[mod.computers.pyramid] another c bug?

brahms@spp5.UUCP.UUCP (08/02/86)

The following problem came up while I was trying to compile larn on a
pyramid.  I'm curious to know if it is a bug with the Pyramid C compiler
or something that just happens to work on a number of other machines but
not the Pyramid?  It appears that the called procedure writes information
on what is believes is the top of the stack.  Anyway, to get around it, I
modifed lprintf from larn to included extra arguments such as in
five_int_args below.

Script started on Fri Aug  1 15:39:12 1986
demo: cat t.c
#include <stdio.h>

five_int_args (args, a, b, c, d)
int args, a, b, c, d;

{
    int     *pargs;
    int     i;

    pargs = &args;
    while (*pargs != 0)
        printf ("%d : ", *pargs++);
    printf ("%d\n", 0);
}

show_int_args (args)
int args;

{
    int     *pargs;
    int     i;

    pargs = &args;
    while (*pargs != 0)
        printf ("%d : ", *pargs++);
    printf ("%d\n", 0);
}

main()
{
    show_int_args (0);
    show_int_args (1, 0);
    show_int_args (1, 2, 0);
    show_int_args (1, 2, 3, 0);
    show_int_args (1, 2, 3, 4, 0);
    show_int_args (1, 2, 3, 4, 5, 0);
    five_int_args (1, 2, 3, 4, 0);
}
demo: cc -o t t.c
demo: t
0
1 : 0
1 : 2 : -1072901880 : 320 : 12 : -1072902016 : 0
1 : 2 : -1072901880 : 0
1 : 2 : -1072901880 : 4 : 0
1 : 2 : -1072901880 : 4 : 5 : 0
1 : 2 : 3 : 4 : 0
demo: what /lib/ccom
/lib/ccom
        CCOMP_3.1.9     Mar 19 16:50:26 1986
        yaccpar 4.1     (Berkeley)      2/11/83
        FLUSH.c 1.2 6/10/81
        NEW.c 1.3 6/10/81
        UNIT.c 1.2 6/10/81
        WRITLN.c 1.2 6/10/81
        ERROR.c 1.10 1/10/83
        PCEXIT.c 1.1 10/30/80
        PCLOSE.c 1.6 1/21/83
        PCSTART.c 1.8 1/10/83
        PFCLOSE.c       1.3     (Berkeley)      1/21/83
        PERROR.c 1.2 1/10/83
        PFLUSH.c 1.2 1/21/83
        EXCEPT.c 1.3 1/10/83
        malloc.c        4.3 (Berkeley) 9/16/83

script done on Fri Aug  1 15:40:23 1986

chris@MIMSY.UMD.EDU.UUCP (08/02/86)

This is not a compiler bug.  It is not legal to take the address
of one argument and use that to compute the address of another
argument.  It works on all pure stack machines as long as you know
the direction of stack growth.  It does not work on mixed stack/register
machines like the Pyramid.

The proper thing to do in larn's case is to use varargs and vprintf.
I am not sure if vprintf is in the `ucb universe' C library, but


	vprintf(fmt, args)
		char *fmt;
		va_list args;
	{

		_doprnt(fmt, args, stdout);
	}

should work (since I know how Pyramid did their varargs).  Given
vprintf, change

	show_int_args (args)
	int args;
	{
	    int     *pargs;
	    int     i;
	
	    pargs = &args;
	    while (*pargs != 0)
	        printf ("%d : ", *pargs++);
	    printf ("%d\n", 0);
	}

to

	#include <varargs.h>

	show_int_args(va_alist)
		va_dcl
	{
		register int i;
		va_list p;

		va_start(p);
		for (;;) {
			i = va_arg(p, int);
			if (i == 0)
				break;
			printf("%d : ", i);
		}
		printf("0\n");	/* simplified */
		va_end(p);
	}

	/* emulate a standard printf() */
	xprintf(va_alist)
		va_dcl
	{
		char *fmt;
		va_list p;

		va_start(p);
		fmt = va_arg(p, char *);
		vprintf(fmt, p);
		va_end(p);
	}

Chris