Tag Archives: armeabi

Android NDK: A guide to deploying apps with native libraries

Libraries written in C++ and C have long been supported in Android through the Native Development Kit (NDK). Even if you haven’t directly built native code libraries, you may have included pre-built native libraries; they can be identified by the .so file extension. The main drivers for Android developers to include native libraries are:

  • Porting existing code to Android
  • Building a library of functionality
  • Increasing performance for intensive algorithms

Note: The process of building a native library is not covered in this article. Head to the Android NDK site for the tools and information you need to build native libraries for Android.

Android NDK

Most apps won’t benefit from native libraries and most apps should be developed exclusively using Java, which is executed in a virtual machine on the device using the Java ‘Write once, run anywhere’ philosophy. Including native libraries in your app results in you losing the write once, run anywhere benefit. Figure 1 shows a summary of the pros and cons of using native libraries:

Figure 1: Pros and cons of using a native library

Figure 1: Pros and cons of using a native library

Android Architectures and the Application Binary Interface (ABI)

Different Android devices use different CPU architectures, which in turn support different instruction sets. An ABI is different for each combination of CPU and instruction set. If you build with native code you need to compile that native code for each ABI that your app supports.

ABIs supported by Android

Android supports architectures from ARM, Intel and MIPS technologies, each in 32 bit and 64 bit flavors. In figure 2, below, you can see each ABI supported along with example devices.

Figure 1: Android ABI support

Figure 2: Android ABI support

Shared native library example

In this article I will use the SQLCipher project as an example use case for native libraries. SQLCipher is a great example of where using native libraries in Android is a necessity, here’s why:

  1. Performance benefits: SQLCipher has to decrypt/encrypt data, which requires very intensive processing
  2. Ease of reuse: The SQLCipher library was already developed in C/C++, using a native library makes is easier than rewriting and maintaining the library in different languages for different platforms

Performance testing

Building a shared native library can provide significant performance increases compared with implementing the same algorithm in Java. During some basic testing to compare some similar algorithms in native code compared to Java, I found the result ranged from double to a 20 times performance increase, however it should be noted these algorithms were heavy image manipulation processing. On less intensive use cases there will be little, if any, performance benefit.

Native library performance benefits

Figure 3: Native library performance benefits

APK size increase with native libraries

The table in Figure 4 shows the effect on the file size of an Android APK that include the .so libraries for SQLCipher. The universal APK in the table is built with libraries for all the tested ABIs.

Native library size increase table

Figure 4: APK size effect of including SQLCipher native library

For more information take a look at how to set your build.gradle file to build APKs that target different ABIs.

64 bit native libraries

When including native libraries you have to make a decision if you are going to ship with native libraries specifically targeting the 64 bit ABIs. Up until the start of 2015 this was an easy decision as 64 bit devices were only just hitting the market.

Note: It is not mandatory to build 64 bit libraries to support 64 bit Android devices. All 64 bit ABIs are backward compatible to support native libraries built for the equivalent 32 bit ABI. This applies to ARM as well as Intel and MIPS devices.

During 2015 sales of 64 bit Android devices could overtake those of 32 bit devices in some markets. If you provide native libraries specifically for 64 bit devices you should see a small performance increase with these new devices. Intel® have done some tests on Android performance using 64 bit compared to 32 bit and found performance is generally increased by 5-10%[1].

In summary, if you include 64 bit native libraries in you app:

  • You will still need to include 32 bit native libraries
  • You will not support any more devices than if you only include 32 bit libraries
  • You may see a small increase in performance for 64 bit devices[1]
  • You will increase the size of your APK (if building one APK with all native libraries)

Play Store Device support for different ABIs

While I have not been able to source official figures for sales figures of devices running each architecture, the breakdown is pretty clear through casual observation. ARM architectures dominate the Android device marketplace. There is a small percentage of x86 devices and virtually no MIPS devices.

When uploading an app to the Play Store feedback is shown with regards to the number of devices that are able to run your APK. Here is an example of the results:

Native Play Store device support

Figure 5: Play store reported supported devices

Notice in the table shown in Figure 5 that uploading an APK with armeabi native code will target 12 more devices than uploading an APK containing armeabi-v7a native code. The reason for this is that all devices running the armeabi-v7a are also able to execute the armeabi native code.

Multiple APK support on Google Play

It is possible to upload separate APKs to the play store for each ABI you are targeting. The best place to learn about this is by reading the Android documentation about Multiple APK support.
It should be noted that generally using multiple APKs to target different device configurations is discouraged. Here is what the official documentation has to say:

Note: You should generally use multiple APKs to support different device configurations only when your APK is too large (greater than 50MB) due to the alternative resources needed for different device configurations. Using a single APK to support different configurations is always the best practice, because it makes the path for application updates simple and clear for users (and also makes your life simpler by avoiding development and publishing complexity).

