These three arguments are partially redundant with what TypedArray element type describes in a single enumeration, but the latter also includes FLOAT{16,32,64} and I couldn't find a clean way to layer separation of just the INT/BIGINT values so I've left that as a potential exercise for the future (but ideas for the present are nonetheless welcome).

I suspect that going that route would involve something like introducing ToFloat{16,32,64} operations, migrating the last four columns of that table into Abstract Operations along with Is{Unsigned,UnclampedInteger,BigInt}ElementType for use by ToFixedSizeInteger, and simplifying e.g. NumericToRawBytes/NumericToRawBytes/etc.

Three separate abstract operations seem a bit nicer, because it more clearly separates conversion from and to Number resp. BigInt.

ToFixedSizeInteger (
  _argument_: an integer,
  _signed_: ~unsigned~ or ~signed~,
  _bitWidth_: a positive integer,
): an integer
  1. Let _fixedInt_ be _int_ modulo 2<sup>_bitWidth_</sup>.
  1. If _signed_ is ~signed~ and _fixedInt_ ≥ 2<sup>_bitWidth_ - 1</sup>, return _fixedInt_ - 2<sup>_bitWidth_</sup>.
  1. Return _fixedInt_.


ToFixedSizeNumber(
  _argument_: an ECMAScript language value,
  _signed_: ~unsigned~ or ~signed~,
  _bitWidth_: a positive integer,
): either a normal completion containing an integral Number or a throw completion
  1. Let _int_ be ? ToIntegerOrInfinity(_argument_).
  1. If _int_ is not finite, return 0.
  1. Return 𝔽(ToFixedSizeInteger(_int_, _signed_, _bitWidth_)).


ToFixedSizeBigInt(
  _argument_: an ECMAScript language value,
  _signed_: ~unsigned~ or ~signed~,
  _bitWidth_: a positive integer,
): either a normal completion containing a BigInt or a throw completion
  1. Let _int_ be ℝ(? ToBigInt(_argument_)).
  1. Return ℤ(ToFixedSizeInteger(_int_, _signed_, _bitWidth_)).

Thanks for the suggestion. I took a half step in this direction, keeping the single new operation but making it infallibly map extended mathematical values and moving initial conversion to call sites.

Since this now performs no type conversion, "IntegerToFixedSize" might be a better name. Opinions are welcome.

I'd prefer us to retain something like the example from the old note. Something like ToFixedSizeInteger(ToFixedSizeInteger(_x_, ~signed~, 32), ~unsigned~, 32) is the same as ToFixedSizeInteger(_x_, ~unsigned~, 32).

Done.

(editorial convention)

To be clear (despite GitHub bug), my suggestion is to insert "either".

I find the comparison conventions terribly confusing/incomplete and possibly contradictory, but since int is a mathematical value, shouldn't this actually be If _int_ = +∞ or _int_ = -∞, return 0. as seen at BinaryOr, BigIntBitwiseOp, and IsWordChar? I almost always see the "is either $A or $B" form with $A and $B being language values and spec enums.

Technically, it's an 'extended mathematical value', which the comparison conventions don't mention, but lumping it in with mathematical values would be reasonable. So yeah, using = seems more consistent.

Done.