base/string16.h

   1 // Copyright (c) 2012 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/base_export.h"
  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 namespace base {
  41 typedef std::char_traits<wchar_t> string16_char_traits;
  42 }
  43
  44 #elif defined(WCHAR_T_IS_UTF32)
  45
  46 typedef uint16 char16;
  47
  48 namespace base {
  49
  50 // char16 versions of the functions required by string16_char_traits; these
  51 // are based on the wide character functions of similar names ("w" or "wcs"
  52 // instead of "c16").
  53 BASE_EXPORT int c16memcmp(const char16* s1, const char16* s2, size_t n);
  54 BASE_EXPORT size_t c16len(const char16* s);
  55 BASE_EXPORT const char16* c16memchr(const char16* s, char16 c, size_t n);
  56 BASE_EXPORT char16* c16memmove(char16* s1, const char16* s2, size_t n);
  57 BASE_EXPORT char16* c16memcpy(char16* s1, const char16* s2, size_t n);
  58 BASE_EXPORT char16* c16memset(char16* s, char16 c, size_t n);
  59
  60 struct string16_char_traits {
  61   typedef char16 char_type;
  62   typedef int int_type;
  63
  64   // int_type needs to be able to hold each possible value of char_type, and in
  65   // addition, the distinct value of eof().
  66   COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width);
  67
  68   typedef std::streamoff off_type;
  69   typedef mbstate_t state_type;
  70   typedef std::fpos<state_type> pos_type;
  71
  72   static void assign(char_type& c1, const char_type& c2) {
  73     c1 = c2;
  74   }
  75
  76   static bool eq(const char_type& c1, const char_type& c2) {
  77     return c1 == c2;
  78   }
  79   static bool lt(const char_type& c1, const char_type& c2) {
  80     return c1 < c2;
  81   }
  82
  83   static int compare(const char_type* s1, const char_type* s2, size_t n) {
  84     return c16memcmp(s1, s2, n);
  85   }
  86
  87   static size_t length(const char_type* s) {
  88     return c16len(s);
  89   }
  90
  91   static const char_type* find(const char_type* s, size_t n,
  92                                const char_type& a) {
  93     return c16memchr(s, a, n);
  94   }
  95
  96   static char_type* move(char_type* s1, const char_type* s2, int_type n) {
  97     return c16memmove(s1, s2, n);
  98   }
  99
 100   static char_type* copy(char_type* s1, const char_type* s2, size_t n) {
 101     return c16memcpy(s1, s2, n);
 102   }
 103
 104   static char_type* assign(char_type* s, size_t n, char_type a) {
 105     return c16memset(s, a, n);
 106   }
 107
 108   static int_type not_eof(const int_type& c) {
 109     return eq_int_type(c, eof()) ? 0 : c;
 110   }
 111
 112   static char_type to_char_type(const int_type& c) {
 113     return char_type(c);
 114   }
 115
 116   static int_type to_int_type(const char_type& c) {
 117     return int_type(c);
 118   }
 119
 120   static bool eq_int_type(const int_type& c1, const int_type& c2) {
 121     return c1 == c2;
 122   }
 123
 124   static int_type eof() {
 125     return static_cast<int_type>(EOF);
 126   }
 127 };
 128
 129 }  // namespace base
 130
 131 // The string class will be explicitly instantiated only once, in string16.cc.
 132 //
 133 // std::basic_string<> in GNU libstdc++ contains a static data member,
 134 // _S_empty_rep_storage, to represent empty strings.  When an operation such
 135 // as assignment or destruction is performed on a string, causing its existing
 136 // data member to be invalidated, it must not be freed if this static data
 137 // member is being used.  Otherwise, it counts as an attempt to free static
 138 // (and not allocated) data, which is a memory error.
 139 //
 140 // Generally, due to C++ template magic, _S_empty_rep_storage will be marked
 141 // as a coalesced symbol, meaning that the linker will combine multiple
 142 // instances into a single one when generating output.
 143 //
 144 // If a string class is used by multiple shared libraries, a problem occurs.
 145 // Each library will get its own copy of _S_empty_rep_storage.  When strings
 146 // are passed across a library boundary for alteration or destruction, memory
 147 // errors will result.  GNU libstdc++ contains a configuration option,
 148 // --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which
 149 // disables the static data member optimization, but it's a good optimization
 150 // and non-STL code is generally at the mercy of the system's STL
 151 // configuration.  Fully-dynamic strings are not the default for GNU libstdc++
 152 // libstdc++ itself or for the libstdc++ installations on the systems we care
 153 // about, such as Mac OS X and relevant flavors of Linux.
 154 //
 155 // See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 .
 156 //
 157 // To avoid problems, string classes need to be explicitly instantiated only
 158 // once, in exactly one library.  All other string users see it via an "extern"
 159 // declaration.  This is precisely how GNU libstdc++ handles
 160 // std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring).
 161 //
 162 // This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2),
 163 // in which the linker does not fully coalesce symbols when dead code
 164 // stripping is enabled.  This bug causes the memory errors described above
 165 // to occur even when a std::basic_string<> does not cross shared library
 166 // boundaries, such as in statically-linked executables.
 167 //
 168 // TODO(mark): File this bug with Apple and update this note with a bug number.
 169
 170 extern template
 171 class BASE_EXPORT std::basic_string<char16, base::string16_char_traits>;
 172
 173 typedef std::basic_string<char16, base::string16_char_traits> string16;
 174
 175 namespace base {
 176 BASE_EXPORT extern std::ostream& operator<<(std::ostream& out,
 177                                             const string16& str);
 178
 179 // This is required by googletest to print a readable output on test failures.
 180 BASE_EXPORT extern void PrintTo(const string16& str, std::ostream* out);
 181 }
 182
 183 #endif  // WCHAR_T_IS_UTF32
 184
 185 #endif  // BASE_STRING16_H_