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 #pragma once
   8
   9 // WHAT:
  10 // A version of std::basic_string that provides 2-byte characters even when
  11 // wchar_t is not implemented as a 2-byte type. You can access this class as
  12 // string16. We also define char16, which string16 is based upon.
  13 //
  14 // WHY:
  15 // On Windows, wchar_t is 2 bytes, and it can conveniently handle UTF-16/UCS-2
  16 // data. Plenty of existing code operates on strings encoded as UTF-16.
  17 //
  18 // On many other platforms, sizeof(wchar_t) is 4 bytes by default. We can make
  19 // it 2 bytes by using the GCC flag -fshort-wchar. But then std::wstring fails
  20 // at run time, because it calls some functions (like wcslen) that come from
  21 // the system's native C library -- which was built with a 4-byte wchar_t!
  22 // It's wasteful to use 4-byte wchar_t strings to carry UTF-16 data, and it's
  23 // entirely improper on those systems where the encoding of wchar_t is defined
  24 // as UTF-32.
  25 //
  26 // Here, we define string16, which is similar to std::wstring but replaces all
  27 // libc functions with custom, 2-byte-char compatible routines. It is capable
  28 // of carrying UTF-16-encoded data.
  29
  30 #include <stdio.h>
  31 #include <string>
  32
  33 #include "base/basictypes.h"
  34
  35 #if defined(WCHAR_T_IS_UTF16)
  36
  37 typedef wchar_t char16;
  38 typedef std::wstring string16;
  39
  40 #elif defined(WCHAR_T_IS_UTF32)
  41
  42 typedef uint16 char16;
  43
  44 namespace base {
  45
  46 // char16 versions of the functions required by string16_char_traits; these
  47 // are based on the wide character functions of similar names ("w" or "wcs"
  48 // instead of "c16").
  49 int c16memcmp(const char16* s1, const char16* s2, size_t n);
  50 size_t c16len(const char16* s);
  51 const char16* c16memchr(const char16* s, char16 c, size_t n);
  52 char16* c16memmove(char16* s1, const char16* s2, size_t n);
  53 char16* c16memcpy(char16* s1, const char16* s2, size_t n);
  54 char16* c16memset(char16* s, char16 c, size_t n);
  55
  56 struct string16_char_traits {
  57   typedef char16 char_type;
  58   typedef int int_type;
  59
  60   // int_type needs to be able to hold each possible value of char_type, and in
  61   // addition, the distinct value of eof().
  62   COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width);
  63
  64   typedef std::streamoff off_type;
  65   typedef mbstate_t state_type;
  66   typedef std::fpos<state_type> pos_type;
  67
  68   static void assign(char_type& c1, const char_type& c2) {
  69     c1 = c2;
  70   }
  71
  72   static bool eq(const char_type& c1, const char_type& c2) {
  73     return c1 == c2;
  74   }
  75   static bool lt(const char_type& c1, const char_type& c2) {
  76     return c1 < c2;
  77   }
  78
  79   static int compare(const char_type* s1, const char_type* s2, size_t n) {
  80     return c16memcmp(s1, s2, n);
  81   }
  82
  83   static size_t length(const char_type* s) {
  84     return c16len(s);
  85   }
  86
  87   static const char_type* find(const char_type* s, size_t n,
  88                                const char_type& a) {
  89     return c16memchr(s, a, n);
  90   }
  91
  92   static char_type* move(char_type* s1, const char_type* s2, int_type n) {
  93     return c16memmove(s1, s2, n);
  94   }
  95
  96   static char_type* copy(char_type* s1, const char_type* s2, size_t n) {
  97     return c16memcpy(s1, s2, n);
  98   }
  99
 100   static char_type* assign(char_type* s, size_t n, char_type a) {
 101     return c16memset(s, a, n);
 102   }
 103
 104   static int_type not_eof(const int_type& c) {
 105     return eq_int_type(c, eof()) ? 0 : c;
 106   }
 107
 108   static char_type to_char_type(const int_type& c) {
 109     return char_type(c);
 110   }
 111
 112   static int_type to_int_type(const char_type& c) {
 113     return int_type(c);
 114   }
 115
 116   static bool eq_int_type(const int_type& c1, const int_type& c2) {
 117     return c1 == c2;
 118   }
 119
 120   static int_type eof() {
 121     return static_cast<int_type>(EOF);
 122   }
 123 };
 124
 125 }  // namespace base
 126
 127 // The string class will be explicitly instantiated only once, in string16.cc.
 128 //
 129 // std::basic_string<> in GNU libstdc++ contains a static data member,
 130 // _S_empty_rep_storage, to represent empty strings.  When an operation such
 131 // as assignment or destruction is performed on a string, causing its existing
 132 // data member to be invalidated, it must not be freed if this static data
 133 // member is being used.  Otherwise, it counts as an attempt to free static
 134 // (and not allocated) data, which is a memory error.
 135 //
 136 // Generally, due to C++ template magic, _S_empty_rep_storage will be marked
 137 // as a coalesced symbol, meaning that the linker will combine multiple
 138 // instances into a single one when generating output.
 139 //
 140 // If a string class is used by multiple shared libraries, a problem occurs.
 141 // Each library will get its own copy of _S_empty_rep_storage.  When strings
 142 // are passed across a library boundary for alteration or destruction, memory
 143 // errors will result.  GNU libstdc++ contains a configuration option,
 144 // --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which
 145 // disables the static data member optimization, but it's a good optimization
 146 // and non-STL code is generally at the mercy of the system's STL
 147 // configuration.  Fully-dynamic strings are not the default for GNU libstdc++
 148 // libstdc++ itself or for the libstdc++ installations on the systems we care
 149 // about, such as Mac OS X and relevant flavors of Linux.
 150 //
 151 // See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 .
 152 //
 153 // To avoid problems, string classes need to be explicitly instantiated only
 154 // once, in exactly one library.  All other string users see it via an "extern"
 155 // declaration.  This is precisely how GNU libstdc++ handles
 156 // std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring).
 157 //
 158 // This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2),
 159 // in which the linker does not fully coalesce symbols when dead code
 160 // stripping is enabled.  This bug causes the memory errors described above
 161 // to occur even when a std::basic_string<> does not cross shared library
 162 // boundaries, such as in statically-linked executables.
 163 //
 164 // TODO(mark): File this bug with Apple and update this note with a bug number.
 165
 166 extern template class std::basic_string<char16, base::string16_char_traits>;
 167
 168 typedef std::basic_string<char16, base::string16_char_traits> string16;
 169
 170 namespace base {
 171 extern std::ostream& operator<<(std::ostream& out, const string16& str);
 172 }
 173
 174 #endif  // WCHAR_T_IS_UTF32
 175
 176 #endif  // BASE_STRING16_H_