[Caml-list] matching GADT option types

[Caml-list] matching GADT option types

[Christopher Zimmermann]

[-- Attachment #1: Type: text/plain, Size: 620 bytes --]

Hi,

why does the f type correctly while g fails to type?

Christopher

type 'a t =
  | A : unit t

let f =
  fun (type a) ~(p :a t option) () -> match p with
    | Some A -> ()
    | None -> ()

let g =
  fun (type a) ~(p :a t option) () -> match p with
    | Some A (* TYPING ERROR HERE *)
    | None -> ()

Error: This pattern matches values of type unit t
       but a pattern was expected which matches values of type a t
       Type unit is not compatible with type a


-- 
http://gmerlin.de
OpenPGP: http://gmerlin.de/christopher.pub
CB07 DA40 B0B6 571D 35E2  0DEF 87E2 92A7 13E5 DEE1

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [Caml-list] matching GADT option types

[Jeremy Yallop]

On Thu, 17 Jan 2019 at 09:46, Christopher Zimmermann
<christopher@gmerlin.de> wrote:
> why does the f type correctly while g fails to type?
>
> Christopher
>
> type 'a t =
>   | A : unit t
>
> let f =
>   fun (type a) ~(p :a t option) () -> match p with
>     | Some A -> ()
>     | None -> ()
>
> let g =
>   fun (type a) ~(p :a t option) () -> match p with
>     | Some A (* TYPING ERROR HERE *)
>     | None -> ()
>
> Error: This pattern matches values of type unit t
>        but a pattern was expected which matches values of type a t
>        Type unit is not compatible with type a

GADT matching under or-patterns isn't currently supported in the
current release.  However, the following PR added some support, and
your code is now accepted with the trunk compiler:

    Allow GADT constructors to introduce equations and existential
types under or-patterns
    https://github.com/ocaml/ocaml/pull/2110

Re: [Caml-list] matching GADT option types

[Christopher Zimmermann]

[-- Attachment #1: Type: text/plain, Size: 1240 bytes --]

On Thu, 17 Jan 2019 10:46:11 +0100
Christopher Zimmermann <christopher@gmerlin.de> wrote:

Hi,

why does the f type correctly while g fails to type?

Christopher

type 'a t =
  | A : unit t

let f =
  fun (type a) ~(p :a t option) () -> match p with
    | Some A -> ()
    | None -> ()

let g =
  fun (type a) ~(p :a t option) () -> match p with
    | Some A (* TYPING ERROR HERE *)
    | None -> ()

Error: This pattern matches values of type unit t
       but a pattern was expected which matches values of type a t
       Type unit is not compatible with type a



to add to the confusion, why does f type while g fails to type ?

type 'a t =
  | A : [`A] t
  | B : [`A|`B] t

let f =
  fun (type a) ~(p :a t) () -> match p with
    | A -> (A: a t)
    | B -> (B: a t)

let g =
  fun (type a) ~(p :a t option) () -> match p with
    | Some B -> (B: a t)
    | Some A -> (A: a t)
    | None -> (A (* TYPE ERROR HERE *) : a t)

Error: This expression has type [ `A ] t
       but an expression was expected of type a t
       Type [ `A ] is not compatible with type a 

-- 
http://gmerlin.de
OpenPGP: http://gmerlin.de/christopher.pub
CB07 DA40 B0B6 571D 35E2  0DEF 87E2 92A7 13E5 DEE1

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [Caml-list] matching GADT option types

[Christopher Zimmermann]

[-- Attachment #1: Type: text/plain, Size: 2435 bytes --]

Thank you for the fast answer!

That means my second example will still be rejected in 4.08?

type 'a t =
  | A : [`A] t
  | B : [`A|`B] t

let f =
  fun (type a) ~(p :a t) () -> match p with
    | A -> (A: a t)
    | B -> (B: a t)

let g =
  fun (type a) ~(p :a t option) () -> match p with
    | Some B -> (B: a t)
    | Some A -> (A: a t)
    | None -> (A (* TYPE ERROR HERE *) : a t)

Error: This expression has type [ `A ] t
       but an expression was expected of type a t
       Type [ `A ] is not compatible with type a 


On Thu, 17 Jan 2019 11:12:02 +0100
Gabriel Scherer <gabriel.scherer@gmail.com> wrote:

> Matching on a constructor of a GADT may introduce a typing equation.
> In your example, matching on `A` introduces the equation `a = unit`.
> For this reason, the status of or-patterns (p1 | p2) containing GADT
> constructors is delicate: to type-check them we have to decide which
> equations from both sides are preserved in the result, computing a
> sort of intersection.
> 
> Released versions of OCaml avoid this difficulty by not supporting
> GADTs in or-patterns at all; you have to expand the pattern into two
> branches, as you did in your function `f` above.
> 
> In the current trunk, a change from Thomas Réfis and Leo White has
> been merged that allows GADTs in or-patterns, but discards the
> equations. Your specific example (the function `g`) is now accepted,
> and will be typeable in 4.08. Other examples, where you need to use
> an equation (provided by both sides of the or-patterns) in the body
> of the clause, will still be rejected.
> 
> 
> On Thu, Jan 17, 2019 at 10:46 AM Christopher Zimmermann <
> christopher@gmerlin.de> wrote:  
> 
> > Hi,
> >
> > why does the f type correctly while g fails to type?
> >
> > Christopher
> >
> > type 'a t =
> >   | A : unit t
> >
> > let f =
> >   fun (type a) ~(p :a t option) () -> match p with
> >     | Some A -> ()
> >     | None -> ()
> >
> > let g =
> >   fun (type a) ~(p :a t option) () -> match p with
> >     | Some A (* TYPING ERROR HERE *)
> >     | None -> ()
> >
> > Error: This pattern matches values of type unit t
> >        but a pattern was expected which matches values of type a t
> >        Type unit is not compatible with type a


