Date: Thu, 31 Jan 2002 10:17:23 +1100
From: Nathan Scott <nathans@sgi.com>
To: "David S. Miller" <davem@redhat.com>
Cc: a.gruenbacher@computer.org, linux-xfs@oss.sgi.com
Subject: Re: extended attributes interface
Message-ID: <20020131101723.E82500@wobbly.melbourne.sgi.com>
References: <20020130.144647.21928212.davem@redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.2.5i
In-Reply-To: <20020130.144647.21928212.davem@redhat.com>; from davem@redhat.com on Wed, Jan 30, 2002 at 02:46:47PM -0800
Sender: owner-linux-xfs@oss.sgi.com
Precedence: bulk
Status: O
Content-Length: 1012
Lines: 28

On Wed, Jan 30, 2002 at 02:46:47PM -0800, David S. Miller wrote:
> 
> All the values that go through these syscalls seem to be
> opaque and filesystem specific.  Therefore can I ask the filesystems
> that use these things use fixed sized types such as "u32" "u16" et al.
> instead of things such as "long" or "int"?

Yes, agreed.  The main consumer of this interface currently
is Andreas' POSIX ACL code, and IIRC he does use fixed size
types for his ACL attributes.  The man pages do refer to this
issue, I will flesh out the wording to be a bit more explicit
in this regard.

thanks.

> The reason I ask is, unless strict sized types are used it is going
> to be a real pain in the ass to translate the types passed to these
> system calls in mixed 32-bit/64-bit environments.  This is thus going
> to be a mess on sparc64, ppc64, mips64, ia64, and probably others I
> have forgotten :-)
> 
> If strict sized types are used for the attributes, then no
> translations will need to occur at all.
> 

-- 
Nathan