Android C/C++ native calls Java APIs - java

There are lots of examples that Android C/C++ native calls Java APIs.
However, all of these examples I have read are the Android Java APIs call native
first and then the native calls other Java APIs by using the passed JNI-ENV.
Without the passed JNI-ENV, how can the C/C++ get it?
Is it possible that C/C++ native calls Java APIs without JNI-ENV?
Can you give an example or a link for it if it is possible?
Thanks!

You need to include jni.h first. This brings a ton of useful calls; with newer android releases you'll also need JniInvocation.h. To enable this:
LOCAL_C_INCLUDES += ${JNI_H_INCLUDE}
That certainly works with source tree, not sure about NDK, but should be fine, too.
Second, pretty important thing is to have proper signal chain lib selected. Art or Dalvik will load libsigchain.so, which is a stub and abort()s every time any of its methods are being called. On Android it's done with a little hack: local symbols are being exported to global symbol table, so that Art picks exec's symbols instead of loading shared lib. Here's how it's done:
# Enable native helper calls
LOCAL_SHARED_LIBRARIES += libnativehelper
# Include all of the Android's libsigchain symbols
LOCAL_WHOLE_STATIC_LIBRARIES += libsigchain
# Export only libsigchain symbols to global symbol table.
LOCAL_LDFLAGS += \
-Wl,--export-dynamic \
-Wl,--version-script,art/sigchainlib/version-script.txt
Link your executable now. Double check that the required symbols are indeed exported:
% readelf -a <output_binary> | grep InitializeSignalChain
654: 0002cab1 211 FUNC GLOBAL PROTECTED 12 InitializeSignalChain
Done? things get simpler now:
Initialize the JNI (so that your code uses proper VM)
JniInvocation invocation;
if (invocation.Init(nullptr)) return;
Create JavaVM:
JavaVM* vm;
JNIEnv* env;
JavaVMInitArgs* args;
args.version = JNI_VERSION_1_4; // _5, _6
args.options = nullptr;
args.nOptions = 0;
args.ignoreUnrecognized = JNI_FALSE;
if (JNI_CreateJavaVM(&vm, &env, &args) < 0) return;
At this point your vm and env are ready to use. have fun.
Check for exceptions, if any
if (env->ExceptionCheck()) {
// ...
}
When you're done, clean up
vm->DetachCurrentThread();
vm->DestroyJavaVM();
More interesting stuff can be found here and DalvikVM Main is probably the best source of knowledge. Good luck!

Related

Google OR-Tools in Intellij: UnsatisfiedLinkError

I am setting up a java framework that should use the Google OR-Tools. The code below compiles successfully, but throws an exception at runtime:
Exception in thread "main" java.lang.UnsatisfiedLinkError: com.google.ortools.linearsolver.operations_research_linear_solverJNI.MPSolver_CLP_LINEAR_PROGRAMMING_get()I
at com.google.ortools.linearsolver.operations_research_linear_solverJNI.MPSolver_CLP_LINEAR_PROGRAMMING_get(Native Method)
at com.google.ortools.linearsolver.MPSolver$OptimizationProblemType.<clinit>(MPSolver.java:221)
at Main.main(Main.java:15)
I am using Intellij 2018.3 on Windows 10. I spent a lot of time trying to get this run, but unsuccessful. Based on what I found on the internet, the exception might be caused by poor linking and/or missing external libraries on which OR-Tools depends. However, I don't have the background to resolve this issue, and also Intellij does not highlight anything. Any idea what the problem is?
For completion, this is the code I run:
import com.google.ortools.linearsolver.MPObjective;
import com.google.ortools.linearsolver.MPSolver;
import com.google.ortools.linearsolver.MPVariable;
public final class Main {
public static void main(String[] args) {
// Create the linear solver with the GLOP backend.
MPSolver solver =
new MPSolver("SimpleLpProgram", MPSolver.OptimizationProblemType.GLOP_LINEAR_PROGRAMMING);
// Create the variables x and y.
MPVariable x = solver.makeNumVar(0.0, 1.0, "x");
MPVariable y = solver.makeNumVar(0.0, 2.0, "y");
System.out.println("Number of variables = " + solver.numVariables());
// Create a linear constraint, 0 <= x + y <= 2.
MPConstraint ct = solver.makeConstraint(0.0, 2.0, "ct");
ct.setCoefficient(x, 1);
ct.setCoefficient(y, 1);
System.out.println("Number of constraints = " + solver.numConstraints());
// Create the objective function, 3 * x + y.
MPObjective objective = solver.objective();
objective.setCoefficient(x, 3);
objective.setCoefficient(y, 1);
objective.setMaximization();
solver.solve();
System.out.println("Solution:");
System.out.println("Objective value = " + objective.value());
System.out.println("x = " + x.solutionValue());
System.out.println("y = " + y.solutionValue());
}
}
In my case solution was simple - I just needed to add this singe line of code:
Loader.loadNativeLibraries();
where loader comes from com.google.ortools.Loader
Disclaimer: more a long comment than an answer...
note: I supposed you are using the github repository of or-tools if you used the binary package it should be more or less the same...
1) You must load the jni library which will load the OR-Tools C++ libraries and its dependencies...
