I'm trying to implement a reusable functional-like version of the double checked lock (DCL) pattern in Java.
Indeed, there are many known problems with the DCL pattern in Java, like those ones. So I'm trying to check if the solution I develop has any flaws.
Here is the DCL executor code:
public class DoubleCheckedLockExecutor {
public <T> T getOrCreate(Supplier<T> supplier, Supplier<T> builder, Consumer<T> consumer, Predicate<T> build) {
if (build.test(supplier.get())) {
synchronized (this) {
if (build.test(supplier.get())) {
consumer.accept(builder.get());
}
}
}
return supplier.get();
}
}
And here a Singleton class that uses it:
public class Singleton {
private static Singleton instance = null;
private static final AtomicInteger instanceCount = new AtomicInteger();
private static final DoubleCheckedLockExecutor dclExec = new DoubleCheckedLockExecutor();
private Singleton() {
instanceCount.incrementAndGet();
}
public static Singleton getInstance() {
return dclExec.getOrCreate(() -> instance, Singleton::new, s -> instance = s, s -> s == null);
}
public static int getInstanceCount() {
return instanceCount.get();
}
}
And finally some test code:
#Test
public final void testGetOrCreate() {
int calls = 1000;
ExecutorService executor = Executors.newFixedThreadPool(Runtime.getRuntime().availableProcessors());
try {
for (int i = 0; i < calls; i++) {
executor.execute(() -> Singleton.getInstance());
}
} catch (Exception e) {
e.printStackTrace();
}
executor.shutdown();
while (!executor.isTerminated()) {
}
assertEquals(1, Singleton.getInstanceCount());
}
All tests and analysis I did showed no problems (duplicated instances, for example). But multi-thread and concurrency test is not such easy task for me. So could you guys help me out with this? Could I say this implementation is thread safe and produce the expected results?
I have read this post: Is there a way in JMockit to call the original method from a mocked method?
but the recommend solution throws a NPE. Here is my source
static Map<String, Boolean> detectDeadlocks(int timeInSeconds) {
final Map<String, Boolean> deadlockMap = new LinkedHashMap<>();
new Timer().schedule(new TimerTask() {
#Override
public void run() {
// I want to check if the method is run.
**deadlockMap.put("deadlock", isDeadlockAfterPeriod());**
}
}, timeInSeconds * 1000);
return deadlockMap;
}
I want to be able to invoke isDeadlockAfterPeriod in my unit test. This is a static method that I have mocked in my unit tests.
My unit test code
#Test
public void testDetectDeadlocks() throws Exception {
new Expectations(){
{
// Called from inside the TimerTask.
ClassUnderTest.isDeadlockAfterPeriod();
result = false;
}
};
TimerMockUp tmu = new TimerMockUp();
Deadlocks.detectDeadlocks(0);
Assert.assertEquals(1, tmu.scheduleCount);
}
class TimerMockUp extends MockUp<Timer> {
int scheduleCount = 0;
#Mock
public void $init() {}
#Mock
public void schedule(Invocation invocation, TimerTask task, long delay) {
scheduleCount ++;
invocation.proceed(); // Trying to call the real method, but this throws a NPE.
}
}
Error stack trace is seen with JUnit in Eclipse.
java.lang.NullPointerException
at com.myproject.test.DeadlocksTest$TimerMockUp.schedule(DeadlocksTest.java:78)
at com.myproject.test.Deadlocks.detectDeadlocks(Deadlocks.java:41)
at com.myproject.test.DeadlocksTest.testDetectDeadlocks(DeadlocksTest.java:86)
Your problem is that you are also faking the Timer's constructor (and not only the schedule method).
By doing so, you are preventing the correct initialization of the Timer, and as you are then using its real implementation, it fails to do so.
Specifically (with the sources I have), you are preventing the initialization of its queue field, which is used on its mainLoop() method (the one that will call to your TimerTask.run()).
Also, you need to do partial mocking of Deadlocks class, as I understand that isDeadlockAfterPeriod is also a static method for the said class.
