From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <Xavier.Leroy@inria.fr>
X-Original-To: caml-list@sympa.inria.fr
Delivered-To: caml-list@sympa.inria.fr
Received: from mail1-relais-roc.national.inria.fr (mail1-relais-roc.national.inria.fr [192.134.164.82])
	by sympa.inria.fr (Postfix) with ESMTPS id 02E917ED5C
	for <caml-list@sympa.inria.fr>; Tue, 31 Jul 2012 16:29:41 +0200 (CEST)
X-IronPort-AV: E=Sophos;i="4.77,686,1336341600"; 
   d="scan'208";a="168686982"
Received: from estephe.inria.fr (HELO [128.93.11.95]) ([128.93.11.95])
  by mail1-relais-roc.national.inria.fr with ESMTP; 31 Jul 2012 16:29:40 +0200
Message-ID: <5017EBD4.9080904@inria.fr>
Date: Tue, 31 Jul 2012 16:29:40 +0200
From: Xavier Leroy <Xavier.Leroy@inria.fr>
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:12.0) Gecko/20120430 Thunderbird/12.0.1
MIME-Version: 1.0
To: caml-list@inria.fr
References: <4350198.FKG0AFFZGv@arges> <1343739554.17140.423.camel@thinkpad> <4123447.LCvNL5nQdx@arges>
In-Reply-To: <4123447.LCvNL5nQdx@arges>
X-Enigmail-Version: 1.5pre
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Subject: Re: [Caml-list] Segmentationfault on 64bit when called from C

On Tuesday 31 July 2012 14:59:14 Gerd Stolpmann wrote:

> The -shared switch generates a library that is prepared to be loaded
> with Dynlink. Dynlink registers the frametable.

You're right that .so libraries produced by -shared should only be
used in conjunction with Dynlink.  However, in this case the frame
table isn't the culprit, as the crash occurs before the GC even has a
chance to run.

On 07/31/2012 04:03 PM, Georg Martius wrote:

> ocamlopt -output-obj -o camlcode.o interface.c m1.ml
> gcc -fPIC -g -Wall  -c -I/usr/local/lib/ocaml interface.c
> gcc -fPIC -o test main.c camlcode.o  interface.o -L/usr/local/lib/ocaml -lunix 
> -lm -lasmrun -ldl
> 
> works, but 
> ocamlopt -o libmytest.so -output-obj  interface.c m1.ml
> gcc -Wall main.c  -lmytest -o test
> does not! Is it clear why it doesn't work?

See my comments on PR#5707.  In the latter case, the dynamic loader is
invoked and trashes registers r10 and r11, which ocamlopt uses to pass
arguments to your "foo" function.  We'll have to come up with a better
choice of parameter registers.

> So I get away with static linking for now.

Please do if at all possible.  ocamlopt was designed for static
linking, and the relatively recent addition of dynamic linking support
isn't completely mature yet.

- Xavier Leroy