492840 - ES5: Implement Object.create

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jim Blandy :jimb]

Object.create is specified by 15.2.3.5 in the ES5 draft:
http://wiki.ecmascript.org/doku.php?id=es3.1:es3.1_proposal_working_draft

Comment 1

[Jeff Walden [:Waldo]]

Attachment: Patch atop latest in bug 430133

This doesn't provide any new functionality, but newborn-proto-setting capability is an important step toward removing __proto__ support at some (likely distant) time in the future.

Comment 2

[Jeff Walden [:Waldo]]

Attachment: Give a better error message for Object.create(2), one other uber-nit

Comment 3

[Jeff Walden [:Waldo]]

Attachment: Updated atop latest patch in bug 430133, add another test or two

Comment 4

[Bob Clary [:bc] (inactive)]

Waldo, please fix the test to call reportCompare() and update ecma_5/Object/jstests.list to remove the skip. Thanks.

Comment 5

[Blake Kaplan (:mrbkap) (inactive)]

Comment on attachment 402997 [details] [diff] [review]
Updated atop latest patch in bug 430133, add another test or two

>+MSG_DEF(JSMSG_NOT_OBJECT_OR_NULL,     244, 0, JSEXN_TYPEERR, "value is not null or an object")

You can use JSMSG_UNEXPECTED_TYPE with type being "null or an object" instead of adding a new message here.

>diff --git a/js/src/jsobj.cpp b/js/src/jsobj.cpp
>--- a/js/src/jsobj.cpp
>+++ b/js/src/jsobj.cpp
>@@ -2673,16 +2673,80 @@ obj_defineProperties(JSContext* cx, uint
>     for (size_t i = 0; i < len; i++) {
>         if (!DefineProperty(cx, obj, (*descs)[i], true, dummy))
>             return JS_FALSE;
>     }
> 
>     return JS_TRUE;
> }
> 
>+/* ES5 15.2.3.5: Object.create(O [, Properties]) */
>+static JSBool
>+obj_create(JSContext *cx, uintN argc, jsval *vp)
>+{
>+    jsval v = argc == 0 ? JSVAL_VOID : vp[2];

Nit: parentheses around argc == 0.

>+    if (argc < 1 || !JSVAL_IS_OBJECT(v)) {

The argc < 1 in this check is redundant with the above assignment (especially since we assign a non-object to v if argc is less than 1). I guess it's a little tricky, but I'd say to remove the redundant test.

>+    /*
>+     * Notwithstanding that Object.create seems more likely to be in run-once
>+     * code than performance-critical code, JS_GetScopeChain(cx) is going to
>+     * throw us off trace -- is there an alternative?

Hmm, so js_NewObjectWithGivenProto will use the parent of the prototype (if it's not null) if parent is null. You could use that if the given prototype is non-null. I don't know what we want the lexical scope of these objects to be. Parented to the prototype object's parent (which may be a different lexical scope)? Always parented to the static scope of Object.create? I assume the spec is silent on this issue because it doesn't address the possibility of multiple global objects...

Comment 6

[Jeff Walden [:Waldo]]

(reversing dependency on Object.defineProperty, whose patch this patch depends on)

Comment 7

[Jeff Walden [:Waldo]]

Attachment: Unbitrotted, possibly with previous comments addressed, atop and waiting upon bug 430133

Comment 8

[Jeff Walden [:Waldo]]

http://hg.mozilla.org/tracemonkey/rev/9a24406ddbe7
http://hg.mozilla.org/tracemonkey/rev/aa8a6d3b5dc8
http://hg.mozilla.org/tracemonkey/rev/8cb0a325a704

Comment 9

[Jeff Walden [:Waldo]]

I wrote up a little bit here, but we certainly need to make the information more visible than simply buried in the JS reference:

https://developer.mozilla.org/En/Core_JavaScript_1.5_Reference/Global_Objects/Object/create

Comment 10

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/9a24406ddbe7