733 lines (469 sloc) 35.7 KB

Summary of changes in version 4.12


Pull request #611: Assert.assertNotEquals() for float parameters

Version 4.11 added Assert.assertEquals() for float parameters with a delta, and Assert.assertNotEquals(). This is the combination of those two features.

Pull request #632: Assert.assertArrayEquals() for boolean[] parameters.

Assert.assertArrayEquals() previously existed for all primitive array types, except boolean[]. This has now been added for boolean[].

Pull request #918: Avoid potentially expensive reflection-based loop in Assert.assertArrayEquals()

In the usual case, where the array elements are in fact exactly equal, the potentially expensive reflection-based loop to compare them is avoided by using Arrays.deepEquals() first. The exact comparison is only executed when deepEquals() returns false.

Command-line options

Pull request #647: Support command-line --filter param.

When running JUnit from the command line, a command-line parameter can be supplied using --filter, which supplies a filter that will restrict which tests and subtests from the rest of the command will be run. For example, this will run only the tests in ExampleTestSuite that are in categories Cat1 or Cat2:

java org.junit.runner.JUnitCore \
  --filter=org.junit.experimental.categories.IncludeCategories=pkg.of.Cat1,pkg.of.Cat2 \

In general, the argument to --filter should be ClassName=param, where ClassName names an implementation of FilterFactory, whose createFilter method will be called with an instance of FilterFactoryParams that contains "param", in order to return the filter to be applied.

Test Runners

Pull request #763: Allow custom test runners to create their own TestClasses and customize the scanning of annotations.

This introduces some extension points to ParentRunner to allow subclasses to control creation of the TestClass instance and to scan for annotations.

Pull request #817: Support for context hierarchies

The AnnotatedBuilder is a strategy for constructing runners for test classes that have been annotated with the @RunWith annotation. All tests within such a class will be executed using the runner that was specified within the annotation.

Prior to JUnit 4.12, this covered only the tests within the annotated test class. With 4.12, the AnnotationBuilder will also support inner member classes. If a custom test runner supports inner member classes (which JUnit does not support out-of-the-box), the member classes will inherit the runner from the enclosing class, e.g.:

public class MyTest {
    // some tests might go here

    public class MyMemberClass {
        public void thisTestRunsWith_MyRunner() {
            // some test logic

        // some more tests might go here

    public class AnotherMemberClass {
        // some tests might go here

        public class DeepInnerClass {
            public void thisTestRunsWith_AnotherRunner() {
                // some test logic

        public class DeepInheritedClass extends SuperTest {
            public void thisTestRunsWith_SuperRunner() {
                // some test logic

public class SuperTest {
    // some tests might go here

The key points to note here are:

  • If there is no @RunWith annotation, no runner will be created.
  • The resolve step is inside-out, e.g. the closest @RunWith annotation wins.
  • @RunWith annotations are inherited and work as if the class was annotated itself.
  • The default JUnit runner does not support inner member classes, so this is only valid for custom runners that support inner member classes.
  • Custom runners with support for inner classes may or may not support @RunWith annotations for member classes. Please refer to the custom runner documentation.

One example of a runner that makes use of this extension is the Hierarchical Context Runner (see

Pull request #716: Fix annotation collection from superclasses of JUnit3 tests.

Previously Description.getAnnotations() would always return an empty list for test* methods derived from superclasses.

Pull request #625 (commit 72af03c49f): Make RunNotifier code concurrent.

When running tests from multiple threads, JUnit will now call RunListener methods from multiple threads if the listener class is annotated with @RunListener.ThreadSafe. In addition, the code in RunNotifier has been modified to not use locks.

Pull request #684: Adding AnnotationValidator framework and validation checks for @Category.

This allows for validation to be added to annotations. Validators should extend AnnotationValidator and be attached to annotations with the @ValidateWith annotation. CategoryValidator extends AnnotationValidator and ensures that incompatible annotations (@BeforeClass, @AfterClass, @Before, @After) are not used in conjunction with @Category.

Exception Testing

Pull request #583: Pull request #720: Fix handling of AssertionError and AssumptionViolatedException in ExpectedException rule.

