Wed, 31 Dec 2014 06:09:35 +0100
Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.
michael@0 | 1 | // Copyright (c) 2006-2008 The Chromium Authors. All rights reserved. |
michael@0 | 2 | // Use of this source code is governed by a BSD-style license that can be |
michael@0 | 3 | // found in the LICENSE file. |
michael@0 | 4 | |
michael@0 | 5 | #ifndef BASE_STRING16_H_ |
michael@0 | 6 | #define BASE_STRING16_H_ |
michael@0 | 7 | |
michael@0 | 8 | // WHAT: |
michael@0 | 9 | // A version of std::basic_string that provides 2-byte characters even when |
michael@0 | 10 | // wchar_t is not implemented as a 2-byte type. You can access this class as |
michael@0 | 11 | // string16. We also define char16, which string16 is based upon. |
michael@0 | 12 | // |
michael@0 | 13 | // WHY: |
michael@0 | 14 | // On Windows, wchar_t is 2 bytes, and it can conveniently handle UTF-16/UCS-2 |
michael@0 | 15 | // data. Plenty of existing code operates on strings encoded as UTF-16. |
michael@0 | 16 | // |
michael@0 | 17 | // On many other platforms, sizeof(wchar_t) is 4 bytes by default. We can make |
michael@0 | 18 | // it 2 bytes by using the GCC flag -fshort-wchar. But then std::wstring fails |
michael@0 | 19 | // at run time, because it calls some functions (like wcslen) that come from |
michael@0 | 20 | // the system's native C library -- which was built with a 4-byte wchar_t! |
michael@0 | 21 | // It's wasteful to use 4-byte wchar_t strings to carry UTF-16 data, and it's |
michael@0 | 22 | // entirely improper on those systems where the encoding of wchar_t is defined |
michael@0 | 23 | // as UTF-32. |
michael@0 | 24 | // |
michael@0 | 25 | // Here, we define string16, which is similar to std::wstring but replaces all |
michael@0 | 26 | // libc functions with custom, 2-byte-char compatible routines. It is capable |
michael@0 | 27 | // of carrying UTF-16-encoded data. |
michael@0 | 28 | |
michael@0 | 29 | #include <stdio.h> |
michael@0 | 30 | #include <string> |
michael@0 | 31 | |
michael@0 | 32 | #include "base/basictypes.h" |
michael@0 | 33 | |
michael@0 | 34 | #if defined(WCHAR_T_IS_UTF16) |
michael@0 | 35 | |
michael@0 | 36 | typedef wchar_t char16; |
michael@0 | 37 | typedef std::wstring string16; |
michael@0 | 38 | |
michael@0 | 39 | #elif defined(WCHAR_T_IS_UTF32) |
michael@0 | 40 | |
michael@0 | 41 | typedef uint16_t char16; |
michael@0 | 42 | |
michael@0 | 43 | namespace base { |
michael@0 | 44 | |
michael@0 | 45 | // char16 versions of the functions required by string16_char_traits; these |
michael@0 | 46 | // are based on the wide character functions of similar names ("w" or "wcs" |
michael@0 | 47 | // instead of "c16"). |
michael@0 | 48 | int c16memcmp(const char16* s1, const char16* s2, size_t n); |
michael@0 | 49 | size_t c16len(const char16* s); |
michael@0 | 50 | const char16* c16memchr(const char16* s, char16 c, size_t n); |
michael@0 | 51 | char16* c16memmove(char16* s1, const char16* s2, size_t n); |
michael@0 | 52 | char16* c16memcpy(char16* s1, const char16* s2, size_t n); |
michael@0 | 53 | char16* c16memset(char16* s, char16 c, size_t n); |
michael@0 | 54 | |
michael@0 | 55 | struct string16_char_traits { |
michael@0 | 56 | typedef char16 char_type; |
michael@0 | 57 | typedef int int_type; |
michael@0 | 58 | |
michael@0 | 59 | // int_type needs to be able to hold each possible value of char_type, and in |
michael@0 | 60 | // addition, the distinct value of eof(). |
michael@0 | 61 | COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width); |
michael@0 | 62 | |
michael@0 | 63 | typedef std::streamoff off_type; |
michael@0 | 64 | typedef mbstate_t state_type; |
michael@0 | 65 | typedef std::fpos<state_type> pos_type; |
michael@0 | 66 | |
michael@0 | 67 | static void assign(char_type& c1, const char_type& c2) { |
michael@0 | 68 | c1 = c2; |
michael@0 | 69 | } |
michael@0 | 70 | |
michael@0 | 71 | static bool eq(const char_type& c1, const char_type& c2) { |
michael@0 | 72 | return c1 == c2; |
michael@0 | 73 | } |
michael@0 | 74 | static bool lt(const char_type& c1, const char_type& c2) { |
michael@0 | 75 | return c1 < c2; |
michael@0 | 76 | } |
michael@0 | 77 | |
michael@0 | 78 | static int compare(const char_type* s1, const char_type* s2, size_t n) { |
michael@0 | 79 | return c16memcmp(s1, s2, n); |
michael@0 | 80 | } |
michael@0 | 81 | |
michael@0 | 82 | static size_t length(const char_type* s) { |
michael@0 | 83 | return c16len(s); |
michael@0 | 84 | } |
michael@0 | 85 | |
michael@0 | 86 | static const char_type* find(const char_type* s, size_t n, |
michael@0 | 87 | const char_type& a) { |
michael@0 | 88 | return c16memchr(s, a, n); |
michael@0 | 89 | } |
michael@0 | 90 | |
michael@0 | 91 | static char_type* move(char_type* s1, const char_type* s2, int_type n) { |