/** Simple linear programming example.*/
public class Main {
static {
System.loadLibrary("jniortools");
}
public static void main(String[] args) throws Exception {
2) Did you manage to run the java samples ?
make run SOURCE=ortools/linear_solver/samples/SimpleLpProgram.java
ref: https://developers.google.com/optimization/introduction/java#simple_example
3) As Kayaman pointed out, you must pass the folder where the java runtime can find the native libraries (i.e. the JNI wrapper jniortools.dll and its dependencies libortools.dll)
if you look at the console log you'll see the full command line:
java -Xss2048k -Djava.library.path=lib -cp lib\sample.jar;lib\com.google.ortools.jar;lib\protobuf.jar ...\sample
Which comes from, the makefiles/Makefile.java file:
JAVAFLAGS = -Djava.library.path=$(LIB_DIR)
...
ifeq ($(SOURCE_SUFFIX),.java) # Those rules will be used if SOURCE contain a .java file
$(CLASS_DIR)/$(SOURCE_NAME): $(SOURCE) $(JAVA_OR_TOOLS_LIBS) | $(CLASS_DIR)
-$(DELREC) $(CLASS_DIR)$S$(SOURCE_NAME)
-$(MKDIR_P) $(CLASS_DIR)$S$(SOURCE_NAME)
"$(JAVAC_BIN)" -d $(CLASS_DIR)$S$(SOURCE_NAME) \
-cp $(LIB_DIR)$Scom.google.ortools.jar$(CPSEP)$(LIB_DIR)$Sprotobuf.jar \
$(SOURCE_PATH)
...
.PHONY: run # Run a Java program.
run: build
"$(JAVA_BIN)" -Xss2048k $(JAVAFLAGS) \
-cp $(LIB_DIR)$S$(SOURCE_NAME)$J$(CPSEP)$(LIB_DIR)$Scom.google.ortools.jar$(CPSEP)$(LIB_DIR)$Sprotobuf.jar \
$(SOURCE_NAME) $(ARGS)
endif
src: https://github.com/google/or-tools/blob/46173008fdb15dae1dca0e8fa42a21ed6190b6e4/makefiles/Makefile.java.mk#L15
and
https://github.com/google/or-tools/blob/46173008fdb15dae1dca0e8fa42a21ed6190b6e4/makefiles/Makefile.java.mk#L328-L333
note: you can run make detect_java to know the flags i.e. value of LIB_DIR
note: if you did use the precompiled package the Makefile is here:
https://github.com/google/or-tools/blob/stable/tools/Makefile.cc.java.dotnet
Then after you can try to add this option in Intellij...
You must understand that or-tools is a set of C++ native libraries which are wrapped to Java using the SWIG generator.
To make it work using Intellij (over a windows machine) you need to:
Install Microsoft Visual C++ Redistributable for Visual Studio
Download and extract the OR-Tools library for Java
In intellij, add jar dependency to the 2 jars under the lib folder of the extracted files (each of the 2 jars separately, do not add to lib folder itself. This is why).
Add the lib library path to VM options. In Intellij edit your run-configuration and add to vm options: -Djava.library.path=<path to the lib folder that hold the jars>
Load the jni library statically by adding the below code to your class (as mentioned here.)
static {
System.loadLibrary("jniortools");
}

Unit test Java class that loads native library

I'm running unit tests in Android Studio. I have a Java class that loads a native library with the following code
static
{
System.loadLibrary("mylibrary");
}
But when I test this class inside my src/test directory I get
java.lang.UnsatisfiedLinkError: no mylibrary in java.library.path
at java.lang.ClassLoader.loadLibrary(ClassLoader.java:1864)
at java.lang.Runtime.loadLibrary0(Runtime.java:870)
at java.lang.System.loadLibrary(System.java:1122)
How can I make it find the path of native .so libraries which is located at src/main/libs in order to unit test without errors?
Note: inside src/main/libs directory I have 3 more subdirectories: armeabi, mips and x86. Each one of those contains the proper .so file. I'm using the Non experimental version for building NDK libs.
I don't wanna use other 3rd party testing libraries as all my other "pure" java classes can be unit tested fine. But if that's not possible then I'm open to alternatives.
Here is my test code which throws the error
#Test
public void testNativeClass() throws Exception
{
MyNativeJavaClass test = new MyNativeJavaClass("lalalal")
List<String> results = test.getResultsFromNativeMethodAndPutThemInArrayList();
assertEquals("There should be only three result", 3, results.size());
}
The only solution I found that works without hacks is to use JUnit through instrumentation testing (androidTest directory).
My class can now be tested fine but with help of the android device or emulator.
If the library is required for your test, use an AndroidTest (under src/androidTest/...) rather than a junit test. This will allow you to load and use the native library like you do elsewhere in your code.
If the library is not required for your test, simply wrap the system load in a try/catch. This will allow the JNI class to still work in junit tests (under src/test/...) and it is a safe workaround, given that it is unlikely to mask the error (something else will certainly fail, if the native lib is actually needed). From there, you can use something like mockito to stub out any method calls that still hit the JNI library.
For example in Kotlin:
companion object {
init {
try {
System.loadLibrary("mylibrary")
} catch (e: UnsatisfiedLinkError) {
// log the error or track it in analytics
}
}
}
I am not sure whether this solves your problem or not but so far nobody has mentioned about strategy pattern for dealing with classes preloading library during their creation.
Let's see the example:
We want to implement Fibonacci solver class. Assuming that we provided implementation in the native code and managed to generate the native library, we can implement the following:
public interface Fibonacci {
long calculate(int steps);
}
Firstly, we provide our native implementation:
public final class FibonacciNative implements Fibonacci {
static {
System.loadLibrary("myfibonacci");
}
public native long calculate(int steps);
}
Secondly, we provide Java implementation for Fibonacci solver:
public final class FibonacciJava implements Fibonacci {
#Override
public long calculate(int steps) {
if(steps > 1) {
return calculate(steps-2) + calculate(steps-1);
}
return steps;
}
}
Thirdly, we wrap the solvers with parental class choosing its own implementation during its instantiation:
public class FibonnaciSolver implements Fibonacci {
private static final Fibonacci STRATEGY;
static {
Fibonacci implementation;
try {
implementation = new FibonnaciNative();
} catch(Throwable e) {
implementation = new FibonnaciJava();
}
STRATEGY = implementation;
}
#Override
public long calculate(int steps) {
return STRATEGY.calculate(steps);
}
}
Thus, the problem with finding path to the library using strategy. This case, however, does not resolve the problem if the native library is really necessary to be included during the test. It does not neither solve the problem if the native library is a third-party library.
