The sharing mechanism in the Marshal module

The sharing mechanism in the Marshal module

[Deokhwan Kim]

Hello,

I thought the following code would produce 2 in (* 1 *), but as it is it
outputs 1:

let incr =
let c = ref 0 in
function () -> c:= !c + 1 ; !c

let save =
Marshal.to_string incr [Marshal.Closures]

let incr1 =
(Marshal.from_string save 0 : unit -> int)

let () =
Printf.printf "%d\n" (incr ()) ;
Printf.printf "%d\n" (incr1 ()) (* 1 *)

Why? In addtion, what should I do if I really want the sharing of the c
counter between incr and incr1?

Thanks.

Re: The sharing mechanism in the Marshal module

[Zheng Li]

Hi,

Deokhwan Kim <dk@ropas.snu.ac.kr> writes:
> I thought the following code would produce 2 in (* 1 *), but as it is it
> outputs 1:
That's the semantic of marshal, it serialize functional values as *closures*,
i.e. functions carrying its whole environments.  In your example, the incr1
also gets a copy of incr's environment, which has a copy of c included.

Despite some flaws being spotted in the marshal module last month [1], as far
as your example goes, you should stick with the closure semantics.

> Why? In addtion, what should I do if I really want the sharing of the c
> counter between incr and incr1?

There are some workarounds, but it depends on why you need marshaling. In
many cases, you don't actually need marshal for a shallow copy.

[1] http://caml.inria.fr/pub/ml-archives/caml-list/2007/02/84333195e90a8f13ac9fea7fb6a706bf.en.html

--
Zheng Li
http://www.pps.jussieu.fr/~li

Re: [Caml-list] The sharing mechanism in the Marshal module

[Roland Zumkeller]

On 03/03/07, Deokhwan Kim <dk@ropas.snu.ac.kr> wrote:
> In addtion, what should I do if I really want the sharing of the c
> counter between incr and incr1?

It suffices to define "c" outside of "incr":

let c = ref 0;;

let incr () = c:= !c + 1 ; !c

-- 
http://www.lix.polytechnique.fr/~zumkeller/