I'll leave you here a working example:
Deadlocks.class
public class Deadlocks {
public static Map<String, Boolean> detectDeadlocks(int timeInSeconds) {
final Map<String, Boolean> deadlockMap = new LinkedHashMap<>();
new Timer()// this will be the real constructor
.schedule( // this will be first mocked, then really executed
new TimerTask() {
#Override
public void run() {
deadlockMap.put("deadlock", isDeadlockAfterPeriod()); // this will put false after the mock is called
}
}, timeInSeconds * 1000);
return deadlockMap;
}
public static Boolean isDeadlockAfterPeriod() {
return true; // this, we will mock it
}
}
Test class
#RunWith(JMockit.class)
public TestClass{
#Test
public void testDetectDeadlocks() throws Exception {
new Expectations(Deadlocks.class){ // do partial mocking of the Deadlock class
{
// Called from inside the TimerTask.
Deadlocks.isDeadlockAfterPeriod();
result = false;
}
};
// prepare the fake
TimerMockUp tmu = new TimerMockUp();
// execute the code
Map<String, Boolean> result = Deadlocks.detectDeadlocks(0);
// assert results
assertThat(tmu.scheduleCount, is(1));
assertThat(result.size(), is(1));
assertThat(result.get("deadlock"), is(false));
}
class TimerMockUp extends MockUp<Timer> {
int scheduleCount = 0;
#Mock
public void schedule(Invocation invocation, TimerTask task, long delay) {
scheduleCount ++;
invocation.proceed();
}
}
}
In general, be very careful when faking constructors, as you may leave the instances in an inconsistent state.
I am testing a simple multithread code using EasyMock:
source code:
public class EasyMockTest {
ExecutorService executorService;
TestObject testObject;
public EasyMockTest(ExecutorService executorService, TestObject testObject)
{
this.executorService = executorService;
this.testObject = testObject;
}
public void test()
{
try
{
executorService.submit(() ->{
testObject.doSomething();
});
}
catch(RejectedExecutionException ex)
{
}
}
}
public class TestObject {
public void doSomething()
{
}
}
Test code with EasyMock:
public class EasyMockTest_test {
private TestObject testObject;
private ExecutorService executorService;
private EasyMockTest easyMockTest;
#Before
public void setUp()
{
executorService = new ThreadPoolExecutor(1, 1, 60, TimeUnit.SECONDS, new LinkedBlockingQueue<Runnable>(1));
testObject = EasyMock.createMock(TestObject.class);
easyMockTest = new EasyMockTest(executorService, testObject);
}
#Test
public void test_easyMockTest()
{
testObject.doSomething();
EasyMock.expectLastCall().andAnswer(new IAnswer<Void>(){
#Override
public Void answer() throws Throwable {
Thread.sleep(100);
return null;
}}).times(2);
EasyMock.replay(testObject);
easyMockTest.test();
easyMockTest.test();
easyMockTest.test();
EasyMock.verify(testObject);
}
}
I think in this case testObject.doSomething() should only be called twice. Since thread pool has one thread and queue size is one, and I let first two thread sleep. So when I submit three task, the third one should be rejected and the first two should be called. But when I run this code
there is error:
java.lang.AssertionError:
Expectation failure on verify:
TestObject.doSomething(): expected: 2, actual: 1
at org.easymock.internal.MocksControl.verify(MocksControl.java:225)
at org.easymock.EasyMock.verify(EasyMock.java:2007)
...
This means the method is only called once, which i can't understand.
I also tried comment Thread.sleep(100); this times actual calling times becomes 2, butI think it should be 3 since no thread is sleeping.
Then I tried move .times() position like this:
EasyMock.expectLastCall().times(2).andAnswer(new IAnswer<Void>(){
#Override
public Void answer() throws Throwable {
Thread.sleep(100);
return null;
}});
This time error becomes:
java.lang.AssertionError: Expectation failure on verify:
TestObject.doSomething(): expected: 3, actual: 2
Why result is expecting 3 when i give it 2?
Sorry I am not expert on EasyMock, really appreciate if someone can help.
Nothing is making sure your tasks are executed before the verify is reached. You need something to pace the execution.