Basically, this gets around the native library load problem by mocking out the native code for java code.
Hope this helps somehow:)
There is a way to configure library path of Gradle-run VM for local unit tests, and I'm going to describe it below, but spoiler: in my expericence, #ThanosFisherman is right: local unit tests for stuff that uses the Android NDK seem to be a fools errand right now.
So, for anyone else looking for a way to load shared (i.e. .so) libraries into unit tests with gradle, here's the somewhat lengthy abstract:
The goal is to set the shared library lookup path for the JVM running the unit tests.
Althoug many people suggest putting the lib path into java.library.path, I found that it doesn't work, at least not on my linux machine. (also, same results in this CodeRanch thread)
What does work though is setting the LD_LIBRARY_PATH os environment variable (or PATH is the closest synonym in Windows)
Using Gradle:
// module-level build.gradle
apply plugin: 'com.android.library' // or application
android {
...
testOptions {
unitTests {
all {
// This is where we have access to the properties of gradle's Test class,
// look it up if you want to customize more test parameters
// next we take our cmake output dir for whatever architecture
// you can also put some 3rd party libs here, or override
// the implicitly linked stuff (libc, libm and others)
def libpath = '' + projectDir + '/build/intermediates/cmake/debug/obj/x86_64/'
+':/home/developer/my-project/some-sdk/lib'
environment 'LD_LIBRARY_PATH', libpath
}
}
}
}
With that, you can run, e.g. ./gradlew :mymodule:testDebugUnitTest and the native libs will be looked for in the paths that you specified.
Using Android Studio JUnit plugin
For the Android Studio's JUnit plugin, you can specify the VM options and the environment variables in the test configuration's settings, so just run a JUnit test (right-clicking on a test method or whatever) and then edit the Run Configuration:
Although it sounds like "mission accomplished", I found that when using libc.so, libm.so and others from my os /usr/lib gives me version errors (probably because my own library is compiled by cmake with the android ndk toolkit against it's own platform libs). And using the platform libs from the ndk packages brought down the JVM wih a SIGSEGV error (due to incompatibility of the ndk platform libs with the host os environment)
Update As #AlexCohn incisively pointed out in the comments, one has to build against the host environment libs for this to work; even though your machine most likely is x86_64, the x86_64 binaries built against NDK environment will not do.
There may be something I overlooked, obviously, and I'll appreciate any feedback, but for now I'm dropping the whole idea in favor of instrumented tests.
Just make sure, the directory containing the library is contained in the java.library.path system property.
From the test you could set it before you load the library:
System.setProperty("java.library.path", "... path to the library .../libs/x86");
You can specify the path hard coded, but this will make the project less portable to other environments. So I suggest you build it up programmatically.
The .so files are to be placed under
src/main/jniLibs
Not under src/main/libs
(Tested with Android Studio 1.2.2)
For reference check the page - http://ph0b.com/android-studio-gradle-and-ndk-integration/, though some portions might be outdated.
This is very, very tricky. Setting java.library.path does not work, but trying to understand someone else’s Mac OSX approach I eventually found a working solution.
Legal release: all code examples directly copied into this post are available under CC0 but it would be appeciated to credit my employer ⮡ tarent, the LLCTO project at Deutsche Telekom, and the author mirabilos.
CAVEATS first:
with this, you’re testing a version of the native code compiled against your system libraries (usually glibc on GNU/Linux, and on BSD, Mac OSX and Windows it’s even trickier) so adding some instrumented tests should be done anyway, use the unittests only for faster testing of things that actually can be tested on the host OS
I’ve only tested this with a GNU/Linux host (and am, in fact, excluding these native tests on all other host OSes, see below)
it should work under unixoid OSes with GNU/BSD-style shared libraries as-is
with small adaptions from the “someone else’s” article linked above, it might probably work on Mac OSX
Windows… no, just no. Use WSL, which is basically Linux anyway and makes things much easier, and so much closer to Android which is also basically Linux just not GNU
IDE integration needs manual steps at each developer’s machine (but these are easily documented, see (much) below)
Prerequisites
You’ll need to make sure that all build dependencies of your native code are also installed in the host system. This includes cmake (because we sadly cannot reuse the NDK cmake) and a host C compiler. Note that these introduce further differences in the build: you’re testing something that has been built with the host C compiler (often GCC, not clang like in Android) against the host C library and other libraries by the host clang. Do consider this when writing your tests. I had to move one of the tests to instrumented because it was impossible to test under glibc.
For filesystem layout, we assume the following:
~/MYPRJ/build.gradle is the top-level build file (generated by IntelliJ / Android Studio)
~/MYPRJ/app/build.gradle is where the Android code in question is built (generated by IntelliJ / Android Studio)
~/MYPRJ/app/src/main/native/CMakeLists.txt is where the native code is situated
This means build.gradle (for the app) has something like this already, by the point where you begin wondering about whether your project can be unittested:
externalNativeBuild {
cmake {
path "src/main/native/CMakeLists.txt"
return void // WTF‽
}
}
Make sure your code builds on the host
Doing this ought to be easy at first glance:
$ rm -rf /tmp/build
$ mkdir /tmp/build
$ cd /tmp/build
$ cmake ~/MYPRJ/app/src/main/native/
$ make
(Make sure you give cmake the path to the directory the main CMakeLists.txt file is in, but not to that file itself!)
