$ export JAVA_HOME=/usr/local/java # or wherever you have the JDK installed
$ make
$ build/linux-i386/avian -cp build/linux-i386/test Hello
$ export JAVA_HOME=/Library/Java/Home
$ make
$ build/darwin-i386/avian -cp build/darwin-i386/test Hello
$ git clone [email protected]:ReadyTalk/win32.git ../win32
$ export JAVA_HOME="C:/Program Files/Java/jdk1.6.0_07"
$ make
$ build/windows-i386/avian -cp build/windows-i386/test Hello
$ git clone [email protected]:ReadyTalk/win32.git ../win32
$ export JAVA_HOME="/cygdrive/c/Program Files/Java/jdk1.6.0_07"
$ make
$ build/windows-i386/avian -cp build/windows-i386/test Hello
$ export JAVA_HOME=/usr/local/openjdk7 # or wherever you have the JDK installed
$ gmake
$ build/freebsd-x86_64/avian -cp build/freebsd-x86_64/test Hello
Adjust JAVA_HOME according to your system, but be sure to use forward slashes in the path.
Avian is a lightweight virtual machine and class library designed to provide a useful subset of Java's features, suitable for building self-contained applications. More information is available at the project web site.
If you have any trouble building, running, or embedding Avian, please post a message to our discussion group.
That's also the place for any other questions, comments, or suggestions you might have.
Avian can currently target the following platforms:
- Linux (i386, x86_64, ARM, and 32-bit PowerPC)
- Windows (i386 and x86_64)
- Mac OS X (i386 and x86_64)
- Apple iOS (i386 and ARM)
- FreeBSD (i386, x86_64)
Build requirements include:
- GNU make 3.80 or later
- GCC 3.4 or later (4.5.1 or later for Windows/x86_64) or LLVM Clang 3.1 or later (see use-clang option below)
- JDK 1.5 or later
- MinGW 3.4 or later (only if compiling for Windows)
- zlib 1.2.3 or later
Earlier versions of some of these packages may also work but have not been tested.
The build is directed by a single makefile and may be influenced via certain flags described below, all of which are optional.
$ make \
platform={linux,windows,darwin,freebsd} \
arch={i386,x86_64,powerpc,arm} \
process={compile,interpret} \
mode={debug,debug-fast,fast,small} \
lzma=<lzma source directory> \
ios={true,false} \
bootimage={true,false} \
heapdump={true,false} \
tails={true,false} \
continuations={true,false} \
use-clang={true,false} \
openjdk=<openjdk installation directory> \
openjdk-src=<openjdk source directory> \
android=<android source directory>
-
platform
- the target platform- default: output of $(uname -s | tr [:upper:] [:lower:]), normalized in some cases (e.g. CYGWIN_NT-5.1 -> windows)
-
arch
- the target architecture- default: output of $(uname -m), normalized in some cases (e.g. i686 -> i386)
-
process
- choice between pure interpreter or JIT compiler- default: compile
-
mode
- which set of compilation flags to use to determine optimization level, debug symbols, and whether to enable assertions- default: fast
-
lzma
- if set, support use of LZMA to compress embedded JARs and boot images. The value of this option should be a directory containing a recent LZMA SDK (available here). Currently, only version 9.20 of the SDK has been tested, but other versions might work.- default: not set
-
ios
- if true, cross-compile for iOS on OS X. Note that non-jailbroken iOS devices do not allow JIT compilation, so only process=interpret or bootimage=true builds will run on such devices. See here for an example of an Xcode project for iOS which uses Avian.- default: false
-
bootimage
- if true, create a boot image containing the pre-parsed class library and ahead-of-time compiled methods. This option is only valid for process=compile builds. Note that you may need to specify both build-arch=x86_64 and arch=x86_64 on 64-bit systems where "uname -m" prints "i386".- default: false
-
heapdump
- if true, implement avian.Machine.dumpHeap(String), which, when called, will generate a snapshot of the heap in a simple, ad-hoc format for memory profiling purposes. See heapdump.cpp for details.- default: false
-
tails
- if true, optimize each tail call by replacing the caller's stack frame with the callee's. This convention ensures proper tail recursion, suitable for languages such as Scheme. This option is only valid for process=compile builds.- default: false