This works:
#Test
public void test_easyMockTest() throws InterruptedException {
CountDownLatch latch = new CountDownLatch(3);
testObject.doSomething();
EasyMock.expectLastCall().andAnswer(new IAnswer<Void>(){
#Override
public Void answer() throws Throwable {
latch.countDown();
return null;
}}).times(2);
EasyMock.replay(testObject);
easyMockTest.test();
easyMockTest.test();
easyMockTest.test();
latch.await(1, TimeUnit.SECONDS);
EasyMock.verify(testObject);
}
Here I am assuming that you really want a RejectedExecutionException to be caught and ignored.
I want to check that a method is not run and tried to do it with an Expectation setting times = 0;, however I don't get the expected behaviour.
For example, the following test passes, although the Session#stop method is called, and the expectation has a times = 0; condition:
public static class Session {
public void stop() {}
}
public static class Whatever {
Session s = new Session();
public synchronized void method() {
s.stop();
}
}
#Test
public void testWhatever () throws Exception {
new Expectations(Session.class) {
#Mocked Session s;
{ s.stop(); times = 0; } //Session#stop must not be called
};
final Whatever w = new Whatever();
w.method(); // this method calls Session#stop => the test should fail...
// ... but it passes
}
Note: If I replace the code with { s.stop(); times = 1; }, the test passes too: I must be missing something obvious here...
The reason of the unexpected mocking behavior is that you inadvertently used partial mocking on an strictly mocked type. In this case, recording an expectation with times = <n> means that the first n matching invocations will be mocked, and after that any additional invocations will execute the original "unmocked" method. With regular mocking instead, you would get the expected behavior (ie, an UnexpectedInvocation getting thrown after n invocations).
The proper way to write the test is:
public static class Session { public void stop() {} }
public static class Whatever {
Session s = new Session();
public synchronized void method() { s.stop(); }
}
#Test
public void testWhatever ()
{
new Expectations() {
#Mocked Session s;
{ s.stop(); times = 0; }
};
final Whatever w = new Whatever();
w.method();
}
Alternatively, it can also be written with a verification block instead, which is usually better for situations like these:
#Test
public void testWhatever (#Mocked final Session s)
{
final Whatever w = new Whatever();
w.method();
new Verifications() {{ s.stop(); times = 0; }};
}
Related to this I had trouble with JMockit, times = 0 and the #Tested annotation.
With the #Tested annotation you still have a 'real' class, so when registering an Expectation or a Verification (even with times = 0) on this real class, JMockit tries to execute the method. Solution is to partially mock the class in the Expectations:
#Tested
Session s;
new Expectations(Session.class) {{
s.stop(); times = 0; } //Session#stop must not be called
};
This is the only way I found to use times=0 on methods from #Tested classes.
I found a workaround with the MockUp class - the test below fails as expected - I would still like to understand why the original approach was not working.
#Test
public void testWhatever () throws Exception {
new MockUp<Session>() {
#Mock
public void stop() {
fail("stop should not have been called");
}
};
final Whatever w = new Whatever();
w.method();
}
Try maxTimes instead, you also can reference stop() in a static way:
#Test
public void test(#Mocked Session mockSession){
final Whatever w = new Whatever();
w.method();
new Verifications(){
{
Session.stop();
maxTimes = 0;
}
};
}
From memory, something like
verify( s , times(0) ).stop();
will work. Trouble is, the Session in Whatever isn't your #Mock'ed one, but another object, so insert a
w.s = s;
just before w.method().
Cheers,
I've got a few methods that should call System.exit() on certain inputs. Unfortunately, testing these cases causes JUnit to terminate! Putting the method calls in a new Thread doesn't seem to help, since System.exit() terminates the JVM, not just the current thread. Are there any common patterns for dealing with this? For example, can I subsitute a stub for System.exit()?
[EDIT] The class in question is actually a command-line tool which I'm attempting to test inside JUnit. Maybe JUnit is simply not the right tool for the job? Suggestions for complementary regression testing tools are welcome (preferably something that integrates well with JUnit and EclEmma).
Indeed, Derkeiler.com suggests:
Why System.exit() ?
Instead of terminating with System.exit(whateverValue), why not throw an unchecked exception? In normal use it will drift all the way out to the JVM's last-ditch catcher and shut your script down (unless you decide to catch it somewhere along the way, which might be useful someday).
In the JUnit scenario it will be caught by the JUnit framework, which will report that
such-and-such test failed and move smoothly along to the next.
