Re: [boost] Policy-based designs considered harmful?

From: Beman Dawes (bdawes_at_[hidden])
Date: 2002-01-22 08:58:12

At 09:53 PM 1/21/2002, Andrei Alexandrescu wrote:

 >> In addition, the predefined policies need to be tested as well. This 
can
 >be
 >> done policy by policy if each policy has a formal specification and 
their
 >> behavior is not affected by the parent SmartPtr (which implies no
 >> interaction between policies.)
 >
 >
 >I think that that's the main point. As long as policies are orthogonal,
 >they
 >can be tested in isolation. For example, if CheckingPolicy doesn't 
interact
 >with any other policy, a suite can test all CheckingPolicy 
implementations
 >using some arbitrary choices for the other policies.
 >
 >Otherwise, I was thinking of a testing strategy similar to Jeremy's - 
tests
 >are generated by templates, and testing specific cases/behaviors are
 >modeled
 >with specialization of the test cases.

Yes, even if policies interact, they can still be tested in isolation if 
the interaction is well-specified and bounded.

For example, let's say StoragePolicy is specified to always supply one type 
(say pointer_type) and one function (say get()) via its public interface.

An OwnershipPolicy which accesses a StoragePolicy can still be tested in 
isolation.  Its test driver will have to supply one or more dummy 
StoragePolicys which meet the requirements (supply well-formed pointer_type 
and get()) but do nothing more.

--Beman