| 1 | // |
| 2 | // Copyright (c) 2015 Artyom Beilis (Tonkikh) |
| 3 | // Copyright (c) 2021-2023 Alexander Grund |
| 4 | // |
| 5 | // Distributed under the Boost Software License, Version 1.0. |
| 6 | // https://www.boost.org/LICENSE_1_0.txt |
| 7 | |
| 8 | #ifndef BOOST_LOCALE_GENERIC_CODECVT_HPP |
| 9 | #define BOOST_LOCALE_GENERIC_CODECVT_HPP |
| 10 | |
| 11 | #include <boost/locale/utf.hpp> |
| 12 | #include <cstdint> |
| 13 | #include <locale> |
| 14 | |
| 15 | namespace boost { namespace locale { |
| 16 | |
| 17 | static_assert(sizeof(std::mbstate_t) >= 2, "std::mbstate_t is to small to store an UTF-16 codepoint" ); |
| 18 | namespace detail { |
| 19 | // Avoid including cstring for std::memcpy |
| 20 | inline void copy_uint16_t(void* dst, const void* src) |
| 21 | { |
| 22 | unsigned char* cdst = static_cast<unsigned char*>(dst); |
| 23 | const unsigned char* csrc = static_cast<const unsigned char*>(src); |
| 24 | cdst[0] = csrc[0]; |
| 25 | cdst[1] = csrc[1]; |
| 26 | } |
| 27 | inline uint16_t read_state(const std::mbstate_t& src) |
| 28 | { |
| 29 | uint16_t dst; |
| 30 | copy_uint16_t(dst: &dst, src: &src); |
| 31 | return dst; |
| 32 | } |
| 33 | inline void write_state(std::mbstate_t& dst, const uint16_t src) |
| 34 | { |
| 35 | copy_uint16_t(dst: &dst, src: &src); |
| 36 | } |
| 37 | } // namespace detail |
| 38 | |
| 39 | /// \brief A base class that used to define constants for generic_codecvt |
| 40 | class generic_codecvt_base { |
| 41 | public: |
| 42 | /// Initial state for converting to or from Unicode code points, used by initial_state in derived classes |
| 43 | enum initial_convertion_state { |
| 44 | to_unicode_state, ///< The state would be used by to_unicode functions |
| 45 | from_unicode_state ///< The state would be used by from_unicode functions |
| 46 | }; |
| 47 | }; |
| 48 | |
| 49 | /// \brief Generic codecvt facet for various stateless encodings to UTF-16 and UTF-32 using wchar_t, char32_t |
| 50 | /// and char16_t |
| 51 | /// |
| 52 | /// Implementations should derive from this class defining itself as CodecvtImpl and provide following members |
| 53 | /// |
| 54 | /// - `state_type` - a type of special object that allows to store intermediate cached data, for example `iconv_t` |
| 55 | /// descriptor |
| 56 | /// - `state_type initial_state(generic_codecvt_base::initial_convertion_state direction) const` - member function |
| 57 | /// that creates initial state |
| 58 | /// - `int max_encoding_length() const` - a maximal length that one Unicode code point is represented, for UTF-8 for |
| 59 | /// example it is 4 from ISO-8859-1 it is 1 |
| 60 | /// - `utf::code_point to_unicode(state_type& state, const char*& begin, const char* end)` - extract first code |
| 61 | /// point from the text in range [begin,end), in case of success begin would point to the next character sequence to |
| 62 | /// be encoded to next code point, in case of incomplete sequence - utf::incomplete shell be returned, and in case |
| 63 | /// of invalid input sequence utf::illegal shell be returned and begin would remain unmodified |
| 64 | /// - `utf::len_or_error from_unicode(state_type &state, utf::code_point u, char* begin, const char* end)` - convert |
| 65 | /// a Unicode code point `u` into a character sequence at [begin,end). Return the length of the sequence in case of |
