From owner-freebsd-arch  Mon Jan 22  2:28:33 2001
Delivered-To: freebsd-arch@freebsd.org
Received: from pcnet1.pcnet.com (pcnet1.pcnet.com [204.213.232.3])
	by hub.freebsd.org (Postfix) with ESMTP id 049B137B400
	for <arch@FreeBSD.ORG>; Mon, 22 Jan 2001 02:28:15 -0800 (PST)
Received: (from eischen@localhost)
	by pcnet1.pcnet.com (8.8.7/PCNet) id FAA18961;
	Mon, 22 Jan 2001 05:27:47 -0500 (EST)
Date: Mon, 22 Jan 2001 05:27:47 -0500 (EST)
From: Daniel Eischen <eischen@vigrid.com>
To: Bruce Evans <bde@zeta.org.au>
Cc: Warner Losh <imp@harmony.village.org>,
	"Jacques A. Vidrine" <n@nectar.com>, arch@FreeBSD.ORG
Subject: Re: Request For Review: libc/libc_r changes to allow -lc_r 
In-Reply-To: <Pine.BSF.4.21.0101221935540.27700-100000@besplex.bde.org>
Message-ID: <Pine.SUN.3.91.1010122051818.17483B@pcnet1.pcnet.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-freebsd-arch@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

On Mon, 22 Jan 2001, Bruce Evans wrote:
> On Sun, 21 Jan 2001, Warner Losh wrote:
> 
> > In message <Pine.SUN.3.91.1010121162703.14751A-100000@pcnet1.pcnet.com> Daniel Eischen writes:
> > : Oops, sorry, I missed the second question.  You need _foo to be
> > : used within libc, so that when libc_r/libpthread is linked in,
> > : it can provide a replacement function for it.  We also need to
> > : determine if the function is a cancellation point or not, so
> > : if you just had foo and __sys_foo, libc_r/libpthread would have
> > : no way of knowing if foo was called from within libc or from
> > : the user application.  The former is not a cancellation point,
> > : while the latter is (if foo is read for example).
> > 
> > I understand that.  I guess my question is why name it _foo instead of
> > __foo?  I see the need for the tripartiteness, just not the need to
> > call it _foo.
> 
> (1) Underscores are verbose and ugly.
> (2) _foo is usually sufficient.  _[a-z] is not entirely in the user
>     namespace like you are claimed.  From the 1990 ISO standard:
>     "All identifiers that begin with an underscore are always
>     reserved for use as identifiers with file scope in both the
>     ordinary identifier and tag name spaces".  In practice, this
>     means that the implementation can use names beginning with
>     _[a-z] except for macro names and global variables that are
>     used in macros.  E.g., errno must be defined as (*__error())
>     and not as (*_error()), since the latter would break the
>     standard-conforming application code:
>       #include <errno.h>
>       void foo(void) { int _error = errno; }
>     A single underscore is sufficient in all other cases.  E.g.,
>     struct member names are in a nested namespace so they don't
>     conflict with variable names at all.  They may still need a
>     single underscore so that they don't conflict with macro
>     names.
> (3) We have some precedence for using _foo.
> (4) NetBSD uses _foo (at least in old versions).

How about this compromise: __sys_foo(T), _foo(W), foo(W).
Noone but libc_r, except perhaps for exit.c, needs to reference
__sys_foo.

Poking around with nm on Solaris seems to show that they use
__foo for system calls, and _bar for library functions.

-- 
Dan Eischen


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message