Verifying native libraries included in APK

It is possible to visually inspect the contents of any APK; remember that an APK is just a standard zip archive, so opening this archive you should find your native libraries in the folders for the ABIs you are targeting. For example, if you are building an APK that includes the SQLCipher native libraries for the x86 ABI, the file structure within the APK will look like:

lib/x86/
    libdatabase_sqlcipher.so
    libsqlcipher_android.so
    libstlport_shared.so

Uploading multiple APKs to the Developer Console

In the Google Play developer console you can set your app to have multiple APKs that target different devices while all being part of the one application. This means that all download counts and reviews for the different APKs still apply to the one listing in the Google Play store.

There are some rules the must be followed in order to have ABI specific APKs active for the same app. The developer documentation describes some of these rules as:

  • All APKs you publish for the same application must have the same package name and be signed with the same certificate key.
  • Each APK must have a different version code, specified by the android:versionCode attribute.
  • Each APK must not exactly match the configuration support of another APK.

The image below shows the Google Play website for the Android Maps app. Note how the Size for the app is listed as Varies with device. This is as a result of uploading multiple APKs for the one application.

Example of Play Store variation per device

Figure 7: Play store entry for multiple APK app

Setting the versionCode

If you decide you need to deliver multiple APKs for the one app you will need to implement a version scheme for each APK. In a previous blog I have suggested extending a scheme suggested in the Android documentation, as shown below.

versionCode format

Figure 6: Android versionCode suggested format reference

Read more about this version scheme and how to automate the generation of different versionCodes for different target ABIs in this article:

https://androidbycode.wordpress.com/2015/06/30/android-ndk-version-code-scheme-for-publishing-apks-per-architecture/

Uploading your APKs

The Google Play Developer Console allows you to upload multiple APKs for the same application through the Advanced Mode in the APK management section.

  1. Go to the Developer Console
  2. Select the APK management menu
  3. Select Switch to Advanced Mode button
  4. Select Upload new APK button for each APK that targets different ABI

Your developer console should look similar to Figure 8, below, once you have uploaded multiple APKs.

Developer Console multiple native versions

Figure 8: Developer console uploading multiple APKs

You will notice in figure 8 the warning message:

Some devices are eligible to run multiple APKs. In such a scenario, the device will receive the APK with the higher version code.

You may think each version uploaded is targeting a different ABI Native platform, so why is this message being shown? Remember that some ABIs can support code targeting more than one different ABI. In this example the armeabi-v7a devices also are backwards compatible to support armeabi, but the reverse is not true.

Conclusion

The most pertinent points that must be considered when deciding to deploy your app with native libraries are:

  • Don’t use native libraries if you don’t need to
  • Consider which ABIs you need to target
  • Consider if specific 64 bit libraries to maximize performance are worthwhile
  • Avoid producing multiple APKs for one app unless APK size is an issue

References

Multiple APK support in Android: 

http://developer.android.com/google/play/publishing/multiple-apks.html

Android apk-splits user-guide:

http://tools.android.com/tech-docs/new-build-system/user-guide/apk-splits

[1]How to Develop and Evaluate 64-bit Android Apps on Intel® x86 Platforms

https://software.intel.com/en-us/android/articles/how-to-develop-and-evaluate-64-bit-android-apps-on-intel-x86-platforms

1 Comment

Filed under Android NDK, Android Studio, Gradle

Android NDK: Version code scheme for publishing APKs per architecture

Distributing native libraries with your APK adds complexity to your build process and can bloat the size of your APK. You have to include a specific build of native libraries for every architecture you want to target, or to be more technically correct every ABI (Application Binary Interface) you are targeting. Different Android devices use different CPU’s architecture, which have support for different instruction sets. An ABI is different for each combination of CPU and the instruction set supported.

The recommended distribution method for your APK is to distribute one APK that contains all the static native libraries for all the ABIs being targeted. This article does not follow that path and discusses how to build, upload and manage separate APKs for your app. Each APK will include only the native libraries for a specific ABI. The following assumptions are made:

  • You are using Android Studio 1.2+ and using the Android gradle plugin build system
  • You are distributing your app through the Google Play store
  • You already have static native libraries (.so) built with the Android NDK
  • You are concerned about APK bloat caused by these native libraries

Universal APK or Multiple APKs

You need to make an educated and considered decision about if you distribute one APK that includes all native libraries or multiple APKs. The advantage of one universal APK is that it is simple to deploy and maintain your app, the disadvantage is that you potentially could have a substantial increase in APK size. I have previously described how to use SQLCipher in an Android app which discussed this tradeoff. Figure 1, below, shows the impact on size when shipping SQLCipher native libraries in a basic APK.