| 66 | /// success, utf::incomplete in case of not enough room to encode the code point, or utf::illegal in case conversion |
| 67 | /// can not be performed |
| 68 | /// |
| 69 | /// |
| 70 | /// For example implementation of codecvt for latin1/ISO-8859-1 character set |
| 71 | /// |
| 72 | /// \code |
| 73 | /// |
| 74 | /// template<typename CharType> |
| 75 | /// class latin1_codecvt: boost::locale::generic_codecvt<CharType,latin1_codecvt<CharType> > |
| 76 | /// { |
| 77 | /// public: |
| 78 | /// |
| 79 | /// /* Standard codecvt constructor */ |
| 80 | /// latin1_codecvt(size_t refs = 0): boost::locale::generic_codecvt<CharType,latin1_codecvt<CharType> >(refs) |
| 81 | /// { |
| 82 | /// } |
| 83 | /// |
| 84 | /// /* State is unused but required by generic_codecvt */ |
| 85 | /// struct state_type {}; |
| 86 | /// |
| 87 | /// state_type initial_state(generic_codecvt_base::initial_convertion_state /*unused*/) const |
| 88 | /// { |
| 89 | /// return state_type(); |
| 90 | /// } |
| 91 | /// |
| 92 | /// int max_encoding_length() const |
| 93 | /// { |
| 94 | /// return 1; |
| 95 | /// } |
| 96 | /// |
| 97 | /// boost::locale::utf::code_point to_unicode(state_type&, const char*& begin, const char* end) const |
| 98 | /// { |
| 99 | /// if(begin == end) |
| 100 | /// return boost::locale::utf::incomplete; |
| 101 | /// return *begin++; |
| 102 | /// } |
| 103 | /// |
| 104 | /// boost::locale::utf::len_or_error from_unicode(state_type&, boost::locale::utf::code_point u, |
| 105 | /// char* begin, const char* end) const |
| 106 | /// { |
| 107 | /// if(u >= 256) |
| 108 | /// return boost::locale::utf::illegal; |
| 109 | /// if(begin == end) |
| 110 | /// return boost::locale::utf::incomplete; |
| 111 | /// *begin = u; |
| 112 | /// return 1; |
| 113 | /// } |
| 114 | /// }; |
| 115 | /// |
| 116 | /// \endcode |
| 117 | /// |
| 118 | /// When external tools used for encoding conversion, the `state_type` is useful to save objects used for |
| 119 | /// conversions. For example, icu::UConverter can be saved in such a state for an efficient use: |
| 120 | /// |
| 121 | /// \code |
| 122 | /// template<typename CharType> |
| 123 | /// class icu_codecvt: boost::locale::generic_codecvt<CharType,icu_codecvt<CharType>> |
| 124 | /// { |
| 125 | /// public: |
| 126 | /// |
| 127 | /// /* Standard codecvt constructor */ |
| 128 | /// icu_codecvt(std::string const &name,refs = 0): |
| 129 | /// boost::locale::generic_codecvt<CharType,icu_codecvt<CharType>>(refs) |
| 130 | /// { ... } |
| 131 | /// |
| 132 | /// using state_type = std::unique_ptr<UConverter,void (*)(UConverter*)>; |
| 133 | /// |
| 134 | /// state_type initial_state(generic_codecvt_base::initial_convertion_state /*unused*/) const |
| 135 | /// { |
| 136 | /// UErrorCode err = U_ZERO_ERROR; |
| 137 | /// return state_type(ucnv_safeClone(converter_,0,0,&err),ucnv_close); |
| 138 | /// } |
| 139 | /// |
| 140 | /// boost::locale::utf::code_point to_unicode(state_type &ptr,char const *&begin,char const *end) const |
| 141 | /// { |
| 142 | /// UErrorCode err = U_ZERO_ERROR; |
| 143 | /// boost::locale::utf::code_point cp = ucnv_getNextUChar(ptr.get(),&begin,end,&err); |
| 144 | /// ... |
| 145 | /// } |
| 146 | /// ... |
| 147 | /// }; |
| 148 | /// \endcode |