ExpectedException didn't handle AssertionErrors and AssumptionViolatedException well. This has been fixed. The new documentation explains the usage of ExpectedException for testing these exceptions. The two methods handleAssertionErrors() and handleAssumptionViolatedExceptions() are not needed anymore. If you have used them, just remove it and read ExpectedException's documentation.

Pull request #818: Pull request #993: External version of AssumptionViolatedException

In JUnit 4.11 and earlier, if you wanted to write a custom runner that handled AssumptionViolatedException or you needed to create an instance of AssumptionViolatedException directly, you needed to import an internal class (org.junit.internal.AssumptionViolatedException). Now you can import org.junit.AssumptionViolatedException (which extends org.junit.internal.AssumptionViolatedException).

The classes in Assume have been modified to throw org.junit.AssumptionViolatedException.

The constructors in the external AssumptionViolatedException are also simpler than the ones in the internal version. That being said, it's recommended that you create AssumptionViolatedException via the methods in Assume.

Pull request #985: Change AssumptionViolatedException to not set the cause to null; fixes issue #494

Previously, the AssumptionViolatedException constructors would explicitly set the cause to null (unless you use a constructor where you provide a Throwable, in which case it would set that as the cause). This prevented code directly creating the exception from setting a cause.

With this change, the cause is only set if you pass in a Throwable.

It's recommended that you create AssumptionViolatedException via the methods in Assume.

Pull request #542: Customized failure message for ExpectedException

ExpectedException now allows customization of the failure message when the test does not throw the expected exception. For example:

thrown.reportMissingExceptionWithMessage("FAIL: Expected exception to be thrown");

If a custom failure message is not provided, a default message is used.

Pull request #1013: Make ErrorCollector#checkSucceeds generic

The method ErrorCollector.checkSucceeds() is now generic. Previously, you could only pass in a Callable<Object> and it returned Object. You can now pass any Callable and the return type will match the type of the callable.

Timeout for Tests

See also Timeout for tests

Pull request #823: Throw TestFailedOnTimeoutException instead of plain Exception on timeout

When a test times out, a org.junit.runners.model.TestTimedOutException is now thrown instead of a plain java.lang.Exception.

Pull request #742: Pull request #986: Timeout exceptions now include stack trace from stuck thread (experimental)

Timeout exceptions try to determine if there is a child thread causing the problem, and if so its stack trace is included in the exception in addition to the one of the main thread. This feature must be enabled with the timeout rule by creating it through the new Timeout.builder() method:

public class HasGlobalTimeout {
    @Rule public final TestRule timeout = Timeout.builder()
            .withTimeout(10, TimeUnit.SECONDS)

    public void testInfiniteLoop() {
        for (;;) {

Pull request #544: New constructor and factories in Timeout

Timeout deprecated the old constructor Timeout(int millis). A new constructor is available: Timeout(long timeout, TimeUnit unit). It enables you to use different granularities of time units like NANOSECONDS, MICROSECONDS, MILLISECONDS, and SECONDS. Examples:

@Rule public final TestRule globalTimeout = new Timeout(50, TimeUnit.MILLISECONDS);
@Rule public final TestRule globalTimeout = new Timeout(10, TimeUnit.SECONDS);

and factory methods in Timeout:

@Rule public final TestRule globalTimeout = Timeout.millis(50);
@Rule public final TestRule globalTimeout = Timeout.seconds(10);

This usage avoids the truncation, which was the problem in the deprecated constructor Timeout(int millis) when casting long to int.

Pull request #549: fixes for #544 and #545

The Timeout rule applies the same timeout to all test methods in a class:

public class HasGlobalTimeout {
    public Timeout globalTimeout = new Timeout(10, TimeUnit.SECONDS);

    public void testInfiniteLoop() {
        for (;;) {

    public synchronized void testInterruptableLock() throws InterruptedException {

    public void testInterruptableIO() throws IOException {
        for (;;) {
            FileChannel channel = new RandomAccessFile(file, "rw").getChannel();

            // Interrupted thread closes channel and throws ClosedByInterruptException.

Each test is run in a new daemon thread. If the specified timeout elapses before the test completes, its execution is interrupted via Thread#interrupt(). This happens in interruptable I/O (operations throwing and java.nio.channels.ClosedByInterruptException), locks (package java.util.concurrent) and methods in java.lang.Object and java.lang.Thread throwing java.lang.InterruptedException.