-
continuations
- if true, support continuations via the avian.Continuations methods callWithCurrentContinuation and dynamicWind. See Continuations.java for details. This option is only valid for process=compile builds.- default: false
-
use-clang
- if true, use LLVM's clang instead of GCC to build. Note that this does not currently affect cross compiles, only native builds.- default: false
-
openjdk
- if set, use the OpenJDK class library instead of the default Avian class library. See "Building with the OpenJDK Class Library" below for details.- default: not set
-
openjdk-src
- if this and the openjdk option above are both set, build an embeddable VM using the OpenJDK class library. The JNI components of the OpenJDK class library will be built from the sources found under the specified directory. See "Building with the OpenJDK Class Library" below for details.- default: not set
-
android
- if set, use the Android class library instead of the default Avian class library. See "Building with the Android Class Library" below for details.- default: not set
These flags determine the name of the directory used for the build. The name always starts with ${platform}-${arch}, and each non-default build option is appended to the name. For example, a debug build with bootimage enabled on Linux/i386 would be built in build/linux-i386-debug-bootimage. This allows you to build with several different sets of options independently and even simultaneously without doing a clean build each time.
If you are compiling for Windows, you may either cross-compile using MinGW or build natively on Windows under MSYS or Cygwin.
1. Download and install the current MinGW and MSYS packages from mingw.org, selecting the C and C++ compilers when prompted. Use the post-install script to create the filesystem link to the compiler.
2. Download GNU Make 3.81 from the MSYS download page (make-3.81-MSYS-1.0.11-2.tar.bz2) and extract the tar file into e.g. c:/msys/1.0.
1. Download and run setup.exe from cygwin's website, installing the base system and these packages: make, gcc-mingw-g++, mingw64-i686-gcc-g++, mingw64-x86_64-gcc-g++, and (optionally) git.
You may also find our win32 repository useful: (run this from the directory containing the avian directory)
$ git clone [email protected]:ReadyTalk/win32.git
This gives you the Windows JNI headers, zlib headers and library, and a few other useful libraries like OpenSSL, libjpeg, and libpng. There's also a win64 repository for 64-bit builds:
$ git clone [email protected]:ReadyTalk/win64.git
You can also build using the MSVC compiler, which makes debugging with tools like WinDbg and Visual Studio much easier. Note that you will still need to have GCC installed - MSVC is only used to compile the C++ portions of the VM, while the assembly code and helper tools are built using GCC.
The MSVC build has been tested with Visual Studio Express Edition versions 8, 9, and 10. Other versions may also work.
To build with MSVC, install Cygwin as described above and set the following environment variables:
$ export PATH="/usr/local/bin:/usr/bin:/bin:/usr/X11R6/bin:/cygdrive/c/Program Files/Microsoft Visual Studio 9.0/Common7/IDE:/cygdrive/c/Program Files/Microsoft Visual Studio 9.0/VC/BIN:/cygdrive/c/Program Files/Microsoft Visual Studio 9.0/Common7/Tools:/cygdrive/c/WINDOWS/Microsoft.NET/Framework/v3.5:/cygdrive/c/WINDOWS/Microsoft.NET/Framework/v2.0.50727:/cygdrive/c/Program Files/Microsoft Visual Studio 9.0/VC/VCPackages:/cygdrive/c/Program Files/Microsoft SDKs/Windows/v6.0A/bin:/cygdrive/c/WINDOWS/system32:/cygdrive/c/WINDOWS:/cygdrive/c/WINDOWS/System32/Wbem"
$ export LIBPATH="C:\WINDOWS\Microsoft.NET\Framework\v3.5;C:\WINDOWS\Microsoft.NET\Framework\v2.0.50727;C:\Program Files\Microsoft Visual Studio 9.0\VC\LIB;"
$ export VCINSTALLDIR="C:\Program Files\Microsoft Visual Studio 9.0\VC"
$ export LIB="C:\Program Files\Microsoft Visual Studio 9.0\VC\LIB;C:\Program Files\Microsoft SDKs\Windows\v6.0A\lib;"
$ export INCLUDE="C:\Program Files\Microsoft Visual Studio 9.0\VC\INCLUDE;C:\Program Files\Microsoft SDKs\Windows\v6.0A\include;"
Adjust these definitions as necessary according to your MSVC installation.
