Re: [Caml-list] What is an applicative functor?

[Till Varoquaux <till@pps.jussieu.fr>]

On Fri, Apr 8, 2011 at 2:50 AM, Andreas Rossberg <rossberg@mpi-sws.org> wrote:
> On Apr 7, 2011, at 23:49, Gerd Stolpmann wrote:
>>
>> module Make(X : sig end) = struct type t = Variant end
>> module M1 = Make(struct end)
>> module M2 = Make(struct end)
>>
>> Now, M1.t and M2.t are incompatible - for nominal types like variants
>> the functors aren't applicative, and each instance is a thing of its
>> own:
>
> Nitpick: the Make functor is still applicative. All functors in OCaml are
> applicative. The fact that you get different results above is caused by the
> way you apply the functor, not by the way it is defined.
>
> "Applicative" distinguishes functor semantics in OCaml from that in SML,
> where functors are "generative", i.e. you always get new types when you
> apply them.
>
> Aside: In general, you really want to have both: functors with impure bodies
> better be generative, while functors with pure bodies should be applicative.
> The absence of generative functors in OCaml also is the reason that you
> cannot unpack a first-class module in a functor body (non-locally), because
> that is unsound with applicative semantics.
>
> /Andreas
I tend to consider using Impure functors as very poor coding hygiene..
I am sure there are compelling use cases but I've yet to come across
one. Providing safer types for impure functors does not seem
compelling enough to justify having several kind of functors.
I am not really sure I want applicative functors (based upon my
experience they are pretty hard to explain and rarely buy me anything)
but I balk at the idea of having both applicative and generative
functors..

Till

>
>
> --
> Caml-list mailing list.  Subscription management and archives:
> https://sympa-roc.inria.fr/wws/info/caml-list
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
>