caml-list - the Caml user's mailing list
 help / color / mirror / Atom feed
From: Markus Mottl <mottl@miss.wu-wien.ac.at>
To: debourse@pascal.enst.fr (Benoit Deboursetty)
Cc: caml-list@inria.fr (OCAML)
Subject: Re: cyclic value construction ("let rec")
Date: Mon, 3 Apr 2000 14:57:19 +0200 (MET DST)	[thread overview]
Message-ID: <200004031257.OAA29741@miss.wu-wien.ac.at> (raw)
In-Reply-To: <Pine.GSO.4.02.10003282041400.6520-100000@pascal.enst.fr> from "Benoit Deboursetty" at Mar 30, 2000 10:12:33 PM

> It would be interesting to know whether people have other workarounds.
> Also, a generic graph module which would implement "map", "iter" etc. in a 
> cycle-safe way (like Marshal) would perhaps be interesting in the
> 'user-contributed stdlib extension'...

You will probably always have to rely on "unsafe" things for such
purposes.  My quick hack for problems like this looks as follows:

  type 'a node = { data : 'a; next : 'a node }

  let make_circle n : 'a node =
    let module Dummy = struct
      type 'a t = { data : 'a; mutable next : 'a t } end in
    let res = { Dummy.data = n; Dummy.next = Obj.magic 0 } in
    res.Dummy.next <- res;
    Obj.magic res

  let rec iter f node = f node.data; iter f node.next

  let _ = iter (fun x -> print_int x; print_newline ()) (make_circle 42)

Type "'a node" is immutable from the outside. However, exploiting the
undocumented (good reason for this ;-) type casting features of OCaml,
one can cheat and define an identical type within a local module -
which is mutable!

Initialisation, too, relies on "unsafe" features - we simply put a
reference to a dummy value into the slot until it can be filled with the
"right" value. The return value is cast back again to the "global type"
as indicated in the type restriction in the definition of "make_circle"
- this is important, because otherwise you can crash your program (it
would have "any" type).

As far as I know, this should be safe, because the GC doesn't care
about the static types of the values - it only looks at the value tags
at runtime. Of course, the above will only work as long as the memory
layout of the "local" type agrees with the one of the "global type".

Thus, you can confine the "unsafe" part to specific functions (e.g.
"make_circle"). The iteration function "iter" does not change anything
and can therefore work on the "global" type.

Best regards,
Markus Mottl

-- 
Markus Mottl, mottl@miss.wu-wien.ac.at, http://miss.wu-wien.ac.at/~mottl



  reply	other threads:[~2000-04-06 13:14 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2000-03-30 20:12 Benoit Deboursetty
2000-04-03 12:57 ` Markus Mottl [this message]
2000-04-06 13:18   ` Pierre Weis
2000-04-06 13:34     ` Markus Mottl
2000-04-06 14:25       ` Xavier Leroy
2000-04-06 15:12         ` Markus Mottl
2000-04-10  0:03         ` Pierre Weis
2000-04-10  1:41           ` Markus Mottl

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=200004031257.OAA29741@miss.wu-wien.ac.at \
    --to=mottl@miss.wu-wien.ac.at \
    --cc=caml-list@inria.fr \
    --cc=debourse@pascal.enst.fr \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).