From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from mail1-relais-roc.national.inria.fr (mail1-relais-roc.national.inria.fr [192.134.164.82])
	by walapai.inria.fr (8.13.6/8.13.6) with ESMTP id p9OL1FtU024897
	for <caml-list@sympa-roc.inria.fr>; Mon, 24 Oct 2011 23:01:15 +0200
X-IronPort-Anti-Spam-Filtered: true
X-IronPort-Anti-Spam-Result: AucCAMzQpU7AbSoIe2dsb2JhbABDhHWiFYIJIgEBFiYEIYFuAQEEASNWEAsJBQoCAiYCAhQYMYgTAqJNkW0SgR6FfDNhBIx3mG8
X-IronPort-AV: E=Sophos;i="4.69,400,1315173600"; 
   d="scan'208";a="125693709"
Received: from einhorn.in-berlin.de ([192.109.42.8])
  by mail1-smtp-roc.national.inria.fr with ESMTP/TLS/DHE-RSA-AES256-SHA; 24 Oct 2011 23:01:10 +0200
X-Envelope-From: oliver@first.in-berlin.de
Received: from first (e178015013.adsl.alicedsl.de [85.178.15.13])
	(authenticated bits=0)
	by einhorn.in-berlin.de (8.13.6/8.13.6/Debian-1) with ESMTP id p9OL18HH002999
	(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NOT);
	Mon, 24 Oct 2011 23:01:08 +0200
Received: by first (Postfix, from userid 1000)
	id 89AF4154012A; Mon, 24 Oct 2011 23:01:08 +0200 (CEST)
Date: Mon, 24 Oct 2011 23:01:08 +0200
From: oliver <oliver@first.in-berlin.de>
To: Gerd Stolpmann <info@gerd-stolpmann.de>
Cc: Diego Olivier Fernandez Pons <dofp.ocaml@gmail.com>,
        Xavier Leroy <Xavier.Leroy@inria.fr>, caml-list@inria.fr
Message-ID: <20111024210108.GE1838@siouxsie>
References: <CAHqiZ-J15s9PiVnvT+rw8KF--OooFLyP8YRk6x+e31dTEGX_SQ@mail.gmail.com>
 <4EA54BD0.8090009@inria.fr>
 <CAHqiZ-LXF8uMX2eib_=YAXHCQrs5Me94PVOkh5_8NwNx+uVZnQ@mail.gmail.com>
 <1319460013.18639.90.camel@thinkpad>
 <20111024124617.GA2287@siouxsie>
 <1319461110.18639.96.camel@thinkpad>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <1319461110.18639.96.camel@thinkpad>
User-Agent: Mutt/1.5.20 (2009-06-14)
X-Scanned-By: MIMEDefang_at_IN-Berlin_e.V. on 192.109.42.8
Subject: Re: [Caml-list] How to write an efficient interpreter

On Mon, Oct 24, 2011 at 02:58:30PM +0200, Gerd Stolpmann wrote:
> Am Montag, den 24.10.2011, 14:46 +0200 schrieb oliver:
> > On Mon, Oct 24, 2011 at 02:40:13PM +0200, Gerd Stolpmann wrote:
> > > Am Montag, den 24.10.2011, 13:50 +0200 schrieb Diego Olivier Fernandez
> > > Pons:
> > > >      Caml-list,
> > > > 
> > > > 
> > > > Xavier Leroy wrote
> > > > > Compiling to bytecode is probably overkill.
> > > > 
> > > > 
> > > > 
> > > > I think that writing my own bytecode interpreter is looking for
> > > > trouble. Same for compiling to an existing bytecode.
> > > > 
> > > > 
> > > > The language being a kind of SQL, most of the work is to properly
> > > > execute the comprehensions (= queries).
> > > > 
> > > > 
> > > > For instance
> > > > 
> > > > 
> > > >      range numbers = 0 .. 100;
> > > >      {int}  sqrtLessThan [k in numbers] = { x | x in numbers : x * x
> > > > <= k };
> > > > 
> > > > 
> > > > There are smarter ways to implement this than a double loop
> > > > 
> > > > 
> > > > I was rather thinking of translating on-the-fly into Caml code and
> > > > letting Caml do the job. Is that technically possible (rewriting a
> > > > toplevel ? a CamlP4 grammar ?). If so guess I would have to license
> > > > the Caml compiler from the INRIA.
> > > 
> > > I don't think you need that, because you can load compiled OCaml code
> > > dynamically (look into the Dynlink library).
> > [...]
> > 
> > Maybe you both talk about different things...
> > 
> > What you seem to talk about is not interpreter but compiler stuff,
> > and later bind it together?!
> 
> Exactly. But what's the big difference? You want to run code of a
> domain-specific language.
[...]

The binding process would not be necessary.
(Or do we talk about different things here?)


> 
> > I would assume, with "translating on-the-fly into Caml code"
> > is something meant that could be done via partial application.
> 
> The interpretative overhead does not go away with this technique. E.g.
> looking up variables in your interpretation environment. I'd consider
> this as a micro-optimization that doesn't bring much improvement.
[...]

But it's an easy-going implementation.
You can just use OCaml directly.
The for-loop of the DSL for example is then directly
translated to a for loop in Ocaml.

For optimisation it maybe is better to work on an AST
and optimize it, before translating it to something that can be called.

I had rather the easy-going in mind, than the performance problem,
referring to the "on-the-fly" conversion.

Ciao,
   Oliver