| 149 | /// |
| 150 | template<typename CharType, typename CodecvtImpl, int CharSize = sizeof(CharType)> |
| 151 | class generic_codecvt; |
| 152 | |
| 153 | /// \brief UTF-16 to/from narrow char codecvt facet to use with char16_t or wchar_t on Windows |
| 154 | /// |
| 155 | /// Note in order to fit the requirements of usability by std::wfstream it uses mbstate_t |
| 156 | /// to handle intermediate states in handling of variable length UTF-16 sequences |
| 157 | /// |
| 158 | /// Its member functions implement standard virtual functions of basic codecvt |
| 159 | template<typename CharType, typename CodecvtImpl> |
| 160 | class generic_codecvt<CharType, CodecvtImpl, 2> : public std::codecvt<CharType, char, std::mbstate_t>, |
| 161 | public generic_codecvt_base { |
| 162 | public: |
| 163 | typedef CharType uchar; |
| 164 | |
| 165 | generic_codecvt(size_t refs = 0) : std::codecvt<CharType, char, std::mbstate_t>(refs) {} |
| 166 | const CodecvtImpl& implementation() const { return *static_cast<const CodecvtImpl*>(this); } |
| 167 | |
| 168 | protected: |
| 169 | std::codecvt_base::result do_unshift(std::mbstate_t& s, char* from, char* /*to*/, char*& next) const override |
| 170 | { |
| 171 | if(*reinterpret_cast<char*>(&s) != 0) |
| 172 | return std::codecvt_base::error; |
| 173 | next = from; |
| 174 | return std::codecvt_base::ok; |
| 175 | } |
| 176 | int do_encoding() const noexcept override { return 0; } |
| 177 | int do_max_length() const noexcept override { return implementation().max_encoding_length(); } |
| 178 | bool do_always_noconv() const noexcept override { return false; } |
| 179 | |
| 180 | int do_length(std::mbstate_t& std_state, const char* from, const char* from_end, size_t max) const override |
| 181 | { |
| 182 | bool state = *reinterpret_cast<char*>(&std_state) != 0; |
| 183 | const char* save_from = from; |
| 184 | |
| 185 | auto cvt_state = implementation().initial_state(to_unicode_state); |
| 186 | while(max > 0 && from < from_end) { |
| 187 | const char* prev_from = from; |
| 188 | const utf::code_point ch = implementation().to_unicode(cvt_state, from, from_end); |
| 189 | if(ch == boost::locale::utf::incomplete || ch == boost::locale::utf::illegal) { |
| 190 | from = prev_from; |
| 191 | break; |
| 192 | } |
| 193 | max--; |
| 194 | if(ch > 0xFFFF) { |
| 195 | if(!state) |
| 196 | from = prev_from; |
| 197 | state = !state; |
| 198 | } |
| 199 | } |
| 200 | *reinterpret_cast<char*>(&std_state) = state; |
| 201 | return static_cast<int>(from - save_from); |
| 202 | } |
| 203 | |
| 204 | std::codecvt_base::result do_in(std::mbstate_t& std_state, |
| 205 | const char* from, |
| 206 | const char* from_end, |
| 207 | const char*& from_next, |
| 208 | uchar* to, |
| 209 | uchar* to_end, |
| 210 | uchar*& to_next) const override |
| 211 | { |
| 212 | std::codecvt_base::result r = std::codecvt_base::ok; |
| 213 | |
| 214 | // mbstate_t is POD type and should be initialized to 0 (i.a. state = stateT()) |
| 215 | // according to standard. We use it to keep a flag 0/1 for surrogate pair writing |
| 216 | // |
| 217 | // if 0/false no codepoint above >0xFFFF observed, else a codepoint above 0xFFFF was observed |
| 218 | // and first pair is written, but no input consumed |
| 219 | bool state = *reinterpret_cast<char*>(&std_state) != 0; |
