I'd like to suggest that we refactor our current test source set, which contains both unit, integration and distributed tests, into distinct source sets, test, integrationTest, distributedTest. These source sets would replace the use of the primary categories UnitTest, IntegrationTest and DistributedTest.
The catalyst for this change is an issue that Gradle's test runner doesn't pre-filter categories when launching tests, so if the tests are launched in separate JVMs or Docker containers, like :distributeTest task, the cost of spinning up those resources is realized only to immediately exit without running any test for all test classes in the module. Switching to separate source sets for each category would remove the need to filter on category and only tests in the corresponding source set would get executed in their external JVM or Docker container. An example of this issue is geode-junit:distributedTest task, which forks all test classes in separate JVMs but never actually runs any tests since there are no DistributedTest tagged tests. The secondary effect is a way too isolate dependencies in each of the source sets. Unit tests in the test set would not have dependencies need for integration tests or distributed test so that if you accidentally tried to import classes from those frameworks you would get a compiler failure. Likewise, integration tests would not include distributed test framework dependencies. Any shared test classes like mock, dummies, fakes, etc. could be shared in a commonTest source set, but would not contain any tests itself. The proposed structure would look like this: test/ - only contains unit tests. integrationTest/ - only contains integration style tests. distributedTest/ - only includes DUnit based tests. commonTest/ - includes commonly shared classes between each test category. Does not contain any classes. Thoughts? -Jake
