coercions issues

coercions issues

[Till Varoquaux]

I still can't wrap my head around the whole polymorphic variant of
Ocaml (although I am making steady progresses). Anyways here is my
latest sore point:

This expression cannot be coerced to type
  O.o Ast.T.expr O.m = O.o Ast.T.expr Mon.m;
it has type O.o Ast_js.T.expr Mon.m but is here used with type
  O.o Ast.T.expr O.m = O.o Ast.T.expr Mon.m
Type
  O.o Ast_js.T.expr =
    [ `Binop of To.binop * To.expr * To.expr
    | `Cst of To.constant
    | `Ecall of To.ident * To.arg list
    | `Fun of To.ident list * To.bloc
    | `Lval of To.lvalue
    | `Unop of To.unop * To.expr ]
is not compatible with type
  O.o Ast.T.expr =
    [ `Binop of To.binop * To.expr * To.expr
    | `Cst of To.constant
    | `Ecall of To.ident * To.arg list
    | `Epos of Ast.location * To.expr
    | `Fun of To.ident list * To.bloc
    | `Lval of To.lvalue
    | `Unop of To.unop * To.expr ]
The first variant type does not allow tag(s) `Epos

I am a little confused as to why this coercion is impossible... It
looks like a simple upward cast to me.
Cheers,
Till

Re: [Caml-list] coercions issues

[Chris King]

On 3/12/07, Till Varoquaux <till.varoquaux@gmail.com> wrote:
> This expression cannot be coerced to type
>  O.o Ast.T.expr O.m = O.o Ast.T.expr Mon.m;
> it has type O.o Ast_js.T.expr Mon.m but is here used with type
>  O.o Ast.T.expr O.m = O.o Ast.T.expr Mon.m
> Type
>  O.o Ast_js.T.expr =
>    [ `Binop of To.binop * To.expr * To.expr
>    | `Cst of To.constant
>    | `Ecall of To.ident * To.arg list
>    | `Fun of To.ident list * To.bloc
>    | `Lval of To.lvalue
>    | `Unop of To.unop * To.expr ]
> is not compatible with type
>  O.o Ast.T.expr =
>    [ `Binop of To.binop * To.expr * To.expr
>    | `Cst of To.constant
>    | `Ecall of To.ident * To.arg list
>    | `Epos of Ast.location * To.expr
>    | `Fun of To.ident list * To.bloc
>    | `Lval of To.lvalue
>    | `Unop of To.unop * To.expr ]
> The first variant type does not allow tag(s) `Epos

What sort of type is 'a O.m (or 'a Mon.m)?  If it's something whose 'a
member is mutable, then such a coercion will be impossible (it would
allow you to set the member to have the `Epos tag, which it can't).
This would also happen if 'a O.m is declared as an abstract type,
since O'Caml assumes invariance by default.  In the latter case, you
need to specify that 'a O.m is covariant with its parameter:

type +'a O.m

This tells Caml that if 'a is a subtype of 'b, then 'a O.m is indeed a
subtype of 'b O.m and you will be allowed to perform the above
coercion.

Hope this helps.

- Chris

Re: [Caml-list] coercions issues

[Till Varoquaux]

Great!

That was indeed really helpfull! I really missed out that last bit, I
sould have payed more attention.

'a Mon.m and 'a O.m are monadic values. O.m and Mon.m are the same
type.  My intuition would be that they should be variant  ((a :> b m)
should behave like a >>= (fun x -> return (x:>b) where a is a value m
a monadic type and b a supertype of a)) can anyone concur that.

Cheers,
Till

On 3/12/07, Chris King <colanderman@gmail.com> wrote:
> On 3/12/07, Till Varoquaux <till.varoquaux@gmail.com> wrote:
> > This expression cannot be coerced to type
> >  O.o Ast.T.expr O.m = O.o Ast.T.expr Mon.m;
> > it has type O.o Ast_js.T.expr Mon.m but is here used with type
> >  O.o Ast.T.expr O.m = O.o Ast.T.expr Mon.m
> > Type
> >  O.o Ast_js.T.expr =
> >    [ `Binop of To.binop * To.expr * To.expr
> >    | `Cst of To.constant
> >    | `Ecall of To.ident * To.arg list
> >    | `Fun of To.ident list * To.bloc
> >    | `Lval of To.lvalue
> >    | `Unop of To.unop * To.expr ]
> > is not compatible with type
> >  O.o Ast.T.expr =
> >    [ `Binop of To.binop * To.expr * To.expr
> >    | `Cst of To.constant
> >    | `Ecall of To.ident * To.arg list
> >    | `Epos of Ast.location * To.expr
> >    | `Fun of To.ident list * To.bloc
> >    | `Lval of To.lvalue
> >    | `Unop of To.unop * To.expr ]
> > The first variant type does not allow tag(s) `Epos
>
> What sort of type is 'a O.m (or 'a Mon.m)?  If it's something whose 'a
> member is mutable, then such a coercion will be impossible (it would
> allow you to set the member to have the `Epos tag, which it can't).
> This would also happen if 'a O.m is declared as an abstract type,
> since O'Caml assumes invariance by default.  In the latter case, you
> need to specify that 'a O.m is covariant with its parameter:
>
> type +'a O.m
>
> This tells Caml that if 'a is a subtype of 'b, then 'a O.m is indeed a
> subtype of 'b O.m and you will be allowed to perform the above
> coercion.
>
> Hope this helps.
>
> - Chris
>