Friday, December 15, 2006

fix svn tagging in sourceforge

I have been having tagging problem with Subversion in Sourceforge. It took me several days to realize that this is not going to be resolved by itself. Once again, Google helped me found the solution:

You need to point your working directory to a different repository URL, through command like:

svn switch --relocate https://svn.sourceforge.net/svnroot/cotta/trunk https://cotta.svn.sourceforge.net/svnroot/cotta/trunk

You just need to replace 'cotta' to the unix name of your project.

Monday, December 04, 2006

So Much for Java Generics

I always thought that Java generics is supported through compiler is a bit weird, and after 5 hours of email and spike, I finally know why.

Don't get me wrong, I love generics. Even though with TDD I can catch my mistakes most of the time, but with generics I can see my mistakes even before I run the tests. I won't say no to that.

Anyway, it all started with me starting a prove-of-concept project with another friend. So I am setting up libraries and figured I'll use jMock's constraint framework, after I convert it to support generics. Hence started my journey.

I checked out jmock code and was surprised to find out that there is a jmock2 directory. By the document that has been written, it is a way to stay with jMock style but don't lose the type safety. Cool! What is more, the constraint framework has been extracted into something called hamcrest.

Through email, Nat pointed me to the project in sourceforge (duh): http://sourceforge.net/projects/hamcrest. It looks very promising, even have the adapter for JUnit3, JUnit 4, and TestNG. It is apparent the intention of this project.

After I calmed down with all the excitement, I noticed one thing strange: The Matcher interface, the one interface that matching framework is evaluating matching logic, takes the "Object" type, rather than the generic type. What this means is that all the matching instance will have to do a type cast.

"That was silly", thought I, "Surely they know better than this". So I modified the interface in hamcrest, ran all the tests and proudly submitted my patch to Joe.

Turns out that the issue is not hamcrest itself, but jMock.

If you have two methods of the same name that you want to mock, say
public void test(int value);
public void test(String value);

And you provide the expectation
mock.expect("test").with(eq(1))

Upon the invocation of a.test("arg"), it will try to find the match by checking the matcher returned by eq(1) as well as that returned by eq("arg"). When comparing to eq(1) it will fail because from the instance there is no way you can figure out what the generic type is, and invoking by force will cause a class cast exception because there is a class cast injected by the compiler.

So in the above invocation, instead of having something like "unexpected invocation error", you end up with a weird ClassCastException error.

Bloody hell.