| 1 | /* utf8.c |
| 2 | * |
| 3 | * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
| 4 | * by Larry Wall and others |
| 5 | * |
| 6 | * You may distribute under the terms of either the GNU General Public |
| 7 | * License or the Artistic License, as specified in the README file. |
| 8 | * |
| 9 | */ |
| 10 | |
| 11 | /* |
| 12 | * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever |
| 13 | * heard of that we don't want to see any closer; and that's the one place |
| 14 | * we're trying to get to! And that's just where we can't get, nohow.' |
| 15 | * |
| 16 | * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"] |
| 17 | * |
| 18 | * 'Well do I understand your speech,' he answered in the same language; |
| 19 | * 'yet few strangers do so. Why then do you not speak in the Common Tongue, |
| 20 | * as is the custom in the West, if you wish to be answered?' |
| 21 | * --Gandalf, addressing Théoden's door wardens |
| 22 | * |
| 23 | * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] |
| 24 | * |
| 25 | * ...the travellers perceived that the floor was paved with stones of many |
| 26 | * hues; branching runes and strange devices intertwined beneath their feet. |
| 27 | * |
| 28 | * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] |
| 29 | */ |
| 30 | |
| 31 | #include "EXTERN.h" |
| 32 | #define PERL_IN_UTF8_C |
| 33 | #include "perl.h" |
| 34 | #include "invlist_inline.h" |
| 35 | |
| 36 | static const char malformed_text[] = "Malformed UTF-8 character"; |
| 37 | static const char unees[] = |
| 38 | "Malformed UTF-8 character (unexpected end of string)"; |
| 39 | |
| 40 | /* |
| 41 | These are various utility functions for manipulating UTF8-encoded |
| 42 | strings. For the uninitiated, this is a method of representing arbitrary |
| 43 | Unicode characters as a variable number of bytes, in such a way that |
| 44 | characters in the ASCII range are unmodified, and a zero byte never appears |
| 45 | within non-zero characters. |
| 46 | */ |
| 47 | |
| 48 | /* helper for Perl__force_out_malformed_utf8_message(). Like |
| 49 | * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than |
| 50 | * PL_compiling */ |
| 51 | |
| 52 | static void |
| 53 | S_restore_cop_warnings(pTHX_ void *p) |
| 54 | { |
| 55 | free_and_set_cop_warnings(PL_curcop, (STRLEN*) p); |
| 56 | } |
| 57 | |
| 58 | |
| 59 | void |
| 60 | Perl__force_out_malformed_utf8_message(pTHX_ |
| 61 | const U8 *const p, /* First byte in UTF-8 sequence */ |
| 62 | const U8 * const e, /* Final byte in sequence (may include |
| 63 | multiple chars */ |
| 64 | const U32 flags, /* Flags to pass to utf8n_to_uvchr(), |
| 65 | usually 0, or some DISALLOW flags */ |
| 66 | const bool die_here) /* If TRUE, this function does not return */ |
| 67 | { |
| 68 | /* This core-only function is to be called when a malformed UTF-8 character |
| 69 | * is found, in order to output the detailed information about the |
| 70 | * malformation before dieing. The reason it exists is for the occasions |
| 71 | * when such a malformation is fatal, but warnings might be turned off, so |
| 72 | * that normally they would not be actually output. This ensures that they |
| 73 | * do get output. Because a sequence may be malformed in more than one |
| 74 | * way, multiple messages may be generated, so we can't make them fatal, as |
| 75 | * that would cause the first one to die. |
| 76 | * |
| 77 | * Instead we pretend -W was passed to perl, then die afterwards. The |
| 78 | * flexibility is here to return to the caller so they can finish up and |
| 79 | * die themselves */ |
| 80 | U32 errors; |
| 81 | |
| 82 | PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE; |
| 83 | |
| 84 | ENTER; |
| 85 | SAVEI8(PL_dowarn); |
| 86 | SAVESPTR(PL_curcop); |
| 87 | |
| 88 | PL_dowarn = G_WARN_ALL_ON|G_WARN_ON; |
| 89 | if (PL_curcop) { |
| 90 | /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather |
| 91 | * than PL_compiling */ |
| 92 | SAVEDESTRUCTOR_X(S_restore_cop_warnings, |
| 93 | (void*)PL_curcop->cop_warnings); |
| 94 | PL_curcop->cop_warnings = pWARN_ALL; |
| 95 | } |
| 96 | |
| 97 | (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors); |
| 98 | |
| 99 | LEAVE; |
| 100 | |
| 101 | if (! errors) { |
| 102 | Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should" |
| 103 | " be called only when there are errors found"); |
| 104 | } |
| 105 | |
| 106 | if (die_here) { |
| 107 | Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | STATIC HV * |
| 112 | S_new_msg_hv(pTHX_ const char * const message, /* The message text */ |
| 113 | U32 categories, /* Packed warning categories */ |
| 114 | U32 flag) /* Flag associated with this message */ |
| 115 | { |
| 116 | /* Creates, populates, and returns an HV* that describes an error message |
| 117 | * for the translators between UTF8 and code point */ |
| 118 | |
| 119 | SV* msg_sv = newSVpv(message, 0); |
| 120 | SV* category_sv = newSVuv(categories); |
| 121 | SV* flag_bit_sv = newSVuv(flag); |
| 122 | |
| 123 | HV* msg_hv = newHV(); |
| 124 | |
| 125 | PERL_ARGS_ASSERT_NEW_MSG_HV; |
| 126 | |
| 127 | (void) hv_stores(msg_hv, "text", msg_sv); |
| 128 | (void) hv_stores(msg_hv, "warn_categories", category_sv); |
| 129 | (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv); |
| 130 | |
| 131 | return msg_hv; |
| 132 | } |
| 133 | |
| 134 | /* |
| 135 | =for apidoc uvoffuni_to_utf8_flags |
| 136 | |
| 137 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 138 | Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or |
| 139 | L<perlapi/uvchr_to_utf8_flags>>. |
| 140 | |
| 141 | This function is like them, but the input is a strict Unicode |
| 142 | (as opposed to native) code point. Only in very rare circumstances should code |
| 143 | not be using the native code point. |
| 144 | |
| 145 | For details, see the description for L<perlapi/uvchr_to_utf8_flags>. |
| 146 | |
| 147 | =cut |
| 148 | */ |
| 149 | |
| 150 | U8 * |
| 151 | Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags) |
| 152 | { |
| 153 | PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS; |
| 154 | |
| 155 | return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL); |
| 156 | } |
| 157 | |
| 158 | /* All these formats take a single UV code point argument */ |
| 159 | const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf; |
| 160 | const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf |
| 161 | " is not recommended for open interchange"; |
| 162 | const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode," |
| 163 | " may not be portable"; |
| 164 | |
| 165 | #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \ |
| 166 | STMT_START { \ |
| 167 | if (flags & UNICODE_WARN_SURROGATE) { \ |
| 168 | U32 category = packWARN(WARN_SURROGATE); \ |
| 169 | const char * format = surrogate_cp_format; \ |
| 170 | if (msgs) { \ |
| 171 | *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ |
| 172 | category, \ |
| 173 | UNICODE_GOT_SURROGATE); \ |
| 174 | } \ |
| 175 | else { \ |
| 176 | Perl_ck_warner_d(aTHX_ category, format, uv); \ |
| 177 | } \ |
| 178 | } \ |
| 179 | if (flags & UNICODE_DISALLOW_SURROGATE) { \ |
| 180 | return NULL; \ |
| 181 | } \ |
| 182 | } STMT_END; |
| 183 | |
| 184 | #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \ |
| 185 | STMT_START { \ |
| 186 | if (flags & UNICODE_WARN_NONCHAR) { \ |
| 187 | U32 category = packWARN(WARN_NONCHAR); \ |
| 188 | const char * format = nonchar_cp_format; \ |
| 189 | if (msgs) { \ |
| 190 | *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ |
| 191 | category, \ |
| 192 | UNICODE_GOT_NONCHAR); \ |
| 193 | } \ |
| 194 | else { \ |
| 195 | Perl_ck_warner_d(aTHX_ category, format, uv); \ |
| 196 | } \ |
| 197 | } \ |
| 198 | if (flags & UNICODE_DISALLOW_NONCHAR) { \ |
| 199 | return NULL; \ |
| 200 | } \ |
| 201 | } STMT_END; |
| 202 | |
| 203 | /* Use shorter names internally in this file */ |
| 204 | #define SHIFT UTF_ACCUMULATION_SHIFT |
| 205 | #undef MARK |
| 206 | #define MARK UTF_CONTINUATION_MARK |
| 207 | #define MASK UTF_CONTINUATION_MASK |
| 208 | |
| 209 | /* |
| 210 | =for apidoc uvchr_to_utf8_flags_msgs |
| 211 | |
| 212 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 213 | |
| 214 | Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly. |
| 215 | |
| 216 | This function is for code that wants any warning and/or error messages to be |
| 217 | returned to the caller rather than be displayed. All messages that would have |
| 218 | been displayed if all lexical warnings are enabled will be returned. |
| 219 | |
| 220 | It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter |
| 221 | placed after all the others, C<msgs>. If this parameter is 0, this function |
| 222 | behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should |
| 223 | be a pointer to an C<HV *> variable, in which this function creates a new HV to |
| 224 | contain any appropriate messages. The hash has three key-value pairs, as |
| 225 | follows: |
| 226 | |
| 227 | =over 4 |
| 228 | |
| 229 | =item C<text> |
| 230 | |
| 231 | The text of the message as a C<SVpv>. |
| 232 | |
| 233 | =item C<warn_categories> |
| 234 | |
| 235 | The warning category (or categories) packed into a C<SVuv>. |
| 236 | |
| 237 | =item C<flag> |
| 238 | |
| 239 | A single flag bit associated with this message, in a C<SVuv>. |
