1// Copyright 2013 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_STRINGS_STRING16_H_
6#define BASE_STRINGS_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 <stddef.h>
30#include <stdint.h>
31#include <stdio.h>
32
33#include <functional>
34#include <string>
35
36#include "base/base_export.h"
37#include "build/build_config.h"
38
39#if defined(WCHAR_T_IS_UTF16)
40
41// Define a macro for wrapping construction of char16 arrays and string16s from
42// a literal string. This indirection allows for an easier migration of
43// base::char16 to char16_t on platforms where WCHAR_T_IS_UTF16, as only a one
44// character change to the macro will be necessary.
45// This macro does not exist when WCHAR_T_IS_UTF32, as it is currently not
46// possible to create a char array form a literal in this case.
47// TODO(https://crbug.com/911896): Remove this macro once base::char16 is
48// char16_t on all platforms.
49#define STRING16_LITERAL(x) L##x
50
51namespace base {
52
53typedef wchar_t char16;
54typedef std::wstring string16;
55
56} // namespace base
57
58#elif defined(WCHAR_T_IS_UTF32)
59
60#include <wchar.h> // for mbstate_t
61
62namespace base {
63
64typedef uint16_t char16;
65
66// char16 versions of the functions required by string16_char_traits; these
67// are based on the wide character functions of similar names ("w" or "wcs"
68// instead of "c16").
69BASE_EXPORT int c16memcmp(const char16* s1, const char16* s2, size_t n);
70BASE_EXPORT size_t c16len(const char16* s);
71BASE_EXPORT const char16* c16memchr(const char16* s, char16 c, size_t n);
72BASE_EXPORT char16* c16memmove(char16* s1, const char16* s2, size_t n);
73BASE_EXPORT char16* c16memcpy(char16* s1, const char16* s2, size_t n);
74BASE_EXPORT char16* c16memset(char16* s, char16 c, size_t n);
75
76// This namespace contains the implementation of base::string16 along with
77// things that need to be found via argument-dependent lookup from a
78// base::string16.
79namespace string16_internals {
80
81struct string16_char_traits {
82 typedef char16 char_type;
83 typedef int int_type;
84
85 // int_type needs to be able to hold each possible value of char_type, and in
86 // addition, the distinct value of eof().
87 static_assert(sizeof(int_type) > sizeof(char_type),
88 "int must be larger than 16 bits wide");
89
90 typedef std::streamoff off_type;
91 typedef mbstate_t state_type;
92 typedef std::fpos<state_type> pos_type;
93
94 static void assign(char_type& c1, const char_type& c2) {
95 c1 = c2;
96 }
97
98 static bool eq(const char_type& c1, const char_type& c2) {
99 return c1 == c2;
100 }
101 static bool lt(const char_type& c1, const char_type& c2) {
102 return c1 < c2;
103 }
104
105 static int compare(const char_type* s1, const char_type* s2, size_t n) {
106 return c16memcmp(s1, s2, n);
107 }
108
109 static size_t length(const char_type* s) {
110 return c16len(s);
111 }
112
113 static const char_type* find(const char_type* s, size_t n,
114 const char_type& a) {
115 return c16memchr(s, a, n);
116 }
117
118 static char_type* move(char_type* s1, const char_type* s2, size_t n) {
119 return c16memmove(s1, s2, n);
120 }
121
122 static char_type* copy(char_type* s1, const char_type* s2, size_t n) {
123 return c16memcpy(s1, s2, n);
124 }
125
126 static char_type* assign(char_type* s, size_t n, char_type a) {
127 return c16memset(s, a, n);
128 }
129
130 static int_type not_eof(const int_type& c) {
131 return eq_int_type(c, eof()) ? 0 : c;
132 }
133
134 static char_type to_char_type(const int_type& c) {
135 return char_type(c);
136 }
137
138 static int_type to_int_type(const char_type& c) {
139 return int_type(c);
140 }
141
142 static bool eq_int_type(const int_type& c1, const int_type& c2) {
143 return c1 == c2;
144 }
145
146 static int_type eof() {
147 return static_cast<int_type>(EOF);
148 }
149};
150
151} // namespace string16_internals
152
153typedef std::basic_string<char16,
154 base::string16_internals::string16_char_traits>
155 string16;
156
157namespace string16_internals {
158
159BASE_EXPORT extern std::ostream& operator<<(std::ostream& out,
160 const string16& str);
161
162// This is required by googletest to print a readable output on test failures.
163BASE_EXPORT extern void PrintTo(const string16& str, std::ostream* out);
164
165} // namespace string16_internals
166
167} // namespace base
168
169// The string class will be explicitly instantiated only once, in string16.cc.
170//
171// std::basic_string<> in GNU libstdc++ contains a static data member,
172// _S_empty_rep_storage, to represent empty strings. When an operation such
173// as assignment or destruction is performed on a string, causing its existing
174// data member to be invalidated, it must not be freed if this static data
175// member is being used. Otherwise, it counts as an attempt to free static
176// (and not allocated) data, which is a memory error.
177//
178// Generally, due to C++ template magic, _S_empty_rep_storage will be marked
179// as a coalesced symbol, meaning that the linker will combine multiple
180// instances into a single one when generating output.
181//
182// If a string class is used by multiple shared libraries, a problem occurs.
183// Each library will get its own copy of _S_empty_rep_storage. When strings
184// are passed across a library boundary for alteration or destruction, memory
185// errors will result. GNU libstdc++ contains a configuration option,
186// --enable-fully-dynamic-string (_GLIBCXX_FULLY_DYNAMIC_STRING), which
187// disables the static data member optimization, but it's a good optimization
188// and non-STL code is generally at the mercy of the system's STL
189// configuration. Fully-dynamic strings are not the default for GNU libstdc++
190// libstdc++ itself or for the libstdc++ installations on the systems we care
191// about, such as Mac OS X and relevant flavors of Linux.
192//
193// See also http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24196 .
194//
195// To avoid problems, string classes need to be explicitly instantiated only
196// once, in exactly one library. All other string users see it via an "extern"
197// declaration. This is precisely how GNU libstdc++ handles
198// std::basic_string<char> (string) and std::basic_string<wchar_t> (wstring).
199//
200// This also works around a Mac OS X linker bug in ld64-85.2.1 (Xcode 3.1.2),
201// in which the linker does not fully coalesce symbols when dead code
202// stripping is enabled. This bug causes the memory errors described above
203// to occur even when a std::basic_string<> does not cross shared library
204// boundaries, such as in statically-linked executables.
205//
206// TODO(mark): File this bug with Apple and update this note with a bug number.
207
208extern template class BASE_EXPORT
209 std::basic_string<base::char16,
210 base::string16_internals::string16_char_traits>;
211
212// Specialize std::hash for base::string16. Although the style guide forbids
213// this in general, it is necessary for consistency with WCHAR_T_IS_UTF16
214// platforms, where base::string16 is a type alias for std::wstring.
215namespace std {
216template <>
217struct hash<base::string16> {
218 std::size_t operator()(const base::string16& s) const {
219 std::size_t result = 0;
220 for (base::char16 c : s)
221 result = (result * 131) + c;
222 return result;
223 }
224};
225} // namespace std
226
227#endif // WCHAR_T_IS_UTF32
228
229#endif // BASE_STRINGS_STRING16_H_
230