Finally, build with the msvc flag set to the MSVC tool directory:
$ make msvc="/cygdrive/c/Program Files/Microsoft Visual Studio 9.0/VC"
By default, Avian uses its own lightweight class library. However, that library only contains a relatively small subset of the classes and methods included in the JRE. If your application requires features beyond that subset, you may want to tell Avian to use OpenJDK's class library instead. To do so, specify the directory where OpenJDK is installed, e.g.:
$ make openjdk=/usr/lib/jvm/java-7-openjdk
This will build Avian as a conventional JVM (e.g. libjvm.so) which loads its boot class library and native libraries (e.g. libjava.so) from /usr/lib/jvm/java-7-openjdk/jre at runtime. Note that you must use an absolute path here, or else the result will not work when run from other directories. In this configuration, OpenJDK needs to remain installed for Avian to work, and you can run applications like this:
$ build/linux-x86_64-openjdk/avian-dynamic -cp /path/to/my/application \
com.example.MyApplication
Alternatively, you can enable a stand-alone build using OpenJDK by specifying the location of the OpenJDK source code, e.g.:
$ make openjdk=$(pwd)/../jdk7/build/linux-amd64/j2sdk-image \
openjdk-src=$(pwd)/../jdk7/jdk/src
You must ensure that the path specified for openjdk-src does not have any spaces in it; make gets confused when dependency paths include spaces, and we haven't found away around that except to avoid paths with spaces entirely.
The result of such a build is a self-contained binary which does not depend on external libraries, jars, or other files. In this case, the specified paths are used only at build time; anything needed at runtime is embedded in the binary. Thus, the process of running an application is simplified:
$ build/linux-x86_64-openjdk-src/avian -cp /path/to/my/application \
com.example.MyApplication
Note that the resulting binary will be very large due to the size of OpenJDK's class library. This can be mitigated using UPX, preferably an LZMA-enabled version:
$ upx --lzma --best build/linux-x86_64-openjdk-src/avian
You can reduce the size futher for embedded builds by using ProGuard and the supplied openjdk.pro configuration file (see "Embedding with ProGuard and a Boot Image" below). Note that you'll still need to use vm.pro in that case -- openjdk.pro just adds additional constraints specific to the OpenJDK port. Also see app.mk in git:https://oss.readytalk.com/avian-swt-examples.git for an example of using Avian, OpenJDK, ProGuard, and UPX in concert.
Here are some examples of how to install OpenJDK and build Avian with it on various OSes:
Conventional build:
$ apt-get install openjdk-7-jdk
$ make openjdk=/usr/lib/jvm/java-7-openjdk test
Stand-alone build:
$ apt-get install openjdk-7-jdk
$ apt-get source openjdk-7-jdk
$ apt-get build-dep openjdk-7-jdk
$ (cd openjdk-7-7~b147-2.0 && dpkg-buildpackage)
$ make openjdk=/usr/lib/jvm/java-7-openjdk \
openjdk-src=$(pwd)/openjdk-7-7~b147-2.0/build/openjdk/jdk/src \
test
####Mac OS X: Prerequisite: Build OpenJDK 7 according to this site.
Conventional build:
$ make openjdk=$(pwd)/../jdk7u-dev/build/macosx-amd64/j2sdk-image test
Stand-alone build:
$ make openjdk=$(pwd)/../jdk7u-dev/build/macosx-amd64/j2sdk-image \
openjdk-src=$(pwd)/../p/jdk7u-dev/jdk/src test
####Windows (Cygwin): Prerequisite: Build OpenJDK 7 according to this site. Alternatively, use https://github.com/alexkasko/openjdk-unofficial-builds.