Pull request #876: The timeout rule never times out if you pass in a timeout of zero.

Parameterized Tests

Pull request #702: Support more return types for the @Parameters method of the Parameterized runner

The return types Iterator<? extends Object>, Object[] and Object[][] are now supported on methods annotated with @Parameters. You don't have to wrap arrays with Iterables and single parameters with Object arrays.

Pull request #773: Allow configurable creation of child runners of parameterized suites

The factory for creating the Runner instance of a single set of parameters is now configurable. It can be specified by the @UseParametersRunnerFactory annotation.


Pull request #552: Pull request #937: Stopwatch rule

The Stopwatch Rule notifies one of its own protected methods of the time spent by a test. Override them to get the time in nanoseconds. For example, this class will keep logging the time spent by each passed, failed, skipped, and finished test:

public static class StopwatchTest {
    private static final Logger logger = Logger.getLogger("");

    private static void logInfo(String testName, String status, long nanos) {"Test %s %s, spent %d microseconds",
            testName, status, Stopwatch.toMicros(nanos)));

    public Stopwatch stopwatch = new Stopwatch() {
        protected void succeeded(long nanos, Description description) {
            logInfo(description.getMethodName(), "succeeded", nanos);

        protected void failed(long nanos, Throwable e, Description description) {
            logInfo(description.getMethodName(), "failed", nanos);

        protected void skipped(long nanos, AssumptionViolatedException e, Description description) {
            logInfo(description.getMethodName(), "skipped", nanos);

        protected void finished(long nanos, Description description) {
            logInfo(description.getMethodName(), "finished", nanos);

    public void succeeds() {

    public void fails() {

    public void skips() {

An example to assert running time:

public void performanceTest() throws InterruptedException {
    long delta = 30;
    assertEquals(300D, stopwatch.runtime(MILLISECONDS), delta);
    assertEquals(800D, stopwatch.runtime(MILLISECONDS), delta);

Pull request #932: Allow static @Rules also annotated with @ClassRule

JUnit 4.11 introduced restrictions requiring @Rule members to be non-static and @ClassRule members to be static. These restrictions have been relaxed slightly, in that a static member annotated with both @Rule and @ClassRule is now considered valid. This means a single rule may be used to perform actions both before/after a class (e.g. setup/tear down an external resource) and between tests (e.g. reset the external resource), without the need for any workarounds mentioned in issue #793.

Note that a non-static @ClassRule annotated member is still considered invalid, even if annotated with @Rule.

public class CommonRuleTest {
    public static MySetupResetAndTearDownRule rule = new MySetupResetAndTearDownRule();

Be warned that if you have static methods or fields annotated with @Rule you will not be able to run your test methods in parallel.

Pull request #956: DisableOnDebug rule

The DisableOnDebug rule allows users to disable other rules when the JVM is launched in debug mode. Prior to this feature the common approach to disable rules that make debugging difficult was to comment them out and remember to revert the change. When using this feature users no longer have to modify their test code nor do they need to remember to revert changes.

This rule is particularly useful in combination with the Timeout rule.

public DisableOnDebug timeout = new DisableOnDebug(Timeout.seconds(1));

See the Javadoc for more detail and limitations. Related to

Pull request #974: Updated TemporaryFolder.newFolder() to give an error message if a path contains a slash.

If you call TemporaryFolder.newFolder("foo/bar") in JUnit 4.10 the method returns a File object for the new folder but actually fails to create it. That is contrary to the expected behaviour of the method which is to actually create the folder. In JUnit 4.11 the same call throws an exception. Nowhere in the documentation does it explain that the String(s) passed to that method can only be single path components.

With this fix, folder names are validated to contain single path name. If the folder name consists of multiple path names, an exception is thrown stating that usage of multiple path components in a string containing folder name is disallowed.

Pull request #1015: Methods annotated with Rule can return a MethodRule.

Methods annotated with @Rule can now return either a TestRule (or subclass) or a MethodRule (or subclass).