Prevent System.exit() to actually exit the JVM:
Try modifying the TestCase to run with a security manager that prevents calling System.exit, then catch the SecurityException.
public class NoExitTestCase extends TestCase
{
protected static class ExitException extends SecurityException
{
public final int status;
public ExitException(int status)
{
super("There is no escape!");
this.status = status;
}
}
private static class NoExitSecurityManager extends SecurityManager
{
#Override
public void checkPermission(Permission perm)
{
// allow anything.
}
#Override
public void checkPermission(Permission perm, Object context)
{
// allow anything.
}
#Override
public void checkExit(int status)
{
super.checkExit(status);
throw new ExitException(status);
}
}
#Override
protected void setUp() throws Exception
{
super.setUp();
System.setSecurityManager(new NoExitSecurityManager());
}
#Override
protected void tearDown() throws Exception
{
System.setSecurityManager(null); // or save and restore original
super.tearDown();
}
public void testNoExit() throws Exception
{
System.out.println("Printing works");
}
public void testExit() throws Exception
{
try
{
System.exit(42);
} catch (ExitException e)
{
assertEquals("Exit status", 42, e.status);
}
}
}
Update December 2012:
Will proposes in the comments using System Rules, a collection of JUnit(4.9+) rules for testing code which uses java.lang.System.
This was initially mentioned by Stefan Birkner in his answer in December 2011.
System.exit(…)
Use the ExpectedSystemExit rule to verify that System.exit(…) is called.
You could verify the exit status, too.
For instance:
public void MyTest {
#Rule
public final ExpectedSystemExit exit = ExpectedSystemExit.none();
#Test
public void noSystemExit() {
//passes
}
#Test
public void systemExitWithArbitraryStatusCode() {
exit.expectSystemExit();
System.exit(0);
}
#Test
public void systemExitWithSelectedStatusCode0() {
exit.expectSystemExitWithStatus(0);
System.exit(0);
}
}
The library System Lambda has a method catchSystemExit.With this rule you are able to test code, that calls System.exit(...):
public class MyTest {
#Test
public void systemExitWithArbitraryStatusCode() {
SystemLambda.catchSystemExit(() -> {
//the code under test, which calls System.exit(...);
});
}
#Test
public void systemExitWithSelectedStatusCode0() {
int status = SystemLambda.catchSystemExit(() -> {
//the code under test, which calls System.exit(0);
});
assertEquals(0, status);
}
}
For Java 5 to 7 the library System Rules has a JUnit rule called ExpectedSystemExit. With this rule you are able to test code, that calls System.exit(...):
public class MyTest {
#Rule
public final ExpectedSystemExit exit = ExpectedSystemExit.none();
#Test
public void systemExitWithArbitraryStatusCode() {
exit.expectSystemExit();
//the code under test, which calls System.exit(...);
}
#Test
public void systemExitWithSelectedStatusCode0() {
exit.expectSystemExitWithStatus(0);
//the code under test, which calls System.exit(0);
}
}
Full disclosure: I'm the author of both libraries.
How about injecting an "ExitManager" into this Methods:
public interface ExitManager {
void exit(int exitCode);
}
public class ExitManagerImpl implements ExitManager {
public void exit(int exitCode) {
System.exit(exitCode);
}
}
public class ExitManagerMock implements ExitManager {
public bool exitWasCalled;
public int exitCode;
public void exit(int exitCode) {
exitWasCalled = true;
this.exitCode = exitCode;
}
}
public class MethodsCallExit {
public void CallsExit(ExitManager exitManager) {
// whatever
if (foo) {
exitManager.exit(42);
}
// whatever
}
}
The production code uses the ExitManagerImpl and the test code uses ExitManagerMock and can check if exit() was called and with which exit code.
You actually can mock or stub out the System.exit method, in a JUnit test.
