Re: [Caml-list] Types look compatible, but they aren't?

[Anthony Tavener <anthony.tavener@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 5357 bytes --]

In this instance it aquired a _ though, or does that have some other
meaning here?

I was getting < .. > at first, but when I called Modifer.attach with three
arguments the inference
changed to < _.. >.

As a side note, I didn't know what to make of < .. > until Kakadu mentioned
objects. I was thinking "what is that, some kind of abstract type"? That's
what I get for rarely using objects!


On Sat, Apr 13, 2013 at 11:14 AM, Gabriel Scherer <gabriel.scherer@gmail.com
> wrote:

> > So, 'a becomes: (< _.. > as 'a) -- I get some monomorphic... object?
>
> Just a small thing, (< .. > as 'a) is not monomorphic, it is still a
> polymorphic type, that may be instantiated with any object type. It is
> "less polymorphic" than 'a (can be instantiated with anything).
>
> On Sat, Apr 13, 2013 at 6:15 PM, Anthony Tavener
> <anthony.tavener@gmail.com> wrote:
> > I forgot to note, that the interesting thing was how the type inferred
> for
> > Modifier.attach when it had
> > one argument applied did not show the < _.. > monomorphic object
> constraint.
> > Modifier.attach
> > is actually a: fun id -> (key -> fn -> deleter), rather than a
> > straightforward three-argument function. Once
> > the (key -> fn -> deleter) function would come into play, the "object"
> was
> > revealed.
> >
> >
> > On Sat, Apr 13, 2013 at 10:07 AM, Anthony Tavener
> > <anthony.tavener@gmail.com> wrote:
> >>
> >> Ohhh... that is interesting. (TL;DR: problem solved, and it was from
> >> inappropriate Oo.id use.)
> >>
> >> Modifier.attach is actually implemented as a function of one argument
> >> which does some stuff,
> >> returning a function of two arguments, to avoid redundant lookups in the
> >> case of multiple "attach"
> >> to the same "id".
> >>
> >> When I remove the let m = ... and just inline "Modifer.attach id ..."
> the
> >> type of Modifier.attach changes to:
> >>
> >>   Db.key -> int * (((< _.. > as 'a) list -> exn) * (exn -> 'a list) ->
> 'a
> >> -> Modifier.deleter
> >>
> >> So, 'a becomes: (< _.. > as 'a) -- I get some monomorphic... object?
> >>
> >> As I wrote this I had an idea and found the problem:
> >>
> >>   ...
> >>   (* return (tbl -> unit) function which deletes this specific function
> *)
> >>   let del_id = Oo.id fn in
> >>   (fun tbl ->
> >>     let lst = List.filter (fun e -> Oo.id e <> del_id) (fn_list tbl) in
> >>     Hashtbl.replace tbl tag (inj lst))
> >>
> >>
> >> Here, "fn" is the provided function, and I want an easy way to remove
> such
> >> functions uniquely from the
> >> mess of Hashtbl, universal embedding, and list. I tried a trick I once
> >> read Alain suggest for getting a
> >> unique id using the object module... and I guess that brought in this
> <..>
> >> thing I was unfamiliar with. :)
> >> Instead of Oo.id I'm using Hashtbl.hash now, which is normally what I'd
> >> do... not sure why I
> >> half-remembered some trick with Oo.id.
> >>
> >> Thank-you for looking at this, both of you. It helped me dig in the
> right
> >> direction!
> >>
> >>
> >> On Sat, Apr 13, 2013 at 1:33 AM, Gabriel Scherer
> >> <gabriel.scherer@gmail.com> wrote:
> >>>
> >>> This looks like a value restriction issue with
> >>>
> >>>   let m = Modifier.attach id
> >>>
> >>>   "A function obtained through partial application is not polymorphic
> >>> enough"
> >>>   http://caml.inria.fr/resources/doc/faq/core.en.html#eta-expansion
> >>>
> >>> If this is indeed the source of your error, you can regain
> >>> type-checking by using instead
> >>>
> >>>   let m total = Modifier.attach id total
> >>>
> >>> Note that this may change the semantics of your code if
> >>> (Modifier.attach id) does a side-effect before getting its next
> >>> parameter: if would have been effected only once with your previous
> >>> definition, and will be effected at each call of 'm' with the new
> >>> definition.
> >>>
> >>> On Sat, Apr 13, 2013 at 8:56 AM, Kakadu <kakadu.hafanana@gmail.com>
> >>> wrote:
> >>> > Maybe function type (int * int -> int * int) is incompatible with
> >>> > object
> >>> > type <..>?
> >>> >
> >>> > Kakadu
> >>> >
> >>> >
> >>> > On Sat, Apr 13, 2013 at 10:50 AM, Anthony Tavener
> >>> > <anthony.tavener@gmail.com> wrote:
> >>> >>
> >>> >> File "virtue.ml", line 462, characters 12-24:
> >>> >> Error: This expression has type
> >>> >>          int * ((int * int -> int * int) list -> exn) *
> >>> >>          (exn -> (int * int -> int * int) list)
> >>> >>        but an expression was expected of type
> >>> >>          int * ((< .. > as 'a) list -> exn) * (exn -> 'a list)
> >>> >>
> >>> >> The code in question:
> >>> >>
> >>> >>   (fun id ->
> >>> >>     let m = Modifier.attach id in
> >>> >>       [ m Cast.total'k (fun (v,b) -> (v, max 1 (b-3)))     (* <--
> line
> >>> >> 462
> >>> >> *)
> >>> >>       ; m Lab.total'k (fun (v,b) -> (v, max 1 (b-3))) ])
> >>> >>
> >>> >> For reference, the signature of Modifier.attach:
> >>> >>   Db.key -> int * ('a list -> exn) * (exn -> 'a list) -> 'a ->
> >>> >> Modifier.deleter
> >>> >>
> >>> >> OCaml version is 4.00.0 -- I know I should upgrade. Keep meaning
> to, I
> >>> >> guess I will if I wake up and there's no helpful soul explaining
> what
> >>> >> could
> >>> >> be wrong here. :)
> >>> >>
> >>> >> Thank-you for any help. My eyes are starting to bug-out looking at
> >>> >> this.
> >>> >>
> >>> >>  -Tony
> >>> >>
> >>> >>
> >>> >
> >>
> >>
> >
>

[-- Attachment #2: Type: text/html, Size: 8051 bytes --]