Prior to this change, all public methods annotated with @Rule were called, but the return value was ignored if it could not be assigned to a TestRule. After this change, the method is only called if the return type could be assigned to TestRule or MethodRule. For methods annotated with @Rule that return other values, see the notes for pull request #1020.

Pull request #1020: Added validation that @ClassRule should only be implementation of TestRule.

Prior to this change, fields annotated with @ClassRule that did not have a type of TestRule (or a class that implements TestRule) were ignored. With this change, the test will fail with a validation error.

Prior to this change, methods annotated with @ClassRule that did specify a return type of TestRule(or a class that implements TestRule) were ignored. With this change, the test will fail with a validation error.

Pull request #1021: JavaDoc of TemporaryFolder: folder not guaranteed to be deleted.

Adjusted JavaDoc of TemporaryFolder to reflect that temporary folders are not guaranteed to be deleted.


Pull request #529: @DataPoints-annotated methods can now yield null values

Up until JUnit 4.11 a @DataPoints-annotated array field could contain null values, but the array returned by a @DataPoints-annotated method could not. This asymmetry has been resolved: both can now provide a null data point.

Pull request #572: Ensuring no-generic-type-parms validator called/tested for theories

The Theories runner now disallows Theory methods with parameters that have "unresolved" generic type parameters (e.g. List<T> where T is a type variable). It is exceedingly difficult for the DataPoint(s) scraper or other ParameterSuppliers to correctly decide values that can legitimately be assigned to such parameters in a type-safe way, so JUnit now disallows them altogether. Theory parameters such as List<String> and Iterable<? extends Number> are still allowed.

The machinery to perform this validation was in the code base for some time, but not used. It now is used.

junit.contrib's rendition of theories performs the same validation.

Pull request #607: Improving theory failure messages

Theory failure messages previously were of the form: ParameterizedAssertionError: theoryTest(badDatapoint, allValues[1], otherVar), where allValues, badDatapoint and otherVar were the variables the datapoints was sourced from. These messages are now of the form:

ParameterizedAssertionError: theoryTest(null <from badDatapoint>, "good value" <from allValues[1]>, 
                                  [toString() threw RuntimeException: Error message] <from otherVar>)

Pull request #601: Allow use of Assume in tests run by Theories runner

If, in a theory, all parameters were "assumed" away, the Theories runner would properly fail, informing you that no parameters were found to actually test something. However, if you had another method in that same class, that was not a theory (annotated with @Test only,) you could not use Assume in that test. Now, the Theories runner will verify the method is annotated with @Theory before failing due to no parameters being found.

public class TheoriesAndTestsTogether {
    public static Object o;

    public void theory(Object o) {
        // this will still fail: java.lang.AssertionError: Never found parameters that satisfied method assumptions.

    public void test() {
        // this will no longer fail

Pull request #623: Ensure data points array fields and methods are public and static in Theory classes.

Previously if a data points array field or method was non-static or non-public it would be silently ignored and the data points not used. Now the Theories runner verifies that all @DataPoint or @DataPoints annotated fields or methods in classes are both public and static, and such classes will fail to run with InitializationErrors if they are not.

Pull request #621: Added mechanism for matching specific data points in theories to specific parameters, by naming data points.

@DataPoints fields or methods can now be given (one or more) names in the annotation, and @Theory method parameters can be annotated with @FromDataPoints(name), to limit the data points considered for that parameter to only the data points with that name:

public static String[] unnamed = new String[] { ... };

public static String[] regexStrings = new String[] { ... };

@DataPoints({"forMatching", "alphanumeric"})
public static String[] testStrings = new String[] { ... }; 

public void stringTheory(String param) {
    // This will be called with every value in 'regexStrings',
    // 'testStrings' and 'unnamed'.

public void regexTheory(@FromDataPoints("regexes") String regex,
                        @FromDataPoints("forMatching") String value) {
    // This will be called with only the values in 'regexStrings' as 
    // regex, only the values in 'testStrings' as value, and none 
    // of the values in 'unnamed'.

Pull request #654: Auto-generation of enum and boolean data points

Any theory method parameters with boolean or enum types that cannot be supplied with values by any other sources will be automatically supplied with default values: true and false, or every value of the given enum. If other explicitly defined values are available (e.g. from a specified ParameterSupplier or some DataPoints method in the theory class), only those explicitly defined values will be used.

