On Sat, 15 Dec 2001 13:38 +0000 (GMT Standard Time),
brangdon / cix.co.uk (Dave Harris) wrote:

>That the function's result should not depend on the order of its arguments 
>is, in my view, an important fact that should be made more explicit than 
>it currently is. We can deduce it from the function's implementation, or 
>from the function's name plus what we know about triangles, but I'm not 
>sure that is enough to satisfy the "Once" part of "Once and Only Once".
>
>I suspect we need some more support from the test rig; some mechanism 
>which will generate all permutations of a collection of arguments and 
>#perform them. There are 6 permutations, and we need to test at least the 
>3 normal cases, but I would tend to regard the whole thing as 1 test 
>rather than 18.

Given the implementation, I'm not clear on why we would want to test
other permutations (since the internal representation is a sorted
collection). Are you suggesting additional tests as protection against
someone reimplementing the function erroneously?

Ronald E Jeffries
http://www.XProgramming.com
http://www.objectmentor.com