Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!mips!cs.uoregon.edu!ogicse!hsdndev!dartvax!mars!nic!bbn.com!cjross
From: cjross@bbn.com (chris ross)
Newsgroups: comp.lang.c
Subject: passing address of floating-point parameter
Message-ID: <64623@bbn.BBN.COM>
Date: 12 Jun 91 02:00:42 GMT
Sender: news@bbn.com
Reply-To: cjross@BBN.COM (chris ross)
Organization: Bolt Beranek and Newman Inc., Cambridge MA
Lines: 44


Here's something to feed to your compiler.

Define a routine which takes a (float) parameter and passes the address to a
routine which expects a (float*).  What value do you get when following the
pointer in the second routine?  Under Vax Ultrix and Masscomp RTU, I get
what I passed the first routine.  Under SunOS and MIPS RISC/OS, I get
something different.

Here's the code.  It should print
  1 1
  1 1


main ()
{
    foo(1.0, 1.0);
}

foo (x, y)
float x, y;
{
    float local_x = x;
    show(&local_x, &y);
    printf("%g %g\n", x, y);
}

show (xp, yp)
float *xp, *yp;
{
    printf("%g %g\n", *xp, *yp);
}


Apparently, the Sun and MIPS compilers do not hide the fact that the
parameter in the first routine is actually on the stack as a (double).
Should they, or, as with va_arg, must the programmer explicitly take the
type promotion into account?  Does ANSI specify the proper behavior?


----------  chris ross  ----------  <cjross@bbn.com>  ----------
lisp in action is like a finely choreographed ballet.
ada in action is like a waltz of drugged elephants.
c in action is like a sword dance on a freshly waxed floor.