From mboxrd@z Thu Jan  1 00:00:00 1970
Received: (from weis@localhost) by pauillac.inria.fr (8.7.6/8.7.3) id BAA01372 for caml-redistribution@pauillac.inria.fr; Mon, 10 Apr 2000 01:25:08 +0200 (MET DST)
Resent-Message-Id: <200004092325.BAA01372@pauillac.inria.fr>
Received: from concorde.inria.fr (concorde.inria.fr [192.93.2.39]) by pauillac.inria.fr (8.7.6/8.7.3) with ESMTP id QAA12082 for <weis@pauillac.inria.fr>; Thu, 6 Apr 2000 16:25:53 +0200 (MET DST)
Received: from pauillac.inria.fr (pauillac.inria.fr [128.93.11.35])
	by concorde.inria.fr (8.8.7/8.8.7) with ESMTP id QAA22065;
	Thu, 6 Apr 2000 16:25:52 +0200 (MET DST)
Received: (from xleroy@localhost) by pauillac.inria.fr (8.7.6/8.7.3) id QAA22815; Thu, 6 Apr 2000 16:25:52 +0200 (MET DST)
Message-ID: <20000406162552.49765@pauillac.inria.fr>
Date: Thu, 6 Apr 2000 16:25:52 +0200
From: Xavier Leroy <Xavier.Leroy@inria.fr>
To: Markus Mottl <mottl@miss.wu-wien.ac.at>,
        Pierre Weis <Pierre.Weis@inria.fr>
Cc: OCAML <caml-list@inria.fr>
Subject: Re: cyclic value construction ("let rec")
References: <20000406151819.42281@pauillac.inria.fr> <200004061334.PAA18159@miss.wu-wien.ac.at>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.89.1
In-Reply-To: <200004061334.PAA18159@miss.wu-wien.ac.at>; from Markus Mottl on Thu, Apr 06, 2000 at 03:34:18PM +0200
Resent-From: weis@pauillac.inria.fr
Resent-Date: Mon, 10 Apr 2000 01:25:08 +0200
Resent-To: caml-redistribution@pauillac.inria.fr

> > [Pierre Weis:]
> > This certainly suggests to allow the export of an immutable view of a
> > record type with mutable fields. This way you could do the
> > initialization in a safe way (no magic) using side effects, and still
> > export a safe immutable type to the external world.
> 
> [Markus Mottl:]
> Sounds like a good idea! Using powerful "magic" is probably too dangerous
> for "everyday"-use and definitely not in accordance with the "zero defect"
> ambitions of the type system...
> Although it would sometimes be nice to even hide specific fields of the
> record, this would probably not work well together with separate
> compilation.  However, the memory layout of the fields does not change by
> just omitting the "mutable" declaration, so this should not do any harm.

Alas, it can do a lot of harm.  For one thing, you could break type
safety this way, just like with polymorphic references:

A.ml:
        type 'a t = { mutable contents: 'a }
        let assign t v = t.contents <- v

A.mli:
        type 'a t = { contents: 'a}
        val assign: 'a t -> 'a -> unit

Client.ml:
        open A
        ...
          let x = { contents = [] } in
          assign x [1];
          x.contents = [true]

When typing Client.ml, since "contents" is assumed immutable, the
definition of x is a syntactic value, hence x receives type
        forall 'a.  'a list t
But of course this typing is invalidated by the call to "assign",
and you end up comparing an int list to a bool list -- a typing violation.

Some compiler optimisations, specific to immutable structures, could
similarly be broken.

So, no, we can't allow exporting a record with different mutability
annotations than in its definition.

- Xavier Leroy