393306 - add JSClass hook for iteration

Status: RESOLVED FIXED

(Core :: JavaScript Engine, enhancement)

Description

[Jason Orendorff [:jorendorff]]

Implementing iteration in C is awkward.  One way is to resolve the "__iterator__" property in NewResolve.  See bug 390947, attachment 275918 [details] [diff] [review] for an example of code that tries to do this.

A JSClass hook for iteration would make things more straightforward for the embedder (and in this case, the embedder is Mozilla).

Comment 1

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: WIP

I'm holding off on asking for review until I can test this (and convince hg to generate a diff -w).

Comment 2

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: jsiter.c -w

These are the jsiter.c changes. I changed the indentation in one place, otherwise this is functionally equivalent.

Comment 3

[Brendan Eich [:brendan]]

Comment on attachment 278227 [details] [diff] [review]
WIP

>+            xclasp->iteratorObject(cx, obj, (flags & JSITER_FOREACH) == 0);
. . .
>+(* JS_DLL_CALLBACK JSIteratorOp)(JSContext *cx, JSObject *obj, JSBool foreach);

So either the arg is notforeach, or you want != 0 in the call's third arg.

/be

Comment 4

[Blake Kaplan (:mrbkap) (inactive)]

Oops, yeah.

Comment 5

[Blake Kaplan (:mrbkap) (inactive)]

Actually, I've done this instead:

-(* JS_DLL_CALLBACK JSIteratorOp)(JSContext *cx, JSObject *obj, JSBool foreach);
+(* JS_DLL_CALLBACK JSIteratorOp)(JSContext *cx, JSObject *obj,
+                                 JSBool keysonly);

since it matches __iterator__'s sense.

Comment 6

[Brendan Eich [:brendan]]

Comment on attachment 278227 [details] [diff] [review]
WIP

Ok, with that keysonly, r+a=me. Still going to regret this when JS2 iterator support goes in and removes keysonly, but compatibility rules for now.

/be

Comment 7

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: What I'm about to check in

Comment 8

[Blake Kaplan (:mrbkap) (inactive)]

Fix checked into trunk.