Stupid question re:modules

Stupid question re:modules

[Brian Hurt]

I should just know this.  So let's say I have two module types defined:

module type Foo = sig
 	type 'a t
 	val foo : 'a -> 'a t
end;;

module type Bar = sig
 	type 'a t
 	val bar : 'a -> 'a t
end;;

Now, I want to define a module that is both a Foo and a Bar without 
cutting and pasting the module definitions around.  I've been trying to 
do:

module Baz : sig
 	type 'a baz
 	include Foo with type 'a t = 'a baz
 	include Bar with type 'a t = 'a baz
end;;

but this blows up on the Bar line (multiple definitions of 'a t).

There is a solution to this, I'm just being stupid and forgetting what it 
is.  Hints would be appreciated.

Brian

Re: [Caml-list] Stupid question re:modules

[Jacques Garrigue]

From: Brian Hurt <bhurt@spnz.org>
> I should just know this.  So let's say I have two module types defined:
> 
> module type Foo = sig
>  	type 'a t
>  	val foo : 'a -> 'a t
> end;;
> 
> module type Bar = sig
>  	type 'a t
>  	val bar : 'a -> 'a t
> end;;
> 
> Now, I want to define a module that is both a Foo and a Bar without 
> cutting and pasting the module definitions around.  I've been trying to 
> do:
> 
> module Baz : sig
>  	type 'a baz
>  	include Foo with type 'a t = 'a baz
>  	include Bar with type 'a t = 'a baz
> end;;
> 
> but this blows up on the Bar line (multiple definitions of 'a t).
> 
> There is a solution to this, I'm just being stupid and forgetting what it 
> is.  Hints would be appreciated.

Unfortunately, there is no solution to this, as you cannot remove a
declaration from a signature, and include cannot overwrite existing
type definitions. There was a paper in ICFP'2005 on how to solve this,
but I'm not aware of any plan to follow it.

If you're ready to write longer signatures, there is a workaround.

module type T = sig type 'a t end

module FooS(X:T) = struct
  module type S = sig
    val foo : 'a -> 'a X.t
  end
end

module BarS(X:T) = struct
  module type S = sig
    val bar : 'a -> 'a X.t
  end
end

module Baz : sig
  module T : sig type 'a t end
  include FooS(T)
  include BarS(T)
end

The basic idea if you want to be able to construct signatures
modularly is to always keep types and function declarations
separated.

Jacques Garrigue

Re: [Caml-list] Stupid question re:modules

[Julien Moutinho]

On Thu, Aug 23, 2007 at 08:32:22PM -0400, Brian Hurt wrote:
>
> I should just know this.  So let's say I have two module types defined:
>
> module type Foo = sig
> 	type 'a t
> 	val foo : 'a -> 'a t
> end;;
>
> module type Bar = sig
> 	type 'a t
> 	val bar : 'a -> 'a t
> end;;
>
> Now, I want to define a module that is both a Foo and a Bar without cutting 
> and pasting the module definitions around.  I've been trying to do:
>
> module Baz : sig
> 	type 'a baz
> 	include Foo with type 'a t = 'a baz
> 	include Bar with type 'a t = 'a baz
> end;;
>
> but this blows up on the Bar line (multiple definitions of 'a t).
>
> There is a solution to this, I'm just being stupid and forgetting what it 
> is.  Hints would be appreciated.
>
> Brian

I might haven't got exactly what you want,
but, with control over Foo and Bar,
and starting from the structure,
I would have written :

module type Foo =
  functor (T: T) ->
  sig val foo : 'a -> 'a T.t end

module type Bar =
  functor (T: T) ->
  sig val bar : 'a -> 'a T.t end

module Baz =
  functor (T: T) ->
  functor (Foo: Foo) ->
  functor (Bar: Bar) ->
  struct
	include T
	type 'a baz = 'a t
	module Foo = Foo(T)
	module Bar = Bar(T)
	include Foo
	include Bar
  end

Then I would have generated the signature automatically with -i.

But if you really want a short .mli without redundancy,
I see no solution, even for the above functorized code.
Because :
  1/ unable to override or delete elements from a signature
  2/ unable to get the signature of a structure
     from a functor in a signature.

I mean, given this :

module type T = sig end
module type F =
  functor (T: T) ->
  sig end

This signature does not work :

module B :
  functor (T: T) ->
  functor (F: F) ->
  sig module M : F(T) end

nor this :

module B :
  functor (T: T) ->
  functor (M: F(T)) ->
  sig end

nor this :

module B :
  functor (T: T) ->
  functor (F: F) ->
  sig include F(T) end

nor this :

module B :
  functor (T: T) ->
  functor (F: F) ->
  sig open F(T) end

Despite the fact that the structured version of
the first two signatures work.
And that I do not understand why they all do not work,
since it's a kind of substitution, isn't it?

Anyway, hope you'll find a way to _avoid_ your problem.

Re: [Caml-list] Stupid question re:modules

[Vincent Aravantinos]

Le 24 août 07 à 02:32, Brian Hurt a écrit :
>
> I should just know this.  So let's say I have two module types  
> defined:
>
> module type Foo = sig
> 	type 'a t
> 	val foo : 'a -> 'a t
> end;;
>
> module type Bar = sig
> 	type 'a t
> 	val bar : 'a -> 'a t
> end;;
>
> Now, I want to define a module that is both a Foo and a Bar without  
> cutting and pasting the module definitions around.  I've been  
> trying to do:
>
> module Baz : sig
> 	type 'a baz
> 	include Foo with type 'a t = 'a baz
> 	include Bar with type 'a t = 'a baz
> end;;
>
> but this blows up on the Bar line (multiple definitions of 'a t).
>
> There is a solution to this, I'm just being stupid and forgetting  
> what it is.  Hints would be appreciated.
>
> Brian

See also :
  http://caml.inria.fr/pub/ml-archives/caml-list/2006/12/ 
c7461312202053f2213a9bb33206fcb8.en.html

BTW, I thought there was some kind of research about inheritance and  
modules (heard about "mixin modules"), is there any plan to implement  
this one day in ocaml ?

The lack of real inheritance with modules is a major restriction IMHO...




--
Vincent Aravantinos
PhD Student - LIG - CAPP Team

Re: [Caml-list] Stupid question re:modules

[David Teller]

Ask Tom Hirschowitz. I'm not sure it's still his research interest, unfortunately.

Cheers,
 David

On Fri, 2007-08-24 at 12:16 +0200, Vincent Aravantinos wrote:

> BTW, I thought there was some kind of research about inheritance and  
> modules (heard about "mixin modules"), is there any plan to implement  
> this one day in ocaml ?
> 
> The lack of real inheritance with modules is a major restriction IMHO...