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/base_api.h" 34 #include "base/basictypes.h" 35 36 #if defined(WCHAR_T_IS_UTF16) 37 38 typedef wchar_t char16; 39 typedef std::wstring string16; 40 41 #elif defined(WCHAR_T_IS_UTF32) 42 43 typedef uint16 char16; 44 45 namespace base { 46 47 // char16 versions of the functions required by string16_char_traits; these 48 // are based on the wide character functions of similar names ("w" or "wcs" 49 // instead of "c16"). 50 BASE_API int c16memcmp(const char16* s1, const char16* s2, size_t n); 51 BASE_API size_t c16len(const char16* s); 52 BASE_API const char16* c16memchr(const char16* s, char16 c, size_t n); 53 BASE_API char16* c16memmove(char16* s1, const char16* s2, size_t n); 54 BASE_API char16* c16memcpy(char16* s1, const char16* s2, size_t n); 55 BASE_API char16* c16memset(char16* s, char16 c, size_t n); 56 57 struct 58 #ifdef ANDROID 59 BASE_API 60 #endif 61 string16_char_traits { 62 typedef char16 char_type; 63 typedef int int_type; 64 65 // int_type needs to be able to hold each possible value of char_type, and in 66 // addition, the distinct value of eof(). 67 COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width); 68 69 typedef std::streamoff off_type; 70 typedef mbstate_t state_type; 71 typedef std::fpos<state_type> pos_type; 72 assignstring16_char_traits73 static void assign(char_type& c1, const char_type& c2) { 74 c1 = c2; 75 } 76 eqstring16_char_traits77 static bool eq(const char_type& c1, const char_type& c2) { 78 return c1 == c2; 79 } ltstring16_char_traits80 static bool lt(const char_type& c1, const char_type& c2) { 81 return c1 < c2; 82 } 83 comparestring16_char_traits84 static int compare(const char_type* s1, const char_type* s2, size_t n) { 85 return c16memcmp(s1, s2, n); 86 } 87 lengthstring16_char_traits88 static size_t length(const char_type* s) { 89 return c16len(s); 90 } 91 findstring16_char_traits92 static const char_type* find(const char_type* s, size_t n, 93 const char_type& a) { 94 return c16memchr(s, a, n); 95 } 96 movestring16_char_traits97 static char_type* move(char_type* s1, const char_type* s2, int_type n) { 98 return c16memmove(s1, s2, n); 99 } 100 copystring16_char_traits101 static char_type* copy(char_type* s1, const char_type* s2, size_t n) { 102 return c16memcpy(s1, s2, n); 103 } 104 assignstring16_char_traits105 static char_type* assign(char_type* s, size_t n, char_type a) { 106 return c16memset(s, a, n); 107 } 108 not_eofstring16_char_traits109 static int_type not_eof(const int_type& c) { 110 return eq_int_type(c, eof()) ? 0 : c; 111 } 112 to_char_typestring16_char_traits113 static char_type to_char_type(const int_type& c) { 114 return char_type(c); 115 } 116 to_int_typestring16_char_traits117 static int_type to_int_type(const char_type& c) { 118 return int_type(c); 119 } 120 eq_int_typestring16_char_traits121 static bool eq_int_type(const int_type& c1, const int_type& c2) { 122 return c1 == c2; 123 } 124 eofstring16_char_traits125 static int_type eof() { 126 return static_cast<int_type>(EOF); 127 } 128 }; 129 130 } // namespace base 131 132 // The string class will be explicitly instantiated only once, in string16.cc. 133 // 134 // std::basic_string<> in GNU libstdc++ contains a static data member, 135 // _S_empty_rep_storage, to represent empty strings. When an operation such 136 // as assignment or destruction is performed on a string, causing its existing 137 // data member to be invalidated, it must not be freed if this static data 138 // member is being used. Otherwise, it counts as an attempt to free static 139 // (and not allocated) data, which is a memory error. 140 // 141 // Generally, due to C++ template magic, _S_empty_rep_storage will be marked 142 // as a coalesced symbol, meaning that the linker will combine multiple 143 // instances into a single one when generating output. 144 // 145 // If a string class is used by multiple shared libraries, a problem occurs. 146 // Each library will get its own copy of _S_empty_rep_storage. When strings 147 // are passed across a library boundary for alteration or destruction, memory 148 // errors will result. GNU libstdc++ contains a configuration option, 149 // --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which 150 // disables the static data member optimization, but it's a good optimization 151 // and non-STL code is generally at the mercy of the system's STL 152 // configuration. Fully-dynamic strings are not the default for GNU libstdc++ 153 // libstdc++ itself or for the libstdc++ installations on the systems we care 154 // about, such as Mac OS X and relevant flavors of Linux. 155 // 156 // See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 . 157 // 158 // To avoid problems, string classes need to be explicitly instantiated only 159 // once, in exactly one library. All other string users see it via an "extern" 160 // declaration. This is precisely how GNU libstdc++ handles 161 // std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring). 162 // 163 // This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2), 164 // in which the linker does not fully coalesce symbols when dead code 165 // stripping is enabled. This bug causes the memory errors described above 166 // to occur even when a std::basic_string<> does not cross shared library 167 // boundaries, such as in statically-linked executables. 168 // 169 // TODO(mark): File this bug with Apple and update this note with a bug number. 170 171 extern template class 172 #ifdef ANDROID 173 BASE_API 174 #endif 175 std::basic_string<char16, base::string16_char_traits>; 176 177 typedef std::basic_string<char16, base::string16_char_traits> string16; 178 179 namespace base { 180 BASE_API extern std::ostream& operator<<(std::ostream& out, const string16& str); 181 } 182 183 #endif // WCHAR_T_IS_UTF32 184 185 #endif // BASE_STRING16_H_ 186