Re: [MO-java-dev] 0.9 porting roadblock

[Barry K. <bk...@in...>]

Tim Mackinnon wrote:
> patient: Ouch doctor it hurts when I hit myself....
> dr: Don't do it then...
> 
> Barry - Its a bit unclear from your example (or lack of an example) but
> MockOjects require an interface. Plain and simple... I don't recall the old
> version supporting this? If it did it was accidental - as we advocate
> programming by interface. In our experience programming by subclassing is
> typically a smell, its hard to test - and compositional programming will in
> the long term give you much clearer tests that are easier to refactor and
> much clearer to test.
> 
> My advice is to convert your abstract class to an interface and proceed as
> normal... The sun io classes are a classic example of how not to do it (I
> figure they must be legacy code that predates them having interfaces in
> java).

Yes, yes, I agree of course (see one of my options for handling the 
situation), but making that change is not possible at this time. I'm 
looking for a way to quickly upgrade with minimal changes to the code 
(ie, I only want to change one thing a time).

But 0.8 /did/ allow for the testing of an abstract class, via the use of 
explicit CallSequence instance variables in a non-dynamic mock class. 
(And you thought the use of an abstract class was smelly!).

1. subclass the abstract class: AbstractClassUnderTest
2. declare explict CallSequence instance variables
3. implement the template hook methods such that they use the CallSequence

But I did come up with a quick workaround. If someone else posts a 
similar problem I'll describe how I did it. But first I would give the 
same advice as you did above.

thanks!