This will fail for everything nōntrivial, of course. Most people would use Android logging. (It will also fail because it cannot find <jni.h>, and because GNU libc requires an extra _GNU_SOURCE definition to access some prototypes, etc…)
So I wrote a header to include instead of <android/log.h> which abstracts the logging away…
#ifndef MYPRJ_ALOG_H
#define MYPRJ_ALOG_H
#ifndef MYPRJ_ALOG_TAG
#define MYPRJ_ALOG_TAG "MYPRJ-JNI"
#endif
#if defined(MYPRJ_ALOG_TYPE) && (MYPRJ_ALOG_TYPE == 1)
#include <android/log.h>
#define ecnlog_err(msg, ...) __android_log_print(ANDROID_LOG_ERROR, \
MYPRJ_ALOG_TAG, msg, ##__VA_ARGS__)
#define ecnlog_warn(msg, ...) __android_log_print(ANDROID_LOG_WARN, \
MYPRJ_ALOG_TAG, msg, ##__VA_ARGS__)
#define ecnlog_info(msg, ...) __android_log_print(ANDROID_LOG_INFO, \
MYPRJ_ALOG_TAG, msg, ##__VA_ARGS__)
#elif defined(MYPRJ_ALOG_TYPE) && (MYPRJ_ALOG_TYPE == 2)
#include <stdio.h>
#define ecnlog_err(msg, ...) fprintf(stderr, \
"E: [" MYPRJ_ALOG_TAG "] " msg "\n", ##__VA_ARGS__)
#define ecnlog_warn(msg, ...) fprintf(stderr, \
"W: [" MYPRJ_ALOG_TAG "] " msg "\n", ##__VA_ARGS__)
#define ecnlog_info(msg, ...) fprintf(stderr, \
"I: [" MYPRJ_ALOG_TAG "] " msg "\n", ##__VA_ARGS__)
#else
# error What logging system to use?
#endif
#endif
… and updated my CMakeLists.txt to indicate whether building for NDK (must be default) or native:
cmake_minimum_required(VERSION 3.10)
project(myprj-native)
option(UNDER_NDK "Build under the Android NDK" ON)
add_compile_options(-fvisibility=hidden)
add_compile_options(-Wall -Wextra -Wformat)
add_library(myprj-native SHARED
alog.h
myprj-jni.c
)
if (UNDER_NDK)
add_definitions(-DECNBITS_ALOG_TYPE=1)
find_library(log-lib log)
target_link_libraries(myprj-native ${log-lib})
else (UNDER_NDK)
add_definitions(-DECNBITS_ALOG_TYPE=2)
include(FindJNI)
include_directories(${JNI_INCLUDE_DIRS})
add_definitions(-D_GNU_SOURCE)
endif (UNDER_NDK)
Note this also already includes the fix for <jni.h> (FindJNI) and the extra definitions.
Now let’s try to build it again:
$ rm -rf /tmp/build
$ mkdir /tmp/build
$ cd /tmp/build
$ cmake -DUNDER_NDK=OFF ~/MYPRJ/app/src/main/native/
$ make
In my case, this was sufficient. If you’re still not there, fix this first before proceeding. If you cannot fix this, give up on buildhost-local unit tests for your JNI code and move the respective tests to instrumented.
Let Gradle build the host-native code
Add the following to the app build.gradle:
def dirForNativeNoNDK = project.layout.buildDirectory.get().dir("native-noNDK")
def srcForNativeNoNDK = project.layout.projectDirectory.dir("src/main/native").asFile
task createNativeNoNDK() {
def dstdir = dirForNativeNoNDK.asFile
if (!dstdir.exists()) dstdir.mkdirs()
}
task buildCMakeNativeNoNDK(type: Exec) {
dependsOn createNativeNoNDK
workingDir dirForNativeNoNDK
commandLine "/usr/bin/env", "cmake", "-DUNDER_NDK=OFF", srcForNativeNoNDK.absolutePath
}
task buildGMakeNativeNoNDK(type: Exec) {
dependsOn buildCMakeNativeNoNDK
workingDir dirForNativeNoNDK
commandLine "/usr/bin/env", "make"
}
project.afterEvaluate {
if (org.gradle.internal.os.OperatingSystem.current().isLinux()) {
testDebugUnitTest {
dependsOn buildGMakeNativeNoNDK
systemProperty "java.library.path", dirForNativeNoNDK.asFile.absolutePath + ":" + System.getProperty("java.library.path")
}
testReleaseUnitTest {
dependsOn buildGMakeNativeNoNDK
systemProperty "java.library.path", dirForNativeNoNDK.asFile.absolutePath + ":" + System.getProperty("java.library.path")
}
}
}
This defines a few new tasks to compile the buildhost-native version of the shared library, and hooks this up if the host OS is “Linux”. (This syntax will also work for other unixoid OSes — BSD, Mac OSX — but not for Windows. But we can probably test this under Linux only anyway. WSL counts as Linux.) It also sets up the JVM library path so that ../gradlew test will let the JVM pick up the library from its path.
Loose ends
There’s a few loose ends you might have noticed here:
In the last paragraph of the previous section, I mentioned that ../gradlew test will pick up the library. Testing from the IDE will not work yet; this involves manual setup.
I mentioned that the relevant unit tests must be skipped if the buildhost OS is not “Linux”; we have yet to do that. Unfortunately, JUnit 4 lacks such facilities, but switching the unit tests to JUnit 5 “Jupiter” will allow us to do that. (We’re not switching the instrumented tests, though; that’d be more invasive.)
You’ll probably not yet have noticed, but the logging output from the native code will not show up thanks to Gradle’s default settings which we’ll need to change.
So, let’s do that. First, edit your app build.gradle file again. There will be a dependencies { block. We’ll need to fill that with suitable dependencies for either JUnit:
dependencies {
testImplementation 'org.junit.jupiter:junit-jupiter-api:5.7.0'
testRuntimeOnly 'org.junit.jupiter:junit-jupiter-engine:5.7.0'
//noinspection GradleDependency
androidTestImplementation 'com.android.support.test:runner:1.0.1'
//noinspection GradleDependency
androidTestImplementation 'com.android.support.test.espresso:espresso-core:3.0.1'
//noinspection GradleDependency
androidTestImplementation 'junit:junit:4.12'
}
You’ll also have a line apply plugin: 'com.android.application' (or perhaps apply plugin: 'com.android.library') at the top. Directly below that line, insert this one:
apply plugin: 'de.mannodermaus.android-junit5'
Also, make sure that, under android { defaultConfig { the testInstrumentationRunner is still "android.support.test.runner.AndroidJUnitRunner" (the default as generated by IntelliJ / Android Studio).