michael@0 | 92 | return c16memmove(s1, s2, n); |
michael@0 | 93 | } |
michael@0 | 94 | |
michael@0 | 95 | static char_type* copy(char_type* s1, const char_type* s2, size_t n) { |
michael@0 | 96 | return c16memcpy(s1, s2, n); |
michael@0 | 97 | } |
michael@0 | 98 | |
michael@0 | 99 | static char_type* assign(char_type* s, size_t n, char_type a) { |
michael@0 | 100 | return c16memset(s, a, n); |
michael@0 | 101 | } |
michael@0 | 102 | |
michael@0 | 103 | static int_type not_eof(const int_type& c) { |
michael@0 | 104 | return eq_int_type(c, eof()) ? 0 : c; |
michael@0 | 105 | } |
michael@0 | 106 | |
michael@0 | 107 | static char_type to_char_type(const int_type& c) { |
michael@0 | 108 | return char_type(c); |
michael@0 | 109 | } |
michael@0 | 110 | |
michael@0 | 111 | static int_type to_int_type(const char_type& c) { |
michael@0 | 112 | return int_type(c); |
michael@0 | 113 | } |
michael@0 | 114 | |
michael@0 | 115 | static bool eq_int_type(const int_type& c1, const int_type& c2) { |
michael@0 | 116 | return c1 == c2; |
michael@0 | 117 | } |
michael@0 | 118 | |
michael@0 | 119 | static int_type eof() { |
michael@0 | 120 | return static_cast<int_type>(EOF); |
michael@0 | 121 | } |
michael@0 | 122 | }; |
michael@0 | 123 | |
michael@0 | 124 | } // namespace base |
michael@0 | 125 | |
michael@0 | 126 | // The string class will be explicitly instantiated only once, in string16.cc. |
michael@0 | 127 | // |
michael@0 | 128 | // std::basic_string<> in GNU libstdc++ contains a static data member, |
michael@0 | 129 | // _S_empty_rep_storage, to represent empty strings. When an operation such |
michael@0 | 130 | // as assignment or destruction is performed on a string, causing its existing |
michael@0 | 131 | // data member to be invalidated, it must not be freed if this static data |
michael@0 | 132 | // member is being used. Otherwise, it counts as an attempt to free static |
michael@0 | 133 | // (and not allocated) data, which is a memory error. |
michael@0 | 134 | // |
michael@0 | 135 | // Generally, due to C++ template magic, _S_empty_rep_storage will be marked |
michael@0 | 136 | // as a coalesced symbol, meaning that the linker will combine multiple |
michael@0 | 137 | // instances into a single one when generating output. |
michael@0 | 138 | // |
michael@0 | 139 | // If a string class is used by multiple shared libraries, a problem occurs. |
michael@0 | 140 | // Each library will get its own copy of _S_empty_rep_storage. When strings |
michael@0 | 141 | // are passed across a library boundary for alteration or destruction, memory |
michael@0 | 142 | // errors will result. GNU libstdc++ contains a configuration option, |
michael@0 | 143 | // --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which |
michael@0 | 144 | // disables the static data member optimization, but it's a good optimization |
michael@0 | 145 | // and non-STL code is generally at the mercy of the system's STL |
michael@0 | 146 | // configuration. Fully-dynamic strings are not the default for GNU libstdc++ |
michael@0 | 147 | // libstdc++ itself or for the libstdc++ installations on the systems we care |
michael@0 | 148 | // about, such as Mac OS X and relevant flavors of Linux. |
michael@0 | 149 | // |
michael@0 | 150 | // See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 . |
michael@0 | 151 | // |
michael@0 | 152 | // To avoid problems, string classes need to be explicitly instantiated only |
michael@0 | 153 | // once, in exactly one library. All other string users see it via an "extern" |
michael@0 | 154 | // declaration. This is precisely how GNU libstdc++ handles |
michael@0 | 155 | // std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring). |
michael@0 | 156 | // |
michael@0 | 157 | // This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2), |
michael@0 | 158 | // in which the linker does not fully coalesce symbols when dead code |
michael@0 | 159 | // stripping is enabled. This bug causes the memory errors described above |
michael@0 | 160 | // to occur even when a std::basic_string<> does not cross shared library |
michael@0 | 161 | // boundaries, such as in statically-linked executables. |
michael@0 | 162 | // |
michael@0 | 163 | // TODO(mark): File this bug with Apple and update this note with a bug number. |
michael@0 | 164 | |
michael@0 | 165 | extern template class std::basic_string<char16, base::string16_char_traits>; |
michael@0 | 166 | |
michael@0 | 167 | typedef std::basic_string<char16, base::string16_char_traits> string16; |
michael@0 | 168 | |
michael@0 | 169 | extern std::ostream& operator<<(std::ostream& out, const string16& str); |
michael@0 | 170 | |
michael@0 | 171 | #endif // WCHAR_T_IS_UTF32 |
michael@0 | 172 | |
michael@0 | 173 | #endif // BASE_STRING16_H_ |