| 240 | The bit corresponds to some bit in the C<*errors> return value, |
| 241 | such as C<UNICODE_GOT_SURROGATE>. |
| 242 | |
| 243 | =back |
| 244 | |
| 245 | It's important to note that specifying this parameter as non-null will cause |
| 246 | any warnings this function would otherwise generate to be suppressed, and |
| 247 | instead be placed in C<*msgs>. The caller can check the lexical warnings state |
| 248 | (or not) when choosing what to do with the returned messages. |
| 249 | |
| 250 | The caller, of course, is responsible for freeing any returned HV. |
| 251 | |
| 252 | =cut |
| 253 | */ |
| 254 | |
| 255 | /* Undocumented; we don't want people using this. Instead they should use |
| 256 | * uvchr_to_utf8_flags_msgs() */ |
| 257 | U8 * |
| 258 | Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs) |
| 259 | { |
| 260 | PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS; |
| 261 | |
| 262 | if (msgs) { |
| 263 | *msgs = NULL; |
| 264 | } |
| 265 | |
| 266 | if (OFFUNI_IS_INVARIANT(uv)) { |
| 267 | *d++ = LATIN1_TO_NATIVE(uv); |
| 268 | return d; |
| 269 | } |
| 270 | |
| 271 | if (uv <= MAX_UTF8_TWO_BYTE) { |
| 272 | *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2)); |
| 273 | *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK); |
| 274 | return d; |
| 275 | } |
| 276 | |
| 277 | /* Not 2-byte; test for and handle 3-byte result. In the test immediately |
| 278 | * below, the 16 is for start bytes E0-EF (which are all the possible ones |
| 279 | * for 3 byte characters). The 2 is for 2 continuation bytes; these each |
| 280 | * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000 |
| 281 | * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC; |
| 282 | * 0x800-0xFFFF on ASCII */ |
| 283 | if (uv < (16 * (1U << (2 * SHIFT)))) { |
| 284 | *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3)); |
| 285 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); |
| 286 | *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); |
| 287 | |
| 288 | #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so |
| 289 | aren't tested here */ |
| 290 | /* The most likely code points in this range are below the surrogates. |
| 291 | * Do an extra test to quickly exclude those. */ |
| 292 | if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) { |
| 293 | if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) |
| 294 | || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) |
| 295 | { |
| 296 | HANDLE_UNICODE_NONCHAR(uv, flags, msgs); |
| 297 | } |
| 298 | else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { |
| 299 | HANDLE_UNICODE_SURROGATE(uv, flags, msgs); |
| 300 | } |
| 301 | } |
| 302 | #endif |
| 303 | return d; |
| 304 | } |
| 305 | |
| 306 | /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII |
| 307 | * platforms, and 0x4000 on EBCDIC. There are problematic cases that can |
| 308 | * happen starting with 4-byte characters on ASCII platforms. We unify the |
| 309 | * code for these with EBCDIC, even though some of them require 5-bytes on |
| 310 | * those, because khw believes the code saving is worth the very slight |
| 311 | * performance hit on these high EBCDIC code points. */ |
| 312 | |
| 313 | if (UNLIKELY(UNICODE_IS_SUPER(uv))) { |
| 314 | if (UNLIKELY( uv > MAX_LEGAL_CP |
| 315 | && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX))) |
| 316 | { |
| 317 | Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv)); |
| 318 | } |
| 319 | if ( (flags & UNICODE_WARN_SUPER) |
| 320 | || ( (flags & UNICODE_WARN_PERL_EXTENDED) |
| 321 | && UNICODE_IS_PERL_EXTENDED(uv))) |
| 322 | { |
| 323 | const char * format = super_cp_format; |
| 324 | U32 category = packWARN(WARN_NON_UNICODE); |
| 325 | U32 flag = UNICODE_GOT_SUPER; |
| 326 | |
| 327 | /* Choose the more dire applicable warning */ |
| 328 | if (UNICODE_IS_PERL_EXTENDED(uv)) { |
| 329 | format = PL_extended_cp_format; |
| 330 | category = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE); |
| 331 | if (flags & (UNICODE_WARN_PERL_EXTENDED |
| 332 | |UNICODE_DISALLOW_PERL_EXTENDED)) |
| 333 | { |
| 334 | flag = UNICODE_GOT_PERL_EXTENDED; |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | if (msgs) { |
| 339 | *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), |
| 340 | category, flag); |
| 341 | } |
| 342 | else if ( ckWARN_d(WARN_NON_UNICODE) |
| 343 | || ( (flag & UNICODE_GOT_PERL_EXTENDED) |
| 344 | && ckWARN(WARN_PORTABLE))) |
| 345 | { |
| 346 | Perl_warner(aTHX_ category, format, uv); |
| 347 | } |
| 348 | } |
| 349 | if ( (flags & UNICODE_DISALLOW_SUPER) |
| 350 | || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED) |
| 351 | && UNICODE_IS_PERL_EXTENDED(uv))) |
| 352 | { |
| 353 | return NULL; |
| 354 | } |
| 355 | } |
| 356 | else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) { |
| 357 | HANDLE_UNICODE_NONCHAR(uv, flags, msgs); |
| 358 | } |
| 359 | |
| 360 | /* Test for and handle 4-byte result. In the test immediately below, the |
| 361 | * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte |
| 362 | * characters). The 3 is for 3 continuation bytes; these each contribute |
| 363 | * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on |
| 364 | * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC; |
| 365 | * 0x1_0000-0x1F_FFFF on ASCII */ |
| 366 | if (uv < (8 * (1U << (3 * SHIFT)))) { |
| 367 | *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4)); |
| 368 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK); |
| 369 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); |
| 370 | *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); |
| 371 | |
| 372 | #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte |
| 373 | characters. The end-plane non-characters for EBCDIC were |
| 374 | handled just above */ |
| 375 | if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) { |
| 376 | HANDLE_UNICODE_NONCHAR(uv, flags, msgs); |
| 377 | } |
| 378 | else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { |
| 379 | HANDLE_UNICODE_SURROGATE(uv, flags, msgs); |
| 380 | } |
| 381 | #endif |
| 382 | |
| 383 | return d; |
| 384 | } |
| 385 | |
| 386 | /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII |
| 387 | * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop |
| 388 | * format. The unrolled version above turns out to not save all that much |
| 389 | * time, and at these high code points (well above the legal Unicode range |
| 390 | * on ASCII platforms, and well above anything in common use in EBCDIC), |
| 391 | * khw believes that less code outweighs slight performance gains. */ |
| 392 | |
| 393 | { |
| 394 | STRLEN len = OFFUNISKIP(uv); |
| 395 | U8 *p = d+len-1; |
| 396 | while (p > d) { |
| 397 | *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK); |
| 398 | uv >>= SHIFT; |
| 399 | } |
| 400 | *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len)); |
| 401 | return d+len; |
| 402 | } |
| 403 | } |
| 404 | |
| 405 | /* |
| 406 | =for apidoc uvchr_to_utf8 |
| 407 | |
| 408 | Adds the UTF-8 representation of the native code point C<uv> to the end |
| 409 | of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to |
| 410 | C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to |
| 411 | the byte after the end of the new character. In other words, |
| 412 | |
| 413 | d = uvchr_to_utf8(d, uv); |
| 414 | |
| 415 | is the recommended wide native character-aware way of saying |
| 416 | |
| 417 | *(d++) = uv; |
| 418 | |
| 419 | This function accepts any code point from 0..C<IV_MAX> as input. |
| 420 | C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. |
| 421 | |
| 422 | It is possible to forbid or warn on non-Unicode code points, or those that may |
| 423 | be problematic by using L</uvchr_to_utf8_flags>. |
| 424 | |
| 425 | =cut |
| 426 | */ |
| 427 | |
| 428 | /* This is also a macro */ |
| 429 | PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv); |
| 430 | |
| 431 | U8 * |
| 432 | Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv) |
| 433 | { |
| 434 | return uvchr_to_utf8(d, uv); |
| 435 | } |
| 436 | |
| 437 | /* |
| 438 | =for apidoc uvchr_to_utf8_flags |
| 439 | |
| 440 | Adds the UTF-8 representation of the native code point C<uv> to the end |
| 441 | of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to |
| 442 | C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to |
| 443 | the byte after the end of the new character. In other words, |
| 444 | |
| 445 | d = uvchr_to_utf8_flags(d, uv, flags); |
| 446 | |
| 447 | or, in most cases, |
| 448 | |
| 449 | d = uvchr_to_utf8_flags(d, uv, 0); |
| 450 | |
| 451 | This is the Unicode-aware way of saying |
| 452 | |
| 453 | *(d++) = uv; |
| 454 | |
| 455 | If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as |
| 456 | input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. |
| 457 | |
| 458 | Specifying C<flags> can further restrict what is allowed and not warned on, as |
| 459 | follows: |
| 460 | |
| 461 | If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set, |
| 462 | the function will raise a warning, provided UTF8 warnings are enabled. If |
| 463 | instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return |
| 464 | NULL. If both flags are set, the function will both warn and return NULL. |
| 465 | |
