Re: Old patches from Red Hat

[Daniel J Sebald]

John W. Eaton wrote:
> On 28-Apr-2005, Daniel J Sebald wrote:
>
> | Can some of these definitions and routines be replaced with
> | something from the header file "ieee754.h"?  In there are a variety
> | of helper routines and structures for accessing the sign bit.
>
> Is that header file defined by any standard?

Good question.  I've searched the web.  My guess is that it might be part of the 
GNU C library definition (not actually in the library though because it is just 
definitions).  As for machines that don't use IEEE-754 floating point, I don't 
know how those are handled.

> On my system, /usr/include/ieee754.h does not include any declarations
> for functions or any macros that appear to be related to accessing the
> sign bit.  All I see are some declarations of some structures for
> defining the format of IEEE floating point values.  But maybe I'm
> missing something.

That does appear to be it.  Perhaps the designers felt a bit field pointer was 
simple enough.  So, lo_ieee_signbit(x) could be defined as casting the address 
of x to the ieee754_float type and using (pointer)->negative.  I.e.,

#define lo_ieee_signbit(x) (((ieee754_float *)&x).ieee->negative)

That gets rid of all the

#ifndef _GNU_SOURCE
#define _GNU_SOURCE 1
#define _UNSET_GNU_SOURCE 1
#endif

contortions.  However, the issue of portability to non-GNU compilers is still 
there.

Dan