Unit test requirements (was Re: [Coin-discuss] OsiCbc)

[Michael Hennebry]

On Wed, 6 Apr 2005, Matthew Saltzman wrote:

> On Wed, 6 Apr 2005, Ted Ralphs wrote:
> > A mailing list for the OSI is probably appropriate. I think written
> > documentation is a good goal, but realistically, it's easier (and perhaps
> > better) just to say that the expected behavior for an interface is simply
> > that it pass the unit test. If we design and comment the unit test carefully,
> > then any deviations from the required behavior will be easy to understand and
> > fix. This way, we don't have to try to keep the documentation in sync with
> > the unit test if we make changes. What do you think?
>
> One mailing list coming up.
>
> I agree that a good unit test is the best spec.  One design goal for OSI
> is to have a single, universal unit test for the base class that will do
> the right thing with any SI layer.  That will act as an enforcement
> mechanism to keep SIs up to date as the API evolves and will help prevent
> feature creep in different SIs.

Perhaps there should be make file entries that look somethinng like this:

UNIT_TESTS += unit_test1
## The optimal solution must be feasible.
unit_test1:
	....

unit_test: unit_test1 ....

Less diffuse documentation could be generated with sed.

-- 
Mike   hennebry at web.cs.ndsu.NoDak.edu
"I AM DEATH, NOT TAXES.  *I* ONLY TURN UP ONCE."  --  Death