Pull request #651: Improvements to Theory parameter and DataPoint type matching

  • Validity of DataPoints for theory parameters for all field data points and multi-valued method data points (i.e. not single-valued method data points) is now done on runtime type, not field/method return type (previously this was the case for multi-valued array methods only).

  • Validity of DataPoints for theory parameters for all data points now correctly handles boxing and unboxing for primitive and wrapper types; e.g. int values will be considered for theory parameters that are Integer assignable, and vice versa.

Pull request #639: Failing theory datapoint methods now cause theory test failures

Previously @DataPoint(s) methods that threw exceptions were quietly ignored and if another DataPoint source was available then those values alone were used, leaving the theory passing using only a subset of the (presumably) intended input values. Now, any data point method failures during invocation of a theory will cause the theory being tested to fail immediately.

This is a non-backward-compatible change, and could potentially break theory tests that depended on failing methods. If that was desired behavior, then the expected exceptions can instead be specifically ignored using the new ignoredExceptions array attribute on @DataPoint and @DataPoints methods. Adding an exception to this ignoredExceptions array will stop theory methods from failing if the given exception, or subclasses of it, are thrown in the annotated method. This attribute has no effect on data point fields.

Pull request #658: Iterables can now be used as data points

Previously, when building sets of data points for theory parameters, the only valid multi-valued @DataPoints types were arrays. This has now been extended to also take parameters from Iterable @DataPoints methods and fields.


Pull request #566: Enables inheritance on Category by adding @Inherited

@interface Category now is annotated with @Inherited itself. This enables inheritance of categories from ancestors (e.g. abstract test-classes). Note that you are able to "overwrite" @Category on inheritors and that this has no effect on method-level categories (see @Inherited).

Pull request #503: Configurable Categories

From a given set of test classes, the Categories runner runs only the classes and methods that are annotated with either the category given with the @IncludeCategory annotation, or a subtype of that category. Either classes or interfaces can be used as categories. Subtyping works, so if you say @IncludeCategory(SuperClass.class), a test marked @Category({SubClass.class}) will be run.

You can also exclude categories by using the @ExcludeCategory annotation; see SlowTestSuiteWithoutFast.

The suite FastOrSmokeTestSuite is an example to run multiple categories.

To execute tests which match all categories, use matchAny = false in annotations. See FastAndSmokeTestSuite.


public static interface FastTests { /* category marker */ }
public static interface SlowTests { /* category marker */ }
public static interface SmokeTests { /* category marker */ }

public static class A {
    public void a() {

    public void b() {

    @Category({FastTests.class, SmokeTests.class})
    public void c() {

@Category({SlowTests.class, FastTests.class})
public static class B {
    public void d() {

@Suite.SuiteClasses({A.class, B.class})
public static class SlowTestSuite {
    // Will run A.b and B.d, but not A.a and A.c

@Categories.IncludeCategory({FastTests.class, SmokeTests.class})
@Suite.SuiteClasses({A.class, B.class})
public static class FastOrSmokeTestSuite {
    // Will run A.c and B.d, but not A.b because it is not any of FastTests or SmokeTests

@Categories.IncludeCategory(value = {FastTests.class, SmokeTests.class}, matchAny = false)
@Suite.SuiteClasses({A.class, B.class})
public static class FastAndSmokeTestSuite {
    // Will run only A.c => match both FastTests AND SmokeTests

@Suite.SuiteClasses({A.class, B.class}) // Note that Categories is a kind of Suite
public class SlowTestSuiteWithoutFast {
    // Will run A.b, but not A.a, A.c or B.d

Use with Maven

Pull request #879: Add the default 'Implementation-*' headers to the manifest

The default Maven-style 'Implementation-*' headers are now present in the manifest of junit.jar. Example:

Implementation-Vendor: JUnit
Implementation-Title: JUnit
Implementation-Version: 4.12
Implementation-Vendor-Id: junit

Pull request #511: Maven project junit:junit:jar

How to install Maven

Download the Maven binary


If you are in the project root, extract the archive (tar xvzf apache-maven-3.0.4-bin.tar.gz). Create directory .m2 in your user home. Then the artifacts and plugins are stored in ~/.m2/repository. ( ~ stands for user home)

How to launch the build from the command line

Clone the project (git clone and navigate to the project root on your local system (cd junit). Clean the previous build in target directory, build the project, and install new artifacts in your local repository:

apache-maven-3.0.4/bin/mvn clean install

On Windows type the command apache-maven-3.0.4\bin\mvn clean install.

