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