Changeset 97827 in webkit for trunk/Source/JavaScriptCore/runtime/JSValue.cpp

Timestamp:

Oct 18, 2011, 7:54:29 PM (14 years ago)

Author:

[email protected]

Message:

Switched ropes from malloc memory to GC memory
​https://bugs.webkit.org/show_bug.cgi?id=70364

Reviewed by Gavin Barraclough.

~1% SunSpider speedup. Neutral elsewhere. Removes one cause for strings
having C++ destructors.

• heap/MarkStack.cpp:

(JSC::visitChildren): Call the JSString visitChildren function now,
since it's no longer a no-op.

• runtime/JSString.cpp:

(JSC::JSString::~JSString): Moved this destructor out of line because
it's called virtually, so there's no value to inlining.

(JSC::JSString::RopeBuilder::expand): Switched RopeBuilder to be a thin
initializing wrapper around JSString. JSString now represents ropes
directly, rather than relying on an underlying malloc object.

(JSC::JSString::visitChildren): Visit our rope fibers, since they're GC
objects now.

(JSC::JSString::resolveRope):
(JSC::JSString::resolveRopeSlowCase):
(JSC::JSString::outOfMemory): Updated for operating on JSStrings instead
of malloc objects.

(JSC::JSString::replaceCharacter): Removed optimizations for substringing
ropes and replacing subsections of ropes. We want to reimplement versions
of these optimizations in the future, but this patch already has good
performance without them.

• runtime/JSString.h:

(JSC::RopeBuilder::JSString):
(JSC::RopeBuilder::finishCreation):
(JSC::RopeBuilder::createNull):
(JSC::RopeBuilder::create):
(JSC::RopeBuilder::createHasOtherOwner):
(JSC::jsSingleCharacterString):
(JSC::jsSingleCharacterSubstring):
(JSC::jsNontrivialString):
(JSC::jsString):
(JSC::jsSubstring):
(JSC::jsOwnedString): Lots of mechanical changes here. The two important
things are: (1) The fibers in JSString::m_fibers are JSStrings now, not
malloc objects; (2) I simplified the JSString constructor interface to
only accept PassRefPtr<StringImpl>, instead of variations on that like
UString, reducing refcount churn.

• runtime/JSValue.h:
• runtime/JSValue.cpp:

(JSC::JSValue::toPrimitiveString): Updated this function to return a
JSString instead of a UString, since that's what clients want now.

• runtime/Operations.cpp:

(JSC::jsAddSlowCase):

• runtime/Operations.h:

(JSC::jsString):

• runtime/SmallStrings.cpp:

(JSC::SmallStrings::createEmptyString): Updated for interface changes above.

• runtime/StringConstructor.cpp:

(JSC::constructWithStringConstructor):

• runtime/StringObject.h:

(JSC::StringObject::create): Don't create a new JSString if we already
have a JSString.

• runtime/StringPrototype.cpp:

(JSC::stringProtoFuncConcat): Updated for interface changes above.

File:

• trunk/Source/JavaScriptCore/runtime/JSValue.cpp (modified) (1 diff)

trunk/Source/JavaScriptCore/runtime/JSValue.cpp

@@ -205 +205 @@
 }
 
+JSString* JSValue::toPrimitiveString(ExecState* exec) const
+{
+    if (isString())
+        return static_cast<JSString*>(asCell());
+    if (isInt32())
+        return jsString(&exec->globalData(), exec->globalData().numericStrings.add(asInt32()));
+    if (isDouble())
+        return jsString(&exec->globalData(), exec->globalData().numericStrings.add(asDouble()));
+    if (isTrue())
+        return jsNontrivialString(exec, exec->propertyNames().trueKeyword.ustring());
+    if (isFalse())
+        return jsNontrivialString(exec, exec->propertyNames().falseKeyword.ustring());
+    if (isNull())
+        return jsNontrivialString(exec, exec->propertyNames().nullKeyword.ustring());
+    if (isUndefined())
+        return jsNontrivialString(exec, exec->propertyNames().undefined.ustring());
+
+    ASSERT(isCell());
+    JSValue v = asCell()->toPrimitive(exec, NoPreference);
+    if (v.isString())
+        return static_cast<JSString*>(v.asCell());
+    return jsString(&exec->globalData(), v.toString(exec));
+}
+
 } // namespace JSC