NaN

Should equal(NaN, NaN) return true?

[epoberezkin]

We can prove first NaN = 1 and that second NaN = 0. If NaN == NaN then we can prove 1==0.

This proof has (or would have) an obvious logical flaw, like all sophisms, discussing which is out of scope here.

I think the right approach is just do what most other libraries do (still need to check it).

[Ygg01]

No. NaN means value you get is indeterminate.

Take for example following expression x*1/x. It’s obvious this value is 1 whatever x is, however look what happens when x approaches really small value (e.g. 1e-324*(1/1e-324)) the value becomes NaN.

Another example could be 0*x. The value is obviously 0 for any x != 0 so when e.g. x is 1e+324*(1/1e+324) it becomes NaN as well.

We can prove first NaN = 1 and that second NaN = 0. If NaN == NaN then we can prove 1==0.