Android NDK Overview

Introduction:

The Android NDK is a set of tools that allows Android application developers
to embed native machine code compiled from C and/or C++ source files into
their application packages.

IMPORTANT:
  The Android NDK can only be used to target Android system images
  running Cupcake (a.k.a 1.5) or later versions of the platform.

  1.0 and 1.1 system images are specifically *not* supported due to
  subtle ABI and toolchain changes that happened for the 1.5 release.


I. Android NDK Goals:
---------------------

The Android VM allows your application's source code to call methods
implemented in native code through the JNI. In a nutshell, this means that:

  - Your application's source code will declare one or more methods
    with the 'native' keyword to indicate that they are implemented through
    native code. E.g.:

      native byte[]  loadFile(String  filePath);

  - You must provide a native shared library that contains the
    implementation of these methods, which will be packaged into your
    application's .apk. This library must be named according to standard
    Unix conventions as lib<something>.so, and shall contain a standard JNI
    entry point (more on this later). For example:

      libFileLoader.so

  - Your application must explicitly load the library. For example, to load
    it at application start-up, simply add the following to its source code:

      static {
        System.loadLibrary("FileLoader");
      }

    Note that you should not use the 'lib' prefix and '.so' suffix here.


The Android NDK is a complement to the Android SDK that helps you to:

  - Generate JNI-compatible shared libraries that can run on the Android
    1.5 platform (and later) running on ARM CPUs.

  - Copy the generated shared libraries to a proper location of your
    application project path, so they will be automatically added to your
    final (and signed) .apks

  - In later revisions of the NDK, we intend to provide tools that help
    debug your native code through a remote gdb connection and as much
    source/symbol information as possible.

Moreover, the Android NDK provides:

  - A set of cross-toolchains (compilers, linkers, etc..) that can
    generate native ARM binaries on Linux, OS X and Windows (with Cygwin)

  - A set of system headers corresponding to the list of stable native APIs
    supported by the Android platform. This corresponds to definitions that
    are guaranteed to be supported in all later releases of the platform.

    They are documented in the file docs/STABLE-APIS.html

    IMPORTANT:
    Keep in mind that most of the native system libraries in Android system
    images are not frozen and might changed drastically, or even deleted,
    in later updates and releases of the platform.

  - A build system that allow developers to only write very short build files
    to describe which sources need to be compiled, and how. The build system
    deals with all the hairy toolchain/platform/CPU/ABI specifics. Moreover,
    later updates of the NDK can add support for more toolchains, platforms,
    system interfaces without requiring changes in the developer's build
    files (more on this later).


II. Android NDK Non-Goals:
--------------------------

The NDK is *not* a good way to write generic native code that runs on Android
devices. In particular, your applications should still be written in the Java
programming language, handle Android system events appropriately to avoid the
"Application Not Responding" dialog or deal with the Android application
life-cycle.

Note however that is is possible to write a sophisticated application in
native code with a small "application wrapper" used to start/stop it
appropriately.

A good understanding of JNI is highly recommended, since many operations
in this environment require specific actions from the developers, that are
not necessarily common in typical native code. These include:

  - Not being able to directly access the content of VM objects through
    direct native pointers. E.g. you cannot safely get a pointer to a
    String object's 16-bit char array to iterate over it in a loop.

  - Requiring explicit reference management when the native code wants to
    keep handles to VM objects between JNI calls.


The NDK only provides system headers for a very limited set of native
APIs and libraries supported by the Android platform. While a typical
Android system image includes many native shared libraries, these should
be considered an implementation detail that might change drastically between
updates and releases of the platform.

If an Android system library is not explicitly supported by the NDK
headers, then applications should not depend on it being available, or
they risk breaking after the next over-the-air system update on various
devices.

Selected system libraries will gradually be added to the set of stable NDK
APIs.


III. NDK development in practice:
---------------------------------

Here's a very rough overview of how you can develop native code with the
Android NDK:

  1/ Place your native sources under $PROJECT/jni/...

  2/ Write $PROJECT/jni/Android.mk to describe your sources
     to the NDK build system

  3/ Optional: write $PROJECT/jni/Application.mk to describe your
     project in more details to the build system. You don't need
     one to get started though, but this allows you to target
     more than one CPU or override compiler/linker flags
     (see docs/APPLICATION-MK.html for all details).

  4/ Build your native code by running "$NDK/ndk-build" from your
     project directory, or any of its sub-directories.

The last step will copy, in case of success, the stripped shared libraries
your application needs to your application's root project directory. You
will then need to generate your final .apk through the usual means.

Now, for a few more details:


III.1/ Configuring the NDK:
- - - - - - - - - - - - - -

Previous releases required that you run the 'build/host-setup.sh'
script to configure your NDK. This step has been removed completely
in release 4 (a.k.a. NDK r4).


III.2/ Placing C and C++ sources:
- - - - - - - - - - - - - - - - -

Place your native sources under the following directory:

    $PROJECT/jni/

Where $PROJECT corresponds to the path of your Android application
project.

You are pretty free to organize the content of 'jni' as you want,
the directory names and structure here will not influence the final
generated application packages, so you don't have to use pseudo-unique
names like com.<mycompany>.<myproject> as is the case for application
package names.

Note that C and C++ sources are supported. The default C++ file extensions
supported by the NDK is '.cpp', but other extensions can be handled as well
(see docs/ANDROID-MK.html for details).

It is possible to store your sources in a different location by adjusting
your Android.mk file (see below).


III.3/ Writing an Android.mk build script:
- - - - - - - - - - - - - - - - - - - - - -

An Android.mk file is a small build script that you write to describe your
sources to the NDK build system. Its syntax is described in details in
the file docs/ANDROID-MK.html.

