js-choi / proposal-bigint-bit-length Goto Github PK

I wrote this in the explainer:

In contrast to other operations – such as abs and sqrt from the BigInt Math proposal – the new bit-length API would probably be useful only on BigInts and not on ordinary numbers.

Is this really true? One potential use case for ordinary-number bit lengths might be conveniently determining whether an ordinary number integer will fit within e.g. an Int16Array.

Of course, this would be complicated by what to do with non-integer numbers (something to do with IEEE floating-point representation? truncate? throw a TypeError?).

Either way, before or even during Stage 1, it is worth explicitly considering bit lengths for ordinary numbers.

I think, for math use cases, ceil(log2(abs(a)+1)) is better, but this may be not expected, different libraries do it differently:
mpz_sizeinbase in GMP ignores the sign;
Dart, Java and .Net use ceil(log2(-a))
, so may be negative bigint should cause exception

Should the result type should be a Number or a BigInt?

We should be consistent with how we treat lengths in the other BigInt operators. In particular, the answer should match the inverse operation: given a bit length, produce a BigInt of that size. That inverse operation is either 1n << length or 1n << length-1n depending on whether we choose the off-by-one version or not, but in either case it wants the length represented as a BigInt. So that points towards returning a BigInt.