Date:	Tue, 6 Apr 93 18:04:38 +1000
From:	graeme AT labtam DOT labtam DOT oz DOT au (Graeme Gill)
To:	djgpp AT sun DOT soe DOT clarkson DOT edu
Subject:	Bug in floating point constants

	

	I have discovered a bug in the handling of
floating point constants that caused a significant error
in a program I am trying to compile. It seems the
compiler has rather inaccurate (and inconsistent) mapping
between decimal and binary floating point constants. This
causes a particularly bad error when using the value
MAXDOUBLE.

for example :

---------------------- cut here --------------------------

/* use the value in include/values.h of djgpp 1.09 */

#define MAXDOUBLE       1.79769313486231570e+308

double dd = MAXDOUBLE;

main()
	{
	printf("MAXDOUBLE = %.20e, dd = %.20e\n",MAXDOUBLE,dd);
	if (dd > MAXDOUBLE)
		printf("ERROR: dd > MAXDOUBLE\n");
	if (dd < MAXDOUBLE)
		printf("ERROR: dd < MAXDOUBLE\n");
	if (MAXDOUBLE > dd)
		printf("ERROR: MAXDOUBLE > dd\n");
	if (MAXDOUBLE < dd)
		printf("ERROR: MAXDOUBLE < dd\n");
	}
---------------------- cut here --------------------------

Compile this and run it, and you'll notice that

MAXDOUBLE != MAXDOUBLE != 1.79769313486231570e+308


This may well be a generic bug in the version of gcc used,
but I will leave it to those who know more about gcc to look
into it.

The workaround is to change line in include/values.h below
#if defined(i396)

from 

#define MAXDOUBLE       1.79769313486231570e+308

to

#define MAXDOUBLE       1.79769313486231470e+308
                                        ^ this digit changes.

	Graeme Gill.

- Raw text -