For example, using JMockit you could write (there are other ways as well):
#Test
public void mockSystemExit(#Mocked("exit") System mockSystem)
{
// Called by code under test:
System.exit(); // will not exit the program
}
EDIT: Alternative test (using latest JMockit API) which does not allow any code to run after a call to System.exit(n):
#Test(expected = EOFException.class)
public void checkingForSystemExitWhileNotAllowingCodeToContinueToRun() {
new Expectations(System.class) {{ System.exit(anyInt); result = new EOFException(); }};
// From the code under test:
System.exit(1);
System.out.println("This will never run (and not exit either)");
}
One trick we used in our code base was to have the call to System.exit() be encapsulated in a Runnable impl, which the method in question used by default. To unit test, we set a different mock Runnable. Something like this:
private static final Runnable DEFAULT_ACTION = new Runnable(){
public void run(){
System.exit(0);
}
};
public void foo(){
this.foo(DEFAULT_ACTION);
}
/* package-visible only for unit testing */
void foo(Runnable action){
// ...some stuff...
action.run();
}
...and the JUnit test method...
public void testFoo(){
final AtomicBoolean actionWasCalled = new AtomicBoolean(false);
fooObject.foo(new Runnable(){
public void run(){
actionWasCalled.set(true);
}
});
assertTrue(actionWasCalled.get());
}
I like some of the answers already given but I wanted to demonstrate a different technique that is often useful when getting legacy code under test. Given code like:
public class Foo {
public void bar(int i) {
if (i < 0) {
System.exit(i);
}
}
}
You can do a safe refactoring to create a method that wraps the System.exit call:
public class Foo {
public void bar(int i) {
if (i < 0) {
exit(i);
}
}
void exit(int i) {
System.exit(i);
}
}
Then you can create a fake for your test that overrides exit:
public class TestFoo extends TestCase {
public void testShouldExitWithNegativeNumbers() {
TestFoo foo = new TestFoo();
foo.bar(-1);
assertTrue(foo.exitCalled);
assertEquals(-1, foo.exitValue);
}
private class TestFoo extends Foo {
boolean exitCalled;
int exitValue;
void exit(int i) {
exitCalled = true;
exitValue = i;
}
}
This is a generic technique for substituting behavior for test cases, and I use it all the time when refactoring legacy code. It not usually where I'm going to leave thing, but an intermediate step to get the existing code under test.
For VonC's answer to run on JUnit 4, I've modified the code as follows
protected static class ExitException extends SecurityException {
private static final long serialVersionUID = -1982617086752946683L;
public final int status;
public ExitException(int status) {
super("There is no escape!");
this.status = status;
}
}
private static class NoExitSecurityManager extends SecurityManager {
#Override
public void checkPermission(Permission perm) {
// allow anything.
}
#Override
public void checkPermission(Permission perm, Object context) {
// allow anything.
}
#Override
public void checkExit(int status) {
super.checkExit(status);
throw new ExitException(status);
}
}
private SecurityManager securityManager;
#Before
public void setUp() {
securityManager = System.getSecurityManager();
System.setSecurityManager(new NoExitSecurityManager());
}
#After
public void tearDown() {
System.setSecurityManager(securityManager);
}
Create a mock-able class that wraps System.exit()
I agree with EricSchaefer. But if you use a good mocking framework like Mockito a simple concrete class is enough, no need for an interface and two implementations.
Stopping test execution on System.exit()
Problem:
// do thing1
if(someCondition) {
System.exit(1);
}
// do thing2
System.exit(0)
A mocked Sytem.exit() will not terminate execution. This is bad if you want to test that thing2 is not executed.
Solution:
You should refactor this code as suggested by martin:
// do thing1
if(someCondition) {
return 1;
}
// do thing2
return 0;
And do System.exit(status) in the calling function. This forces you to have all your System.exit()s in one place in or near main(). This is cleaner than calling System.exit() deep inside your logic.
Code
Wrapper:
public class SystemExit {
public void exit(int status) {
System.exit(status);
}
}
Main:
public class Main {
private final SystemExit systemExit;
Main(SystemExit systemExit) {
this.systemExit = systemExit;
}
public static void main(String[] args) {
SystemExit aSystemExit = new SystemExit();
Main main = new Main(aSystemExit);
main.executeAndExit(args);
}
void executeAndExit(String[] args) {
int status = execute(args);
systemExit.exit(status);
}
private int execute(String[] args) {
System.out.println("First argument:");
if (args.length == 0) {
return 1;
}
System.out.println(args[0]);
return 0;
}
}
Test:
public class MainTest {
private Main main;
private SystemExit systemExit;
#Before
public void setUp() {
systemExit = mock(SystemExit.class);
main = new Main(systemExit);
}
#Test
public void executeCallsSystemExit() {
String[] emptyArgs = {};
// test
main.executeAndExit(emptyArgs);
verify(systemExit).exit(1);
}
}
System Stubs - https://github.com/webcompere/system-stubs - is also able to solve this problem. It shares System Lambda's syntax for wrapping around code that we know will execute System.exit, but that can lead to odd effects when other code unexpectedly exits.
