[olivernn]

Using `charCodeAt()` for numeric keys is an interesting optimisation, I'm definitely going to try that out with Lunr.js.

I'm also interested in _why_ that is quicker though, aren't all object keys converted to strings anyway?

[vanderZwan]

That is what the spec says, but one way to interpret that is to say that they only have to behave as if they were.

Hashing an integer or double can be much simpler than hashing a string, and like the other comment said: for small enough keys you might even get away with using an array.

edit: to give another example, the spec also says all number are doubles, and yet all JS modern engines use optimisations in places where it can guarantee a value is a 32-bit integer; in some cases adding `|0` (a forced truncation by bitmask-or zero) as a compiler hint can speed up code.

edit2: this is also part of why you want your objects to be as type-stable as possible (it also makes code less bug-prone that way because easier to reason about). This includes the shape of objects. See these slides from 2011:

https://www.slideshare.net/newmovie/know-yourengines-velocit...

[tomsmeding]

Not if they can reasonably be an array. I believe that the integer keys of an object are usually stored in an array in V8 unless they grow really large, or something.