Smaller CSSParser UTF16 buffers for escaped strings.

Source/core/css/CSSTokenizer.h

 /*
  * Copyright (C) 2003 Lars Knoll (knoll@kde.org)
  * Copyright (C) 2004, 2005, 2006, 2008, 2009, 2010 Apple Inc. All rights reserved.
  * Copyright (C) 2008 Eric Seidel <eric@webkit.org>
  * Copyright (C) 2009 - 2010  Torch Mobile (Beijing) Co. Ltd. All rights reserved.
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Library General Public
  * License as published by the Free Software Foundation; either
  * version 2 of the License, or (at your option) any later version.
@@ skipping 56 matching lines @@
 
     bool is8BitSource() const { return m_is8BitSource; }
 
     // FIXME: These 2 functions should be private so that we don't need the definitions below.
     template <typename CharacterType>
     inline CharacterType* tokenStart();
 
     inline unsigned tokenStartOffset();
 
 private:
-    UChar*& currentCharacter16();
+    UChar* getStringBuffer16(size_t len);
 
     template <typename CharacterType>
     inline CharacterType*& currentCharacter();
 
     template <typename CharacterType>
     inline CharacterType* dataStart();
 
     template <typename CharacterType>
     inline void setTokenStart(CharacterType*);
 
     template <typename CharacterType>
     inline bool isIdentifierStart();
 
     template <typename CharacterType>
     inline CSSParserLocation tokenLocation();
 
     template <typename CharacterType>
-    unsigned parseEscape(CharacterType*&);
+    static unsigned parseEscape(CharacterType*&);
     template <typename DestCharacterType>
-    inline void UnicodeToChars(DestCharacterType*&, unsigned);
+    static inline void UnicodeToChars(DestCharacterType*&, unsigned);
+
     template <typename SrcCharacterType, typename DestCharacterType>
-    inline bool parseIdentifierInternal(SrcCharacterType*&, DestCharacterType*&, bool&);
-
+    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>
     bool parseNthChild();
     template <typename CharacterType>
     bool parseNthChildExtra();
     template <typename CharacterType>
     inline bool detectFunctionTypeToken(int);
     template <typename CharacterType>
     inline void detectMediaQueryToken(int);
@@ skipping 20 matching lines @@
         SupportsMode,
         NthChildMode
     };
 
     ParsingMode m_parsingMode;
     bool m_is8BitSource;
     OwnPtr<LChar[]> m_dataStart8;
     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;
     } m_tokenStart;
     unsigned m_length;
     int m_token;
     int m_lineNumber;
     int m_tokenStartLineNumber;
 
     // FIXME: This boolean is misnamed. Also it would be nice if we could consolidate it
@@ skipping 18 matching lines @@
 
 template <>
 inline UChar* CSSTokenizer::tokenStart<UChar>()
 {
     return m_tokenStart.ptr16;
 }
 
 } // namespace WebCore
 
 #endif // CSSTokenizer_h