base/string16.h

   1 // Copyright (c) 2006-2008 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #ifndef BASE_STRING16_H_
   6 #define BASE_STRING16_H_
   7
   8 // WHAT:
   9 // A version of std::basic_string that provides 2-byte characters even when
  10 // wchar_t is not implemented as a 2-byte type. You can access this class as
  11 // string16. We also define char16, which string16 is based upon.
  12 //
  13 // WHY:
  14 // On Windows, wchar_t is 2 bytes, and it can conveniently handle UTF-16/UCS-2
  15 // data. Plenty of existing code operates on strings encoded as UTF-16.
  16 //
  17 // On many other platforms, sizeof(wchar_t) is 4 bytes by default. We can make
  18 // it 2 bytes by using the GCC flag -fshort-wchar. But then std::wstring fails
  19 // at run time, because it calls some functions (like wcslen) that come from
  20 // the system's native C library -- which was built with a 4-byte wchar_t!
  21 // It's wasteful to use 4-byte wchar_t strings to carry UTF-16 data, and it's
  22 // entirely improper on those systems where the encoding of wchar_t is defined
  23 // as UTF-32.
  24 //
  25 // Here, we define string16, which is similar to std::wstring but replaces all
  26 // libc functions with custom, 2-byte-char compatible routines. It is capable
  27 // of carrying UTF-16-encoded data.
  28
  29 #include <stdio.h>
  30 #include <string>
  31
  32 #include "base/basictypes.h"
  33
  34 #if defined(WCHAR_T_IS_UTF16)
  35
  36 typedef wchar_t char16;
  37 typedef std::wstring string16;
  38
  39 #elif defined(WCHAR_T_IS_UTF32)
  40
  41 typedef uint16 char16;
  42
  43 namespace base {
  44
  45 // char16 versions of the functions required by string16_char_traits; these
  46 // are based on the wide character functions of similar names ("w" or "wcs"
  47 // instead of "c16").
  48 int c16memcmp(const char16* s1, const char16* s2, size_t n);
  49 size_t c16len(const char16* s);
  50 const char16* c16memchr(const char16* s, char16 c, size_t n);
  51 char16* c16memmove(char16* s1, const char16* s2, size_t n);
  52 char16* c16memcpy(char16* s1, const char16* s2, size_t n);
  53 char16* c16memset(char16* s, char16 c, size_t n);
  54
  55 struct string16_char_traits {
  56   typedef char16 char_type;
  57   typedef int int_type;
  58
  59   // int_type needs to be able to hold each possible value of char_type, and in
  60   // addition, the distinct value of eof().
  61   COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width);
  62
  63   typedef std::streamoff off_type;
  64   typedef mbstate_t state_type;
  65   typedef std::fpos<state_type> pos_type;
  66
  67   static void assign(char_type& c1, const char_type& c2) {
  68     c1 = c2;
  69   }
  70
  71   static bool eq(const char_type& c1, const char_type& c2) {
  72     return c1 == c2;
  73   }
  74   static bool lt(const char_type& c1, const char_type& c2) {
  75     return c1 < c2;
  76   }
  77
  78   static int compare(const char_type* s1, const char_type* s2, size_t n) {
  79     return c16memcmp(s1, s2, n);
  80   }
  81
  82   static size_t length(const char_type* s) {
  83     return c16len(s);
  84   }
  85
  86   static const char_type* find(const char_type* s, size_t n,
  87                                const char_type& a) {
  88     return c16memchr(s, a, n);
  89   }
  90
  91   static char_type* move(char_type* s1, const char_type* s2, int_type n) {
  92     return c16memmove(s1, s2, n);
  93   }
  94
  95   static char_type* copy(char_type* s1, const char_type* s2, size_t n) {
  96     return c16memcpy(s1, s2, n);
  97   }
  98
  99   static char_type* assign(char_type* s, size_t n, char_type a) {
 100     return c16memset(s, a, n);
 101   }
 102
 103   static int_type not_eof(const int_type& c) {
 104     return eq_int_type(c, eof()) ? 0 : c;
 105   }
 106
 107   static char_type to_char_type(const int_type& c) {
 108     return char_type(c);
 109   }
 110
 111   static int_type to_int_type(const char_type& c) {
 112     return int_type(c);
 113   }
 114
 115   static bool eq_int_type(const int_type& c1, const int_type& c2) {
 116     return c1 == c2;
 117   }
 118
 119   static int_type eof() {
 120     return static_cast<int_type>(EOF);
 121   }
 122 };
 123
 124 }  // namespace base
 125
 126 // The string class will be explicitly instantiated only once, in string16.cc.
 127 //
 128 // std::basic_string<> in GNU libstdc++ contains a static data member,
 129 // _S_empty_rep_storage, to represent empty strings.  When an operation such
 130 // as assignment or destruction is performed on a string, causing its existing
 131 // data member to be invalidated, it must not be freed if this static data
 132 // member is being used.  Otherwise, it counts as an attempt to free static
 133 // (and not allocated) data, which is a memory error.
 134 //
 135 // Generally, due to C++ template magic, _S_empty_rep_storage will be marked
 136 // as a coalesced symbol, meaning that the linker will combine multiple
 137 // instances into a single one when generating output.
 138 //
 139 // If a string class is used by multiple shared libraries, a problem occurs.
 140 // Each library will get its own copy of _S_empty_rep_storage.  When strings
 141 // are passed across a library boundary for alteration or destruction, memory
 142 // errors will result.  GNU libstdc++ contains a configuration option,
 143 // --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which
 144 // disables the static data member optimization, but it's a good optimization
 145 // and non-STL code is generally at the mercy of the system's STL
 146 // configuration.  Fully-dynamic strings are not the default for GNU libstdc++
 147 // libstdc++ itself or for the libstdc++ installations on the systems we care
 148 // about, such as Mac OS X and relevant flavors of Linux.
 149 //
 150 // See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 .
 151 //
 152 // To avoid problems, string classes need to be explicitly instantiated only
 153 // once, in exactly one library.  All other string users see it via an "extern"
 154 // declaration.  This is precisely how GNU libstdc++ handles
 155 // std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring).
 156 //
 157 // This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2),
 158 // in which the linker does not fully coalesce symbols when dead code
 159 // stripping is enabled.  This bug causes the memory errors described above
 160 // to occur even when a std::basic_string<> does not cross shared library
 161 // boundaries, such as in statically-linked executables.
 162 //
 163 // TODO(mark): File this bug with Apple and update this note with a bug number.
 164
 165 extern template class std::basic_string<char16, base::string16_char_traits>;
 166
 167 typedef std::basic_string<char16, base::string16_char_traits> string16;
 168
 169 extern std::ostream& operator<<(std::ostream& out, const string16& str);
 170
 171 #endif  // WCHAR_T_IS_UTF32
 172
 173 #endif  // BASE_STRING16_H_