444787 - Object.getPrototypeOf

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Robert Sayre]

Added by ES3.1

Comment 1

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: Easy fix

We don't need to worry about call or block objects here, because only the parent can expose those.

Comment 2

[Brendan Eich [:brendan]]

Comment on attachment 329100 [details] [diff] [review]
Easy fix

>+    obj = js_ValueToNonNullObject(cx, vp[2]);
>+    if (!obj)
>+        return JS_FALSE;

15-July-08 ES3.1 spec says

15.2.3.2	Object.getPrototypeOf ( O )
When the getPrototypeOf method is called with argument O, the following steps are taken:
1.	If Type(O) is not Object throw a TypeError exception.
2.	Return the [[Prototype]] property of O.

So you do not wan to convert vp[2] to Object type -- rather throw if JSVAL_IS_PRIMITIVE(vp[2]).

>+    vp[2] = OBJECT_TO_JSVAL(obj);

So no need to update the rooted object argument.

/be

Comment 3

[Blake Kaplan (:mrbkap) (inactive)]

Yeah, the version of the spec that I was writing against hadn't quite defined the semantics yet, so I guessed. Updated patch coming.

Comment 4

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: Updated to spec

I'm not sure if the way I've done the error reporting here is the cleanest (should I be able to pass an index to the DVG?). I could convert js_ReportIsNullOrUndefined to be more general as well, but I didn't think it was necessary.

Comment 5

[Brendan Eich [:brendan]]

Comment on attachment 329798 [details] [diff] [review]
Updated to spec

>+    if (JSVAL_IS_PRIMITIVE(vp[2])) {
>+        char *bytes;
>+
>+        bytes = js_DecompileValueGenerator(cx, JSDVG_SEARCH_STACK,
>+                                           vp[2], NULL);

Of course you should pass the spindex when you know which stack item is being blamed. In this case, we're in a fast native that has no stack frame, so the first argument (vp[2]) is top of stack. But someone could pass extra args, so you want cx->fp->sp - cx->fp->argc, I think. Get me on IRC if this doesn't work.

/be

Comment 6

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: With updated error handling

-argc provides access to the first argument. argc = 0 is a weird case, since that maps to JSDVG_IGNORE_STACK.

Comment 7

[Brendan Eich [:brendan]]

Hot from the TC39 meeting: 3.1 folks agree to relax the type check to throw only on null and undefined, for primitive string/number/boolean get the prototype of the wrapper (String.prototype, Number.prototype, Boolean.prototype).

/be

Comment 8

[Blake Kaplan (:mrbkap) (inactive)]

Brendan: is that going to apply to Object.defineProperty as well?

Comment 9

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: With even more updated error handling

Comment 10

[Brendan Eich [:brendan]]

Comment on attachment 332235 [details] [diff] [review]
With even more updated error handling

Looks good.

Object.defineProperty can't do useful work on what we hope will be sealed/final string, double, and boolean "value type" objects. But the fact that these are sealed may be the reason, not lack of trying to wrap the ES1-3/JS1.x "primitive type" with a corresponding capitalized-name object wrapper.

Anyway, this is for the other bug, which awaits a better spec.

/be

Comment 11

[Blake Kaplan (:mrbkap) (inactive)]

Pushed as http://hg.mozilla.org/mozilla-central/index.cgi/rev/9959a21c4309

Comment 12

[Bob Clary [:bc] (inactive)]

Attachment: ecma_3_1/Object/regress-444787.js

Comment 13

[Bob Clary [:bc] (inactive)]

test added in Bug 450386 

Comment 14

[Eric Shepherd [:sheppy]]

This is documented here; feel free to make fixes as needed:

http://developer.mozilla.org/En/Core_JavaScript_1.5_Reference/Global_Objects/Object/GetPrototypeOf

Comment 15

[Brendan Eich [:brendan]]

ES5 draft still says to throw if O is not of type Object. I've pinged Allen Wirfs-Brock about this.

/be