this boundary.
This is visible in the compiled class file for Outer .
Historically this has been at most a small annoyance on the JVM. For Android, though, these synthetic accessor methods contribute to the method count in our dex files, increase APK size, slow down class loading and verification, and degrade performance by turning a field lookup into a method call!
In Java 11, the class file format was updated to introduce the concept of nests to describe these nesting relationships.
The output here has been trimmed significantly, but the two class files are still produced except without an access$000 in Outer and with new NestMembers and NestHost attributes. These allow the VM to enforce a level of access control between package-private and private called nestmates. As a result, Inner can directly access Outer ’s name field.
ART does not understand the concept of nestmates so it needs to be desugared back into synthetic accessor methods.
Unfortunately, at the time of writing, this does not work. The version of ASM, the library used to read Java class files, predates the final implementation of nestmates. Beyond that, though, D8 does not support desugaring of nest mates. You can star the D8 feature request on the Android issue tracker to convey your support for this feature.
Without support for desugaring nestmates it is currently impossible to use Java 11 for Android. Even if you avoid accessing things across the nested boundary, the mere presence of nesting will fail to compile.
Without the APIs from Java 11 in the Android SDK, its single language feature of lambda parameter type inference isn’t compelling. For now, Android developers are not missing anything by being stuck on Java 10. That is, until we start looking forward…
Java 12
With a release date of March 2019, Java 12 is quickly approaching. The language features and APIs of this release have been in development for a few months already. Through early-access builds, we can download and experiment with these today.
In the current EA build, number 20, there are two new language features available: expression switch and string literals.
Once again, both of these features are implemented entirely as part of the Java compiler without any new bytecodes or APIs.
We can push this to a device to ensure that it actually works at runtime.
This works because the bytecode for expression switch is the same as the “regular” switch we would otherwise write with an uninitialized local, case blocks with break , and a separate return statement. And a multi-line string literal is just a string with newlines in it, something we’ve been able to do with escape characters forever.
As with all the other releases covered, there will be new APIs in Java 12 and it’s the same story as before. They’ll need added to the Android SDK and evaluated for desugaring capability.
Hopefully by the time Java 12 is actually released D8 will have implemented desugaring for Java 11’s nestmates. Otherwise the pain of being stuck on Java 10 will go up quite a bit!
Java 8 language features are here and desugaring of its APIs are coming (star the issue!). As the larger Java ecosystem moves forward to newer versions, it’s reassuring that every language feature between 8 and 12 is already available on Android.
With Java 9 work seemingly happening in AOSP (cross your fingers for Android P+1), hopefully we’ll have a new batch of APIs in the summer as candidates for desugaring. Once that lands, the smaller releases of Java will hopefully yield faster integration into the Android SDK.
Despite this, the end advice remains the same as in the last post. It’s vitally important to maintain pressure on Android for supporting the new APIs and VM features from newer versions of Java. Without APIs being integrated into the SDK they can’t (easily) be made available for use via desugaring. Without VM features being integrated into ART D8 bears a desugaring burden for all API levels instead of only to provide backwards compatibility.
Before these posts move on to talk about R8, the optimizing version of D8, the next one will cover how D8 works around version-specific and vendor-specific bugs in the VM.
(This post was adapted from a part of my Digging into D8 and R8 talk that was never presented. Watch the video and look out for future blog posts for more content like this.)
Источник
Java Enterprise vs Android в 2019 — что выбрать новичку?
Решил поделиться своими мыслями на тему того, в какую отрасль разработки стоит пойти человеку, освоившему Java core и основы computer science. А дороги как известно две: Java Enterprise или Android-разработка. Под Java Enterprise программированием я понимаю разработку, вакансии которой на hh имеют заголовок «Java-разработчик». Вакансии, связанные с android-разработкой можно найти на том же сайте по запросу, соответственно, «android-разработчик».
Будем считать, что вы изучили на базовом уровне core языка Java, а также ознакомились с темами алгоритмов, SQL и другими базовыми вещами из computer science, и теперь выбираете путь куда двигаться дальше с прицелом на то, чтобы как можно скорее устроиться на должность junior-программиста. Обсудим какие есть плюсы и минусы в выборе между двумя обозначенными выше ветками разработки.
1) Стек технологий
Java-программирование пришло в бизнес уже почти 20 лет назад. Как следствие, в сфере появилось большое разнообразие фреймворков. На каждом проекте используется свой стек технологий, и бывает достаточно сложно понять, какие технологии еще будут жить, какие вот-вот умрут, а какие уже давно не используются. При этом, кроме джавовских вещей от джависта также требуют знать UI технологии: JS с фреймворками, html, css
Так как android относительно молодая ОС, а в бизнес она вошла еще позже, то тут нет такого зоопарка фреймворков как в большой джаве. Нет тут ни спринга, ни хибернейта, ни других более экзотических вещей. Работу тут можно начинать имея в багаже знаний лишь android sdk и java core. UI, насколько мне известно, предоставляется прямо «из коробки» (android studio), средствами drag and drop. То есть GUI часть дополнительно изучать не нужно
2) Особенности сферы
Как это ни печально, в России Java-программирования нет (за небольшим исключением — банки и гос порталы). То что мы пишем тут отправляется заказчикам в Европу и Америку. Следствием того, что заказчиками являются крупные компании является сложность самой бизнес-логики. Минимальный порог вхождения предполагает не только знание основных фреймворков, но также и специфические требования по распределенным системам, big data, глубокому пониманию многопоточности, машинному обучению
Android-приложения пишутся как для малого бизнеса, так и для крупных игроков по всему миру. Как следствие, есть возможность стартовать с простых проектов и развиваться в сторону более сложных. Другими словами, android предоставляет более низкий порог входа, но не ограничивает потолок — наряду с простыми проектами на рынке присутствуют также и достаточно сложные.
3) Есть ли будущее?
Насчет промышленного программирования на большой джаве — не знаю. Хотя она и держится в первых строчках рейтинга языков программирования, чем это вызвано для меня не понятно. Раньше это можно было объяснить наличием JVM и ее переносимостью, но теперь, когда десктопные приложения больше не разрабатываются и весь функционал выносится в веб — будущее джавы как языка для enterprise программирования для меня под вопросом. Наверное, раз джаву не очень активно используют российские компании, предпочитая ей другие языки программирования, видимо джава не очень современный язык, отвечающий требованиям реальной жизни.
На рынке смартфонов OS Android нет конкурентов. Вероятность того, что iOS поглотит android близка к нулю. Следовательно, ближайшие пару десятков лет в отрасли вряд ли произойдут существенные изменения. Может будут появляться новые фреймворки, но тот кто начинает с нуля сейчас сможет последовательно их изучать и расти профессионально вместе с развитием самой отрасли. В отличие от большой джавы не нужно пытаться изучить все и сразу, чтобы хоть куда-то устроиться — можно спокойно изучать тот небольшой scope технологий, что используются на рынке сейчас
Источник