I wholeheartedly agree with zwon, especially since I have been giving the subject a good coating of thinking about (and not a little discussion/argument) of late - with hardened Java programmers who have been loaded with the task to develop a product test harness/environment in perl.

In my case the discussions have had occasion to plumb the depths of argument since the preferred implementation style...

• implements a package in a single .pm file i.e. it doesn't (currently) utilise Test::More or any other of that good stuff supplied for free merely by utilising h2xs.
• utilises a Test::Unit style implementation of a test regime - which to my mind is a white box test environment i.e. it avoids (proper) black box/interface testing as provided for by Test::More (A corollary of this is that, my agreement with zwon is tempered by the fact that, IMO, more thorough module testing could be acheived thro' the use of both i.e. Test::More & Test::Unit, approaches)

As a perl script can be written to be essentially self-modifying, to my mind there may just be an advantage in run-time checking - since, together with the test harness, belt-and-braces checking of the interface implementation would be facilitated.