Should find within be entirely implementation defined?

[jasnell]

This one is important: should the find / contains algorithm be entirely implementation defined or should we defined a fall-back search algorithm? I lean towards entirely implementation defined. Of course, there are caveats given to how to treat different needle types... what I'm talking about here is the actual search algorithm not the validation/preparation of the inputs.

[ljharb]

In theory it wouldn't be observable what algorithm the host uses, and as such, I think it should be entirely in the spec, and they can then do whatever they want.

[bakkot]

+1 ljharb. More precisely, I think it should basically be

1. Throw if the receiver is not a TA
2. Throw if the needle is not an object
3. Get the .length of the needle (and throw if the result is not a nonnegative integral number)
4. Early exit with false if the length of the needle is longer than the target TA
5. Read the entire needle, and throw an exception if any value is not a Number
6. Do the search (which now is not observable, and can be implemented however the engine wants)

When the argument is another TA, or is known to be a dense Array of numbers with no own getters/setters (engines sometimes have a type meaning precisely that), then engines can skip doing the read of the needle. But that's an unobservable optimization, not something we need to write down.