Minor cleanup of CompareIC state

src/ic.h

Index: src/ic.h
diff --git a/src/ic.h b/src/ic.h
index cb316407c0563c340e143bc621c69d75642916dc..12328d511fb3b0176c428c6fe9f130ea1faed31e 100644
--- a/src/ic.h
+++ b/src/ic.h
@@ -752,14 +752,20 @@ class BinaryOpIC: public IC {
 
 class CompareIC: public IC {
  public:
+  // The type/state lattice is defined by the following inequations:
+  //   UNINITIALIZED < ...
+  //   ... < GENERIC
+  //   SMI < NUMBER
+  //   SYMBOL < STRING
+  //   KNOWN_OBJECT < OBJECT
   enum State {
     UNINITIALIZED,
     SMI,
-    HEAP_NUMBER,
+    NUMBER,

[Jakob Kummerow, 2013/02/15 09:12:53]

I don't think this renaming is an improvement, for two reasons:
(1) All ICs call this state HEAP_NUMBER, so renaming it for CompareICs
introduces inconsistency.
(2) The previous name does make sense, as generally a stub's type name indicates
the most generic type it can handle. The "SMI < [HEAP_]NUMBER" inequation above
implies that the HEAP_NUMBER stub can handle Smis.
That said, if you insist on this change, I don't feel strongly enough about it
to argue.

[rossberg, 2013/02/19 11:32:13]

Good point, I fixed it for Unary and Binary ops as well. :)
As discussed offline, your very argument is why I changed it. HEAP_NUMBER names
a specific representation type, but a stub in the respective state can also
handle SMI representations, so NUMBER is the actual type it handles, in
accordance to our type hierarchy.

     SYMBOL,
     STRING,
-    OBJECT,
-    KNOWN_OBJECTS,
+    OBJECT,         // JSObject
+    KNOWN_OBJECT,   // JSObject with specific map (faster check)
     GENERIC
   };