Version française
Home     About     Download     Resources     Contact us    
Browse thread
camlp4 & free variables
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: -- (:)
From: Nicolas Pouillard <nicolas.pouillard@i...>
Subject: Re: [Caml-list] camlp4 & free variables
On 7/20/06, Christophe TROESTLER <Christophe.Troestler@umh.ac.be> wrote:
> On Thu, 20 Jul 2006, "Nicolas Pouillard" <nicolas.pouillard@gmail.com> wrote:
> >
> > On 7/20/06, Christophe TROESTLER <Christophe.Troestler@umh.ac.be> wrote:
> > > Hi,
> > >
> > > Is it possible to write a camlp4 syntax extension that "extract" the
> > > free variables of an expression?
> > You are quite lucky, the camlp4 version has one module to do that!
>
> Thank you very much for your help!  Is there an updated manual for
> camlp4 (if possible written more from a user perspective :)?
>

Alas not now.

> If I understand well your code (I really ought to learn camlp4!)
> FV.free_vars first argument is a set of names which should net be
> considered as free variables.  Is it possible to do it contextually,
> e.g. in
>
>      open_out("X" ^ x)
>
> only "x" is returned as free because "open_out" and "^" are known from
> Pervasives?  Or in
>
>      let z = 1
>      ...
>      FUNCTION(x + z)
>
> only "x" is returned as free because "z" is known from the context?
>

In my previous program, I provide a function f that takes an AST:

f <<let z = x + 2 in x + 2 * y * x * z>>

Here the AST is given using a quotation but there is no relation
between the code outside and inside the quotation so:

let x = 42 in f <<let z = x + 2 in x + 2 * y * x * z>>

This code does not bind x inside the quotation.

Now if you want to use FUNCTION(...) as a sort of macro function there
is a camlp4 filter for that.

Let's call close_expr the this macro function (instead of FUNCTION)

$ cat expression_closure_filter.ml
(* camlp4r *)
#default_quotation "expr";

open Camlp4.PreCast;
open Format;

module FV = Camlp4.Struct.FreeVars.Make Ast;
module S = FV.S;

value _loc = Loc.ghost;

value pervasives =
  let list =
    [ "+"; "-"; "/"; "*" (* ... *) ]
  in List.fold_right S.add list S.empty;

value collect_free_vars_sets =
  object (self)
    inherit FV.fold_free_vars [S.t] S.add ~env_init:pervasives S.empty as super;
    value free_sets = [];
    method set_free free = {< free = free >};
    method expr =
      fun
      [ << close_expr $e$ >> -> (self#expr e)#add_current_free#set_free free
      | e -> super#expr e ];
    method add_current_free = {< free_sets = [ free :: free_sets ] >};
    method free_sets = free_sets;
  end;

value apply_close_expr next_free_set =
  object (self)
    inherit Ast.map as super;
    method expr =
      fun
      [ << close_expr $e$ >> ->
          let e = self#expr e in
          let fv = next_free_set () in
          S.fold (fun x acc -> << fun ~ $x$ -> $acc$ >>) fv e
      | e -> super#expr e ];
  end;

value f st =
  let fv_sets = ref (collect_free_vars_sets#str_item st)#free_sets in
  let next_free_set () =
    match fv_sets.val with
    [ [] -> assert False
    | [x::xs] -> let () = fv_sets.val := xs in x ]
  in (apply_close_expr next_free_set)#str_item st;

AstFilters.register_str_item_filter f;

$ ocamlc -c -pp camlp4rf expression_closure_filter.ml

$ cat test_expression_closure_filter.ml
(* x and y are free *)
close_expr(x y);;

(* bind x *)
let x = 42;;

(* y is free *)
close_expr(x y);;

(* bind y locally so the expr is closed *)
close_expr(let y = x in x y);;

(* bind y locally but outside, z is free *)
let y = x in close_expr(x z y);;

$ camlp4o ./expression_closure_filter.cmo test_expression_closure_filter.ml
(* x and y are free *)
let _ = fun ~y ~x -> x y
(* bind x *)
let x = 42
(* y is free *)
let _ = fun ~y -> x y
(* bind y locally so the expr is closed *)
let _ = let y = x in x y
(* bind y locally but outside, z is free *)
let _ = let y = x in fun ~z -> x z y


> Thanks again -- the new camlp4 version seem to really shine
> w.r.t. the previous one !

Thanks, I'm glad to see camlp4 receive appreciations :)

Cheers,

-- 
Nicolas Pouillard