| 220 | auto cvt_state = implementation().initial_state(to_unicode_state); |
| 221 | while(to < to_end && from < from_end) { |
| 222 | const char* from_saved = from; |
| 223 | |
| 224 | utf::code_point ch = implementation().to_unicode(cvt_state, from, from_end); |
| 225 | |
| 226 | if(ch == boost::locale::utf::illegal) { |
| 227 | from = from_saved; |
| 228 | r = std::codecvt_base::error; |
| 229 | break; |
| 230 | } |
| 231 | if(ch == boost::locale::utf::incomplete) { |
| 232 | from = from_saved; |
| 233 | r = std::codecvt_base::partial; |
| 234 | break; |
| 235 | } |
| 236 | // Normal codepoints go directly to stream |
| 237 | if(ch <= 0xFFFF) |
| 238 | *to++ = static_cast<uchar>(ch); |
| 239 | else { |
| 240 | // For other codepoints we do the following |
| 241 | // |
| 242 | // 1. We can't consume our input as we may find ourselves |
| 243 | // in state where all input consumed but not all output written,i.e. only |
| 244 | // 1st pair is written |
| 245 | // 2. We only write first pair and mark this in the state, we also revert back |
| 246 | // the from pointer in order to make sure this codepoint would be read |
| 247 | // once again and then we would consume our input together with writing |
| 248 | // second surrogate pair |
| 249 | ch -= 0x10000; |
| 250 | std::uint16_t w1 = static_cast<std::uint16_t>(0xD800 | (ch >> 10)); |
| 251 | std::uint16_t w2 = static_cast<std::uint16_t>(0xDC00 | (ch & 0x3FF)); |
| 252 | if(!state) { |
| 253 | from = from_saved; |
| 254 | *to++ = w1; |
| 255 | } else |
| 256 | *to++ = w2; |
| 257 | state = !state; |
| 258 | } |
| 259 | } |
| 260 | from_next = from; |
| 261 | to_next = to; |
| 262 | if(r == std::codecvt_base::ok && (from != from_end || state)) |
| 263 | r = std::codecvt_base::partial; |
| 264 | *reinterpret_cast<char*>(&std_state) = state; |
| 265 | return r; |
| 266 | } |
| 267 | |
| 268 | std::codecvt_base::result do_out(std::mbstate_t& std_state, |
| 269 | const uchar* from, |
| 270 | const uchar* from_end, |
| 271 | const uchar*& from_next, |
| 272 | char* to, |
| 273 | char* to_end, |
| 274 | char*& to_next) const override |
| 275 | { |
| 276 | std::codecvt_base::result r = std::codecvt_base::ok; |
| 277 | // mbstate_t is POD type and should be initialized to 0 (i.a. state = stateT()) |
| 278 | // according to standard. We assume that sizeof(mbstate_t) >=2 in order |
| 279 | // to be able to store first observed surrogate pair |
| 280 | // |
| 281 | // State: state!=0 - a first surrogate pair was observed (state = first pair), |
| 282 | // we expect the second one to come and then zero the state |
| 283 | std::uint16_t state = detail::read_state(src: std_state); |
| 284 | auto cvt_state = implementation().initial_state(from_unicode_state); |
| 285 | while(to < to_end && from < from_end) { |
| 286 | utf::code_point ch = 0; |
| 287 | if(state != 0) { |
| 288 | // if the state indicates that 1st surrogate pair was written |
| 289 | // we should make sure that the second one that comes is actually |
| 290 | // second surrogate |
| 291 | std::uint16_t w1 = state; |
| 292 | std::uint16_t w2 = *from; |
| 293 | // we don't forward from as writing may fail to incomplete or |
| 294 | // partial conversion |
| 295 | if(0xDC00 <= w2 && w2 <= 0xDFFF) { |
| 296 | std::uint16_t vh = w1 - 0xD800; |
| 297 | std::uint16_t vl = w2 - 0xDC00; |
