Number::toString seems to require accuracy

[bergus]

I think #2007 or #1135 broke the Number::toString abstract operation, now specifying a mathematical value equality where originally a floating point ("Number value") equality was meant.

In ES6, step 5 did still read as

Otherwise, let n, k, and s be integers such that k ≥ 1, 10k−1 ≤ s < 10k, the Number value for s × 10n−k is m, and k is as small as possible.

At some point this changed to the current reading of

Otherwise, let n, k, and s be integers such that k ≥ 1, 10k−1 ≤ s < 10k, s × 10n - k is ℝ(x), and k is as small as possible.

This sounds like s × 10n - k is a mathematical value, and to fulfill this equality exactly, s would need to be a quite long integer - no rounding permitted in choosing these values, and therefore no inaccuracy. This doesn't make sense to me in light of Note 1 ("The least significant digit of s is not always uniquely determined") and Note 2 ("For implementations that provide more accurate conversions […], choose the value of s for which s × 10n - k is closest in value to ℝ(x)").

I believe that instead this sentence should be

Otherwise, let n, k, and s be integers such that k ≥ 1, 10k−1 ≤ s < 10k, 𝔽(s × 10n - k) is x, and k is as small as possible.

and similiar for the more accurate version in Note 2

Otherwise, let n, k, and s be integers such that k ≥ 1, 10k−1 ≤ s < 10k, 𝔽(s × 10n - k) is x, and k is as small as possible. If there are multiple possibilities for s, choose the value of s for which s × 10n - k is closest in value to ℝ(x).

---

For reference, this came up in answering the StackOverflow question Why does Number(“x”) == BigInt(“x”) … only sometimes?.