On Tue, Oct 08, 2002 at 05:52:46PM +0900, Mauricio Fern?ndez wrote:
> > would submit the object (or rather its class) to a small ad-hoc test
> > suite, which would not only test whether it responded to (for example)
> > #read, #write and #seek, but also check whether what gets written with
> > #write can be later read with #read.

> What if 

Oh, I knew it was coming... ;-)

> the object represents /dev/null and you cannot read back what you 
> wrote previously? 

If the method needs to read back, you've just caught a bug; if it does
not need to read back, you're running the wrong test.

> It is hard to specify the right minimal semantic requirements...

Anything in particular makes you think so?

> > It would impose quite an overhead at runtime, but could be executed
> > only when a global switch was turned on (--inline-tests), and would
> > guarantee that an object really provides some functionality, not just
> > `looks like to' as with respond_to? or Java interfaces.
> 
> IMHO it is a good idea if some provision is made for "non-standard"
> behaving objects. 

I wasn't suggesting standard tests -- that would be equivalent to
going back to type names; I was suggesting ad-hoc tests made up
according to the needs of the method.


Massimiliano