| 466 | Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags |
| 467 | affect how the function handles a Unicode non-character. |
| 468 | |
| 469 | And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags |
| 470 | affect the handling of code points that are above the Unicode maximum of |
| 471 | 0x10FFFF. Languages other than Perl may not be able to accept files that |
| 472 | contain these. |
| 473 | |
| 474 | The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of |
| 475 | the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all |
| 476 | three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the |
| 477 | allowed inputs to the strict UTF-8 traditionally defined by Unicode. |
| 478 | Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and |
| 479 | C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the |
| 480 | above-Unicode and surrogate flags, but not the non-character ones, as |
| 481 | defined in |
| 482 | L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>. |
| 483 | See L<perlunicode/Noncharacter code points>. |
| 484 | |
| 485 | Extremely high code points were never specified in any standard, and require an |
| 486 | extension to UTF-8 to express, which Perl does. It is likely that programs |
| 487 | written in something other than Perl would not be able to read files that |
| 488 | contain these; nor would Perl understand files written by something that uses a |
| 489 | different extension. For these reasons, there is a separate set of flags that |
| 490 | can warn and/or disallow these extremely high code points, even if other |
| 491 | above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED> |
| 492 | and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see |
| 493 | C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will |
| 494 | treat all above-Unicode code points, including these, as malformations. (Note |
| 495 | that the Unicode standard considers anything above 0x10FFFF to be illegal, but |
| 496 | there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1)) |
| 497 | |
| 498 | A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is |
| 499 | retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly, |
| 500 | C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named |
| 501 | C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC |
| 502 | platforms,these flags can apply to code points that actually do fit in 31 bits. |
| 503 | The new names accurately describe the situation in all cases. |
| 504 | |
| 505 | =cut |
| 506 | */ |
| 507 | |
| 508 | /* This is also a macro */ |
| 509 | PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags); |
| 510 | |
| 511 | U8 * |
| 512 | Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) |
| 513 | { |
| 514 | return uvchr_to_utf8_flags(d, uv, flags); |
| 515 | } |
| 516 | |
| 517 | #ifndef UV_IS_QUAD |
| 518 | |
| 519 | STATIC int |
| 520 | S_is_utf8_cp_above_31_bits(const U8 * const s, |
| 521 | const U8 * const e, |
| 522 | const bool consider_overlongs) |
| 523 | { |
| 524 | /* Returns TRUE if the first code point represented by the Perl-extended- |
| 525 | * UTF-8-encoded string starting at 's', and looking no further than 'e - |
| 526 | * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31. |
| 527 | * |
| 528 | * The function handles the case where the input bytes do not include all |
| 529 | * the ones necessary to represent a full character. That is, they may be |
| 530 | * the intial bytes of the representation of a code point, but possibly |
| 531 | * the final ones necessary for the complete representation may be beyond |
| 532 | * 'e - 1'. |
| 533 | * |
| 534 | * The function also can handle the case where the input is an overlong |
| 535 | * sequence. If 'consider_overlongs' is 0, the function assumes the |
| 536 | * input is not overlong, without checking, and will return based on that |
| 537 | * assumption. If this parameter is 1, the function will go to the trouble |
| 538 | * of figuring out if it actually evaluates to above or below 31 bits. |
| 539 | * |
| 540 | * The sequence is otherwise assumed to be well-formed, without checking. |
| 541 | */ |
| 542 | |
| 543 | const STRLEN len = e - s; |
| 544 | int is_overlong; |
| 545 | |
| 546 | PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS; |
| 547 | |
| 548 | assert(! UTF8_IS_INVARIANT(*s) && e > s); |
| 549 | |
| 550 | #ifdef EBCDIC |
| 551 | |
| 552 | PERL_UNUSED_ARG(consider_overlongs); |
| 553 | |
| 554 | /* On the EBCDIC code pages we handle, only the native start byte 0xFE can |
| 555 | * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can |
| 556 | * also be the start byte for a 31-bit code point; we need at least 2 |
| 557 | * bytes, and maybe up through 8 bytes, to determine that. (It can also be |
| 558 | * the start byte for an overlong sequence, but for 30-bit or smaller code |
| 559 | * points, so we don't have to worry about overlongs on EBCDIC.) */ |
| 560 | if (*s != 0xFE) { |
| 561 | return 0; |
| 562 | } |
| 563 | |
| 564 | if (len == 1) { |
| 565 | return -1; |
| 566 | } |
| 567 | |
| 568 | #else |
| 569 | |
| 570 | /* On ASCII, FE and FF are the only start bytes that can evaluate to |
| 571 | * needing more than 31 bits. */ |
| 572 | if (LIKELY(*s < 0xFE)) { |
| 573 | return 0; |
| 574 | } |
| 575 | |
| 576 | /* What we have left are FE and FF. Both of these require more than 31 |
| 577 | * bits unless they are for overlongs. */ |
| 578 | if (! consider_overlongs) { |
| 579 | return 1; |
| 580 | } |
| 581 | |
| 582 | /* Here, we have FE or FF. If the input isn't overlong, it evaluates to |
| 583 | * above 31 bits. But we need more than one byte to discern this, so if |
| 584 | * passed just the start byte, it could be an overlong evaluating to |
| 585 | * smaller */ |
| 586 | if (len == 1) { |
| 587 | return -1; |
| 588 | } |
| 589 | |
| 590 | /* Having excluded len==1, and knowing that FE and FF are both valid start |
| 591 | * bytes, we can call the function below to see if the sequence is |
| 592 | * overlong. (We don't need the full generality of the called function, |
| 593 | * but for these huge code points, speed shouldn't be a consideration, and |
| 594 | * the compiler does have enough information, since it's static to this |
| 595 | * file, to optimize to just the needed parts.) */ |
| 596 | is_overlong = is_utf8_overlong_given_start_byte_ok(s, len); |
| 597 | |
| 598 | /* If it isn't overlong, more than 31 bits are required. */ |
| 599 | if (is_overlong == 0) { |
| 600 | return 1; |
| 601 | } |
| 602 | |
| 603 | /* If it is indeterminate if it is overlong, return that */ |
| 604 | if (is_overlong < 0) { |
| 605 | return -1; |
| 606 | } |
| 607 | |
| 608 | /* Here is overlong. Such a sequence starting with FE is below 31 bits, as |
| 609 | * the max it can be is 2**31 - 1 */ |
| 610 | if (*s == 0xFE) { |
| 611 | return 0; |
| 612 | } |
| 613 | |
| 614 | #endif |
| 615 | |
| 616 | /* Here, ASCII and EBCDIC rejoin: |
| 617 | * On ASCII: We have an overlong sequence starting with FF |
| 618 | * On EBCDIC: We have a sequence starting with FE. */ |
| 619 | |
| 620 | { /* For C89, use a block so the declaration can be close to its use */ |
| 621 | |
| 622 | #ifdef EBCDIC |
| 623 | |
| 624 | /* U+7FFFFFFF (2 ** 31 - 1) |
| 625 | * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13 |
| 626 | * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73 |
| 627 | * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72 |
| 628 | * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75 |
| 629 | * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF |
| 630 | * U+80000000 (2 ** 31): |
| 631 | * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 |
| 632 | * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 |
| 633 | * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 |
| 634 | * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0 |
| 635 | * |
| 636 | * and since we know that *s = \xfe, any continuation sequcence |
| 637 | * following it that is gt the below is above 31 bits |
| 638 | [0] [1] [2] [3] [4] [5] [6] */ |
| 639 | const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42"; |
| 640 | |
| 641 | #else |
| 642 | |
| 643 | /* FF overlong for U+7FFFFFFF (2 ** 31 - 1) |
| 644 | * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF |
| 645 | * FF overlong for U+80000000 (2 ** 31): |
| 646 | * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80 |
| 647 | * and since we know that *s = \xff, any continuation sequcence |
| 648 | * following it that is gt the below is above 30 bits |
| 649 | [0] [1] [2] [3] [4] [5] [6] */ |
| 650 | const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81"; |
| 651 | |
| 652 | |
| 653 | #endif |
| 654 | const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1; |
| 655 | const STRLEN cmp_len = MIN(conts_len, len - 1); |
| 656 | |
| 657 | /* Now compare the continuation bytes in s with the ones we have |
| 658 | * compiled in that are for the largest 30 bit code point. If we have |
| 659 | * enough bytes available to determine the answer, or the bytes we do |
| 660 | * have differ from them, we can compare the two to get a definitive |
| 661 | * answer (Note that in UTF-EBCDIC, the two lowest possible |
| 662 | * continuation bytes are \x41 and \x42.) */ |
