Ashley Yakeley wrote:
Bertram Felgenhauer wrote:
This is correct according to the IEEE 754 standard, which defines that NaN compares unequal to everything, including itself.
This is numerically useful, perhaps, but nonetheless disturbing. For it would be helpful to expect that any type that presumes to be an instance of Eq has a == a = True.
There really are two different notions of equality here. Perhaps they should be given two different names in the language?
I think that the problem is not in equality, but in NaN: equality on numbers makes perfect sense (although rounding/lack of precision can render it useless), but NaN is *not a number*, so you can't properly compare it with numbers. IMO, NaN should not have the same type as real numbers; illegal operations should return NaN in the form of an error in some monad or by raising an exception (I don't know the behaviour of NaN in other aspects, but this strange equality sounds like propagation of errors/exceptions). Greetings, Arie