leaky arguments - broad conclusion insufficiently supported

[mncharity]

Further in Node 8.3+ we won't be punished for exposing the arguments object to other functions

This seems too strong a conclusion to be supported by bench/arguments.js's

function leakyArguments () {
  return other(arguments)
}

In the past, f.apply(o, arguments) was special cased. But to avoid arguments arrayification, there was the additional requirement that the call site be monomorphic - that there was no diversity in function shape (normal, strict, bound, arrow, etc).

This one benchmark case explores passing arguments to a named inlineable function.

Any conclusions broader than that seem unsupported. The inlining may be required, or the named-ness, or the monomorphism. And given the historical "fragility" of avoiding arguments arrayification, there may be additional requirements. Examining source, or authoritative guidance, or more extensive tests, seem needed to support the current conclusion.

[mcollina]

@mncharity thanks for reporting! Would you like to provide some counter examples were arguments is getting slower again?

As a side note: arrow functions do not have arguments, and the usage of apply with mixed normal/strict mode is probably not common. I check the bound case and I can confirm the same behavior.