Re: Do people who use Ada also use ocaml or F#?

[Robert A Duff <bobduff@shell01.TheWorld.com>]

Stephen Leake <stephen_leake@stephe-leake.org> writes:

> Robert A Duff <bobduff@shell01.TheWorld.com> writes:
>
>> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
>>
>>> How do you execute a generic body? Testable are generic instances.
>>
>>     You can't test a generic without instantiating it.
>>     
>>     You can't test a procedure without calling it.
>>     
>>     You can't test an abstract type without declaring
>>     a concrete type, and some objects of that type.
>>
>> What am I missing?
>
> To truly test a generic, you need to test with all possible
> instantiations.

To "truly" test a procedure, you have to call it with all
possible parameter values (and all possible globals,
if it uses those).

> Or at least with a representative subset.

Right, it's usually impossible to try all possibilities,
so we pick some subset -- for generics and for procedures.

>...If the generic parameters are
> well structured, it may be possible to show that a few instances are
> sufficient.

For something like Ada.Containers.Vectors, it's fairly easy
to know that the code doesn't much care what the element type is.

If you find a bug in a generic, it is almost certainly
a bug in every instance, in my experience.  Only rarely
does the bug depend on specific properties of some
instance.

> That's much more effort than testing a single concrete procedure.

Well, I agree that testing a generic is typically harder
than testing a single procedure.  That's because generics
are typically more, um, "general".  But the claim was
that generics are untestable, which seems too extreme to me.
And if you replace generics with some other feature,
then if that feature has the generality desired,
it will be equally hard to test.

> It may be worth it, of course; I often use generics precisely to reduce
> testing burden.

Right, it's certainly easier to test a generic, than to write
similar code 20 times for 20 different types, and test all of
those.  Just like it's easier to test a single procedure than to
test 20 cut&pasted versions of the same code (each of which
has been slightly modified so they're not quite identical
anymore).

- Bob