Figure 1: Size increase when including SQLCipher

Figure 1: Size increase when including SQLCipher

Play Store requirement for unique versionCode per ABI

The only way for a developer to distribute an app, with different APKs, to users on different devices is to have a unique versionCode and to ensure each APK is targeting different devices. The versionCode needs to be unique as it is possible that a users device is compatible with more than one APK, when this happens they will receive the version with the highest versionCode.

When I mention the version I am not referring to the version that is visible to the user, which is specified as the versionName, I am talking about the internal version which is a numerical versionCode that indicates an ascending order of releases. The snippet from a build.grade script below shows the default versionCode and versionName that Android Studio will apply to a new project.

defaultConfig {
 applicationId "com.hookedonplay.androidbycode.mynewapp"
 minSdkVersion 15
 targetSdkVersion 21
 versionCode 1
 versionName "1.0"
}

It is possible that you are maintaining existing apps which only specify the version in the AndroidManifest.xml. While this works well for static version information we need to use the new Android Gradle build system to be able to dynamically generate version information. Here is an example of version information added directly into the AndroidManifest.xml.

<manifest xmlns:android="http://schemas.android.com/apk/res/android"
      package="com.hookedonplay.androidbycode.mynewapp"
      android:versionCode="1"
      android:versionName="1.0">
      ...
</manifest>

Format for individual versionCodes per ABI

Each ABI needs to have an unique versionCode before it can be uploaded to the play store. We are going alter the build.gradle file to automatically generate a unique version number for each ABI. Assuming we want to build our app with versionCode 101, Our aim is to produce the following results:

ABI versionCode table

Figure 2: Example versionCode values for each ABI

The order of the ABI numbers are important as is the length of the 8 digit versionCode.

The order is based on way the Play store works with multiple APKs. Remember, if a users device is compatible with more than one APK for an application the device will receive the version with the highest version code. This is why an armeabi build will be given a prefix of 1 as this build will also be able to run a armeabi-v7a device (2) and in some cases a x86 device (8) due to a proprietary ARM translation layer called libhoudini.

The versionCode we are going to generate is 8 digits long to cater for multiple APKs that differ not only by target ABI, but also different Android SDK versions and screen sizes. An example of how to incorporate all these details into the 8 digit versionCode is shown in Figure 3 below. This is an extension of a numbering scheme suggested in the Android developer documentation where we prefix the number with the ABI id value that was listed in Figure 2.

versionCode format

Figure 3: Android versionCode suggested format reference

You should be aware that this is only a suggested method for generating version information. It is not mandatory and it can be tailored to me your needs. For example, if you know that you are not going to distribute different APKs based on Android SDK API level then you could drop those two digits from the version.

Automating unique versionCode generation

Before continuing and changing your versionCode naming convention you need to understand the following:

If you are distributing one APK containing all flavors of native libraries then you do not need to worry about implementing the following version scheme.

To automate the version process we need to customize the build process. The build.gradle script for your app is where you can configure the build to produce a separate APK for each ABI type and also set the versionCode to conform with the desired 8 digit format discussed earlier.

Here is an example of a complete build.gradle that will achieve this:

apply plugin: 'com.android.application'
 
 android {
    compileSdkVersion 22
    buildToolsVersion "22.0.1"
 
    defaultConfig {
        applicationId "com.hookedonplay.androidbycode.androiddatabaseencryption"
        minSdkVersion 15
        targetSdkVersion 22
        versionCode 101
        versionName "1.0"
    }
 
    splits {
        abi {
            enable true
            reset()
            include 'x86', 'armeabi', 'armeabi-v7a'
            universalApk true
        }
    }
     
    project.ext.versionCodes = ['armeabi': 1, 'armeabi-v7a': 2, 'arm64-v8a': 3, 'mips': 5, 'mips64': 6, 'x86': 8, 'x86_64': 9]
 
    android.applicationVariants.all { variant ->
        variant.outputs.each { output ->
            output.versionCodeOverride =
                project.ext.versionCodes.get(output.getFilter(
                    com.android.build.OutputFile.ABI), 0) * 10000000 + android.defaultConfig.versionCode
        }
    }
 
    buildTypes {
        release {
            minifyEnabled false
            proguardFiles getDefaultProguardFile('proguard-android.txt'), 'proguard-rules.pro'
        }
    }
 }
 
 dependencies {
    compile fileTree(dir: 'libs', include: ['*.jar'])
    compile 'com.android.support:appcompat-v7:22.2.0'
 }

Building separate APK files

The splits mechanism for the Android Gradle Plugin allows you to generate multiple files as shown by the example below. You can include only the ABI types you want to build and it is also possible to build an APK that includes all possible native libraries through the command universalApk true.

splits {
        abi {
            enable true
            reset()
            include 'x86', 'armeabi', 'armeabi-v7a'
            universalApk true
        }
}

