Math.round() implementation does not match spec

[A. Sundararajan]

Details on Oracle SCA here: http://openjdk.java.net/contribute/

Thanks
-Sundar

On Thursday 28 March 2013 10:02 PM, Marcus Lagergren wrote:
> André,
>
> It appears you are correct. I've filed bug number JDK-8011023 to track this.  I agree that the fix is correct.  If you have signed a third party source contribution agreement for the OpenJDK I can push this and credit you as contributor.
>
> /M
>
>
> On Mar 28, 2013, at 5:09 PM, André Bargull <andrebargull at googlemail.com> wrote:
>
>> The implementation for Math.round() does not match what's specified in the ES spec.
>>
>> A few test cases which currently fail:
>> assertEq(-Infinity, 1/Math.round(-0.5))
>> assertEq(9007199254740991, Math.round(9007199254740991))
>> assertEq(18446744073709552000, Math.round(9223372036854775807*2))
>>
>> The NativeMath#round() method should be changed as follows:
>>> @Function(attributes = Attribute.NOT_ENUMERABLE, where = Where.CONSTRUCTOR)
>>> public static Object round(final Object self, final Object x) {
>>>     double d = JSType.toNumber(x);
>>>     int e = Math.getExponent(d);
>>>     if (e >= 52) {
>>>         return d;
>>>     }
>>>     return Math.copySign(Math.floor(d + 0.5), d);
>>> }
>>
>> - André