Line data Source code
1 : /* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
2 : /* vim: set ts=8 sts=2 et sw=2 tw=80: */
3 : // Copyright (c) 2006-2008 The Chromium Authors. All rights reserved.
4 : // Use of this source code is governed by a BSD-style license that can be
5 : // found in the LICENSE file.
6 :
7 : #ifndef BASE_STRING16_H_
8 : #define BASE_STRING16_H_
9 :
10 : // WHAT:
11 : // A version of std::basic_string that provides 2-byte characters even when
12 : // wchar_t is not implemented as a 2-byte type. You can access this class as
13 : // string16. We also define char16, which string16 is based upon.
14 : //
15 : // WHY:
16 : // On Windows, wchar_t is 2 bytes, and it can conveniently handle UTF-16/UCS-2
17 : // data. Plenty of existing code operates on strings encoded as UTF-16.
18 : //
19 : // On many other platforms, sizeof(wchar_t) is 4 bytes by default. We can make
20 : // it 2 bytes by using the GCC flag -fshort-wchar. But then std::wstring fails
21 : // at run time, because it calls some functions (like wcslen) that come from
22 : // the system's native C library -- which was built with a 4-byte wchar_t!
23 : // It's wasteful to use 4-byte wchar_t strings to carry UTF-16 data, and it's
24 : // entirely improper on those systems where the encoding of wchar_t is defined
25 : // as UTF-32.
26 : //
27 : // Here, we define string16, which is similar to std::wstring but replaces all
28 : // libc functions with custom, 2-byte-char compatible routines. It is capable
29 : // of carrying UTF-16-encoded data.
30 :
31 : #include <stdio.h>
32 : #include <string>
33 :
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_t 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 : int c16memcmp(const char16* s1, const char16* s2, size_t n);
51 : size_t c16len(const char16* s);
52 : const char16* c16memchr(const char16* s, char16 c, size_t n);
53 : char16* c16memmove(char16* s1, const char16* s2, size_t n);
54 : char16* c16memcpy(char16* s1, const char16* s2, size_t n);
55 : char16* c16memset(char16* s, char16 c, size_t n);
56 :
57 : struct string16_char_traits {
58 : typedef char16 char_type;
59 : typedef int int_type;
60 :
61 : // int_type needs to be able to hold each possible value of char_type, and in
62 : // addition, the distinct value of eof().
63 : COMPILE_ASSERT(sizeof(int_type) > sizeof(char_type), unexpected_type_width);
64 :
65 : typedef std::streamoff off_type;
66 : typedef mbstate_t state_type;
67 : typedef std::fpos<state_type> pos_type;
68 :
69 0 : static void assign(char_type& c1, const char_type& c2) {
70 0 : c1 = c2;
71 0 : }
72 :
73 0 : static bool eq(const char_type& c1, const char_type& c2) {
74 0 : return c1 == c2;
75 : }
76 : static bool lt(const char_type& c1, const char_type& c2) {
77 : return c1 < c2;
78 : }
79 :
80 0 : static int compare(const char_type* s1, const char_type* s2, size_t n) {
81 0 : return c16memcmp(s1, s2, n);
82 : }
83 :
84 0 : static size_t length(const char_type* s) {
85 0 : return c16len(s);
86 : }
87 :
88 0 : static const char_type* find(const char_type* s, size_t n,
89 : const char_type& a) {
90 0 : return c16memchr(s, a, n);
91 : }
92 :
93 0 : static char_type* move(char_type* s1, const char_type* s2, int_type n) {
94 0 : return c16memmove(s1, s2, n);
95 : }
96 :
97 0 : static char_type* copy(char_type* s1, const char_type* s2, size_t n) {
98 0 : return c16memcpy(s1, s2, n);
99 : }
100 :
101 0 : static char_type* assign(char_type* s, size_t n, char_type a) {
102 0 : return c16memset(s, a, n);
103 : }
104 :
105 : static int_type not_eof(const int_type& c) {
106 : return eq_int_type(c, eof()) ? 0 : c;
107 : }
108 :
109 : static char_type to_char_type(const int_type& c) {
110 : return char_type(c);
111 : }
112 :
113 : static int_type to_int_type(const char_type& c) {
114 : return int_type(c);
115 : }
116 :
117 : static bool eq_int_type(const int_type& c1, const int_type& c2) {
118 : return c1 == c2;
119 : }
120 :
121 : static int_type eof() {
122 : return static_cast<int_type>(EOF);
123 : }
124 : };
125 :
126 : } // namespace base
127 :
128 : // The string class will be explicitly instantiated only once, in string16.cc.
129 : //
130 : // std::basic_string<> in GNU libstdc++ contains a static data member,
131 : // _S_empty_rep_storage, to represent empty strings. When an operation such
132 : // as assignment or destruction is performed on a string, causing its existing
133 : // data member to be invalidated, it must not be freed if this static data
134 : // member is being used. Otherwise, it counts as an attempt to free static
135 : // (and not allocated) data, which is a memory error.
136 : //
137 : // Generally, due to C++ template magic, _S_empty_rep_storage will be marked
138 : // as a coalesced symbol, meaning that the linker will combine multiple
139 : // instances into a single one when generating output.
140 : //
141 : // If a string class is used by multiple shared libraries, a problem occurs.
142 : // Each library will get its own copy of _S_empty_rep_storage. When strings
143 : // are passed across a library boundary for alteration or destruction, memory
144 : // errors will result. GNU libstdc++ contains a configuration option,
145 : // --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which
146 : // disables the static data member optimization, but it's a good optimization
147 : // and non-STL code is generally at the mercy of the system's STL
148 : // configuration. Fully-dynamic strings are not the default for GNU libstdc++
149 : // libstdc++ itself or for the libstdc++ installations on the systems we care
150 : // about, such as Mac OS X and relevant flavors of Linux.
151 : //
152 : // See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 .
153 : //
154 : // To avoid problems, string classes need to be explicitly instantiated only
155 : // once, in exactly one library. All other string users see it via an "extern"
156 : // declaration. This is precisely how GNU libstdc++ handles
157 : // std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring).
158 : //
159 : // This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2),
160 : // in which the linker does not fully coalesce symbols when dead code
161 : // stripping is enabled. This bug causes the memory errors described above
162 : // to occur even when a std::basic_string<> does not cross shared library
163 : // boundaries, such as in statically-linked executables.
164 : //
165 : // TODO(mark): File this bug with Apple and update this note with a bug number.
166 :
167 : extern template class std::basic_string<char16, base::string16_char_traits>;
168 :
169 : typedef std::basic_string<char16, base::string16_char_traits> string16;
170 :
171 : extern std::ostream& operator<<(std::ostream& out, const string16& str);
172 :
173 : #endif // WCHAR_T_IS_UTF32
174 :
175 : #endif // BASE_STRING16_H_
|