比较Kotlin的NaN

发布于06月17日

所以我最近开始喜欢kotlin的语言.今天，在比较双打时，我遇到了不可避免的NaN分.

fun main(args: Array<String>) {
    val nan = Double.NaN
    println("1: " + (nan == nan))
    println("2: " + (nan == (nan as Number)))
    println("3: " + ((nan as Number) == nan))
}

N.B: (Double is a subtype of Number)

运行上述代码会产生:

1: false
2: true
3: true

I understand that comparing with NaN in Java returns false, so I would expect false for all expressions.

How can this behavior be explained? What is the rationale behind it?

推荐答案

这是因为(2)和(3)被编译为装箱一个基元，然后进行Double.equals判断:在JVM上，基元double不能与装箱的基元进行比较.

Double.equals, in turn, checks equality by comparing doubleToLongBits(...) of the two Doubles, and for the latter there's a guarantee that

If the argument is NaN, the result is 0x7ff8000000000000L.

因此，两个NaN返回的位相等，这里忽略规则NaN != NaN.

Also, as @miensol mentioned, there's another consequence of this equality check: +0 and -0 are equal according to == check and not to equals check.

Equivalent code in Java would be:

double nan = Double.NaN;
System.out.println("1: " + (nan == nan)) //false 
System.out.println("2: " + ((Double) nan).equals(((Number) nan)))
System.out.println("3: " + ((Number) nan).equals(nan));

The last two lines call Double.equals, comparing doubleToLongBits(...).