runtime: fix useArrowFunction and noArguments

[smorimoto]

Note: this partially introduces ES6!

[smorimoto]

You can use this trick to change the length of an array. I will comment the benchmark later.

[smorimoto]

https://jsben.ch/AKrQh

[TyOverby]

This PR looks like it contains some formatting changes (using arrow functions) and also rewrites some of the most foundational primitives in the jsoo runtime. Could you split these up so that that latter can be thouroughly reviewed and benchmarked?

[TyOverby]

Also, why are you interested in using arrow functions and avoid using arguments?

[smorimoto]

Context Differences

• The Arrow function inherits this from its parent. This may reduce the overhead of binding this by using the Arrow function.
• The function function binds this to its own scope, so you must explicitly bind this to prevent unintended behaviour in situations where this is treated differently.

The arrow function has no binding processing, so you might see some performance improvement in this case. That said, you should not see much difference in general.

[TyOverby]

The function function binds this to its own scope, so you must explicitly bind this to prevent unintended behaviour in situations where this is treated differently.

Does this ever come up in jsoo?

[smorimoto]

@TyOverby See runtime/fs_fake.js

[smorimoto]

Also, why are you interested in using arrow functions and avoid using arguments?

arguments is not Array, it's ArrayLike, and converting it to Array has some cost. Also, newer JS engines have better optimisation for the rest parameter.

[smorimoto]

The rest parameter and arguments show almost no difference in the benchmark results for major browsers: https://jsben.ch/BQEVR

[TyOverby]

arguments is not Array, it's ArrayLike, and converting it to Array has some cost

But Chrome is converting it to an array under the hood too. I benchmarked this a few years ago and found that using the spread operator had no impact on speed

[smorimoto]

The rest parameter seems to be faster for many arguments: https://jsben.ch/Krmit

[smorimoto]

arguments is not Array, it's ArrayLike, and converting it to Array has some cost

But Chrome is converting it to an array under the hood too. I benchmarked this a few years ago and found that using the spread operator had no impact on speed

It's true, but it has a certain advantage here because it simplifies the runtime code. If we are going to introduce const/let, ES6 is required, and if we don't see any performance regression, I think we can consider this kind of change positively.

[TyOverby]

The rest parameter and arguments show almost no difference in the benchmark results for major browsers: https://jsben.ch/BQEVR

On my machine at least, the new code is reliably 4% slower.

The rest parameter seems to be faster for many arguments: https://jsben.ch/Krmit

I think we should prioritize calling functions that have a small number of arguments, as small-arity functions are more common in OCaml than functions with 100+ parameters.

It's true, but it has a certain advantage here because it simplifies the runtime code. If we are going to introduce const/let, ES6 is required, and if we don't see any performance regression, I think we can consider this kind of change positively.

I modified your benchmark so that the arrays aren't empty, and to do some fast computation with them instead of logging (logging is slow, and mucks with benchmark results), and this new benchmark shows that even for 100-parameter calls, the old method is twice as fast (on Chrome at least) than using the new syntax: https://jsben.ch/sGBgk

[smorimoto]

I modified your benchmark so that the arrays aren't empty, and to do some fast computation with them instead of logging (logging is slow, and mucks with benchmark results), and this new benchmark shows that even for 100-parameter calls, the old method is twice as fast (on Chrome at least) than using the new syntax: https://jsben.ch/sGBgk

I'm seeing a strange result 🤔 (I'm also on the latest Chrome btw)

[TyOverby]

Yeah, very strange.

[smorimoto]

I ran it on the latest Chrome with Apple Silicon. What's your environment?

[smorimoto]

Here is the result from Safari on the latest iOS:

[TyOverby]

I ran it on the latest Chrome with Apple Silicon. What's your environment?

Latest Chrome x86-64