| 663 | if (cmp_len >= conts_len || memNE(s + 1, |
| 664 | conts_for_highest_30_bit, |
| 665 | cmp_len)) |
| 666 | { |
| 667 | return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len)); |
| 668 | } |
| 669 | |
| 670 | /* Here, all the bytes we have are the same as the highest 30-bit code |
| 671 | * point, but we are missing so many bytes that we can't make the |
| 672 | * determination */ |
| 673 | return -1; |
| 674 | } |
| 675 | } |
| 676 | |
| 677 | #endif |
| 678 | |
| 679 | PERL_STATIC_INLINE int |
| 680 | S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len) |
| 681 | { |
| 682 | /* Returns an int indicating whether or not the UTF-8 sequence from 's' to |
| 683 | * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if |
| 684 | * it isn't, and -1 if there isn't enough information to tell. This last |
| 685 | * return value can happen if the sequence is incomplete, missing some |
| 686 | * trailing bytes that would form a complete character. If there are |
| 687 | * enough bytes to make a definitive decision, this function does so. |
| 688 | * Usually 2 bytes sufficient. |
| 689 | * |
| 690 | * Overlongs can occur whenever the number of continuation bytes changes. |
| 691 | * That means whenever the number of leading 1 bits in a start byte |
| 692 | * increases from the next lower start byte. That happens for start bytes |
| 693 | * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal |
| 694 | * start bytes have already been excluded, so don't need to be tested here; |
| 695 | * ASCII platforms: C0, C1 |
| 696 | * EBCDIC platforms C0, C1, C2, C3, C4, E0 |
| 697 | */ |
| 698 | |
| 699 | const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); |
| 700 | const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); |
| 701 | |
| 702 | PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK; |
| 703 | assert(len > 1 && UTF8_IS_START(*s)); |
| 704 | |
| 705 | /* Each platform has overlongs after the start bytes given above (expressed |
| 706 | * in I8 for EBCDIC). What constitutes an overlong varies by platform, but |
| 707 | * the logic is the same, except the E0 overlong has already been excluded |
| 708 | * on EBCDIC platforms. The values below were found by manually |
| 709 | * inspecting the UTF-8 patterns. See the tables in utf8.h and |
| 710 | * utfebcdic.h. */ |
| 711 | |
| 712 | # ifdef EBCDIC |
| 713 | # define F0_ABOVE_OVERLONG 0xB0 |
| 714 | # define F8_ABOVE_OVERLONG 0xA8 |
| 715 | # define FC_ABOVE_OVERLONG 0xA4 |
| 716 | # define FE_ABOVE_OVERLONG 0xA2 |
| 717 | # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41" |
| 718 | /* I8(0xfe) is FF */ |
| 719 | # else |
| 720 | |
| 721 | if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) { |
| 722 | return 1; |
| 723 | } |
| 724 | |
| 725 | # define F0_ABOVE_OVERLONG 0x90 |
| 726 | # define F8_ABOVE_OVERLONG 0x88 |
| 727 | # define FC_ABOVE_OVERLONG 0x84 |
| 728 | # define FE_ABOVE_OVERLONG 0x82 |
| 729 | # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80" |
| 730 | # endif |
| 731 | |
| 732 | |
| 733 | if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG)) |
| 734 | || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG)) |
| 735 | || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG)) |
| 736 | || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG))) |
| 737 | { |
| 738 | return 1; |
| 739 | } |
| 740 | |
| 741 | /* Check for the FF overlong */ |
| 742 | return isFF_OVERLONG(s, len); |
| 743 | } |
| 744 | |
| 745 | PERL_STATIC_INLINE int |
| 746 | S_isFF_OVERLONG(const U8 * const s, const STRLEN len) |
| 747 | { |
| 748 | /* Returns an int indicating whether or not the UTF-8 sequence from 's' to |
| 749 | * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if |
| 750 | * it isn't, and -1 if there isn't enough information to tell. This last |
| 751 | * return value can happen if the sequence is incomplete, missing some |
| 752 | * trailing bytes that would form a complete character. If there are |
| 753 | * enough bytes to make a definitive decision, this function does so. */ |
| 754 | |
| 755 | PERL_ARGS_ASSERT_ISFF_OVERLONG; |
| 756 | |
| 757 | /* To be an FF overlong, all the available bytes must match */ |
| 758 | if (LIKELY(memNE(s, FF_OVERLONG_PREFIX, |
| 759 | MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1)))) |
| 760 | { |
| 761 | return 0; |
| 762 | } |
| 763 | |
| 764 | /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must |
| 765 | * be there; what comes after them doesn't matter. See tables in utf8.h, |
| 766 | * utfebcdic.h. */ |
| 767 | if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) { |
| 768 | return 1; |
| 769 | } |
| 770 | |
| 771 | /* The missing bytes could cause the result to go one way or the other, so |
| 772 | * the result is indeterminate */ |
| 773 | return -1; |
| 774 | } |
| 775 | |
| 776 | #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */ |
| 777 | # ifdef EBCDIC /* Actually is I8 */ |
| 778 | # define HIGHEST_REPRESENTABLE_UTF8 \ |
| 779 | "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" |
| 780 | # else |
| 781 | # define HIGHEST_REPRESENTABLE_UTF8 \ |
| 782 | "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" |
| 783 | # endif |
| 784 | #endif |
| 785 | |
| 786 | PERL_STATIC_INLINE int |
| 787 | S_does_utf8_overflow(const U8 * const s, |
| 788 | const U8 * e, |
| 789 | const bool consider_overlongs) |
| 790 | { |
| 791 | /* Returns an int indicating whether or not the UTF-8 sequence from 's' to |
| 792 | * 'e' - 1 would overflow an IV on this platform; that is if it represents |
| 793 | * a code point larger than the highest representable code point. It |
| 794 | * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't |
| 795 | * enough information to tell. This last return value can happen if the |
| 796 | * sequence is incomplete, missing some trailing bytes that would form a |
| 797 | * complete character. If there are enough bytes to make a definitive |
| 798 | * decision, this function does so. |
| 799 | * |
| 800 | * If 'consider_overlongs' is TRUE, the function checks for the possibility |
| 801 | * that the sequence is an overlong that doesn't overflow. Otherwise, it |
| 802 | * assumes the sequence is not an overlong. This can give different |
| 803 | * results only on ASCII 32-bit platforms. |
| 804 | * |
| 805 | * (For ASCII platforms, we could use memcmp() because we don't have to |
| 806 | * convert each byte to I8, but it's very rare input indeed that would |
| 807 | * approach overflow, so the loop below will likely only get executed once.) |
| 808 | * |
| 809 | * 'e' - 1 must not be beyond a full character. */ |
| 810 | |
| 811 | |
| 812 | PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW; |
| 813 | assert(s <= e && s + UTF8SKIP(s) >= e); |
| 814 | |
| 815 | #if ! defined(UV_IS_QUAD) |
| 816 | |
| 817 | return is_utf8_cp_above_31_bits(s, e, consider_overlongs); |
| 818 | |
| 819 | #else |
| 820 | |
| 821 | PERL_UNUSED_ARG(consider_overlongs); |
| 822 | |
| 823 | { |
| 824 | const STRLEN len = e - s; |
| 825 | const U8 *x; |
| 826 | const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8; |
| 827 | |
| 828 | for (x = s; x < e; x++, y++) { |
| 829 | |
| 830 | if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) { |
| 831 | continue; |
| 832 | } |
| 833 | |
| 834 | /* If this byte is larger than the corresponding highest UTF-8 |
| 835 | * byte, the sequence overflow; otherwise the byte is less than, |
| 836 | * and so the sequence doesn't overflow */ |
| 837 | return NATIVE_UTF8_TO_I8(*x) > *y; |
| 838 | |
| 839 | } |
| 840 | |
| 841 | /* Got to the end and all bytes are the same. If the input is a whole |
| 842 | * character, it doesn't overflow. And if it is a partial character, |
| 843 | * there's not enough information to tell */ |
| 844 | if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) { |
| 845 | return -1; |
| 846 | } |
| 847 | |
| 848 | return 0; |
| 849 | } |
| 850 | |
| 851 | #endif |
| 852 | |
| 853 | } |
| 854 | |
| 855 | #if 0 |
| 856 | |
| 857 | /* This is the portions of the above function that deal with UV_MAX instead of |
| 858 | * IV_MAX. They are left here in case we want to combine them so that internal |
| 859 | * uses can have larger code points. The only logic difference is that the |
| 860 | * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has |
| 861 | * different logic. |
| 862 | */ |
| 863 | |
| 864 | /* Anything larger than this will overflow the word if it were converted into a UV */ |
| 865 | #if defined(UV_IS_QUAD) |
| 866 | # ifdef EBCDIC /* Actually is I8 */ |
| 867 | # define HIGHEST_REPRESENTABLE_UTF8 \ |
| 868 | "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" |
| 869 | # else |
| 870 | # define HIGHEST_REPRESENTABLE_UTF8 \ |
| 871 | "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" |
| 872 | # endif |
| 873 | #else /* 32-bit */ |
| 874 | # ifdef EBCDIC |
| 875 | # define HIGHEST_REPRESENTABLE_UTF8 \ |
| 876 | "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF" |
| 877 | # else |
| 878 | # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF" |
| 879 | # endif |
| 880 | #endif |
| 881 | |
| 882 | #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC) |
| 883 | |
| 884 | /* On 32 bit ASCII machines, many overlongs that start with FF don't |
| 885 | * overflow */ |