Via the JUnit 5 plugin, we can provide insurance that any exit will be converted to an exception:
#ExtendWith(SystemStubsExtension.class)
class SystemExitUseCase {
// the presence of this in the test means System.exit becomes an exception
#SystemStub
private SystemExit systemExit;
#Test
void doSomethingThatAccidentallyCallsSystemExit() {
// this test would have stopped the JVM, now it ends in `AbortExecutionException`
// System.exit(1);
}
#Test
void canCatchSystemExit() {
assertThatThrownBy(() -> System.exit(1))
.isInstanceOf(AbortExecutionException.class);
assertThat(systemExit.getExitCode()).isEqualTo(1);
}
}
Alternatively, the assertion-like static method can also be used:
assertThat(catchSystemExit(() -> {
//the code under test
System.exit(123);
})).isEqualTo(123);
A quick look at the api, shows that System.exit can throw an exception esp. if a securitymanager forbids the shutdown of the vm. Maybe a solution would be to install such a manager.
You can use the java SecurityManager to prevent the current thread from shutting down the Java VM. The following code should do what you want:
SecurityManager securityManager = new SecurityManager() {
public void checkPermission(Permission permission) {
if ("exitVM".equals(permission.getName())) {
throw new SecurityException("System.exit attempted and blocked.");
}
}
};
System.setSecurityManager(securityManager);
You can test System.exit(..) with replacing Runtime instance.
E.g. with TestNG + Mockito:
public class ConsoleTest {
/** Original runtime. */
private Runtime originalRuntime;
/** Mocked runtime. */
private Runtime spyRuntime;
#BeforeMethod
public void setUp() {
originalRuntime = Runtime.getRuntime();
spyRuntime = spy(originalRuntime);
// Replace original runtime with a spy (via reflection).
Utils.setField(Runtime.class, "currentRuntime", spyRuntime);
}
#AfterMethod
public void tearDown() {
// Recover original runtime.
Utils.setField(Runtime.class, "currentRuntime", originalRuntime);
}
#Test
public void testSystemExit() {
// Or anything you want as an answer.
doNothing().when(spyRuntime).exit(anyInt());
System.exit(1);
verify(spyRuntime).exit(1);
}
}
There are environments where the returned exit code is used by the calling program (such as ERRORLEVEL in MS Batch). We have tests around the main methods that do this in our code, and our approach has been to use a similar SecurityManager override as used in other tests here.
Last night I put together a small JAR using Junit #Rule annotations to hide the security manager code, as well as add expectations based on the expected return code. http://code.google.com/p/junitsystemrules/
Most solutions will
terminate the test (method, not the entire run) the moment System.exit() is called
ignore an already installed SecurityManager
Sometimes be quite specific to a test framework
restrict to be used at max once per test case
Thus, most solutions are not suited for situations where:
Verification of side-effects are to be performed after the call to System.exit()
An existing security manager is part of the testing.
A different test framework is used.
You want to have multiple verifications in a single test case. This may be strictly not recommended, but can be very convenient at times, especially in combination with assertAll(), for example.
I was not happy with the restrictions imposed by the existing solutions presented in the other answers, and thus came up with something on my own.
The following class provides a method assertExits(int expectedStatus, Executable executable) which asserts that System.exit() is called with a specified status value, and the test can continue after it. It works the same way as JUnit 5 assertThrows. It also respects an existing security manager.
