Spark Shen wrote: > 2007/7/5, Alexei Zakharov <[EMAIL PROTECTED]>: >> >> Paulex Yang wrote: >> > What I meant is the test cases may need to be written against API >> factory >> > method instead of construct implementation classes directly via >> > implementation specific constructor/class, and the compatibility tests >> like >> > serialization tests are needed. I have no preference in where the >> > implementation classes locate. :) >> >> Look, in beans we have special (package-protected) entities that are >> responsible for handling all persistence-related problems. If we like >> to test these non-public entities we have two ways: write an >> implementation test that can call protected methods directly to check >> the desired behavior, or write an API test that will walk rather long >> code path - create an appropriate content, instantiate XMLEncoder and >> so on. In the last case our testing method is less accurate and >> flexible comparing to the first one. Moreover, the real persistent >> delegate issue can be hidden by bugs in complex logic of XMLEncoder. > > > But, the problem is some internal tests will fail on RI. Then what's these > test cases test against?
That's the definition of an implementation test -- it is implementation-specific. We would only expect the API tests to pass on the RI, but API + implementation tests to pass against Harmony. HTH. Regards, Tim
