Smaller CSSParser UTF16 buffers for escaped strings.

Source/core/css/CSSTokenizer.h

Index: Source/core/css/CSSTokenizer.h
diff --git a/Source/core/css/CSSTokenizer.h b/Source/core/css/CSSTokenizer.h
index 5d8fae12e7f2bb08845b13c100431a4ce12411da..05356b5fda378e72212760d65173190a07fa2005 100644
--- a/Source/core/css/CSSTokenizer.h
+++ b/Source/core/css/CSSTokenizer.h
@@ -74,7 +74,7 @@ public:
     inline unsigned tokenStartOffset();
 
 private:
-    UChar*& currentCharacter16();
+    UChar* getStringBuffer16(size_t len);
 
     template <typename CharacterType>
     inline CharacterType*& currentCharacter();
@@ -92,29 +92,33 @@ private:
     inline CSSParserLocation tokenLocation();
 
     template <typename CharacterType>
-    unsigned parseEscape(CharacterType*&);
+    static unsigned parseEscape(CharacterType*&);
     template <typename DestCharacterType>
-    inline void UnicodeToChars(DestCharacterType*&, unsigned);
-    template <typename SrcCharacterType, typename DestCharacterType>
-    inline bool parseIdentifierInternal(SrcCharacterType*&, DestCharacterType*&, bool&);
+    static inline void UnicodeToChars(DestCharacterType*&, unsigned);
 
+    template <typename SrcCharacterType, typename DestCharacterType>
+    static inline bool parseIdentifierInternal(SrcCharacterType*&, DestCharacterType*&, bool&);
+    template <typename SrcCharacterType>
+    static size_t peekMaxIdentifierLen(SrcCharacterType*);
     template <typename CharacterType>
     inline void parseIdentifier(CharacterType*&, CSSParserString&, bool&);
 
+    template <typename SrcCharacterType>
+    static size_t peekMaxStringLen(SrcCharacterType*, UChar quote);
     template <typename SrcCharacterType, typename DestCharacterType>
-    inline bool parseStringInternal(SrcCharacterType*&, DestCharacterType*&, UChar);
-
+    static inline bool parseStringInternal(SrcCharacterType*&, DestCharacterType*&, UChar);
     template <typename CharacterType>
     inline void parseString(CharacterType*&, CSSParserString& resultString, UChar);
 
     template <typename CharacterType>
     inline bool findURI(CharacterType*& start, CharacterType*& end, UChar& quote);
-
+    template <typename SrcCharacterType>
+    static size_t peekMaxURILen(SrcCharacterType*, UChar quote);
     template <typename SrcCharacterType, typename DestCharacterType>
-    inline bool parseURIInternal(SrcCharacterType*&, DestCharacterType*&, UChar quote);
-
+    static inline bool parseURIInternal(SrcCharacterType*&, DestCharacterType*&, UChar quote);
     template <typename CharacterType>
     inline void parseURI(CSSParserString&);
+
     template <typename CharacterType>
     inline bool parseUnicodeRange();
     template <typename CharacterType>
@@ -155,6 +159,7 @@ private:
     OwnPtr<UChar[]> m_dataStart16;
     LChar* m_currentCharacter8;
     UChar* m_currentCharacter16;
+    Vector<OwnPtr<UChar[]> > m_cssStrings16;

[Julien - ping for review, 2014/03/20 20:57:39]

Really not a huge fan of the Vector approach but we need to hold onto the
characters as the intermediate strings don't do it.

Would you add a comment to explain this optimization in the code and why this is
needed? (if we revisit how we handle CSSParserString's, we could remove this
giant Vector).

[Daniel Bratell, 2014/03/21 15:14:39]

Done. - Though I wouldn't call it giant. It will be 20 bytes and empty in most
cases. For Ars Technica it will also contain a 1 character string for another
~20 bytes. Those 40 bytes should be compared to the 700 KB we used to allocate
into m_currentCharacter16 before this patch.

[Julien - ping for review, 2014/03/21 17:56:26]

I meant giant *hack* Vector but forgot one word. Ideally we shouldn't require
that but due to the way CSSParserString works, we have to retain the memory
until parsing is done :-(

     union {
         LChar* ptr8;
         UChar* ptr16;