| 886 | if (consider_overlongs && isFF_OVERLONG(s, len) > 0) { |
| 887 | |
| 888 | /* To be such an overlong, the first bytes of 's' must match |
| 889 | * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we |
| 890 | * don't have any additional bytes available, the sequence, when |
| 891 | * completed might or might not fit in 32 bits. But if we have that |
| 892 | * next byte, we can tell for sure. If it is <= 0x83, then it does |
| 893 | * fit. */ |
| 894 | if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) { |
| 895 | return -1; |
| 896 | } |
| 897 | |
| 898 | return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83; |
| 899 | } |
| 900 | |
| 901 | /* Starting with the #else, the rest of the function is identical except |
| 902 | * 1. we need to move the 'len' declaration to be global to the function |
| 903 | * 2. the endif move to just after the UNUSED_ARG. |
| 904 | * An empty endif is given just below to satisfy the preprocessor |
| 905 | */ |
| 906 | #endif |
| 907 | |
| 908 | #endif |
| 909 | |
| 910 | #undef F0_ABOVE_OVERLONG |
| 911 | #undef F8_ABOVE_OVERLONG |
| 912 | #undef FC_ABOVE_OVERLONG |
| 913 | #undef FE_ABOVE_OVERLONG |
| 914 | #undef FF_OVERLONG_PREFIX |
| 915 | |
| 916 | STRLEN |
| 917 | Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags) |
| 918 | { |
| 919 | STRLEN len; |
| 920 | const U8 *x; |
| 921 | |
| 922 | /* A helper function that should not be called directly. |
| 923 | * |
| 924 | * This function returns non-zero if the string beginning at 's' and |
| 925 | * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a |
| 926 | * code point; otherwise it returns 0. The examination stops after the |
| 927 | * first code point in 's' is validated, not looking at the rest of the |
| 928 | * input. If 'e' is such that there are not enough bytes to represent a |
| 929 | * complete code point, this function will return non-zero anyway, if the |
| 930 | * bytes it does have are well-formed UTF-8 as far as they go, and aren't |
| 931 | * excluded by 'flags'. |
| 932 | * |
| 933 | * A non-zero return gives the number of bytes required to represent the |
| 934 | * code point. Be aware that if the input is for a partial character, the |
| 935 | * return will be larger than 'e - s'. |
| 936 | * |
| 937 | * This function assumes that the code point represented is UTF-8 variant. |
| 938 | * The caller should have excluded the possibility of it being invariant |
| 939 | * before calling this function. |
| 940 | * |
| 941 | * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags |
| 942 | * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return |
| 943 | * 0 if the code point represented is well-formed Perl-extended-UTF-8, but |
| 944 | * disallowed by the flags. If the input is only for a partial character, |
| 945 | * the function will return non-zero if there is any sequence of |
| 946 | * well-formed UTF-8 that, when appended to the input sequence, could |
| 947 | * result in an allowed code point; otherwise it returns 0. Non characters |
| 948 | * cannot be determined based on partial character input. But many of the |
| 949 | * other excluded types can be determined with just the first one or two |
| 950 | * bytes. |
| 951 | * |
| 952 | */ |
| 953 | |
| 954 | PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER; |
| 955 | |
| 956 | assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE |
| 957 | |UTF8_DISALLOW_PERL_EXTENDED))); |
| 958 | assert(! UTF8_IS_INVARIANT(*s)); |
| 959 | |
| 960 | /* A variant char must begin with a start byte */ |
| 961 | if (UNLIKELY(! UTF8_IS_START(*s))) { |
| 962 | return 0; |
| 963 | } |
| 964 | |
| 965 | /* Examine a maximum of a single whole code point */ |
| 966 | if (e - s > UTF8SKIP(s)) { |
| 967 | e = s + UTF8SKIP(s); |
| 968 | } |
| 969 | |
| 970 | len = e - s; |
| 971 | |
| 972 | if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) { |
| 973 | const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); |
| 974 | |
| 975 | /* Here, we are disallowing some set of largish code points, and the |
| 976 | * first byte indicates the sequence is for a code point that could be |
| 977 | * in the excluded set. We generally don't have to look beyond this or |
| 978 | * the second byte to see if the sequence is actually for one of the |
| 979 | * excluded classes. The code below is derived from this table: |
| 980 | * |
| 981 | * UTF-8 UTF-EBCDIC I8 |
| 982 | * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate |
| 983 | * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate |
| 984 | * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode |
| 985 | * |
| 986 | * Keep in mind that legal continuation bytes range between \x80..\xBF |
| 987 | * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't |
| 988 | * continuation bytes. Hence, we don't have to test the upper edge |
| 989 | * because if any of those is encountered, the sequence is malformed, |
| 990 | * and would fail elsewhere in this function. |
| 991 | * |
| 992 | * The code here likewise assumes that there aren't other |
| 993 | * malformations; again the function should fail elsewhere because of |
| 994 | * these. For example, an overlong beginning with FC doesn't actually |
| 995 | * have to be a super; it could actually represent a small code point, |
| 996 | * even U+0000. But, since overlongs (and other malformations) are |
| 997 | * illegal, the function should return FALSE in either case. |
| 998 | */ |
| 999 | |
| 1000 | #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */ |
| 1001 | # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA |
| 1002 | # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2) |
| 1003 | |
| 1004 | # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \ |
| 1005 | /* B6 and B7 */ \ |
| 1006 | && ((s1) & 0xFE ) == 0xB6) |
| 1007 | # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF)) |
| 1008 | #else |
| 1009 | # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5 |
| 1010 | # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90) |
| 1011 | # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0) |
| 1012 | # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE) |
| 1013 | #endif |
| 1014 | |
| 1015 | if ( (flags & UTF8_DISALLOW_SUPER) |
| 1016 | && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) |
| 1017 | { |
| 1018 | return 0; /* Above Unicode */ |
| 1019 | } |
| 1020 | |
| 1021 | if ( (flags & UTF8_DISALLOW_PERL_EXTENDED) |
| 1022 | && UNLIKELY(isUTF8_PERL_EXTENDED(s))) |
| 1023 | { |
| 1024 | return 0; |
| 1025 | } |
| 1026 | |
| 1027 | if (len > 1) { |
| 1028 | const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); |
| 1029 | |
| 1030 | if ( (flags & UTF8_DISALLOW_SUPER) |
| 1031 | && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1))) |
| 1032 | { |
| 1033 | return 0; /* Above Unicode */ |
| 1034 | } |
| 1035 | |
| 1036 | if ( (flags & UTF8_DISALLOW_SURROGATE) |
| 1037 | && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1))) |
| 1038 | { |
| 1039 | return 0; /* Surrogate */ |
| 1040 | } |
| 1041 | |
| 1042 | if ( (flags & UTF8_DISALLOW_NONCHAR) |
| 1043 | && UNLIKELY(UTF8_IS_NONCHAR(s, e))) |
| 1044 | { |
| 1045 | return 0; /* Noncharacter code point */ |
| 1046 | } |
| 1047 | } |
| 1048 | } |
| 1049 | |
| 1050 | /* Make sure that all that follows are continuation bytes */ |
| 1051 | for (x = s + 1; x < e; x++) { |
| 1052 | if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) { |
| 1053 | return 0; |
| 1054 | } |
| 1055 | } |
| 1056 | |
| 1057 | /* Here is syntactically valid. Next, make sure this isn't the start of an |
| 1058 | * overlong. */ |
| 1059 | if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) { |
| 1060 | return 0; |
| 1061 | } |
| 1062 | |
| 1063 | /* And finally, that the code point represented fits in a word on this |
| 1064 | * platform */ |
| 1065 | if (0 < does_utf8_overflow(s, e, |
| 1066 | 0 /* Don't consider overlongs */ |
| 1067 | )) |
| 1068 | { |
| 1069 | return 0; |
| 1070 | } |
| 1071 | |
| 1072 | return UTF8SKIP(s); |
| 1073 | } |
| 1074 | |
| 1075 | char * |
| 1076 | Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format) |
| 1077 | { |
| 1078 | /* Returns a mortalized C string that is a displayable copy of the 'len' |
| 1079 | * bytes starting at 'start'. 'format' gives how to display each byte. |
| 1080 | * Currently, there are only two formats, so it is currently a bool: |
| 1081 | * 0 \xab |
| 1082 | * 1 ab (that is a space between two hex digit bytes) |
| 1083 | */ |
| 1084 | |
| 1085 | const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a |
| 1086 | trailing NUL */ |
| 1087 | const U8 * s = start; |
| 1088 | const U8 * const e = start + len; |
| 1089 | char * output; |
| 1090 | char * d; |
| 1091 | |
| 1092 | PERL_ARGS_ASSERT__BYTE_DUMP_STRING; |
| 1093 | |
| 1094 | Newx(output, output_len, char); |
| 1095 | SAVEFREEPV(output); |
| 1096 | |
| 1097 | d = output; |
| 1098 | for (s = start; s < e; s++) { |
| 1099 | const unsigned high_nibble = (*s & 0xF0) >> 4; |
| 1100 | const unsigned low_nibble = (*s & 0x0F); |
| 1101 | |
| 1102 | if (format) { |
| 1103 | if (s > start) { |
| 1104 | *d++ = ' '; |
| 1105 | } |
| 1106 | } |
| 1107 | else { |
| 1108 | *d++ = '\\'; |
| 1109 | *d++ = 'x'; |
| 1110 | } |
| 1111 | |
| 1112 | if (high_nibble < 10) { |
