Path: utzoo!attcan!uunet!mcvax!hp4nl!kunivv1!phoibos!ge
From: ge@phoibos.UUCP (Ge Weijers)
Newsgroups: comp.lang.c
Subject: Re: Passing variable #of args
Summary: portability of var. arg#
Message-ID: <517@phoibos.UUCP>
Date: 23 Jan 89 12:22:45 GMT
References: <470@marob.MASA.COM>
Organization: University of Nijmegen, The Netherlands.
Lines: 34

In article <470@marob.MASA.COM>, daveh@marob.MASA.COM (Dave Hammond) writes:
> I was surprised to see the following in a "GNU-something" source file:
> 
> lprintf(fmt,a1,a2,a3,a4,a5,a6,a7,a8,a9,10)
> char *fmt;
> {
> 	char buf[BUFSIZ];
> 	sprintf(buf,fmt,a1,a2,a3,a4,a5,a6,a7,a8,a9,10)
> 	...
> }
> 
> Given the availability of varargs, is this style still acceptable and,
> more importantly, is it portable?
> 
> --
> Dave Hammond
> ...!uunet!masa.com!{marob,dsix2}!daveh

This is NOT portable. Some compilers use strange calling conventions which
will cause this example to break. An example:

The Turbo C compiler for the Atari ST (sold in Europe) uses registers to pass
arguments. The calling convention is the following (except for details, I'm
producing this from memory):

The first 3 scalar (int, long, char etc.) variables are passed in registers
D0-D2, and the first 2 pointer variables in A0-A1. The call
f(3,&var) and f(&var,3) will yield the same code. The above example breaks.
It is possible to use the ANSI <stdarg.h> variable argument passing, because
the compiler does not pass arguments in registers that correspond to the ...
in an ANSI prototype (line lprintf(char *format, ...))

You'd be better off using <varargs.h> or <stdarg.h>, then it is at least obvious
where you have to change a program.