Conventional build:
$ make openjdk=$(pwd)/../jdk7u-dev/build/windows-i586/j2sdk-image test
Stand-alone build:
$ make openjdk=$(pwd)/../jdk7u-dev/build/windows-i586/j2sdk-image \
openjdk-src=$(pwd)/../p/jdk7u-dev/jdk/src test
Currently, only OpenJDK 7 is supported. Later versions might work, but have not yet been tested.
As an alternative to both the Avian and OpenJDK class libaries, you can also build with the Android class library on some platforms (currently Linux works and OS X mostly works). To build this way, do the following, starting from the Avian directory:
cd ..
mkdir -p android/system android/external
cd android
git clone https://android.googlesource.com/platform/bionic
(cd bionic && \
git checkout 84983592ade3ec7d72d082262fb6646849979bfc)
git clone https://android.googlesource.com/platform/system/core \
system/core
(cd system/core && \
git checkout fafcabd0dd4432de3c7f5956edec23f6ed241b56)
git clone https://android.googlesource.com/platform/external/fdlibm \
external/fdlibm
(cd external/fdlibm && \
git checkout 0da5f683c9ddc9442af3b389b4220e91ccffb320)
git clone https://android.googlesource.com/platform/external/icu4c \
external/icu4c
(cd external/icu4c && \
git checkout 8fd45e08f1054d80a356ef8aa05659a2ba84707c)
git clone https://android.googlesource.com/platform/libnativehelper
(cd libnativehelper && \
git checkout cf5ac0ec696fce7fac6b324ec7d4d6da217e501c)
git clone https://android.googlesource.com/platform/external/openssl \
external/openssl
(cd external/openssl && \
git checkout 7b972f1aa23172c4430ada7f3236fa1fd9b31756)
git clone https://android.googlesource.com/platform/external/zlib \
external/zlib
(cd external/zlib && \
git checkout 15b6223aa57a347ce113729253802cb2fdeb4ad0)
git clone git:https://git.openssl.org/openssl.git openssl-upstream
(cd openssl-upstream && \
git checkout OpenSSL_1_0_1e)
git clone https://github.com/dicej/android-libcore64 libcore
curl -Of https://readytalk.github.io/avian/expat-2.1.0.tar.gz
(cd external && tar xzf ../expat-2.1.0.tar.gz && mv expat-2.1.0 expat)
(cd external/expat && CFLAGS=-fPIC CXXFLAGS=-fPIC ./configure \
--enable-static && make)
(cd external/fdlibm && (mv makefile.in Makefile.in || true) \
&& CFLAGS=-fPIC bash configure && make)
(cd external/icu4c && CFLAGS=-fPIC CXXFLAGS=-fPIC ./configure \
--enable-static && make)
NB: use 'CC="gcc -fPIC" ./Configure darwin64-x86_64-cc' when building for x86_64 OS X instead of 'CC="gcc -fPIC" ./config':
(cd openssl-upstream \
&& (for x in \
progs \
handshake_cutthrough \
jsse \
channelid \
eng_dyn_dirs \
fix_clang_build \
tls12_digests \
alpn; \
do patch -p1 < ../external/openssl/patches/$x.patch; done) \
&& CC="gcc -fPIC" ./config && make)
cd ../avian
make android=$(pwd)/../android test
Note that we use https://github.com/dicej/android-libcore64 above instead of the upstream https://android.googlesource.com/platform/libcore repository, since the former has patches to provide better support for non-Linux platforms.
Also note that we use the upstream OpenSSL repository and apply the Android patches to it. This is because it is not clear how to build the Android fork of OpenSSL directly without checking out and building the entire platform. As of this writing, the patches apply cleanly against OpenSSL 1.0.1e, so that's the tag we check out, but this may change in the future when the Android fork rebases against a new OpenSSL version.
Finally, we specify specific commit hashes for each repository which are known to work. Later versions may also work, but have not been tested.
Installing Avian is as simple as copying the executable to the desired directory:
$ cp build/${platform}-${arch}/avian ~/bin/
The following series of commands illustrates how to produce a stand-alone executable out of a Java application using Avian.
Note: if you are building on Cygwin, prepend "x86_64-w64-mingw32-" or "i686-w64-mingw32-" to the ar, g++, gcc, strip, and dlltool commands below (e.g. x86_64-w64-mingw32-gcc).