Set the environment variables M2_HOME and PATH when frequently building via command line mvn clean install.

How to install and build the Maven project in Eclipse

I made a clone of JUnit project from GitHub to local folder C:\cygwin\usr\local\etc\junit.

In menu go to File -> Import...

In the popup menu open section Maven, click on Existing Maven Projects and click on Next. In Import Maven Projects specify the project root, and next proceed further with installing maven support plugin in Eclipse.

You have created the Maven project, and now build the project.

In the Package Explorer click on pom.xml. In the menu Run -> Run As -> 2 Maven build... open the popup Edit Configuration and specify the build phase clean install in section Goals. Click on Run and build the project.

How to install and build the Maven project in IntelliJ IDEA

In IDEA menu create a new project File -> New Project....

Select Create Java project from existing sources, then click on Next and specify Project file location.

On the right-hand side is the Maven Projects tab. Click on + and add pom.xml into the project. Then click on the icon Maven Settings, and set Maven home directory as the location of extracted Maven archive on your system. Click on the green triangle and launch the build.

See the IntelliJ IDEA Web help

How to install the Maven project with documentation

Use the profile generate-docs to build sources.jar and javadoc.jar. Building Maven site is not yeat supported.

Example: mvn -Pgenerate-docs install

How to activate and deactivate Maven profiles in Integrated Development Environments:

In Eclipse, from the main menu navigate to Run -> Run As -> 2 Maven build..., open the popup Edit Configuration and specify the profiles.

Follow this link for IntelliJ IDEA:


Pull request #776: Add support for Travis CI

Travis CI is a free CI server for public Github repositories. Every pull request is run by Travis CI and Github's web interface shows the CI result for each pull request. Every user can use Travis CI for testing her branches, too.

Pull request #921: Apply Google Code Style

JUnit is now using the well documented Google Code Style

Pull request #939 Renamed license file

While using JUnit in Android apps, if any other referenced library has a file named LICENSE.txt, the APK generation failed with the following error -

Error generating final archive: Found duplicate file for APK: LICENSE.txt

To avoid this, the license file has been renamed to LICENSE-junit.txt

Pull request #962: Do not include thread start time in test timeout measurements.

The time it takes to start a thread can be surprisingly large. Especially in virtualized cloud environments where noisy neighbours. This change reduces the probability of non-deterministic failures of tests with timeouts (@Test(timeout=…)) by not beginning the timeout clock until we have observed the starting of the task thread – the thread that runs the actual test. This should make tests with small timeout values more reliable in general, and especially in cloud CI environments.

Fixes to issues introduced in JUnit 4.12

The following section lists fixes to problems introduced in the first release candidates for JUnit 4.12. You can ignore this section if you are trying to understand the changes between 4.11 and 4.12.

Pull request #961: Restore field names with f prefix.

In order to make the JUnit code more consistent with current coding practices, we changed a number of field names to not start with the prefix "f". Unfortunately, at least one IDE referenced a private field via reflection. This change reverts the field names for fields known to be read via reflection.

Pull request #988: Revert "Delete classes that are deprecated for six years."

In 745ca05 we removed classes that were deprecated for many releases. There was some concern that people might not expect classes to be removed in a 4.x release. Even though we are not aware of any problems from the deletion, we decided to add them back.

These classes may be removed in JUnit 5.0 or later.

Pull request #989: Add JUnitSystem.exit() back.

In 917a88f the exit() method in JUnit was removed. This caused problems for at least one user. Even though this class is in an internal package, we decided to add it back, and deprecated it.

This method may be removed in JUnit 5.0 or later.

Pull request #994: Pull request #1000: Ensure serialization compatibility where possible.

JUnit 4.12 RC1 introduced serilization incompatibilities with some of the classes. For example, these pre-release versions of JUnit could not read instances of Result that were serialized in JUnit 4.11 and earlier. These changes fix that problem.