From: Nat Pryce <nat.pryce@...13media.com>
Date: 2002-09-23T18:31:23+09:00
Subject: Re: The problem with run-time type checking

On Sun, 2002-09-22 at 19:36, Philip Mak wrote:
> I'm asking when I write a function that depends on its arguments to be a
> specific type, how can I make them produce informative stack dumps if the
> arguments are the wrong type? I don't see how unit testing would help the
> person writing a library function.

Testing that an object makes correct calls to a library is not the
responsibility of the library writer, it is the responsibility of the
programmer who is using the library.

A programmer should test that their objects implement their
postconditions and invariants correctly, and that their objects meet the
preconditions of other object that they call.  One way to do this is to
mock up the environment of the object in order to assert that outgoing
calls are correct when the object is exercised in a test case.

This kind of testing is easiest if libraries are written as collections
of classes that are used through *composition*, rather than through
inheritance.  That is, an object should use your library through an
object reference, rather than through a module of free-floating
functions.  This way, programmers using your library can create mock
versions of your classes in their unit tests to ensure that the correct
values are being passed out of their objects.

If you really want to help clients of your library, you can provide a
suite of mock objects that can be used for unit testing client code. 

Cheers,
	Nat.
-- 
Dr. Nathaniel Pryce, Technical Director, B13media Ltd.
Studio 3a, 22-24 Highbury Grove, London N5 2EA, UK
http://www.b13media.com