Sam Tregar wrote:
> 
> On Wed, 26 Jul 2000, Dave Paris wrote:
> 
> > A patch test should do a minimum of two things:
> > (a) prove the existance of the problem *before* the patch.
> 
> If the patch solves a problem (versus adding a feature), this is true.

just to be consistent, a test to prove the lack of the feature would be
nice.

> > (b) prove the problem has been fixed *after* the patch - without
> > breaking anything else.
> 
> This should be proved by the existing body of regression tests
> successfully passing.  You don't rewrite a general test of your module
> each time you add a test to test.pl, do you?

the condition of possibly breaking a core module does extend this a
bit.  I see and agree with your take on simply adding a test to test.pl.
(and alluded to this with "may be able to be pared down by calling other
existing tests")

> Also, you need;
> 
> (c) proves that the patch deals elegently with "out-of-bounds" input,
> when such is possible.

a good addition.

--a-mused

Reply via email to