Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83 SMI; site sun.uucp
Path: utzoo!linus!decvax!decwrl!sun!gnu
From: gnu@sun.uucp (John Gilmore)
Newsgroups: net.micro.68k
Subject: nargs not possible with optimized code
Message-ID: <401@sun.uucp>
Date: Wed, 28-Dec-83 11:57:16 EST
Article-I.D.: sun.401
Posted: Wed Dec 28 11:57:16 1983
Date-Received: Thu, 29-Dec-83 01:19:20 EST
References: <363@sii.UUCP>
Organization: Sun Microsystems, Inc.
Lines: 25

We used to run Unisoft Unix and even a year ago their C optimizer would
optimize out the stack-adjusting instructions where possible.  For example,
it would turn:

	movl	arg,sp@-
	jsr	subr
	addql	#4,sp		<-- the instruction nargs() looks for
	movl	arg2,sp@-
	jsr	subr2
	addql	#4,sp		<-- the instruction nargs() looks for

into:

	movl	arg,sp@-
	jsr	subr
	movl	arg2,sp@	<-- the instruction is gone...
	jsr	subr2
	addql	#4,sp

Unless your calling sequence is specifically designed to support
"nargs", it can't be relied on.  The "standard" 68000 C calling
sequence (designed by Jack Test at MIT?) doesn't support it -- rightly,
I think.  Part of the reason subroutine calls were so expensive on the
Vax was this kind of unused generality.