Generate versionCode per APK

The section of the build script that sets the versionCode of each APK to a unique number is shown in the code snippet below. Note the project.ext.versionCodes which sets the first digit of the versionCode when multiplied by 10000000 we generate the 8 digit number we require. Finally the initial versionCode of the app, accessed with android.defaultConfig.versionCode, is added to complete the full numerical version for each APK.

   
project.ext.versionCodes = ['armeabi': 1, 'armeabi-v7a': 2, 'arm64-v8a': 3, 'mips': 5, 'mips64': 6, 'x86': 8, 'x86_64': 9]
 
android.applicationVariants.all { variant ->
        variant.outputs.each { output ->
            output.versionCodeOverride =
                project.ext.versionCodes.get(output.getFilter(
                    com.android.build.OutputFile.ABI), 0) * 10000000 + android.defaultConfig.versionCode
    }
}

Verify the versionCode included in each APK

The Android Asset Packaging Tool (aapt) is part of the Android build tools and is required to build APK files. You can also use it to verify the contents of the APK files you have generated.

You will find the aapt executable in your latest version of the build tools, for example ..\sdk\build-tools\22.0.1

Execute the following command on each of the APKs you have produced:

 aapt dump badging AndroidApp.apk

You will find a wealth of information about the APK but the following items in the output will be of most interest:

versionCode='20000101'
...
native-code: 'armeabi-v7a'

Below is a more detailed, but still trimmed sample output from the aapt badging request:

package: name='com.hookedonplay.androidbycode.androiddatabaseencryption' 
versionCode='2000101' versionName='1.0' platformBuildVersionName='5.0.1-1624448'
sdkVersion:'15'
targetSdkVersion:'21'
application-label:'Android Database Encryption'
application-icon-160:'res/mipmap-mdpi-v4/ic_launcher.png'
application-icon-240:'res/mipmap-hdpi-v4/ic_launcher.png'
application-icon-320:'res/mipmap-xhdpi-v4/ic_launcher.png'
application-icon-480:'res/mipmap-xxhdpi-v4/ic_launcher.png'
application-icon-640:'res/mipmap-xxxhdpi-v4/ic_launcher.png'
application: label='Android Database Encryption' icon='res/mipmap-mdpi-v4/ic_launcher.png'
application-debuggable
uses-feature: name='android.hardware.touchscreen'
uses-implied-feature: name='android.hardware.touchscreen' reason='default feature for all apps'
main
supports-screens: 'small' 'normal' 'large' 'xlarge'
supports-any-density: 'true'
locales: '--_--' 'ca' 'da' 'fa' 'ja' 'nb' 'de' 'af' 'bg' 'th' 'fi' 'hi' 'vi' 'sk' 'uk' 'el' 'nl' 'pl' 'sl' 'tl' 'am' 'in' 'ko' 'ro' 'ar' 'fr' 'hr' 'sr' 'tr' 'cs' 'es' 'it' 'lt' 'pt' 'hu' 'ru' 'zu' 'lv' 'sv' 'iw' 'sw' 'fr-CA' 'lo-LA' 'en-GB' 'bn-BD' 'et-EE' 'ka-GE' 'ky-KG' 'km-KH' 'zh-HK' 'si-LK' 'mk-MK' 'ur-PK' 'hy-AM' 'my-MM' 'zh-CN' 'ta-IN' 'te-IN' 'ml-IN' 'en-IN' 'kn-IN' 'mr-IN' 'mn-MN' 'ne-NP' 'gl-ES' 'eu-ES' 'is-IS' 'es-US' 'pt-PT' 'zh-TW' 'ms-MY' 'kk-KZ' 'uz-UZ'
densities: '160' '240' '320' '480' '640'
native-code: 'armeabi-v7a'

This same dump can be also used to verify other attributes of your APK, such as the build SDK version, locales and supported screen densities.

You will also be able to verify the targeted ABI and versionCode once you upload to the Play Store via the developer console.

Distribution of multiple APKs via alternative app stores

This article has assumed that you are distributing your app through the Google Play Store. It may be possible that you are using multiple distribution methods, such as promoting your app on the Amazon Appstore. According to the documentation the same version methodology will work for the multiple APK support within the Amazon Appstore. There are also a number of smaller app marketplaces that may automatically add your app. Each of these app store may or may not handle multiple APKs correctly which may result in some users downloading an APK that is incompatible with their device.

References

Multiple APK support in Android

http://developer.android.com/google/play/publishing/multiple-apks.html

Android apk-splits user-guide

http://tools.android.com/tech-docs/new-build-system/user-guide/apk-splits

Amazon Device Targeting

https://developer.amazon.com/public/solutions/devices/fire-tv/docs/using-device-targeting

4 Comments

Filed under Android NDK, Android Studio, Automation, Google Play, Gradle