Next, edit the top-level ~/MYPRJ/build.gradle file. You’ll already have a buildscript { dependencies { and will have to add a line to that section to make the JUnit5 plugin available in the first place:
//noinspection GradleDependency
classpath 'de.mannodermaus.gradle.plugins:android-junit5:1.5.2.0'
Then, add a new section under allprojects {:
tasks.withType(Test) {
testLogging {
outputs.upToDateWhen { false }
showStandardStreams = true
exceptionFormat = 'full'
}
systemProperty 'java.util.logging.config.file', file('src/test/resources/logging.properties').getAbsolutePath()
}
This ensures that…
tests are never skipped because Gradle thinks them up-to-date
logging output and exceptions are shown in full
if you have a ~/MYPRJ/app/src/test/resources/logging.properties it will set up java.util.logging with this (recommended)
Now see to your test, something like ~/MYPRJ/app/src/test/java/org/example/packagename/JNITest.java. First, you should add a “test” that can always run (I use one that merely tests whether my JNI class can be loaded), and ensure it displays some information first:
// or Lombok #Log
private static final java.util.logging.Logger LOGGER = java.util.logging.Logger.getLogger(JNITest.class.getName());
#Test
public void testClassBoots() {
LOGGER.info("running on " + System.getProperty("os.name"));
if (!LINUX.isCurrentOs()) {
LOGGER.warning("skipping JNI tests");
}
// for copy/paste into IntelliJ run options
LOGGER.info("VM options: -Djava.library.path=" +
System.getProperty("java.library.path"));
LOGGER.info("testing Java™ part of JNI class…");
[…]
}
Then, annotate the actual JNI tests that need to be skipped on other OSes:
#Test
#EnabledOnOs(LINUX)
public void testJNIBoots() {
LOGGER.info("testing JNI part of JNI class…");
final long tid;
try {
tid = JNI.n_gettid();
} catch (Throwable t) {
LOGGER.log(Level.SEVERE, "it failed", t);
Assertions.fail("JNI does not work");
return;
}
LOGGER.info("it also works: " + tid);
assertNotEquals(0, tid, "but is 0");
}
For comparison, instrumented tests (unittests that run on the Android device or emulator) — e.g. ~/MYPRJ/app/src/androidTest/java/org/example/packagename/JNIInstrumentedTest.java — look like this:
#RunWith(AndroidJUnit4.class)
public class JNIInstrumentedTest {
#Test
public void testJNIBoots() {
Log.i("ECN-Bits-JNITest", "testing JNI part of JNI class…");
final long tid;
try {
tid = JNI.n_gettid();
} catch (Throwable t) {
Log.e("ECN-Bits-JNITest", "it failed", t);
fail("JNI does not work");
return;
}
Log.i("ECN-Bits-JNITest", "it also works: " + tid);
assertNotEquals("but is 0", 0, tid);
}
}
See Testable.java if you need an assertThrows for instrumented tests (JUnit 5 already comes with one), by the way. (Note that this does not fall under the CC0 grant above but comes under a permissive licence.)
Now, you can run both tests, unittests and (if an Android emulator is started or device commected) instrumented tests:
../gradlew test connectedAndroidTest
Do so. Note the output of the VM options: logger call from the buildhost-native unit tests; in fact, copy it to the clipboard. You’ll now need it to set up testing in the IDE.
In the Project view (left-side tree), right-click either on your JNITest class or the entire src/test/java/ directory. Click on Run 'JNITest' (or Run 'Tests in 'java''), it will fail with an UnsatisfiedLinkError as in the original post.
Now click on the arrow in the test drop-down below the menu bar, then select Save JNITest configuration, then do it again and select Edit configurations… and select your configuration. Append the entire pasted thing to VM options: so the field will now look like -ea -Djava.library.path=/home/USER/MYPRJ/app/build/native-noNDK:/usr/java/packages/lib:/usr/lib/x86_64-linux-gnu/jni:/lib/x86_64-linux-gnu:/usr/lib/x86_64-linux-gnu:/usr/lib/jni:/lib:/usr/lib (of course, the actual value will differ) and click OK. Re-run the test, and it will succeed.
Unfortunately, you’ll have to do this for every native test class once and for the entire directory, so all possible ways of invocation will be covered. You’ll also have to do this manually, by clicking around, for every IDE instance, and these values depend on the path the code was checked out into. I’ve not found a way to automate these (if you know of one, do tell).
Exception backtraces
If you’re throwing custom exceptions from your code, you’ll most likely wish to include file/lineno/function information. Use a constructor like MyprjNativeException(final String file, final int line, final String func, final String msg, … /* custom data */, final Throwable cause) and, after calling super(msg, cause) (possibly with a changed message), do this:
StackTraceElement[] currentStack = getStackTrace();
StackTraceElement[] newStack = new StackTraceElement[currentStack.length + 1];
System.arraycopy(currentStack, 0, newStack, 1, currentStack.length);
newStack[0] = new StackTraceElement("<native>", func, file, line);
setStackTrace(newStack);
Then, to throw an exception like this from native code:
#define throw(env,...) vthrow(__FILE__, __func__, env, __LINE__, __VA_ARGS__)
static void vthrow(const char *loc_file, const char *loc_func, JNIEnv *env,
int loc_line, /* custom args */ const char *msg, ...);
Use as follows:
if (func() != expected)
throw(env, /* custom args */ "foo");
Implementation (assuming you cache class and constructor method references) looks as follows (adjust for custom args):
static void vthrow(const char *loc_file, const char *loc_func, JNIEnv *env,
int loc_line, const char *fmt, ...)
