[Bug]: A problem about rounding double precision floating point numbers

[ayuan0828]

ChakraCore Version

36becec

Steps to reproduce

When I use the 'print()' method to output an array containing a specific double-precision floating-point number, there is an unexpected rounding error in the last digit. Should the binary representation of the value be stored accurately?

Proof of concept

const testNumber = 9.758231785433683;
const numberArray = [testNumber];
print(numberArray);

Exception or Error

9.758231785433682

Additional Context

No response

[Owaiseimdad]

@ayuan0828 this is because of small rounding error. Where in the code we are using the above example and relying on more than 10th precision?

[ayuan0828]

It was discovered during use and I was just a little curious as to why ChakraCore was implemented this way.

[rhuanjl]

This is a duplicate of issue #149, never closed because fixing it proved to be a lot of work for a fairly small gain - cases where the difference matters is very rare. Though I'd love to accept a PR that fixes it.

To answer some curiosity/give background Floating points are stored as a native type that is not displayable as a string; (this is done in all JS engines) because much faster maths can be done with the native Float than a string of digits.

Converting that native type to a display string is actually quite complex see issue #149 for more.