| 298 | ch = ((uint32_t(vh) << 10) | vl) + 0x10000; |
| 299 | } else { |
| 300 | // Invalid surrogate |
| 301 | r = std::codecvt_base::error; |
| 302 | break; |
| 303 | } |
| 304 | } else { |
| 305 | ch = *from; |
| 306 | if(0xD800 <= ch && ch <= 0xDBFF) { |
| 307 | // if this is a first surrogate pair we put |
| 308 | // it into the state and consume it, note we don't |
| 309 | // go forward as it should be illegal so we increase |
| 310 | // the from pointer manually |
| 311 | state = static_cast<uint16_t>(ch); |
| 312 | from++; |
| 313 | continue; |
| 314 | } else if(0xDC00 <= ch && ch <= 0xDFFF) { |
| 315 | // if we observe second surrogate pair and |
| 316 | // first only may be expected we should break from the loop with error |
| 317 | // as it is illegal input |
| 318 | r = std::codecvt_base::error; |
| 319 | break; |
| 320 | } |
| 321 | } |
| 322 | if(!boost::locale::utf::is_valid_codepoint(v: ch)) { |
| 323 | r = std::codecvt_base::error; |
| 324 | break; |
| 325 | } |
| 326 | const utf::code_point len = implementation().from_unicode(cvt_state, ch, to, to_end); |
| 327 | if(len == boost::locale::utf::incomplete) { |
| 328 | r = std::codecvt_base::partial; |
| 329 | break; |
| 330 | } else if(len == boost::locale::utf::illegal) { |
| 331 | r = std::codecvt_base::error; |
| 332 | break; |
| 333 | } else |
| 334 | to += len; |
| 335 | state = 0; |
| 336 | from++; |
| 337 | } |
| 338 | from_next = from; |
| 339 | to_next = to; |
| 340 | if(r == std::codecvt_base::ok && (from != from_end || state != 0)) |
| 341 | r = std::codecvt_base::partial; |
| 342 | detail::write_state(dst&: std_state, src: state); |
| 343 | return r; |
| 344 | } |
| 345 | }; |
| 346 | |
| 347 | /// \brief UTF-32 to/from narrow char codecvt facet to use with char32_t or wchar_t on POSIX platforms |
| 348 | /// |
| 349 | /// Its member functions implement standard virtual functions of basic codecvt. |
| 350 | /// mbstate_t is not used for UTF-32 handling due to fixed length encoding |
| 351 | template<typename CharType, typename CodecvtImpl> |
| 352 | class generic_codecvt<CharType, CodecvtImpl, 4> : public std::codecvt<CharType, char, std::mbstate_t>, |
| 353 | public generic_codecvt_base { |
| 354 | public: |
| 355 | typedef CharType uchar; |
| 356 | |
| 357 | generic_codecvt(size_t refs = 0) : std::codecvt<CharType, char, std::mbstate_t>(refs) {} |
| 358 | |
| 359 | const CodecvtImpl& implementation() const { return *static_cast<const CodecvtImpl*>(this); } |
| 360 | |
| 361 | protected: |
| 362 | std::codecvt_base::result |
| 363 | do_unshift(std::mbstate_t& /*s*/, char* from, char* /*to*/, char*& next) const override |
| 364 | { |
| 365 | next = from; |
| 366 | return std::codecvt_base::ok; |
| 367 | } |
| 368 | int do_encoding() const noexcept override { return 0; } |
| 369 | int do_max_length() const noexcept override { return implementation().max_encoding_length(); } |
| 370 | bool do_always_noconv() const noexcept override { return false; } |
| 371 | |
| 372 | int do_length(std::mbstate_t& /*state*/, const char* from, const char* from_end, size_t max) const override |
| 373 | { |
| 374 | const char* start_from = from; |
| 375 | auto cvt_state = implementation().initial_state(to_unicode_state); |
| 376 | while(max > 0 && from < from_end) { |
