Re: [MO-java-users] Re: [MO-java-dev] Cretion of the dynamic mocks for classes vs. interfaces

[Nat P. <nat...@b1...>]

On Mon, 2003-12-01 at 05:53, Eugene Kuleshov wrote:
> Nat Pryce wrote:
>   I faced that issue too. But my understanding was that I can and should 
> use "match" for such calls. So all "expect"-like calls will be in order 
> and all "match"-like calls can appear anywhere. Otherwise it will be just 
> not possible to handle different returns for the sequential calls 
> without or with the same parameters.

The issue with an ordered InvocationDispatcher is that it makes tests
too brittle.  In general, one needs to specify that some calls are
stubbed (matched in the 0.09 parlance), some are expected after others
and some are expected but not in any order.  Requiring that *all* calls
are expected in a fixed order makes tests easy to break when code is
changed over time.  It is also quite a rare case -- I've only needed it
when testing some quite baroque event announcement code. And even then
it broke tests later!

Our new approach is to provide an API that is used to specify that one
invocation is expected to occur after another.  This has the benefits
that it highlights the ordering in the test and allows *partial*
ordering, which gives much needed flexibility to the test code.

Cheers,
	Nat.