I’m currently adding a new feature to LibFX, for which I create some custom collections akin to those from the Java Collections Framework. I went looking for tests I could run against them and was delighted to find out that Google’s Guava contains just what I need: a massive test suite which verifies every nook and cranny of my implementation for all collection interfaces from the JDK and Guava.
Let’s have a quick look at it.
Overview
This post will first show how to set up the project before looking at how to get started with the actual tests.
I created no dedicated examples but you can see how I used this in LibFX.
Setup
For this to work we need JUnit, the Guava-Testlib and a little boilerplate code.
Get JUnit
In case you’re not already using JUnit in your project, get it here. If you use Maven or Gradle:
Dependency Information for Maven
<dependency> <groupId>junit</groupId> <artifactId>junit</artifactId> <version>4.12</version> <scope>test</scope> </dependency>
Dependency Information for Gradle
testCompile 'junit:junit:4.12'
Get Guava
What we actually need is not Guava itself but the Guava-Testlib. You can download it from the central repository, which also contains the dependency information for different managers.
For your convenience:
Dependency Information for Maven
<dependency> <groupId>com.google.guava</groupId> <artifactId>guava-testlib</artifactId> <version>18.0</version> <scope>test</scope> </dependency>
Dependency Information for Gradle
testCompile 'com.google.guava:guava-testlib:18.0'
Write Some Boilerplate
Assume you want to write a MySet and the corresponding MySetTest.
Doing this the JUnit-3.8.x-way, create a method public static Test suite();. JUnit looks for this method and uses it to identify all tests which it will run for that class. Inside that method create a TestSuite and add the tests we’re going to write further down:
Boilerplate in ‘MySetTest’
public class MySetTest {
public static Test suite() {
return new MySetTest().allTests();
}
public Test allTests() {
TestSuite suite =
new TestSuite("package.name.of.MySetTest");
suite.addTest(testForOneToWayUseMySet());
suite.addTest(testForAnotherWayToUseMySet());
return suite;
}
}(I did not try to do this with JUnit 4’s annotations. If you did, ping me and I will include it here.)
With this boilerplate in place you can run this class with JUnit, e.g. from inside your IDE or on your CI server.
Test Your Implementations
Now that that’s done we can start actually creating tests for our implementations. Or, more precisely, tell Guava how to do that for us. This is a two part process: one creates a generator for the elements in the collection and the unit under test, the other uses one of Guava’s test suite builders to create a comprehensive set of tests tailored to the implementation.
We will continue to test an implementation of Set. Below we will see for what other interfaces test suites are available.
Generator For Elements And The Unit Under Test
The test suite builder requires you to give it a possibility to create the sample elements in the collection and instantiate your collection. To do this you have to implement the TestSetGenerator<E> (where E is the type of the elements).
This is straight forward with order(List<E>) being the only method which may require some thought. Note that contrary to the documentation the current versoin of the testlib (18.0) does call this method even when CollectionFeature.KNOWN_ORDER is not reported (see below for details about features). In my case it suffices to return the insertion order.
Test Suite Builder
Now this is were the real magic happens. You take your generator from above, pass it to the correct test suite builder, specify which features your collection has and it will create a tailored and comprehensive suite of tests:
Using the ‘SetTestSuiteBuilder’ to Create Tests
public Test testForOneToWayUseMySet() {
return SetTestSuiteBuilder
.using(new MySetGenerator())
.named("one way to use MySet")
.withFeatures(
CollectionSize.ANY,
CollectionFeature.ALLOWS_NULL_VALUES,
CollectionFeature.FAILS_FAST_ON_CONCURRENT_MODIFICATION,
CollectionFeature.SUPPORTS_ADD,
CollectionFeature.SUPPORTS_ITERATOR_REMOVE,
CollectionFeature.SUPPORTS_REMOVE,
)
.createTestSuite();
}Features
It is important to specify the correct features. Take a look at the two enums CollectionSize and CollectionFeatures to see which possibilities exist to describe your collection’s behavior.
Note that the created tests verify the features both ways! E.g. if ALLOWS_NULL_VALUES is left out, the builder will generate tests which verify that adding null to the collection throws a NullPointerException.
Suppressing Tests
By calling suppressing on the builder, you can specify test methods which will not be run. It seems to exist as a last resort when the features do not suffice to precisely specify the behavior. I did not use it.
Setup & Teardown
If you have to run code before or after each test, you can hand it as a Runnable to withSetUp or withTearDown, respectively (can both be called on the builder).
Available Test Suites
Of course you can generate tests suites for other interfaces as well. A first glance yields these possibilities:
Java’s collections:
- Collection
- Iterator
- List
- Map
- NavigableMap
- NavigableSet
- Queue
- Set
- SortedMap
- SortedSet
Guava’s collections:
- BiMap
- ListMultimap
- Multimap
- Multiset
- SetMultimap
- SortedMultiset
- SortedSetMultimap
A type search for *TestSuiteBuilder (note the wildcard) yields some other builders. I did not investigate them but it is possible that those can be used to create tests for other cases.
In order to use these, simply implement the according Test...Generator and hand it to the respective ...TestSuiteBuilder.
Reflection
We have seen how to test collection implementations with Guava’s Testlib: how to include it and JUnit in our project, what boilerplate we need to make it run and an overview over the generator and test suite builder. The latter is where all the magic happens as it creates comprehensive tests, tailored to our description of our implementation and its features.
| Reference: | Test Collection Implementations with Guava from our JCG partner Nicolai Parlog at the CodeFx blog. |