| 1113 | *d++ = high_nibble + '0'; |
| 1114 | } |
| 1115 | else { |
| 1116 | *d++ = high_nibble - 10 + 'a'; |
| 1117 | } |
| 1118 | |
| 1119 | if (low_nibble < 10) { |
| 1120 | *d++ = low_nibble + '0'; |
| 1121 | } |
| 1122 | else { |
| 1123 | *d++ = low_nibble - 10 + 'a'; |
| 1124 | } |
| 1125 | } |
| 1126 | |
| 1127 | *d = '\0'; |
| 1128 | return output; |
| 1129 | } |
| 1130 | |
| 1131 | PERL_STATIC_INLINE char * |
| 1132 | S_unexpected_non_continuation_text(pTHX_ const U8 * const s, |
| 1133 | |
| 1134 | /* Max number of bytes to print */ |
| 1135 | STRLEN print_len, |
| 1136 | |
| 1137 | /* Which one is the non-continuation */ |
| 1138 | const STRLEN non_cont_byte_pos, |
| 1139 | |
| 1140 | /* How many bytes should there be? */ |
| 1141 | const STRLEN expect_len) |
| 1142 | { |
| 1143 | /* Return the malformation warning text for an unexpected continuation |
| 1144 | * byte. */ |
| 1145 | |
| 1146 | const char * const where = (non_cont_byte_pos == 1) |
| 1147 | ? "immediately" |
| 1148 | : Perl_form(aTHX_ "%d bytes", |
| 1149 | (int) non_cont_byte_pos); |
| 1150 | const U8 * x = s + non_cont_byte_pos; |
| 1151 | const U8 * e = s + print_len; |
| 1152 | |
| 1153 | PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT; |
| 1154 | |
| 1155 | /* We don't need to pass this parameter, but since it has already been |
| 1156 | * calculated, it's likely faster to pass it; verify under DEBUGGING */ |
| 1157 | assert(expect_len == UTF8SKIP(s)); |
| 1158 | |
| 1159 | /* As a defensive coding measure, don't output anything past a NUL. Such |
| 1160 | * bytes shouldn't be in the middle of a malformation, and could mark the |
| 1161 | * end of the allocated string, and what comes after is undefined */ |
| 1162 | for (; x < e; x++) { |
| 1163 | if (*x == '\0') { |
| 1164 | x++; /* Output this particular NUL */ |
| 1165 | break; |
| 1166 | } |
| 1167 | } |
| 1168 | |
| 1169 | return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x," |
| 1170 | " %s after start byte 0x%02x; need %d bytes, got %d)", |
| 1171 | malformed_text, |
| 1172 | _byte_dump_string(s, x - s, 0), |
| 1173 | *(s + non_cont_byte_pos), |
| 1174 | where, |
| 1175 | *s, |
| 1176 | (int) expect_len, |
| 1177 | (int) non_cont_byte_pos); |
| 1178 | } |
| 1179 | |
| 1180 | /* |
| 1181 | |
| 1182 | =for apidoc utf8n_to_uvchr |
| 1183 | |
| 1184 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 1185 | Most code should use L</utf8_to_uvchr_buf>() rather than call this |
| 1186 | directly. |
| 1187 | |
| 1188 | Bottom level UTF-8 decode routine. |
| 1189 | Returns the native code point value of the first character in the string C<s>, |
| 1190 | which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than |
| 1191 | C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to |
| 1192 | the length, in bytes, of that character. |
| 1193 | |
| 1194 | The value of C<flags> determines the behavior when C<s> does not point to a |
| 1195 | well-formed UTF-8 character. If C<flags> is 0, encountering a malformation |
| 1196 | causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) |
| 1197 | is the next possible position in C<s> that could begin a non-malformed |
| 1198 | character. Also, if UTF-8 warnings haven't been lexically disabled, a warning |
| 1199 | is raised. Some UTF-8 input sequences may contain multiple malformations. |
| 1200 | This function tries to find every possible one in each call, so multiple |
| 1201 | warnings can be raised for the same sequence. |
| 1202 | |
| 1203 | Various ALLOW flags can be set in C<flags> to allow (and not warn on) |
| 1204 | individual types of malformations, such as the sequence being overlong (that |
| 1205 | is, when there is a shorter sequence that can express the same code point; |
| 1206 | overlong sequences are expressly forbidden in the UTF-8 standard due to |
| 1207 | potential security issues). Another malformation example is the first byte of |
| 1208 | a character not being a legal first byte. See F<utf8.h> for the list of such |
| 1209 | flags. Even if allowed, this function generally returns the Unicode |
| 1210 | REPLACEMENT CHARACTER when it encounters a malformation. There are flags in |
| 1211 | F<utf8.h> to override this behavior for the overlong malformations, but don't |
| 1212 | do that except for very specialized purposes. |
| 1213 | |
| 1214 | The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other |
| 1215 | flags) malformation is found. If this flag is set, the routine assumes that |
| 1216 | the caller will raise a warning, and this function will silently just set |
| 1217 | C<retlen> to C<-1> (cast to C<STRLEN>) and return zero. |
| 1218 | |
| 1219 | Note that this API requires disambiguation between successful decoding a C<NUL> |
| 1220 | character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as |
| 1221 | in both cases, 0 is returned, and, depending on the malformation, C<retlen> may |
| 1222 | be set to 1. To disambiguate, upon a zero return, see if the first byte of |
| 1223 | C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an |
| 1224 | error. Or you can use C<L</utf8n_to_uvchr_error>>. |
| 1225 | |
| 1226 | Certain code points are considered problematic. These are Unicode surrogates, |
| 1227 | Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF. |
| 1228 | By default these are considered regular code points, but certain situations |
| 1229 | warrant special handling for them, which can be specified using the C<flags> |
| 1230 | parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all |
| 1231 | three classes are treated as malformations and handled as such. The flags |
| 1232 | C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and |
| 1233 | C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to |
| 1234 | disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE> |
| 1235 | restricts the allowed inputs to the strict UTF-8 traditionally defined by |
| 1236 | Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness |
| 1237 | definition given by |
| 1238 | L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>. |
| 1239 | The difference between traditional strictness and C9 strictness is that the |
| 1240 | latter does not forbid non-character code points. (They are still discouraged, |
| 1241 | however.) For more discussion see L<perlunicode/Noncharacter code points>. |
| 1242 | |
| 1243 | The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, |
| 1244 | C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>, |
| 1245 | C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be |
| 1246 | raised for their respective categories, but otherwise the code points are |
| 1247 | considered valid (not malformations). To get a category to both be treated as |
| 1248 | a malformation and raise a warning, specify both the WARN and DISALLOW flags. |
| 1249 | (But note that warnings are not raised if lexically disabled nor if |
| 1250 | C<UTF8_CHECK_ONLY> is also specified.) |
| 1251 | |
| 1252 | Extremely high code points were never specified in any standard, and require an |
| 1253 | extension to UTF-8 to express, which Perl does. It is likely that programs |
| 1254 | written in something other than Perl would not be able to read files that |
| 1255 | contain these; nor would Perl understand files written by something that uses a |
| 1256 | different extension. For these reasons, there is a separate set of flags that |
| 1257 | can warn and/or disallow these extremely high code points, even if other |
| 1258 | above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and |
| 1259 | C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see |
| 1260 | C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all |
| 1261 | above-Unicode code points, including these, as malformations. |
| 1262 | (Note that the Unicode standard considers anything above 0x10FFFF to be |
| 1263 | illegal, but there are standards predating it that allow up to 0x7FFF_FFFF |
| 1264 | (2**31 -1)) |
| 1265 | |
| 1266 | A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is |
| 1267 | retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly, |
| 1268 | C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named |
| 1269 | C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags |
| 1270 | can apply to code points that actually do fit in 31 bits. This happens on |
| 1271 | EBCDIC platforms, and sometimes when the L<overlong |
| 1272 | malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately |
| 1273 | describe the situation in all cases. |
| 1274 | |
| 1275 | |
| 1276 | All other code points corresponding to Unicode characters, including private |
| 1277 | use and those yet to be assigned, are never considered malformed and never |
| 1278 | warn. |
| 1279 | |
| 1280 | =for apidoc Amnh||UTF8_CHECK_ONLY |
| 1281 | =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE |
| 1282 | =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE |
| 1283 | =for apidoc Amnh||UTF8_DISALLOW_SURROGATE |
| 1284 | =for apidoc Amnh||UTF8_DISALLOW_NONCHAR |
| 1285 | =for apidoc Amnh||UTF8_DISALLOW_SUPER |
| 1286 | =for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE |
| 1287 | =for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE |
| 1288 | =for apidoc Amnh||UTF8_WARN_SURROGATE |
