On Sat, Dec 08, 2001 at 03:35:11AM +0900, John Roth wrote:
> Let's think about this for a moment. 5 test cases appears to
> be adequate for a black box test. You need one for equilateral,
> three for isosceles and one for scalene. If you're deriving
> more test cases, you're doing an open box test, based on the
> actual implementation (or on some set of assumptions about
> frequent defects, or some such.)

I think you need more tests than that, but more so because the
requirements are wrong than anything else.  A combination of three input
values does not necessarily represent a valid triangle, so the function
really cannot return "scalar" just because the three values do not
represent an isosceles or equilateral triangle.

So that's six more test cases (a+b=c, a+b<c, a+c=b, a+c<b, b+c=a,
b+c<a).

I don't think the requirements state that negative values should be
disallowed (a negative value may simply represent a triangle that is
upside-down).  So we can't add more bad data tests without having
better requirements.

Paul