Normative: Set "name" property for anonymous functions

[anba]

Fixes #1049

692cdb2

• [[SourceText]] was not assigned for class expressions with an inferred name and async arrow functions. Fixed here because the next commits will change the same algorithms.

91d3ac1

• Changes the NamedEvaluation operations to no longer forward to Evaluation and splits Evaluation for ClassExpression. This in preparation for the next commit where NamedEvaluation will directly pass the function name to FunctionCreate.

99bf205

• Add name property even if the function doesn't have an explicit or inferred name (Should anonymous functions get a name data property? #1049).
• Set the function name property as part of FunctionInitialize. This matches the behaviour of JSC, Graal and V8, which all three create (resp. iterate) the name property before prototype for functions.

// Use a strict function to omit non-standard "caller" and "arguments".
print(Reflect.ownKeys(function f(){"use strict"}));

Engine	Output
ES2019	length,prototype,name
SpiderMonkey	prototype,length,name
Chakra	prototype,length,name
JSC	length,name,prototype
Graal	length,name,prototype
V8	length,name,prototype

With this PR the iteration order will be changed to "length, name, prototype".

V8 was the only engine returning the correct property iteration order for classes, so it seems to be safe to change it, too.

print(Reflect.ownKeys(class C {}));

Engine	Output
ES2019	length,prototype,name
SpiderMonkey	prototype,length,name
Chakra	prototype,length,name
JSC	length,name,prototype
Graal	length,name,prototype
V8	length,prototype,name (← matches spec!)

c6d0113

• Changes the definition of anonymous function to mean a function with a name whose value is the empty string value.
• The explicit name property attributes for %ThrowTypeError% are needed to ensure the function stays frozen.

Note:

• I didn't add ! in abstract operations which didn't already use ! to keep the usage within a single operation consistent.
• There's function name property and function "length" property, so name is always used without double-quotes whereas "length" always uses quotes. Is this intentional?
• And I wonder if CreateBuiltinFunction should be changed to directly set name and length instead of relying on the declarative description of their values.

[gsathya]

Would it be possible to remove NamedEvaluation completely and always pass an appropriate parameter (the correct name or "") to Evaluation?

[gsathya]

This PR got consensus at the TC39 meeting today. It'd be great to get some reviews so we can merge this.

[ljharb]

Rebased this, and on top of #1569, since i hope to land that first.

[rkirsling]

Note that with the policy of #1302 we'll want to write

The `"name"` property

here and in a handful of newly-added lines below.

[ljharb]

@anba this needs a rebase, when you have a moment

[ExE-Boss]

Do note that both SpiderMonkey and V8 already include the name property as an own property on %ThrowTypeError%.

I haven’t checked other engines, but I expect it to be the same.

[syg]

Ah I see, I had missed that we already call %ThrowTypeError% an "anonymous function". In that case option 1 seems fine to me.

[jmdyck]

The original comment says:

91d3ac1

• Changes the NamedEvaluation operations to no longer forward to Evaluation and splits Evaluation for ClassExpression. This in preparation for the next commit where NamedEvaluation will directly pass the function name to FunctionCreate.

But the PR no longer has the change in which the function name is passed to OrdinaryFunctionCreate, and I don't see any discussion above that would have resulted in its disappearance. So I'm wondering why.

[anba]

I've changed the PR to no longer modify OrdinaryFunctionCreate, instead the function name is set after OrdinaryFunctionCreate has been called. That way the PR should be less invasive. Any refactoring to invoke SetFunctionName directly within OrdinaryFunctionCreate can happen at a later time, if there's interest in a such a change.

[syg]

Looks great! Thanks for the change.

[bakkot]

LGTM other than comment.