In a nutshell, the NDK groups your sources into "modules", where each module
can be one of the following:

  - a static library
  - a shared library

You can define several modules in a single Android.mk, or you can write
several Android.mk files, each one defining a single module.

Note that a single Android.mk might be parsed several times by the build
system so don't assume that certain variables are not defined in them.
By default, the NDK will look for the following build script:

   $PROJECT/jni/Android.mk

If you want to define Android.mk files in sub-directories, you should
include them explicitly in your top-level Android.mk. There is even
a helper function to do that, i.e. use:

   include $(call all-subdir-makefiles)

This will include all Android.mk files in sub-directories of the current
build file's path.


III.4/ Writing an Application.mk build file (optional):
- - - - - - - - - - - - - - - - - - - - - - - - - - - -

While an Android.mk file describes your modules to the build system, the
Application.mk file describes your application itself. See the
docs/APPLICATION-MK.html document to understand what this file allows you
to do. This includes, among others:

   - The exact list of modules required by your application.

   - The CPU architecture(s) to generate machine code for.

  - Optional information, like whether you want a release or debug
    build, specific C or C++ compiler flags and others that should
    apply to all modules being built.

This file is optional: by default the NDK will provide one that simply
builds *all* the modules listed from your Android.mk (and all the makefiles
it includes) and target the default CPU ABI (armeabi).

There are two ways to use an Application.mk:

  - Place it under $PROJECT/jni/Application.mk, and it will be picked
    up automatically by the 'ndk-build' script (more on this later)

  - Place it under $NDK/apps/<name>/Application.mk, where $NDK
    points to your NDK installation path. After that, launch
    "make APP=<name>" from the NDK directory.

    This was the way this file was used before Android NDK r4.
    It is still supported for compatibility reasons, but we strongly
    encourage you to use the first method instead, since it is much
    simpler and doesn't need modifying / changing directories of the
    NDK installation tree.

Again, see docs/APPLICATION-MK.html for a complete description of its
content.


III.5/ Invoke the NDK build system:
- - - - - - - - - - - - - - - - - -

The preferred way to build machine code with the NDK is to use the
'ndk-build' script introduced with Android NDK r4. You can also use
a second, legacy, method that depends on creating a '$NDK/apps' subdirectory.

In both cases, a successful build will copy the final stripped binary modules
(i.e. shared libraries) required by your application to your application's
project path (Note that unstripped versions are kept for debugging
purposes, there is no need to copy unstripped binaries to a device).


  1: Using the 'ndk-build' command:
  ---------------------------------

  The 'ndk-build' script, located at the top of the NDK installation path
  can be invoked directly from your application project directory (i.e. the
  one where your AndroidManifest.xml is located) or any of its sub-directories.
  For example:

    cd $PROJECT
    $NDK/ndk-build

  This will launch the NDK build scripts, which will automatically probe your
  development system and application project file to determine what to build.

  For example:

    ndk-build
    ndk-build  clean    --> clean generated binaries
    ndk-build  -B V=1   --> force complete rebuild, showing commands

  By default, it expects an optional file under $PROJECT/jni/Application.mk,
  and a required $PROJECT/jni/Android.mk.

  On success, this will copy the generated binary modules (i.e. shared
  libraries) to the appropriate location in your project tree. You can later
  rebuild the full Android application package either through the usual
  'ant' command, or the ADT Eclipse plug-in.

  See docs/NDK-BUILD.html for a more complete description of what this script
  does and which options it can take.


  2: Using the $NDK/apps/<name>/Application.mk:
  ---------------------------------------------

  This build method was the only one before Android NDK r4 and is only
  supported for compatibility reason. We strongly recommend you to migrate
  to using the 'ndk-build' command as soon as possible, since we may remove
  legacy support in a later NDK release.

  It requires the following:

    1. Creating a sub-directory named $NDK/apps/<name>/ under
       your NDK installation directory (not your project path).

       Where <name> is an arbitrary name to describe your application
       to the NDK build system (no spaces allowed).

    2. Write $NDK/apps/<name>/Application.mk, which then requires
       a definition for APP_PROJECT_PATH that points to your
       application project directory.

    3. Go to the NDK installation path on the command line then
       invoke the top-level GNUMakefile, as in:

         cd $NDK
         make APP=<name>

  The result will be equivalent to the first method, except for the fact
  that intermediate generated files will be placed under $NDK/out/apps/<name>/


IV. Rebuild your application package:
- - - - - - - - - - - - - - - - - - -

After generating the binaries with the NDK, you need to rebuild your
Android application package files (.apk) using the normal means, i.e.
either using the 'ant' command or the ADT Eclipse plug-in.

See the Android SDK documentation for more details. The new .apk will
embed your shared libraries, and they will be extracted automatically
at installation time by the system when you install the package on a
target device.


V. Debugging support:
- - - - - - - - - - -

The NDK provides a helper script, named 'ndk-gdb' to very easily launch
a native debugging session of your applications.

Native debugging can *ONLY* be performed on production devices running
Android 2.2 or higher, and does not require root or privileged access, as
long as your application is debuggable.

For more information, read docs/NDK-GDB.html. In a nutshell, native debugging
follows this simple scheme:

   1. Ensure your application is debuggable (e.g. set android:debuggable
      to "true" in your AndroidManifest.xml)

   2. Build your application with 'ndk-build', then install it on your
      device/emulator.

   3. Launch your application.

   4. Run 'ndk-gdb' from your application project directory.

You will get a gdb prompt. See the GDB User Manual for a list of useful
commands.