Date: Mon, 11 Aug 2003 09:28:05 -0400
Message-Id: <200308111328.h7BDS5un031026@envy.delorie.com>
From: DJ Delorie <dj AT delorie DOT com>
To: djgpp-workers AT delorie DOT com
In-reply-to: <uu18osptc.fsf@elta.co.il> (message from Eli Zaretskii on 11 Aug
	2003 07:53:19 +0200)
Subject: Re: Anomaly in printf()
References: <1e0 DOT eca6e87 DOT 2c67e363 AT aol DOT com> <200308101817 DOT h7AIHDhr019129 AT envy DOT delorie DOT com> <9003-Sun10Aug2003222306+0300-eliz AT elta DOT co DOT il> <200308102312 DOT h7ANC6mQ021365 AT envy DOT delorie DOT com> <uu18osptc DOT fsf AT elta DOT co DOT il>
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk


> In a long double, the mantissa includes the integer bit, unlike in a
> float or a double.  Se we lose one bit.

The 80387 book claims that there's 64 bits of mantissa (0..63), one
bit of sign (79), and 15 bits of exponent (64..78).

> (I'd be interested in seeing the actual bit patterns of the numbers
> as shown by a debugger; I'm quite sure these bit patterns are okay.)

Linux prints this:

00 00 40 3e ff ff ff ff ff ff ff ff = 18446744073709551615.000
00 00 40 3e ff ff ff ff ff ff ff fe = 18446744073709551614.000
00 00 40 3e ff ff ff ff ff ff ff fd = 18446744073709551613.000
00 00 40 3e ff ff ff ff ff ff ff fc = 18446744073709551612.000

> Since _doprnt does some amount of math on its argument, loss of
> precision beyond the significant digits is inevitable.

If that's our excuse, it's still a bug.  We should choose better
methods so that we don't lose precision.