From owner-cvs-sys  Sun Aug 27 11:59:04 1995
Return-Path: cvs-sys-owner
Received: (from majordom@localhost)
          by freefall.FreeBSD.org (8.6.11/8.6.6) id LAA00785
          for cvs-sys-outgoing; Sun, 27 Aug 1995 11:59:04 -0700
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.34])
          by freefall.FreeBSD.org (8.6.11/8.6.6) with ESMTP id LAA00778
          ; Sun, 27 Aug 1995 11:58:50 -0700
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id EAA26062; Mon, 28 Aug 1995 04:56:01 +1000
Date: Mon, 28 Aug 1995 04:56:01 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199508271856.EAA26062@godzilla.zeta.org.au>
To: bde@zeta.org.au, wollman@lcs.mit.edu
Subject: Re: cvs commit: src/sys/vm vm_map.h vm_map.c
Cc: CVS-commiters@freefall.FreeBSD.org, bde@freefall.FreeBSD.org,
        cvs-sys@freefall.FreeBSD.org
Sender: cvs-sys-owner@FreeBSD.org
Precedence: bulk

>> 1) for syscalls.  The conversion from the application's args is
>[...]
>>       bar_args *'.  Perhaps the right method is to pass a `void *'
>>       or `int *' and convert it in each function, like you have to
>>       do for qsort().

>This is precisely the approach I took when doing this rationalization
>in 1.x.  I would have preferred to continue it (and fix the current
>syscalls to match the old model), but other people wanted to do it the
>new (broken) 4.4-Lite way, and I let them have their way.

I'm now automatically generating prototypes from the data
syscalls.master.  I cast the initializers in init_sysent.c to hide
the warnings.  I had to add some new optional fields to handle
non-orthogonal names arg struct tags such as `args' instead of
getfoo_args for getfoo(), and non- int return types for exit() and
umask().  I included the prototypes in <sys/systm.h>.  They'll have
to be moved, because this increases the kernel build time by 10%!
The arg struct declarations should be automatically generated too.
syscalls.master will have to look more like vnode_if.src for that.
I could generate a nicer visible interface, even one with args on
the stack (it's not clear whether the overhead for copying the args
again would be > 0).

Bruce