commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From John Bollinger <thinma...@yahoo.com>
Subject Re: [all] Testing interface implementations
Date Sun, 31 May 2009 02:50:21 GMT



Emmanuel Bourg <ebourg@apache.org> wrote:
> In Commons CLI there are 3 parsers implementing the same interface, and soon 
> another implementation will be added. There is an abstract test case with the 
> test methods, and a concrete subclass for each parser. The concrete test case 
> instantiates the parser and disables some tests by overwriting them if they are 
> known to fail.
> 
> The abstract test is growing and becomes difficult to manage. I may split it 
> into smaller classes grouping the tests by category, but I don't want to have 4 
> subclasses for every abstract class. This is where multi inheritance would be 
> useful.
> 
> Anyone knows a trick or a new test frameworks that may help testing the 
> implementations of an interface?

The part about "The concrete test case [...] disables some tests [...] if they are
known to fail" seems a bit dodgy to me from a design perspective.  I offer as
evidence the very problem you describe.

I have not looked at the test code, but your plan to split up the abstract test
class is a good opportunity to solve this design problem.  Here is what I
would try to do (assuming JUnit 3.x; otherwise YMMV):

1) Attempt to split the base class so that each resulting test class is either
100% go or 100% no-go for each parser.  I would think that adopting Phil's
"semantic integrity" idea ought to do the job*, though I don't see a need for
the test classes to form a hierarchy**.

* If the known failures that are hidden by the current scheme are not directly
related to semantic considerations then why can't the tests be rewritten
so that all the implementations pass?

2) Instead of creating subclasses for (test class, implementation) pairs,
create a parser factory for each parser implementation, and give each test
class a list of the parser factories it should test.  IOW, use composition
instead of inheritance.  This isn't all that different from what you have
already, where the most important function of each concrete test class is
to function as a parser factory.

3) In implementing (2), build dynamic TestSuites in order to provide a
separate test for each (test method, parser implementation) pair.  This
requires a static suite() method and puts much more burden on it than
in most test classes, but that is the price you pay for splitting up the
tests without either reducing test granularity or creating n * m
subclasses**.

Here's a schematic example:

public class CommonParserTests /* no need to extend TestCase */ {

    private final ParserFactory factory;

    public CommonParserTests(ParserFactory factory) {
        this.factory = factory;
    }

    public void testDontHaveACow() {
        CommandLineParser parser = factory.createParser();

        CommandLine line = parser.parse(new Options(),
                new String[] {"Don't", "have", "a", "cow,", "man!"});

        // assertXXX...
    }

    // additional test methods as needed...

    public static Test suite() {
        TestSuite suite = new TestSuite();

        for (ParserFactory factory : new ParserFactory[] {
                new BasicParser(), new GnuParser(), new PosixParser()}) {
            CommonParserTests tests = new CommonParserTests(factory);

            suite.addTest(new TestCase("Don't have a cow, man!") {
                protected void runTest() {
                    tests.testDontHaveACow();
                }
            });

            // additional inner TestCases, one for each test method
        }

        return suite;
    }
}


** Phil's suggestion of arranging the abstract test classes as a
hierarchy provides a simpler way of avoiding n * m concrete subclasses,
PROVIDED THAT there are no tests that crosscut part of the hierarchy.
That is, tests that want to appear in at least two, but not all, of the
abstract test classes don't fit well in an hierarchical scheme.


Regards,

John


      


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Mime
View raw message