-- 
http://gmerlin.de
OpenPGP: http://gmerlin.de/christopher.pub
CB07 DA40 B0B6 571D 35E2  0DEF 87E2 92A7 13E5 DEE1

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [Caml-list] matching GADT option types

[Gabriel Scherer]

[-- Attachment #1: Type: text/plain, Size: 1711 bytes --]

Matching on a constructor of a GADT may introduce a typing equation. In
your example, matching on `A` introduces the equation `a = unit`.
For this reason, the status of or-patterns (p1 | p2) containing GADT
constructors is delicate: to type-check them we have to decide which
equations from both sides are preserved in the result, computing a sort of
intersection.

Released versions of OCaml avoid this difficulty by not supporting GADTs in
or-patterns at all; you have to expand the pattern into two branches, as
you did in your function `f` above.

In the current trunk, a change from Thomas Réfis and Leo White has been
merged that allows GADTs in or-patterns, but discards the equations. Your
specific example (the function `g`) is now accepted, and will be typeable
in 4.08. Other examples, where you need to use an equation (provided by
both sides of the or-patterns) in the body of the clause, will still be
rejected.


On Thu, Jan 17, 2019 at 10:46 AM Christopher Zimmermann <
christopher@gmerlin.de> wrote:

> Hi,
>
> why does the f type correctly while g fails to type?
>
> Christopher
>
> type 'a t =
>   | A : unit t
>
> let f =
>   fun (type a) ~(p :a t option) () -> match p with
>     | Some A -> ()
>     | None -> ()
>
> let g =
>   fun (type a) ~(p :a t option) () -> match p with
>     | Some A (* TYPING ERROR HERE *)
>     | None -> ()
>
> Error: This pattern matches values of type unit t
>        but a pattern was expected which matches values of type a t
>        Type unit is not compatible with type a
>
>
> --
> http://gmerlin.de
> OpenPGP: http://gmerlin.de/christopher.pub
> CB07 DA40 B0B6 571D 35E2  0DEF 87E2 92A7 13E5 DEE1
>

[-- Attachment #2: Type: text/html, Size: 2304 bytes --]

Re: [Caml-list] matching GADT option types

[Gabriel Scherer]

[-- Attachment #1: Type: text/plain, Size: 1899 bytes --]

Your function g is incorrect; in the None case, it claims that the
constructor A has the type (a t) for an unknown (arbitrary) type parameter
(a). A only has the type (a t) for a specific type (a = [`A]), not for all
types.
If you want g to be able to return a value of type (a t) for *some* type
(a) that g chooses, instead of for *any* type (a) chosen by the
context/caller, then you should use an existential type wrapping:

  type any_t = Any : a t -> any_t.

  [...]
  | Some A -> Any A
  | Some B -> Any B
  | None -> Any A

On Thu, Jan 17, 2019 at 11:02 AM Christopher Zimmermann <
christopher@gmerlin.de> wrote:

> On Thu, 17 Jan 2019 10:46:11 +0100
> Christopher Zimmermann <christopher@gmerlin.de> wrote:
>
> Hi,
>
> why does the f type correctly while g fails to type?
>
> Christopher
>
> type 'a t =
>   | A : unit t
>
> let f =
>   fun (type a) ~(p :a t option) () -> match p with
>     | Some A -> ()
>     | None -> ()
>
> let g =
>   fun (type a) ~(p :a t option) () -> match p with
>     | Some A (* TYPING ERROR HERE *)
>     | None -> ()
>
> Error: This pattern matches values of type unit t
>        but a pattern was expected which matches values of type a t
>        Type unit is not compatible with type a
>
>
>
> to add to the confusion, why does f type while g fails to type ?
>
> type 'a t =
>   | A : [`A] t
>   | B : [`A|`B] t
>
> let f =
>   fun (type a) ~(p :a t) () -> match p with
>     | A -> (A: a t)
>     | B -> (B: a t)
>
> let g =
>   fun (type a) ~(p :a t option) () -> match p with
>     | Some B -> (B: a t)
>     | Some A -> (A: a t)
>     | None -> (A (* TYPE ERROR HERE *) : a t)
>
> Error: This expression has type [ `A ] t
>        but an expression was expected of type a t
>        Type [ `A ] is not compatible with type a
>
> --
> http://gmerlin.de
> OpenPGP: http://gmerlin.de/christopher.pub
> CB07 DA40 B0B6 571D 35E2  0DEF 87E2 92A7 13E5 DEE1
>

[-- Attachment #2: Type: text/html, Size: 2780 bytes --]