There is one remaining problem: When the code under test installs a new security manager which completely replaces the security manager set by the test. All other SecurityManager-based solutions known to me suffer the same problem.
import java.security.Permission;
import static java.lang.System.getSecurityManager;
import static java.lang.System.setSecurityManager;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.fail;
public enum ExitAssertions {
;
public static <E extends Throwable> void assertExits(final int expectedStatus, final ThrowingExecutable<E> executable) throws E {
final SecurityManager originalSecurityManager = getSecurityManager();
setSecurityManager(new SecurityManager() {
#Override
public void checkPermission(final Permission perm) {
if (originalSecurityManager != null)
originalSecurityManager.checkPermission(perm);
}
#Override
public void checkPermission(final Permission perm, final Object context) {
if (originalSecurityManager != null)
originalSecurityManager.checkPermission(perm, context);
}
#Override
public void checkExit(final int status) {
super.checkExit(status);
throw new ExitException(status);
}
});
try {
executable.run();
fail("Expected System.exit(" + expectedStatus + ") to be called, but it wasn't called.");
} catch (final ExitException e) {
assertEquals(expectedStatus, e.status, "Wrong System.exit() status.");
} finally {
setSecurityManager(originalSecurityManager);
}
}
public interface ThrowingExecutable<E extends Throwable> {
void run() throws E;
}
private static class ExitException extends SecurityException {
final int status;
private ExitException(final int status) {
this.status = status;
}
}
}
You can use the class like this:
#Test
void example() {
assertExits(0, () -> System.exit(0)); // succeeds
assertExits(1, () -> System.exit(1)); // succeeds
assertExits(2, () -> System.exit(1)); // fails
}
The code can easily be ported to JUnit 4, TestNG, or any other framework, if necessary. The only framework-specific element is failing the test. This can easily be changed to something framework-independent (other than a Junit 4 Rule
There is room for improvement, for example, overloading assertExits() with customizable messages.
Use Runtime.exec(String command) to start JVM in a separate process.
There is a minor problem with the SecurityManager solution. Some methods, such as JFrame.exitOnClose, also call SecurityManager.checkExit. In my application, I didn't want that call to fail, so I used
Class[] stack = getClassContext();
if (stack[1] != JFrame.class && !okToExit) throw new ExitException();
super.checkExit(status);
A generally useful approach that can be used for unit and integration testing, is to have a package private (default access) mockable runner class that provides run() and exit() methods. These methods can be overridden by Mock or Fake test classes in the test modules.
The test class (JUnit or other) provides exceptions that the exit() method can throw in place of System.exit().
package mainmocked;
class MainRunner {
void run(final String[] args) {
new MainMocked().run(args);
}
void exit(final int status) {
System.exit(status);
}
}
the class with main() below, also has an altMain() to receive the mock or fake runner, when unit or integration testing:
package mainmocked;
public class MainMocked {
private static MainRunner runner = new MainRunner();
static void altMain(final String[] args, final MainRunner inRunner) {
runner = inRunner;
main(args);
}
public static void main(String[] args) {
try {
runner.run(args);
} catch (Throwable ex) {
// Log("error: ", ex);
runner.exit(1);
}
runner.exit(0);
} // main
public void run(String[] args) {
// do things ...
}
} // class
A simple mock (with Mockito) would be:
#Test
public void testAltMain() {
String[] args0 = {};
MainRunner mockRunner = mock(MainRunner.class);
MainMocked.altMain(args0, mockRunner);
verify(mockRunner).run(args0);
verify(mockRunner).exit(0);
}
A more complex test class would use a Fake, in which run() could do anything, and an Exception class to replace System.exit():
private class FakeRunnerRuns extends MainRunner {
#Override
void run(String[] args){
new MainMocked().run(args);
}
#Override
void exit(final int status) {
if (status == 0) {
throw new MyMockExitExceptionOK("exit(0) success");
}
else {
throw new MyMockExitExceptionFail("Unexpected Exception");
} // ok
} // exit
} // class
Another technique here is to introduce additional code into the (hopefully small number of) places where the logic does the System.exit(). This additional code then avoids doing the System.exit() when the logic is being called as part of unit test. For example, define a package private constant like TEST_MODE which is normally false. Then have the unit test code set this true and add logic to the code under test to check for that case and bypass the System.exit call and instead throw an exception that the unit test logic can catch. By the way, in 2021 and using something like spotbugs, this problem can manifest itself in the obscure error that "java.io.IOException: An existing connection was forcibly closed by the remote host".
Calling System.exit() is a bad practice, unless it's done inside a main(). These methods should be throwing an exception which, ultimately, is caught by your main(), who then calls System.exit with the appropriate code.