Skip to content

sociomantic-tsunami/turtle

Repository files navigation

Turtle's versioning follows Neptune.

This means that the major version is increased for breaking changes, the minor version is increased for feature releases, and the patch version is increased for bug fixes that don't cause breaking changes.

  • Major branch development period: 3 months
  • Maintained minor versions: 1 most recent
Major Initial release date Supported until
v10.x.x v10.0.0: 08/04/2019 TBD

turtle is a utility library intended to help with the creation of application level black box tests. It consists of several major feature blocks:

  • Spawning a tested application as a separate external process
  • Automated facility for finding and running a test case
  • Creation of a temporary sandbox for tested application

If needed these features can be used independently, but the recommended approach is to base your test suite on turtle.runner.Runner which combines all the features and avoids annoying boilerplate code.

Refer to example projects for a documented example on how to define a test suite based on turtle.runner.Runner.

Imagine you have a project called water and you want to add some higher level tests based on turtle to it. This is suggested order of actions:

The recommended practice is to have a main test runner with the same name as the tested application. Thus, for this example of the water app, the matching runner will be placed in test/water/main.d. Makd will automatically compile it and put the resulting binary in build/last/tmp/test-water.

Create a test runner class which inherits from turtle.runner.Runner.TurtleRunner, specifying the tested application kind as a template argument:

class WaterTestRunner : TurtleRunner!(TestedAppKind.Daemon)
{
    ...
}

There are two mains supported kinds of tested applications - Daemon and CLI. The former is for a persistent application which keeps running in the background all the time while different test cases are executed and only exits when all tests are finished. The latter is for a short-lived CLI application which gets executed each time a test case is run, and after it has terminated, the console output is verified.

To configure your test runner, you may need to override several TurtleRunner methods. configureTestedApplication, prepare and reset are the ones that are almost always needed.

TurtleRunner will automatically find all classes derived from TestCase which are defined in the compiled modules and create them using Object.create. The example projects show some example test cases - you have to override description() to define any metadata (i.e. a test name), run() to define actual testing sequence and optionally prepare() to set up some data (commonly used if you have your own custom test case base class).

Any unhandled exception within run() will be considered a test failure - it is recommended to use the same ocean.core.Test function as you do with unit tests.

Usually any test case looks like a sequence of these actions:

  1. Prepare some data in the mock environment or filesystem.
  2. Wait for the tested application to process it (or run the tested application if it is a CLI one),
  3. Verify that the tested application has made expected changes in the mock environment or filesystem.

Other tests (which are generally uncommon) may verify the console output of the tested application or any files that it may generate. It is advisable to try and make test cases as small and straightforward as possible. Ideally, if a test case fails, the reason for the failure should be apparent even for someone who isn't very familiar with the application. When it comes to testing, clarity regarding the verified scenario is even more important than DRY. Avoid complicated class hierarchies and prefer writing the test code in a verbose and "dumb" manner if that helps to make the intention clearer.

It is perfectly fine for a project to even have hundreds of classes derived from TestCase if needed.

It is important to make sure that all your modules which define test cases are imported from a module which defines test runner. Suggested layout to make it simple:

test/
    water/
        cases/
            basic.d
            complex.d
            regressions.d
            all.d
        main.d

Then make sure all.d imports all other modules:

module test.water.cases.all;
// public import is not necessary, runtime reflection ignores
// protection attributes
import test.water.cases.basic;
import test.water.cases.complex;
import test.water.cases.regressions;

And import it from main.d / runner module:

module test.water.main;
import turtle.runner.Runner;
import test.water.cases.all;

class MyTurtleTests : TurtleRunner!(TestedAppKind.Daemon)

This way you can add new test at any time without ever having to modify a module with the test runner (and only having to modify all.d if you add a new module in test.water.cases).

If you use custom abstract base classes derived from TestCase it is important to ensure that they won't be used by TurtleRunner as an actual test case (trying to create an instance of such a class via runtime reflection will result in a mysterious crash).

When all test cases are put into a dedicated package as suggested above, it is trivial to tell TurtleRunner to only search for test cases there:

class MyTurtleTests : TurtleRunner!(TestedAppKind.Daemon)
{
    this ( )
    {
        // second argument is the package name to use
        super("appbinary", "test.water.cases");
    }

This allows for putting an abstract base class for custom test cases anywhere other than in the test.water package.

Sometimes it is very hard or even impossible to statically define a dedicated class for each test. One common case is automated generation of every single combination from a test matrix, for example, testing a bunch of scenarios with different starting data.

Turtle supports a special kind of TestCase which is defined in the same module (turtle.TestCase) and is called MultiTestCase. It is identical to the plain test case but has a default empty non-abstract implementation of run() and defines a new abstract method TestCase[] getNestedCases().

TurtleRunner recognizes MultiTestCase as a special base class and will recursively run all tests returned by getNestedCases() in the same way as it processes all tests found by runtime reflection. This means that test cases returned by getNestedCases() can in turn also be MultiTestCase.

Note that it is recommended to only use this feature if you have to generate tests in an automated manner and not to define manual nested hierarchies. This makes adding new tests more error-prone (easy to add a new test and forget to add it to the manually maintained list).