Float class>>nan broken

[matthias-springer]

In Squeak: Float nan = Float nan is false and Float nan == Float nan is true. In RSqueak: both are false

[j4yk]

Is it reasonable to state something is identical but not equal?

[matthias-springer]

Not sure about semantics here, but just found out that x != x in Python if x = float('nan') (what???).

[krono]

Thats perfectly ok. Holds in Smalltalk and also Ruby:

irb(main):002:0> x = 0 / 0.0
=> NaN
irb(main):003:0> x == x
=> false
irb(main):004:0> 

[matthias-springer]

== is object identity (defined as "pointer address" in Squeak), so why would x == x ever be false regardless of the value of x?

[krono]

Explanation:

• x == x in Smalltalk is compiled to a pointer check (or value check for integers only), Since on image level there is only one NaN, Float nan == Float nan compares the same objects.
• x = x is a message send for #=, which is a primitive for float. This one explicitly makes sure that two NaNs never compare equal (by delegating to C semantics)

[krono]

On 16.06.2015, at 20:41, Matthias Springer notifications@github.com wrote:

== is object identity, so why would x == x ever be false regardless of the value of x?

�

Boxed objects would do that. But there are ways around that.