Comparing an NSSignedIntegerNumber to an NSDecimalNumber storing NaN

[Ondřej Hošek]

Dear fellow GNUsteppers,

I have another problem. :-D This time, it is NSNumberFormatter's
tendency to compare with NaN.

In [NSNumberFormatter-attributedStringForObjectValue:withDefaultAttributes:],
the message [anObject isEqual:nanNumber] is issued. In my code,
anObject is an NSSignedIntegerNumber, while nanNumber is by design an
NSDecimalNumber.

[NSNumber-isEqual:] calls [NSNumber-isEqualToNumber:], which calls
[NSSignedIntegerNumber-compare:]. Case 'd' is taken, the double values
(99.0 and NaN) are obtained, and the two values are compared using the
macro COMPARE.

Is 99.0 less than NaN? No. Is 99.0 greater than NaN? No. The macro
decides that 99.0 orders the same as NaN, and the callers conclude
that 99.0 equals NaN. Thus, even for such a beautiful number such as
99, the NSNumberFormatter returns "NaN".

My limited experimenting shows that Apple (OS X 10.6.6, Foundation
751.42.0) orders NaNs below any numeric value. Do we wish to follow
suit? (Further experiments, especially with +inf and -inf, are
welcome; I'm too sleepy to continue experimenting right now.)

I hope this is not a compiler or configuration but a code bug. ;-)

Cheers,
~~ Ondra