| 1289 | =for apidoc Amnh||UTF8_WARN_NONCHAR |
| 1290 | =for apidoc Amnh||UTF8_WARN_SUPER |
| 1291 | =for apidoc Amnh||UTF8_WARN_PERL_EXTENDED |
| 1292 | =for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED |
| 1293 | |
| 1294 | =cut |
| 1295 | |
| 1296 | Also implemented as a macro in utf8.h |
| 1297 | */ |
| 1298 | |
| 1299 | UV |
| 1300 | Perl_utf8n_to_uvchr(const U8 *s, |
| 1301 | STRLEN curlen, |
| 1302 | STRLEN *retlen, |
| 1303 | const U32 flags) |
| 1304 | { |
| 1305 | PERL_ARGS_ASSERT_UTF8N_TO_UVCHR; |
| 1306 | |
| 1307 | return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL); |
| 1308 | } |
| 1309 | |
| 1310 | /* |
| 1311 | |
| 1312 | =for apidoc utf8n_to_uvchr_error |
| 1313 | |
| 1314 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 1315 | Most code should use L</utf8_to_uvchr_buf>() rather than call this |
| 1316 | directly. |
| 1317 | |
| 1318 | This function is for code that needs to know what the precise malformation(s) |
| 1319 | are when an error is found. If you also need to know the generated warning |
| 1320 | messages, use L</utf8n_to_uvchr_msgs>() instead. |
| 1321 | |
| 1322 | It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after |
| 1323 | all the others, C<errors>. If this parameter is 0, this function behaves |
| 1324 | identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer |
| 1325 | to a C<U32> variable, which this function sets to indicate any errors found. |
| 1326 | Upon return, if C<*errors> is 0, there were no errors found. Otherwise, |
| 1327 | C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some |
| 1328 | of these bits will be set if a malformation is found, even if the input |
| 1329 | C<flags> parameter indicates that the given malformation is allowed; those |
| 1330 | exceptions are noted: |
| 1331 | |
| 1332 | =over 4 |
| 1333 | |
| 1334 | =item C<UTF8_GOT_PERL_EXTENDED> |
| 1335 | |
| 1336 | The input sequence is not standard UTF-8, but a Perl extension. This bit is |
| 1337 | set only if the input C<flags> parameter contains either the |
| 1338 | C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags. |
| 1339 | |
| 1340 | Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, |
| 1341 | and so some extension must be used to express them. Perl uses a natural |
| 1342 | extension to UTF-8 to represent the ones up to 2**36-1, and invented a further |
| 1343 | extension to represent even higher ones, so that any code point that fits in a |
| 1344 | 64-bit word can be represented. Text using these extensions is not likely to |
| 1345 | be portable to non-Perl code. We lump both of these extensions together and |
| 1346 | refer to them as Perl extended UTF-8. There exist other extensions that people |
| 1347 | have invented, incompatible with Perl's. |
| 1348 | |
| 1349 | On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing |
| 1350 | extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower |
| 1351 | than on ASCII. Prior to that, code points 2**31 and higher were simply |
| 1352 | unrepresentable, and a different, incompatible method was used to represent |
| 1353 | code points between 2**30 and 2**31 - 1. |
| 1354 | |
| 1355 | On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if |
| 1356 | Perl extended UTF-8 is used. |
| 1357 | |
| 1358 | In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still |
| 1359 | may use for backward compatibility. That name is misleading, as this flag may |
| 1360 | be set when the code point actually does fit in 31 bits. This happens on |
| 1361 | EBCDIC platforms, and sometimes when the L<overlong |
| 1362 | malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately |
| 1363 | describes the situation in all cases. |
| 1364 | |
| 1365 | =item C<UTF8_GOT_CONTINUATION> |
| 1366 | |
| 1367 | The input sequence was malformed in that the first byte was a UTF-8 |
| 1368 | continuation byte. |
| 1369 | |
| 1370 | =item C<UTF8_GOT_EMPTY> |
| 1371 | |
| 1372 | The input C<curlen> parameter was 0. |
| 1373 | |
| 1374 | =item C<UTF8_GOT_LONG> |
| 1375 | |
| 1376 | The input sequence was malformed in that there is some other sequence that |
| 1377 | evaluates to the same code point, but that sequence is shorter than this one. |
| 1378 | |
| 1379 | Until Unicode 3.1, it was legal for programs to accept this malformation, but |
| 1380 | it was discovered that this created security issues. |
| 1381 | |
| 1382 | =item C<UTF8_GOT_NONCHAR> |
| 1383 | |
| 1384 | The code point represented by the input UTF-8 sequence is for a Unicode |
| 1385 | non-character code point. |
| 1386 | This bit is set only if the input C<flags> parameter contains either the |
| 1387 | C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags. |
| 1388 | |
| 1389 | =item C<UTF8_GOT_NON_CONTINUATION> |
| 1390 | |
| 1391 | The input sequence was malformed in that a non-continuation type byte was found |
| 1392 | in a position where only a continuation type one should be. See also |
| 1393 | C<L</UTF8_GOT_SHORT>>. |
| 1394 | |
| 1395 | =item C<UTF8_GOT_OVERFLOW> |
| 1396 | |
| 1397 | The input sequence was malformed in that it is for a code point that is not |
| 1398 | representable in the number of bits available in an IV on the current platform. |
| 1399 | |
| 1400 | =item C<UTF8_GOT_SHORT> |
| 1401 | |
| 1402 | The input sequence was malformed in that C<curlen> is smaller than required for |
| 1403 | a complete sequence. In other words, the input is for a partial character |
| 1404 | sequence. |
| 1405 | |
| 1406 | |
| 1407 | C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short |
| 1408 | sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always |
| 1409 | that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete |
| 1410 | sequence was looked at. If no other flags are present, it means that the |
| 1411 | sequence was valid as far as it went. Depending on the application, this could |
| 1412 | mean one of three things: |
| 1413 | |
| 1414 | =over |
| 1415 | |
| 1416 | =item * |
| 1417 | |
| 1418 | The C<curlen> length parameter passed in was too small, and the function was |
| 1419 | prevented from examining all the necessary bytes. |
| 1420 | |
| 1421 | =item * |
| 1422 | |
| 1423 | The buffer being looked at is based on reading data, and the data received so |
| 1424 | far stopped in the middle of a character, so that the next read will |
| 1425 | read the remainder of this character. (It is up to the caller to deal with the |
| 1426 | split bytes somehow.) |
| 1427 | |
| 1428 | =item * |
| 1429 | |
| 1430 | This is a real error, and the partial sequence is all we're going to get. |
| 1431 | |
| 1432 | =back |
| 1433 | |
| 1434 | =item C<UTF8_GOT_SUPER> |
| 1435 | |
| 1436 | The input sequence was malformed in that it is for a non-Unicode code point; |
| 1437 | that is, one above the legal Unicode maximum. |
| 1438 | This bit is set only if the input C<flags> parameter contains either the |
| 1439 | C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags. |
| 1440 | |
| 1441 | =item C<UTF8_GOT_SURROGATE> |
| 1442 | |
| 1443 | The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate |
| 1444 | code point. |
| 1445 | This bit is set only if the input C<flags> parameter contains either the |
| 1446 | C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags. |
| 1447 | |
| 1448 | =back |
| 1449 | |
| 1450 | To do your own error handling, call this function with the C<UTF8_CHECK_ONLY> |
| 1451 | flag to suppress any warnings, and then examine the C<*errors> return. |
| 1452 | |
| 1453 | =cut |
| 1454 | |
| 1455 | Also implemented as a macro in utf8.h |
| 1456 | */ |
| 1457 | |
| 1458 | UV |
| 1459 | Perl_utf8n_to_uvchr_error(const U8 *s, |
| 1460 | STRLEN curlen, |
| 1461 | STRLEN *retlen, |
| 1462 | const U32 flags, |
| 1463 | U32 * errors) |
| 1464 | { |
| 1465 | PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR; |
| 1466 | |
| 1467 | return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL); |
| 1468 | } |
| 1469 | |
| 1470 | /* |
| 1471 | |
| 1472 | =for apidoc utf8n_to_uvchr_msgs |
| 1473 | |
| 1474 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 1475 | Most code should use L</utf8_to_uvchr_buf>() rather than call this |
| 1476 | directly. |
| 1477 | |
| 1478 | This function is for code that needs to know what the precise malformation(s) |
| 1479 | are when an error is found, and wants the corresponding warning and/or error |
| 1480 | messages to be returned to the caller rather than be displayed. All messages |
| 1481 | that would have been displayed if all lexical warnings are enabled will be |
| 1482 | returned. |
| 1483 | |
| 1484 | It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter |
| 1485 | placed after all the others, C<msgs>. If this parameter is 0, this function |
| 1486 | behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should |
| 1487 | be a pointer to an C<AV *> variable, in which this function creates a new AV to |
| 1488 | contain any appropriate messages. The elements of the array are ordered so |
| 1489 | that the first message that would have been displayed is in the 0th element, |
| 1490 | and so on. Each element is a hash with three key-value pairs, as follows: |
| 1491 | |
| 1492 | =over 4 |
| 1493 | |
| 1494 | =item C<text> |
| 1495 | |
| 1496 | The text of the message as a C<SVpv>. |