1. Build Avian, create a new directory, and populate it with the VM object files and bootstrap classpath jar.
$ make
$ mkdir hello
$ cd hello
$ ar x ../build/${platform}-${arch}/libavian.a
$ cp ../build/${platform}-${arch}/classpath.jar boot.jar
2. Build the Java code and add it to the jar.
$ cat >Hello.java <<EOF
public class Hello {
public static void main(String[] args) {
System.out.println("hello, world!");
}
}
EOF
$ javac -bootclasspath boot.jar Hello.java
$ jar u0f boot.jar Hello.class
3. Make an object file out of the jar.
$ ../build/${platform}-${arch}/binaryToObject/binaryToObject boot.jar \
boot-jar.o _binary_boot_jar_start _binary_boot_jar_end ${platform} ${arch}
If you've built Avian using the lzma
option, you may optionally
compress the jar before generating the object:
../build/$(platform}-${arch}-lzma/lzma/lzma encode boot.jar boot.jar.lzma
&& ../build/${platform}-${arch}-lzma/binaryToObject/binaryToObject \
boot.jar.lzma boot-jar.o _binary_boot_jar_start _binary_boot_jar_end \
${platform} ${arch}
Note that you'll need to specify "-Xbootclasspath:[lzma:bootJar]" instead of "-Xbootclasspath:[bootJar]" in the next step if you've used LZMA to compress the jar.
4. Write a driver which starts the VM and runs the desired main method. Note the bootJar function, which will be called by the VM to get a handle to the embedded jar. We tell the VM about this jar by setting the boot classpath to "[bootJar]".
$ cat >embedded-jar-main.cpp <<EOF
#include "stdint.h"
#include "jni.h"
#include "stdlib.h"
#if (defined __MINGW32__) || (defined _MSC_VER)
# define EXPORT __declspec(dllexport)
#else
# define EXPORT __attribute__ ((visibility("default"))) \
__attribute__ ((used))
#endif
#if (! defined __x86_64__) && ((defined __MINGW32__) || (defined _MSC_VER))
# define SYMBOL(x) binary_boot_jar_##x
#else
# define SYMBOL(x) _binary_boot_jar_##x
#endif
extern "C" {
extern const uint8_t SYMBOL(start)[];
extern const uint8_t SYMBOL(end)[];
EXPORT const uint8_t*
bootJar(unsigned* size)
{
*size = SYMBOL(end) - SYMBOL(start);
return SYMBOL(start);
}
} // extern "C"
extern "C" void __cxa_pure_virtual(void) { abort(); }
int
main(int ac, const char** av)
{
JavaVMInitArgs vmArgs;
vmArgs.version = JNI_VERSION_1_2;
vmArgs.nOptions = 1;
vmArgs.ignoreUnrecognized = JNI_TRUE;
JavaVMOption options[vmArgs.nOptions];
vmArgs.options = options;
options[0].optionString = const_cast<char*>("-Xbootclasspath:[bootJar]");
JavaVM* vm;
void* env;
JNI_CreateJavaVM(&vm, &env, &vmArgs);
JNIEnv* e = static_cast<JNIEnv*>(env);
jclass c = e->FindClass("Hello");
if (not e->ExceptionCheck()) {
jmethodID m = e->GetStaticMethodID(c, "main", "([Ljava/lang/String;)V");
if (not e->ExceptionCheck()) {
jclass stringClass = e->FindClass("java/lang/String");
if (not e->ExceptionCheck()) {
jobjectArray a = e->NewObjectArray(ac-1, stringClass, 0);
if (not e->ExceptionCheck()) {
for (int i = 1; i < ac; ++i) {
e->SetObjectArrayElement(a, i-1, e->NewStringUTF(av[i]));
}
e->CallStaticVoidMethod(c, m, a);
}
}
}
}
int exitCode = 0;
if (e->ExceptionCheck()) {
exitCode = -1;
e->ExceptionDescribe();
}
vm->DestroyJavaVM();
return exitCode;
}
EOF
on Linux:
$ g++ -I$JAVA_HOME/include -I$JAVA_HOME/include/linux \
-D_JNI_IMPLEMENTATION_ -c embedded-jar-main.cpp -o main.o
on Mac OS X:
$ g++ -I$JAVA_HOME/include -D_JNI_IMPLEMENTATION_ -c embedded-jar-main.cpp \
-o main.o
on Windows:
$ g++ -fno-exceptions -fno-rtti -I"$JAVA_HOME/include" -I"$JAVA_HOME/include/win32" \