| 377 | const char* save_from = from; |
| 378 | const utf::code_point ch = implementation().to_unicode(cvt_state, from, from_end); |
| 379 | if(ch == boost::locale::utf::incomplete || ch == boost::locale::utf::illegal) { |
| 380 | from = save_from; |
| 381 | break; |
| 382 | } |
| 383 | max--; |
| 384 | } |
| 385 | |
| 386 | return static_cast<int>(from - start_from); |
| 387 | } |
| 388 | |
| 389 | std::codecvt_base::result do_in(std::mbstate_t& /*state*/, |
| 390 | const char* from, |
| 391 | const char* from_end, |
| 392 | const char*& from_next, |
| 393 | uchar* to, |
| 394 | uchar* to_end, |
| 395 | uchar*& to_next) const override |
| 396 | { |
| 397 | std::codecvt_base::result r = std::codecvt_base::ok; |
| 398 | |
| 399 | auto cvt_state = implementation().initial_state(to_unicode_state); |
| 400 | while(to < to_end && from < from_end) { |
| 401 | const char* from_saved = from; |
| 402 | |
| 403 | const utf::code_point ch = implementation().to_unicode(cvt_state, from, from_end); |
| 404 | |
| 405 | if(ch == boost::locale::utf::illegal) { |
| 406 | r = std::codecvt_base::error; |
| 407 | from = from_saved; |
| 408 | break; |
| 409 | } |
| 410 | if(ch == boost::locale::utf::incomplete) { |
| 411 | r = std::codecvt_base::partial; |
| 412 | from = from_saved; |
| 413 | break; |
| 414 | } |
| 415 | *to++ = ch; |
| 416 | } |
| 417 | from_next = from; |
| 418 | to_next = to; |
| 419 | if(r == std::codecvt_base::ok && from != from_end) |
| 420 | r = std::codecvt_base::partial; |
| 421 | return r; |
| 422 | } |
| 423 | |
| 424 | std::codecvt_base::result do_out(std::mbstate_t& /*std_state*/, |
| 425 | const uchar* from, |
| 426 | const uchar* from_end, |
| 427 | const uchar*& from_next, |
| 428 | char* to, |
| 429 | char* to_end, |
| 430 | char*& to_next) const override |
| 431 | { |
| 432 | std::codecvt_base::result r = std::codecvt_base::ok; |
| 433 | auto cvt_state = implementation().initial_state(from_unicode_state); |
| 434 | while(to < to_end && from < from_end) { |
| 435 | const std::uint32_t ch = *from; |
| 436 | if(!boost::locale::utf::is_valid_codepoint(v: ch)) { |
| 437 | r = std::codecvt_base::error; |
| 438 | break; |
| 439 | } |
| 440 | const utf::code_point len = implementation().from_unicode(cvt_state, ch, to, to_end); |
| 441 | if(len == boost::locale::utf::incomplete) { |
| 442 | r = std::codecvt_base::partial; |
| 443 | break; |
| 444 | } else if(len == boost::locale::utf::illegal) { |
| 445 | r = std::codecvt_base::error; |
| 446 | break; |
| 447 | } |
| 448 | to += len; |
| 449 | from++; |
| 450 | } |
| 451 | from_next = from; |
| 452 | to_next = to; |
| 453 | if(r == std::codecvt_base::ok && from != from_end) |
| 454 | r = std::codecvt_base::partial; |
| 455 | return r; |
| 456 | } |
| 457 | }; |
| 458 | |
| 459 | template<typename CodecvtImpl> |
| 460 | class generic_codecvt<char, CodecvtImpl, 1> : public std::codecvt<char, char, std::mbstate_t>, |
| 461 | public generic_codecvt_base { |
| 462 | public: |
| 463 | typedef char uchar; |
| 464 | |
| 465 | const CodecvtImpl& implementation() const { return *static_cast<const CodecvtImpl*>(this); } |
| 466 | |
| 467 | generic_codecvt(size_t refs = 0) : std::codecvt<char, char, std::mbstate_t>(refs) {} |
| 468 | }; |
| 469 | |
| 470 | }} // namespace boost::locale |
| 471 | |
| 472 | #endif |
| 473 | |