[webkit-reviews] review denied: [Bug 15884] JavaScriptCore should use simple type inferencing to speed-up AddNode : [Attachment 17170] Better fix (including LessNode), SunSpider claims this is > 0.5% speedup

[bugzilla-daemon at webkit.org]

Darin Adler <darin at apple.com> has denied Eric Seidel <eric at webkit.org>'s
request for review:
Bug 15884: JavaScriptCore should use simple type inferencing to speed-up
AddNode
http://bugs.webkit.org/show_bug.cgi?id=15884

Attachment 17170: Better fix (including LessNode), SunSpider claims this is >
0.5% speedup
http://bugs.webkit.org/attachment.cgi?id=17170&action=edit

------- Additional Comments from Darin Adler <darin at apple.com>
You have this expression in a few places:

    const JSValue*&

If this is an out parameter of a JSValue*, it should be JSValue*&. The const
doesn't make sense in this case.

I'm a little concerned about adding m_expectedReturnType to every Node -- isn't
that going to make the node tree significantly bigger? If we are going to add
it perhaps it could go in ExpressionNode rather than Node?

You'll also need to merge this with my evaluateToBoolean change.

+    NullNode() KJS_FAST_CALL { m_expectedReturnType = NullType; }

The base class constructor should take this type as a parameter. Otherwise,
we're setting m_expectedReturnType twice, once in the base class constructor
and once in the derived class.

Otherwise looks good.