-D_JNI_IMPLEMENTATION_ -c embedded-jar-main.cpp -o main.o
5. Link the objects produced above to produce the final executable, and optionally strip its symbols.
on Linux:
$ g++ -rdynamic *.o -ldl -lpthread -lz -o hello
$ strip --strip-all hello
on Mac OS X:
$ g++ -rdynamic *.o -ldl -lpthread -lz -o hello -framework CoreFoundation
$ strip -S -x hello
on Windows:
$ dlltool -z hello.def *.o
$ dlltool -d hello.def -e hello.exp
$ gcc hello.exp *.o -L../../win32/lib -lmingwthrd -lm -lz -lws2_32 \
-mwindows -mconsole -o hello.exe
$ strip --strip-all hello.exe
The following illustrates how to embed an application as above, except this time we preprocess the code using ProGuard and build a boot image from it for quicker startup. The pros and cons of using ProGuard are as follow:
-
Pros: ProGuard will eliminate unused code, optimize the rest, and obfuscate it as well for maximum space savings
-
Cons: increased build time, especially for large applications, and extra effort needed to configure it for applications which rely heavily on reflection and/or calls to Java from native code
For boot image builds:
-
Pros: the boot image build pre-parses all the classes and compiles all the methods, obviating the need for JIT compilation at runtime. This also makes garbage collection faster, since the pre-parsed classes are never visited.
-
Cons: the pre-parsed classes and AOT-compiled methods take up more space in the executable than the equivalent class files. In practice, this can make the executable 30-50% larger. Also, AOT compilation does not yet yield significantly faster or smaller code than JIT compilation. Finally, floating point code may be slower on 32-bit x86 since the compiler cannot assume SSE2 support will be available at runtime, and the x87 FPU is not supported except via out-of-line helper functions.
Note you can use ProGuard without using a boot image and vice-versa, as desired.
The following instructions assume we are building for Linux/i386. Please refer to the previous example for guidance on other platforms.
1. Build Avian, create a new directory, and populate it with the VM object files.
$ make bootimage=true
$ mkdir hello
$ cd hello
$ ar x ../build/linux-i386-bootimage/libavian.a
2. Create a stage1 directory and extract the contents of the class library jar into it.
$ mkdir stage1
$ (cd stage1 && jar xf ../../build/linux-i386-bootimage/classpath.jar)
3. Build the Java code and add it to stage1.
$ cat >Hello.java <<EOF
public class Hello {
public static void main(String[] args) {
System.out.println("hello, world!");
}
}
EOF
$ javac -bootclasspath stage1 -d stage1 Hello.java
4. Create a ProGuard configuration file specifying Hello.main as the entry point.
$ cat >hello.pro <<EOF
-keep class Hello {
public static void main(java.lang.String[]);
}
EOF
5. Run ProGuard with stage1 as input and stage2 as output.
$ java -jar ../../proguard4.6/lib/proguard.jar \
-dontusemixedcaseclassnames -injars stage1 -outjars stage2 \
@../vm.pro @hello.pro
(note: The -dontusemixedcaseclassnames option is only needed when building on systems with case-insensitive filesystems such as Windows and OS X. Also, you'll need to add -ignorewarnings if you use the OpenJDK class library since the openjdk-src build does not include all the JARs from OpenJDK, and thus ProGuard will not be able to resolve all referenced classes. If you actually plan to use such classes at runtime, you'll need to add them to stage1 before running ProGuard. Finally, you'll need to add @../openjdk.pro to the above command when using the OpenJDK library.)
