993026 - Merge JSResolveOp and JSNewResolveOp; remove JSCLASS_NEW_RESOLVE

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jason Orendorff [:jorendorff]]

Current signatures:

typedef bool
(* JSResolveOp)(JSContext *cx, JS::HandleObject obj, JS::HandleId id);

typedef bool
(* JSNewResolveOp)(JSContext *cx, JS::HandleObject obj, JS::HandleId id,
                   unsigned flags, JS::MutableHandleObject objp);

After bug 547140, the 'flags' parameter will be gone from the latter, and I think the only proper use of 'objp' is to set it to 'obj' (or leave it null). Thus we can merge the two and have only:

typedef bool
(* JSResolveOp)(JSContext *cx, JS::HandleObject obj, JS::HandleId id,
                bool *resolvedp);

Comment 1

[Tom S [:evilpie]]

Attachment: v1 - Change XPIDLScriptable NewResolve to simpler Resolve

Comment 2

[Tom S [:evilpie]]

Attachment: v1 - Change JSAPI resolve hook

Comment 3

[Tom S [:evilpie]]

Attachment: v1 - Change XPIDLScriptable NewResolve to simpler Resolve

Comment 4

[Tom S [:evilpie]]

Attachment: v1 - Change resolve hook in JS

Comment 5

[Tom S [:evilpie]]

Attachment: v1 - Change resolve hook in browser/XPC

Comment 6

[Bobby Holley (:bholley)]

Comment on attachment 8515453 [details] [diff] [review]
v1 - Change resolve hook in browser/XPC

Review of attachment 8515453 [details] [diff] [review]:
-----------------------------------------------------------------

\o/

Comment 7

[Tom S [:evilpie]]

Comment on attachment 8515451 [details] [diff] [review]
v1 - Change XPIDLScriptable NewResolve to simpler Resolve

Review of attachment 8515451 [details] [diff] [review]:
-----------------------------------------------------------------

Bobby thank you for that super quick review, can you take a look at this one as well? Bz seems to have tons of review requests at the moment.

Comment 8

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 8515453 [details] [diff] [review]
v1 - Change resolve hook in browser/XPC

Could you pleas fix the naming, probably in a separate patch?  So s/NEWRESOLVE_HOOK_NAME/RESOLVE_HOOK_NAME/, s/newResolveHook/resolveHook/, s/CGNewResolveHook/CGResolveHook/, s/NeedNewResolve/NeedResolve/ (including in the documentation), s/NPObjWrapper_NewResolve/NPObjWrapper_Resolve/.

Comment 9

[Bobby Holley (:bholley)]

Comment on attachment 8515451 [details] [diff] [review]
v1 - Change XPIDLScriptable NewResolve to simpler Resolve

Review of attachment 8515451 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/xpconnect/idl/nsIXPCScriptable.idl
@@ +114,5 @@
>      boolean newEnumerate(in nsIXPConnectWrappedNative wrapper,
>                          in JSContextPtr cx, in JSObjectPtr obj,
>                          in uint32_t enum_op, in JSValPtr statep, out jsid idp);
>  
> +    boolean resolve(in nsIXPConnectWrappedNative wrapper,

Please rev the IID of this interface.

Comment 10

[Tom S [:evilpie]]

Attachment: v1 - Rename NewResolve to Resolve in browser

Comment 11

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 8517499 [details] [diff] [review]
v1 - Rename NewResolve to Resolve in browser

>+    Resolve hook for objects with custom hooks.

"for objects that have the NeedResolve extended attribute".

r=me, but please update https://developer.mozilla.org/en-US/docs/Mozilla/WebIDL_bindings#NeedNewResolve when you land

Comment 12

[Tom S [:evilpie]]

Comment on attachment 8515452 [details] [diff] [review]
v1 - Change resolve hook in JS

Review of attachment 8515452 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/shell/js.cpp
@@ +4537,5 @@
>      JS_FN_HELP("shapeOf", ShapeOf, 1, 0,
>  "shapeOf(obj)",
>  "  Get the shape of obj (an implementation detail)."),
>  
> +// XXX this actually seems unused?

Well this code is gone now :)

Comment 13

[Jason Orendorff [:jorendorff]]

Comment on attachment 8515452 [details] [diff] [review]
v1 - Change resolve hook in JS

Review of attachment 8515452 [details] [diff] [review]:
-----------------------------------------------------------------

Very nice simplification. Love it.

::: js/public/Class.h
@@ +118,5 @@
>  // Lazy properties are those reflected from some peer native property space
>  // (e.g., the DOM attributes for a given node reflected as obj) on demand.
>  //
>  // JS looks for a property in an object, and if not found, tries to resolve
> +// the given id. *resolvedp should be set to true iff the propertie was

spelling: "property"

::: js/src/jsapi-tests/testFreshGlobalEvalRedefinition.cpp
@@ +18,2 @@
>  {
> +    return JS_ResolveStandardClass(cx, obj, id, resolvedp);

Hmm. Can this function be eliminated altogether, then?

::: js/src/shell/js.cpp
@@ +2520,1 @@
>              return false;

The if statement could be changed to just:

    return JS_ResolveStandardClass(cx, obj, id, resolvedp);

::: js/src/vm/ArgumentsObject.cpp
@@ -363,2 @@
>  {
> -    objp.set(nullptr);

OK, either this has to be retained, as `*resolvedp = false;`, because there are a bunch of `return true;` exits in the body of the function --- or the contract of resolve hooks has to change to promise that *resolvedp is set to false on entry. I prefer the former.

@@ -479,2 @@
>  {
> -    objp.set(nullptr);

Same here.

::: js/src/vm/NativeObject-inl.h
@@ +511,2 @@
>  
> +    // ToDo: See if we can assert that the property exists?

Bah! Either give it a shot or don't bother with the comment!

Inversely, if (!resolved), you could assert above that the property does not exist. That's more work. Still, it's worth a comment here if for some reason either assertion doesn't hold.

@@ +516,5 @@
>          return true;
>      }
>  
>      Shape *shape;
> +    if (!obj->empty() && (shape = obj->lookup(cx, id)))

Is that obj->empty() necessary? Surely not:

    if (Shape *shape = obj->lookup(cx, id))
        propp.set(shape);
    else
        ...

Comment 14

[Tom S [:evilpie]]

(In reply to Jason Orendorff [:jorendorff] from comment #13)
> ::: js/src/vm/ArgumentsObject.cpp
> @@ -363,2 @@
> >  {
> > -    objp.set(nullptr);
> 
> OK, either this has to be retained, as `*resolvedp = false;`, because there
> are a bunch of `return true;` exits in the body of the function --- or the
> contract of resolve hooks has to change to promise that *resolvedp is set to
> false on entry. I prefer the former.

So the bindings already made use of obj2 is nullptr on entry and I just changed those to resolvedp as well. So I would prefer sticking to what I have got now.

Comment 15

[Tom S [:evilpie]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/b22c5e82467b
https://hg.mozilla.org/integration/mozilla-inbound/rev/2c6e370c940c
https://hg.mozilla.org/integration/mozilla-inbound/rev/466732e6be01
https://hg.mozilla.org/integration/mozilla-inbound/rev/3c8f81efb9a7
https://hg.mozilla.org/integration/mozilla-inbound/rev/08c816590369

Comment 16

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/b22c5e82467b
https://hg.mozilla.org/mozilla-central/rev/2c6e370c940c
https://hg.mozilla.org/mozilla-central/rev/466732e6be01
https://hg.mozilla.org/mozilla-central/rev/3c8f81efb9a7
https://hg.mozilla.org/mozilla-central/rev/08c816590369