bin/109692: printing -NaN

[David Schultz]

The following reply was made to PR kern/109692; it has been noted by GNATS.

From: David Schultz <das at FreeBSD.ORG>
To: Denis Koreshkov <dynamic-wind at mail.ru>
Cc: freebsd-gnats-submit at FreeBSD.ORG
Subject: Re: bin/109692: printing -NaN
Date: Thu, 10 Jan 2008 00:46:37 -0500

 On Wed, Feb 28, 2007, Denis Koreshkov wrote:
 > The vfprintf() function correctly formats +Inf and -Inf but it
 > has no provision for distinguishing between +NaN and -NaN. Hence
 > all functions of printf(3) family output all -NaNs as NaNs.
 
 Thanks for pointing this out. I believe that this is historical
 and deliberate behavior, and I do not think it violates C99 or
 IEEE P754. To quote from IEEE P754:
 
    Conversion of a quiet NaN in a supported format to an external
    character sequence shall produce a language-defined one of
    "nan" or a sequence that is equivalent except for case (e.g.,
    "NaN"), with an optional preceding sign. The sign of a NaN has
    no meaning. (5.12.1)
 
 In an abstract sense, NaNs aren't signed; none of the usual
 mathematical operators can be used to distinguish a "positive NaN"
 from a "negative NaN", even though the underlying representation
 includes a sign bit. Only operations that operate directly on the
 bit strings (IEEE P754 defines assignment, negation, fabs, and
 copysign) have a well-defined effect on the sign bit. For all
 other operations, the hardware is granted free license to set the
 sign bit of a NaN however it wants (6.3). Thus, printing the sign
 for NaNs would probably reduce portability, not to mention
 violating POLA.