6. Build the boot and code images.
$ ../build/linux-i386-bootimage/bootimage-generator
-cp stage2 \
-bootimage bootimage-bin.o \
-codeimage codeimage-bin.o
Note that you can override the default names for the start and end symbols in the boot/code image by also passing:
-bootimage-symbols my_bootimage_start:my_bootimage_end \
-codeimage-symbols my_codeimage_start:my_codeimage_end
7. Write a driver which starts the VM and runs the desired main method. Note the bootimageBin function, which will be called by the VM to get a handle to the embedded boot image. We tell the VM about this function via the "avian.bootimage" property.
Note also that this example includes no resources besides class files. If our application loaded resources such as images and properties files via the classloader, we would also need to embed the jar file containing them. See the previous example for instructions.
$ cat >bootimage-main.cpp <<EOF
#include "stdint.h"
#include "jni.h"
#if (defined __MINGW32__) || (defined _MSC_VER)
# define EXPORT __declspec(dllexport)
#else
# define EXPORT __attribute__ ((visibility("default")))
#endif
#if (! defined __x86_64__) && ((defined __MINGW32__) || (defined _MSC_VER))
# define BOOTIMAGE_BIN(x) binary_bootimage_bin_##x
# define CODEIMAGE_BIN(x) binary_codeimage_bin_##x
#else
# define BOOTIMAGE_BIN(x) _binary_bootimage_bin_##x
# define CODEIMAGE_BIN(x) _binary_codeimage_bin_##x
#endif
extern "C" {
extern const uint8_t BOOTIMAGE_BIN(start)[];
extern const uint8_t BOOTIMAGE_BIN(end)[];
EXPORT const uint8_t*
bootimageBin(unsigned* size)
{
*size = BOOTIMAGE_BIN(end) - BOOTIMAGE_BIN(start);
return BOOTIMAGE_BIN(start);
}
extern const uint8_t CODEIMAGE_BIN(start)[];
extern const uint8_t CODEIMAGE_BIN(end)[];
EXPORT const uint8_t*
codeimageBin(unsigned* size)
{
*size = CODEIMAGE_BIN(end) - CODEIMAGE_BIN(start);
return CODEIMAGE_BIN(start);
}
} // extern "C"
int
main(int ac, const char** av)
{
JavaVMInitArgs vmArgs;
vmArgs.version = JNI_VERSION_1_2;
vmArgs.nOptions = 2;
vmArgs.ignoreUnrecognized = JNI_TRUE;
JavaVMOption options[vmArgs.nOptions];
vmArgs.options = options;
options[0].optionString
= const_cast<char*>("-Davian.bootimage=bootimageBin");
options[1].optionString
= const_cast<char*>("-Davian.codeimage=codeimageBin");
JavaVM* vm;
void* env;
JNI_CreateJavaVM(&vm, &env, &vmArgs);
JNIEnv* e = static_cast<JNIEnv*>(env);
jclass c = e->FindClass("Hello");
if (not e->ExceptionCheck()) {
jmethodID m = e->GetStaticMethodID(c, "main", "([Ljava/lang/String;)V");
if (not e->ExceptionCheck()) {
jclass stringClass = e->FindClass("java/lang/String");
if (not e->ExceptionCheck()) {
jobjectArray a = e->NewObjectArray(ac-1, stringClass, 0);
if (not e->ExceptionCheck()) {
for (int i = 1; i < ac; ++i) {
e->SetObjectArrayElement(a, i-1, e->NewStringUTF(av[i]));
}
e->CallStaticVoidMethod(c, m, a);
}
}
}
}
int exitCode = 0;
if (e->ExceptionCheck()) {
exitCode = -1;
e->ExceptionDescribe();
}
vm->DestroyJavaVM();
return exitCode;
}
EOF
$ g++ -I$JAVA_HOME/include -I$JAVA_HOME/include/linux \
-D_JNI_IMPLEMENTATION_ -c bootimage-main.cpp -o main.o
8. Link the objects produced above to produce the final executable, and optionally strip its symbols.
$ g++ -rdynamic *.o -ldl -lpthread -lz -o hello
$ strip --strip-all hello
Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners.
The Avian project is not affiliated with Oracle.