{
jthrowable e;
va_list ap;
jstring jfile = NULL;
jint jline = loc_line;
jstring jfunc = NULL;
jstring jmsg = NULL;
jthrowable cause = NULL;
const char *msg;
char *msgbuf;
if ((*env)->PushLocalFrame(env, /* adjust for amount you need */ 5)) {
cause = (*env)->ExceptionOccurred(env);
(*env)->ExceptionClear(env);
(*env)->Throw(env, (*env)->NewObject(env, classreference, constructorreference,
jfile, jline, jfunc, jmsg, /* custom */…, cause));
return;
}
if ((cause = (*env)->ExceptionOccurred(env))) {
/* will be treated as cause */
(*env)->ExceptionClear(env);
}
va_start(ap, fmt);
if (vasprintf(&msgbuf, fmt, ap) == -1) {
msgbuf = NULL;
msg = fmt;
} else
msg = msgbuf;
va_end(ap);
jmsg = (*env)->NewStringUTF(env, msg);
free(msgbuf);
if (!jmsg)
goto onStringError;
if (!(jfunc = (*env)->NewStringUTF(env, loc_func)))
goto onStringError;
/* allocate NewStringUTF for any custom things you need */
/* exactly like the one for loc_func above */
/* increase PushLocalFrame argument for each */
jfile = (*env)->NewStringUTF(env, loc_file);
if (!jfile) {
onStringError:
(*env)->ExceptionClear(env);
}
e = (*env)->PopLocalFrame(env, (*env)->NewObject(env, classreference, constructorreference,
jfile, jline, jfunc, jmsg, /* custom */…, cause));
if (e)
(*env)->Throw(env, e);
}
Now using __FILE__ will put the full absolute path into the messages and backtraces. This is not very nice. There’s a compiler option to fix that, but NDK r21’s clang is much too old, so we need a workaround.
CMakeLists.txt:
if (NOT TOPLEV)
message(FATAL_ERROR "setting the top-level directory is mandatory")
endif (NOT TOPLEV)
[…]
if (UNDER_NDK)
[…]
execute_process(COMMAND ${CMAKE_CXX_COMPILER} --version OUTPUT_VARIABLE cxx_version_full)
string(REGEX REPLACE "^Android [^\n]* clang version ([0-9]+)\\.[0-9].*$" "\\1" cxx_version_major ${cxx_version_full})
if (${cxx_version_major} VERSION_GREATER_EQUAL 10)
add_definitions("-ffile-prefix-map=${TOPLEV}=«MyPrj»")
else (${cxx_version_major} VERSION_GREATER_EQUAL 10)
add_definitions(-DOLD_CLANG_SRCDIR_HACK="${TOPLEV}/")
endif (${cxx_version_major} VERSION_GREATER_EQUAL 10)
else (UNDER_NDK)
[…]
add_definitions("-ffile-prefix-map=${TOPLEV}=«MyPrj»")
endif (UNDER_NDK)
app build.gradle:
(straight after the apply plugin lines)
def dirToplev = project.layout.projectDirectory.asFile.absolutePath
(inside android { defaultConfig { add a new block)
externalNativeBuild {
cmake {
//noinspection GroovyAssignabilityCheck because Gradle and the IDE have different world views…
arguments "-DTOPLEV=" + dirToplev
}
return void // WTF‽
}
(later, where you call cmake)
commandLine "/usr/bin/env", "cmake", "-DTOPLEV=" + dirToplev, "-DUNDER_NDK=OFF", srcForNativeNoNDK.absolutePath
Then, replace the line jfile = (*env)->NewStringUTF(env, loc_file); with the following snippet:
#ifdef OLD_CLANG_SRCDIR_HACK
if (!strncmp(loc_file, OLD_CLANG_SRCDIR_HACK, sizeof(OLD_CLANG_SRCDIR_HACK) - 1) &&
asprintf(&msgbuf, "«ECN-Bits»/%s", loc_file + sizeof(OLD_CLANG_SRCDIR_HACK) - 1) != -1) {
msg = msgbuf;
} else {
msg = loc_file;
msgbuf = NULL;
}
#else
#define msg loc_file
#endif
jfile = (*env)->NewStringUTF(env, msg);
#ifdef OLD_CLANG_SRCDIR_HACK
free(msgbuf);
#else
#undef msg
#endif
Tieing it all together
This all is implemented in the ECN-Bits project. I’m posting a permalink because it’s currently on a nōn-default branch but expected to be merged (once the actual functionality is no longer WIP), so be sure to check master at some point in time as well (although this permalink is probably a better example as it has the testing down and there’s not as much “actual” code to get in the way). Note that these links do not fall under the CC0 grant from above; the files all have a permissive licence though (the files which don’t have it explicit (gradle/cmake files) have the same as the unittest class permalink), but enough of it was reposted in this article so that should not be a problem for you; these only serve to show an actually-compiling-and-testing example.
In this project, it’s not in app/ but as a separate library module.
top-level build.gradle
library build.gradle
instrumented tests
unittest class
unittest logging configuration
native CMakeLists.txt
native alog.h
native code including caching the references
Java™ code to which the JNI native code attaches including the Exception class
Try running your test code with java -XshowSettings:properties option and make sure your destination path for system libraries and in the output of this command, library path values are the same
Just to clarify, the System.loadlibrary() call was failing because the junit unit test uses host/system environment which was windows in my case. Hence the loadlibrary() call was trying to search for the .so files in standard shared libary folders. But this isn't what I was expecting to happen. Instead I wanted the libxxx.so files to be loaded from .aar file(contains android resources, jars, jni libs).
This can only happen by two ways:
Baking the libxxx.so into the APK manually: we ourselves place the libxxx.so files into jnilibs dir(which the system will search for when loading the shared lib) under src/java/ root of the apk by copying the required files.
adding the .aar as an external dependency to module level gradle build script (implementation/androidTestImplementation), by doing this we make this aar available for link editing used by the apk build process.
