Subj : Re: "named" vs. anonymous function
To   : Jens Thiele
From : Brendan Eich
Date : Mon Dec 13 2004 10:02 am

Jens Thiele wrote:
> Brendan Eich schrieb:
> 
>>To clarify further, a named function object, per ECMA-262 Edition 3
>>11.2.5 and 13, the production:
>>
>>FunctionExpression :
>>    function Identifier ( FormalParameterListopt ) { FunctionBody }
>>
>>A named function expression is scoped by an Object created to hold its
>>name only while it is active.  That is its activation's scope parent, so
>>that cannot match the compiler-visible static scope parent (the global
>>object in your js shell example).  Hence the clone.
>>
>>Named and anonymous function objects do have different scope chains, per
>>ECMA-262 Edition 3, but that does not map directly onto the __proto__
>>difference you discovered.  The latter is really an implementation
>>detail allowed by the spec.
>>
>>/be
> 
> 
> So this this is not really a matter of named vs. anonymous function but
> a matter of scope?


The spec mandates a difference in scope between named and unnamed 
functions: an Object instance needed to hold the name must scope an 
activation of the named function.


> Test:
> js> (function(){return function(){};})().__proto__===Function.prototype
> false
> js> (function(){return
> function(){};})().__proto__.__proto__===Function.prototype
> true
> 
> though this example strictly wouldn't require a scope object?
> 
> perhaps better?
> js> (function (){var i=10;return function(){return
> i;};})().__proto__===Function.prototype
> false
> js> (function (){var i=10;return function(){return
> i;};})().__proto__.__proto__===Function.prototype
> true
> 
> looking at the source (jsinterp.c)
> the relevant parts seem to be:
> JSOP_ANONFUNOBJ, JSOP_NAMEDFUNOBJ and JSOP_CLOSURE
> (probably not JSOP_DEFFUN)
> 
> in any way it is the call to js_CloneFunctionObject
> creating the clone / additional prototype
> 
> I think I better stop now and get back to work again. Curiosity killed
> the cat ;-)

As you've observed, independent of the named vs. anonymous scope 
(parent-linked chain) difference mandated by ECMA, SpiderMonkey uses a 
proto-linked chain difference to share precompiled function objects 
among clones and the compiler-created "clone parent."

Don't mix up these cases, or the sense of "parent" used by the engine 
and ECMA with the "clone parent" sense I just introduced! ;-)

/be

.