Date: Mon, 07 Dec 1998 10:25:21 -0600
From: Eric Rudd <rudd AT cyberoptics DOT com>
Subject: Re: libc math function upgrade work
To: djgpp-workers AT delorie DOT com
Message-id: <366C0171.AACE7F0@cyberoptics.com>
Organization: CyberOptics
MIME-version: 1.0
X-Mailer: Mozilla 4.05 [en] (Win95; U)
Content-type: text/plain; charset=us-ascii
Content-transfer-encoding: 7bit
References: <199810191414 DOT KAA25670 AT indy DOT delorie DOT com>
 <36696383 DOT 3E66693A AT cyberoptics DOT com> <199812051723 DOT MAA30930 AT envy DOT delorie DOT com>
Reply-To: djgpp-workers AT delorie DOT com

DJ Delorie wrote:

> First off, I don't have a copy of the C9x spec.  Is it official?  How
> do I get a copy?

The home page for activities related to C9x is
http://anubis.dkuug.dk/JTC1/SC22/WG14/

The standard has not yet issued, but the current draft is available at the
above URL in several forms, pointed to at the above URL under the name
"C9X FCD"  The latest plain text version is at

    http://anubis.dkuug.dk/JTC1/SC22/WG14/www/docs/n843.txt

It's about 1.4 MB.  It's unlikely that any of the function names will
change at this point.

> >         .globl  ___pow2, _exp2
> > ___pow2:
> > _exp2:
> >    ...code
>
> Normally, I'd say this was OK if both symbols were defined by the same
> spec (i.e. rand() and srand() are in the same object, localtime() and
> gmtime() are together), but in this case there might be a program that
> is emulating C9x by providing its own exp2() function that calls
> pow2().  If you tried to build such a program, you'd get link
> failures.

I can't see where there's a problem.  Consider the following program:

#include <math.h>
#include <stdio.h>

double exp2(double x) {
   return pow2(x);
}

int main(void) {
   printf("exp2(3) = %f\n", exp2(3.));
   return 0;
}

The program compiled without errors using

   gcc -Wall test.c -o test.exe

even though libc.a contained the function with the proposed double label
_exp2 and ___pow2.  The output was 8., as expected.  If one changes the
call inside the local exp2 to

   return pow10(x);

one gets back 1000., instead of 8., which proves that it's linking in the
__powXX labels, rather than the library _exp2 label.  Since the user's
function satisfies the external reference, the linker simply leaves the
other libc function alone.

I even tried putting the user exp2 function in a user library libtest.a,
and linked it in with

   gcc -Wall test.c -o test.o -ltest

and it still the same as discussed above.  So I'm still puzzled by your
reservations.

If, indeed, there is a problem with a global label, I would expect the
stub mechanism to fail also, since the stub would contain the same global
label _exp2 that you think may cause problems.

Am I missing something obvious?

-Eric Rudd