But in both cases the app runs in the android environment/vm and hence the System.loadlibrary() call will resolve to correct libxxx.so which would be part of the apk. Hence NO ISSUES.
However in case of unit tests, which are does not require instrument(ie, android device) and runs on the JVM running on the host system where the tests are running (ex: windows/linux/mac), The call to System.loadlibrary() resolves only the standard lib paths of host system for finding shared libs/exe and doesn't refer to android system environment. Hence the ISSUES.
Fixes:
unpack the libxxx.so into some temp dir and add this dir to the system's library search path (ex: java.library.path, %PATH% variable on windows etc). Now run the required unit tests which doesn't require the android environment but involves the native code testing using JNI if any. This should work!!
(Efficient method) Simply move these type of unit tests to androidTest(ie, Instrumentation tests) so that above explained loading and packing are intact and System.loadlibrary() can successfully find the libxxx.so when running inside the instrument(android device/os). This way you ensure appropriate lib type (x86, x86-64, arm-v7a, arm-v8a(AARCH64)) is invoked on target device and tests are run on specific target devices.

Write & call Swift code using Java’s JNI

As you see here one can perform native C/C++ method calls from Java code, thanks to JVM’s JNI. But how about performing calls of Swift methods? Is this possible or going to be possible (can be implemented in a reasonable time frame) as Swift becomes more popular?
I want to access Apple’s native API that is only accessible when one writes apps in either Objective-C or Swift. As the JVM is just being ported to ARMv8 (64-bit), I could also imagine the JVM as an alternative runtime for iOS apps in the future. But this might be the future... The present is: The JVM runs on Mac OS X and one can write apps for Mac OS X in Swift that can access some APIs that might be not accessible to Java apps in the same way.
Well, roughly 5½ years later, it turned out, it wasn't the future... No JVMs on iOS.
But you can definitely do it, i.e., call Swift APIs from Java. That said, it's quite cumbersome, because AFAIK, you have to take a detour via C/C++.
Here's the deal:
As you know, you can use JNI to call C code.
From C you can call Swift.
The following relies heavily on this question.
Code
Your Java code is in helloworld/SwiftHelloWorld.java:
package helloworld;
public class SwiftHelloWorld {
static {
System.loadLibrary("SwiftHelloWorld");
}
public static native void printHelloWorldImpl();
public static void main(final String[] args) {
printHelloWorldImpl();
}
}
Now write the native C code (file helloworld_SwiftHelloWorld.c):
#include <jni.h>
#include <stdio.h>
#include "helloworld_SwiftHelloWorld.h"
#include "helloworld_SwiftHelloWorld_swift.h"
JNIEXPORT void JNICALL Java_helloworld_SwiftHelloWorld_printHelloWorldImpl
(JNIEnv *env, jclass clazz) {
int result = swiftHelloWorld(42);
printf("%s%i%s", "Hello World from JNI! ", result, "\n");
}
Which uses a header file named helloworld_SwiftHelloWorld_swift.h for our (yet to be written) Swift code:
int swiftHelloWorld(int);
Finally, our Swift code resides in SwiftCode.swift:
import Foundation
// force the function to have a name callable by the c code
#_silgen_name("swiftHelloWorld")
public func swiftHelloWorld(number: Int) -> Int {
print("Hello world from Swift: \(number)")
return 69
}
Building
To build all this, we first have to compile the Swift code to a dynamic library:
swiftc SwiftCode.swift -emit-library -o libSwiftCode.dylib -Xlinker -install_name -Xlinker libSwiftCode.dylib
We use the -Xlinker directives to ensure that the dylib's location is relative.
Before we can create the C dylib, we first have to generate the Java headers:
javac -h . helloworld/SwiftHelloWorld.java
Now that we have the Java headers and the Swift dylib, we can compile the C dylib, which links against the Swift dylib:
gcc -I"$JAVA_HOME/include" -I"$JAVA_HOME/include/darwin/" -o libSwiftHelloWorld.dylib -dynamiclib helloworld_SwiftHelloWorld.c libSwiftCode.dylib
Now that everything is in place, we must make sure that both dylibs are in the same directory and that that directory can be found by Java, i.e., you may need to set -Djava.library.path=<dir of your dylibs>.
Et voilà!
Swift called from Java!

Porting a CPP application to Android

I need to port a CPP project to Android but I somehow got stuck because of the following things that I am not sure of:
Do I need to use some kind of java wrapper for my CPP project at all, i.e is it necessarily that I use Android SDK to integrate my application with Android? If there is another way, which one would that be?
I have seen some people claiming they have been able to manipulate their cmake file and some custom android-cmake toolchain to build their “.so” or eventually an “.apk” from their project. Would it be possible without a java wrapper to manipulate the cmake files of the cpp project to build your project? (source: Build Android NDK project with Cmake)
From my experience, I would go with using the android java entry point, otherwise you will most likely bump into problems (you can make full native android apps , but I strongly advise against it).
One of the reasons is that you will want SDK function calls from inside your CPP, and using reflection on the java environment from CPP isn't trivial.
The steps would be the following :
Create a very simple C code that will server as bridge to your CPP code . Here are some sample C functions that I've commonly used :
OnApplicationStart
OnApplicationPaused
OnApplicationResumed
OnApplicationUpdate
OnTouchReceived
Export these functions and load them in your Java code (lookup JNI on how to do this)
Handle all Android-specific actions in Java, and all application specific actions in cpp
So the answer to your 1st question is that a java wrapper isn't mandatory, but it's HIGHLY recommended.
To answer your 2nd question :
Yes, you can use cmkae, ant, cygwin , allot of command tools that will end up creating your apk (It's up to you on how you feel comfortable).You will still use the android compiler since you are targeting arm.
The difference is that you need to change in your manifest.xml the entry point of the application from a normal activity to a native activity ( http://developer.android.com/reference/android/app/NativeActivity.html ) .
As you can see, the difference isn't the build system, it's the way you define your entry point.
