Remove arguments.caller from the spec

[claudepache]

Note. Not to be confused with Function#caller.

---

The spec acknowledges that arguments.caller “has historically been provided as an implementation-defined extension by some ECMAScript implementations”.

However, nowadays this historical wart is not found in any browser since long ago, e.g.:

• removed from Mozilla before Firefox was born ;
• removed from IE9 Standards mode.

I think one can clean the spec by removing all special cases around arguments.caller. As far as I can see, the modifications would be:

• 9.4.4. Arguments Exotic Objects, Note 3: remove the mention of "caller".
• 9.4.4.1 [[GetOwnProperty]] of Arguments objects: remove step 7, that special-cases "caller", together with the remark ”If an implementation does not provide a built-in caller property for argument exotic objects then step 7 of this algorithm must be skipped.”
• 9.4.4.5 [[HasProperty]] of Arguments objects, use the same implementation as ordinary objects, removing the "caller" special case and the remark ”If an implementation does not provide a built-in caller property for argument exotic objects then step 2 of this algorithm must be skipped.”
• 9.4.4.7. CreateUnmappedArgumentsObject, remove step 9, that adds a "caller" poison-pill to unmapped arguments objects.
• 16.2 Forbidden Extensions, add: “Arguments objects, both mapped and unmapped, must not be created with an own property named "caller".”
• Annex C The Strict Mode of ECMAScript, modify item: “Arguments objects for strict functions define non-configurable accessor properties named "caller" and "callee" a non-configurable accessor property named "callee" which throws a TypeError exception on access” — (Also, at the time of writing, backreference to section 9.2.7 is incorrect, it should be 9.4.4.7; was already reported in A line in Annex C references the wrong section #572.)

Concretely, for the implementations not supporting the arguments.caller extension, the changes resume to removing the "caller" poison-pill from their unmapped Arguments objects.

[erights]

I just tested and am surprised to see that this is correct. Good riddance! Now that arguments.caller is totally and permanently dead, I agree we no longer need to guard against it. I support this removal.

[leobalter]

I'll add this to the agenda for the next meeting for the consensus formalities.

@claudepache do you want to prepare a PR for this? Otherwise I can do it, also checking test262.