As a friendly advice, try using the minimal java wrapper approach. You might get to better results sooner, and it won't take you more then 1 week of research on the web on how to link java code to cpp.
EDIT :
As of demand, I will shortly explain how I would approach the process of porting a CPP application to Android :
Rethink your application to work as a shared library , using a C entry point :
a.Create a simple C application that will load yourCPPApp.dll (or .so if you are on linux)
b. In your .dll create the minimum necessary extern "C" functions to be exported in order for you to give the necessary information to your dll
For simplicity, we'll assume we have 3 methods :
void StartApplication();
bool OnApplicationUpdate();
void OnScreenTouched(int x, int y);
c. Implement the simple C project that will make the calls to these methods externaly (so the .exe will call the methods from the .dll ! )
Sample code :
#include "myCPPapp.h"
int main(int arg, char** argv)
{
StartApplication();
srand(time(NULL));
while (OnApplicationUpdate())
{
// we assume we have a 480x640 resolution
OnScreenTouched(rand()%480,rand()%640);
}
return 0;
}
Now that we have things working in full native with a .exe and a .dll , time to make it work with a .apk and a .so
a. Rename the exposed methods from myCppApp into java compatible prototypes
extern "C" {
JNIEXPORT void JNICALL Java_com_sample_nativebridge_OnApplicationStart(JNIEnv env, jobject thiz);
JNIEXPORT jboolean JNICALL Java_com_sample_nativebridge_OnApplcationUpdate(JNIEnv env, jobject thiz);
JNIEXPORT void JNICALL Java_com_sample_nativebridge_OnScreenTouched(JNIEnv env, jobject thiz, jint x, jint y);
}
b. create the java class nativebridge (case sensitive) in the package com.sample (you need t respect the names in order for correct linkage to native)
class nativebridge {
public static native void OnApplicationStart();
public static native boolean OnApplicationUpdate();
public static native void OnScreenTouched(int x, int y);
}
c. add the load library statement in the native bridge in order to have your library loaded at runtime
class nativebridge {
....
static {
System.loadLibrary("myCppApp");
// notice that the lib prefix and .so sufix aren't part of the name
}
}
d. compile your CPP code with the armeabi compiler and create libmyCPPApp.so (ndk build system , or whatever you'd like... i would go with the ndk, it's quite easy , just go into the folder with Android.mk and call $ANDROID_NDK_BUILD_PATH/build )
At this point you will need to createa a .mk file that will compile your myCppApp code, but this is out of scope, you will need to do this research on your own (it's quite trivial once you get the hang of it).
c. use the native methods from the bridge inside your java app wherever you see fit.
A very good tip would be to go through a hello world sample of ndk :
http://www.ntu.edu.sg/home/ehchua/programming/android/android_ndk.html
Enjoy.

Calling Haskell from Java with C in between

This probably sounds like a nightmare, but I'd really like to get this working. I am using this example for the most part: Calling C from Haskell and am trying to get this working on ubuntu.
I am running this in java:
package test;
public class JniTest {
public native int fib(int x);
}
this in c after creating the .h file with javah: (test_JniTest.c)
#include "test_JniTest.h"
#include "Safe_stub.h"
JNIEXPORT jint JNICALL Java_test_JniTest_fib(JNIEnv * e, jobject o, jint f)
{
return fibonacci_hs(f);
}
and then for reference in haskell (before stub): (Safe.hs)
module Safe where
import Foreign.C.Types
fibonacci :: Int -> Int
fibonacci n = fibs !! n
where fibs = 0 : 1 : zipWith (+) fibs (tail fibs)
fibonacci_hs :: CInt -> CInt
fibonacci_hs = fromIntegral . fibonacci . fromIntegral
foreign export ccall fibonacci_hs :: CInt -> CInt
and this is what i'm trying to compile it with:
ghc -c -O Safe.hs
followed by:
ghc -shared -o libTest.jnilib -optc-O test_JniTest.c
-I/usr/lib/jvm/java-6-sun-1.6.0.26/include -I/usr/lib/jvm/java-6-sun-1.6.0.26/include/linux
and I am getting this error:
/usr/bin/ld: test_JniTest.o: relocation R_X86_64_PC32 against
undefined symbol `fibonacci_hs' can not be used when making a shared
object; recompile with -fPIC /usr/bin/ld: final link failed: Bad value
collect2: ld returned 1 exit status
I am not a c expert by any means and have no idea what to do about this. I tried compiling various ways with -fPIC, but I kept on getting the same error. Any idea what I might be doing wrong?
Thanks!
Although I've pretty much answered this question here: Communication between Java and Haskell, since this issue is more about the error itself, I will be adding the details for that here. The issue stems from Haskell not supporting shared libraries very well, while Java requires them.
Buildings plugins as Haskell shared libs gives us this insight and workaround:
In principle you can use -shared without -dynamic in the link step. That would mean to statically link the rts all the base libraries into your new shared library. This would make a very big, but standalone shared library. However that would require all the static libraries to have been built with -fPIC so that the code is suitable to include into a shared library and we don't do that at the moment.
If we use ldd again to look at the libfoo.so that we've made we will notice that it is missing a dependency on the rts library. This is problem that we've yet to sort out, so for the moment we can just add the dependency ourselves:
$ ghc --make -dynamic -shared -fPIC Foo.hs -o libfoo.so \
-lHSrts-ghc6.11 -optl-Wl,-rpath,/opt/ghc/lib/ghc-6.11/
This is a workaround because it requires us to know the version of the rts library at build time.
If your goal is to actually get something done (as opposed to just playing around with JNI) I suggest tackling this as a garden variety RPC problem and utilizing one of the many framework/protocols for it:
Protocol Buffers from Google
Thrift from Facebook
Avro (well this is mostly a wire protocol)
From what you are trying to do, Thrift might be your best bet since it describes a full client/server RPC stack but I'm pretty sure any of them would pretty much work over a simple socket.

Categories