| 1497 | |
| 1498 | =item C<warn_categories> |
| 1499 | |
| 1500 | The warning category (or categories) packed into a C<SVuv>. |
| 1501 | |
| 1502 | =item C<flag> |
| 1503 | |
| 1504 | A single flag bit associated with this message, in a C<SVuv>. |
| 1505 | The bit corresponds to some bit in the C<*errors> return value, |
| 1506 | such as C<UTF8_GOT_LONG>. |
| 1507 | |
| 1508 | =back |
| 1509 | |
| 1510 | It's important to note that specifying this parameter as non-null will cause |
| 1511 | any warnings this function would otherwise generate to be suppressed, and |
| 1512 | instead be placed in C<*msgs>. The caller can check the lexical warnings state |
| 1513 | (or not) when choosing what to do with the returned messages. |
| 1514 | |
| 1515 | If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence |
| 1516 | no AV is created. |
| 1517 | |
| 1518 | The caller, of course, is responsible for freeing any returned AV. |
| 1519 | |
| 1520 | =cut |
| 1521 | */ |
| 1522 | |
| 1523 | UV |
| 1524 | Perl__utf8n_to_uvchr_msgs_helper(const U8 *s, |
| 1525 | STRLEN curlen, |
| 1526 | STRLEN *retlen, |
| 1527 | const U32 flags, |
| 1528 | U32 * errors, |
| 1529 | AV ** msgs) |
| 1530 | { |
| 1531 | const U8 * const s0 = s; |
| 1532 | const U8 * send = s0 + curlen; |
| 1533 | U32 possible_problems; /* A bit is set here for each potential problem |
| 1534 | found as we go along */ |
| 1535 | UV uv; |
| 1536 | STRLEN expectlen; /* How long should this sequence be? */ |
| 1537 | STRLEN avail_len; /* When input is too short, gives what that is */ |
| 1538 | U32 discard_errors; /* Used to save branches when 'errors' is NULL; this |
| 1539 | gets set and discarded */ |
| 1540 | |
| 1541 | /* The below are used only if there is both an overlong malformation and a |
| 1542 | * too short one. Otherwise the first two are set to 's0' and 'send', and |
| 1543 | * the third not used at all */ |
| 1544 | U8 * adjusted_s0; |
| 1545 | U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this |
| 1546 | routine; see [perl #130921] */ |
| 1547 | UV uv_so_far; |
| 1548 | dTHX; |
| 1549 | |
| 1550 | PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER; |
| 1551 | |
| 1552 | /* Here, is one of: a) malformed; b) a problematic code point (surrogate, |
| 1553 | * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul |
| 1554 | * syllables that the dfa doesn't properly handle. Quickly dispose of the |
| 1555 | * final case. */ |
| 1556 | |
| 1557 | #ifndef EBCDIC |
| 1558 | |
| 1559 | /* Each of the affected Hanguls starts with \xED */ |
| 1560 | |
| 1561 | if (is_HANGUL_ED_utf8_safe(s0, send)) { |
| 1562 | if (retlen) { |
| 1563 | *retlen = 3; |
| 1564 | } |
| 1565 | if (errors) { |
| 1566 | *errors = 0; |
| 1567 | } |
| 1568 | if (msgs) { |
| 1569 | *msgs = NULL; |
| 1570 | } |
| 1571 | |
| 1572 | return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT)) |
| 1573 | | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT) |
| 1574 | | (s0[2] & UTF_CONTINUATION_MASK); |
| 1575 | } |
| 1576 | |
| 1577 | #endif |
| 1578 | |
| 1579 | /* In conjunction with the exhaustive tests that can be enabled in |
| 1580 | * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely |
| 1581 | * what it is intended to do, and that no flaws in it are masked by |
| 1582 | * dropping down and executing the code below |
| 1583 | assert(! isUTF8_CHAR(s0, send) |
| 1584 | || UTF8_IS_SURROGATE(s0, send) |
| 1585 | || UTF8_IS_SUPER(s0, send) |
| 1586 | || UTF8_IS_NONCHAR(s0,send)); |
| 1587 | */ |
| 1588 | |
| 1589 | s = s0; |
| 1590 | uv = *s0; |
| 1591 | possible_problems = 0; |
| 1592 | expectlen = 0; |
| 1593 | avail_len = 0; |
| 1594 | discard_errors = 0; |
| 1595 | adjusted_s0 = (U8 *) s0; |
| 1596 | uv_so_far = 0; |
| 1597 | |
| 1598 | if (errors) { |
| 1599 | *errors = 0; |
| 1600 | } |
| 1601 | else { |
| 1602 | errors = &discard_errors; |
| 1603 | } |
| 1604 | |
| 1605 | /* The order of malformation tests here is important. We should consume as |
| 1606 | * few bytes as possible in order to not skip any valid character. This is |
| 1607 | * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also |
| 1608 | * https://unicode.org/reports/tr36 for more discussion as to why. For |
| 1609 | * example, once we've done a UTF8SKIP, we can tell the expected number of |
| 1610 | * bytes, and could fail right off the bat if the input parameters indicate |
| 1611 | * that there are too few available. But it could be that just that first |
| 1612 | * byte is garbled, and the intended character occupies fewer bytes. If we |
| 1613 | * blindly assumed that the first byte is correct, and skipped based on |
| 1614 | * that number, we could skip over a valid input character. So instead, we |
| 1615 | * always examine the sequence byte-by-byte. |
| 1616 | * |
| 1617 | * We also should not consume too few bytes, otherwise someone could inject |
| 1618 | * things. For example, an input could be deliberately designed to |
| 1619 | * overflow, and if this code bailed out immediately upon discovering that, |
| 1620 | * returning to the caller C<*retlen> pointing to the very next byte (one |
| 1621 | * which is actually part of the overflowing sequence), that could look |
| 1622 | * legitimate to the caller, which could discard the initial partial |
| 1623 | * sequence and process the rest, inappropriately. |
| 1624 | * |
| 1625 | * Some possible input sequences are malformed in more than one way. This |
| 1626 | * function goes to lengths to try to find all of them. This is necessary |
| 1627 | * for correctness, as the inputs may allow one malformation but not |
| 1628 | * another, and if we abandon searching for others after finding the |
| 1629 | * allowed one, we could allow in something that shouldn't have been. |
| 1630 | */ |
| 1631 | |
| 1632 | if (UNLIKELY(curlen == 0)) { |
| 1633 | possible_problems |= UTF8_GOT_EMPTY; |
| 1634 | curlen = 0; |
| 1635 | uv = UNICODE_REPLACEMENT; |
| 1636 | goto ready_to_handle_errors; |
| 1637 | } |
| 1638 | |
| 1639 | expectlen = UTF8SKIP(s); |
| 1640 | |
| 1641 | /* A well-formed UTF-8 character, as the vast majority of calls to this |
| 1642 | * function will be for, has this expected length. For efficiency, set |
| 1643 | * things up here to return it. It will be overriden only in those rare |
| 1644 | * cases where a malformation is found */ |
| 1645 | if (retlen) { |
| 1646 | *retlen = expectlen; |
| 1647 | } |
| 1648 | |
| 1649 | /* A continuation character can't start a valid sequence */ |
| 1650 | if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) { |
| 1651 | possible_problems |= UTF8_GOT_CONTINUATION; |
| 1652 | curlen = 1; |
| 1653 | uv = UNICODE_REPLACEMENT; |
| 1654 | goto ready_to_handle_errors; |
| 1655 | } |
| 1656 | |
| 1657 | /* Here is not a continuation byte, nor an invariant. The only thing left |
| 1658 | * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START |
| 1659 | * because it excludes start bytes like \xC0 that always lead to |
| 1660 | * overlongs.) */ |
| 1661 | |
| 1662 | /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits |
| 1663 | * that indicate the number of bytes in the character's whole UTF-8 |
| 1664 | * sequence, leaving just the bits that are part of the value. */ |
| 1665 | uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen); |
| 1666 | |
| 1667 | /* Setup the loop end point, making sure to not look past the end of the |
| 1668 | * input string, and flag it as too short if the size isn't big enough. */ |
| 1669 | if (UNLIKELY(curlen < expectlen)) { |
| 1670 | possible_problems |= UTF8_GOT_SHORT; |
| 1671 | avail_len = curlen; |
| 1672 | } |
| 1673 | else { |
| 1674 | send = (U8*) s0 + expectlen; |
| 1675 | } |
| 1676 | |
| 1677 | /* Now, loop through the remaining bytes in the character's sequence, |
| 1678 | * accumulating each into the working value as we go. */ |
| 1679 | for (s = s0 + 1; s < send; s++) { |
| 1680 | if (LIKELY(UTF8_IS_CONTINUATION(*s))) { |
| 1681 | uv = UTF8_ACCUMULATE(uv, *s); |
| 1682 | continue; |
| 1683 | } |
| 1684 | |
| 1685 | /* Here, found a non-continuation before processing all expected bytes. |
| 1686 | * This byte indicates the beginning of a new character, so quit, even |
| 1687 | * if allowing this malformation. */ |
| 1688 | possible_problems |= UTF8_GOT_NON_CONTINUATION; |
| 1689 | break; |
| 1690 | } /* End of loop through the character's bytes */ |
| 1691 | |
| 1692 | /* Save how many bytes were actually in the character */ |
| 1693 | curlen = s - s0; |
| 1694 | |
| 1695 | /* Note that there are two types of too-short malformation. One is when |
| 1696 | * there is actual wrong data before the normal termination of the |
| 1697 | * sequence. The other is that the sequence wasn't complete before the end |
| 1698 | * of the data we are allowed to look at, based on the input 'curlen'. |
| 1699 | * This means that we were passed data for a partial character, but it is |
| 1700 | * valid as far as we saw. The other is definitely invalid. This |
| 1701 | * distinction could be important to a caller, so the two types are kept |
| 1702 | * separate. |
| 1703 | * |
| 1704 | * A convenience macro that matches either of the too-short conditions. */ |
| 1705 | # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION) |
| 1706 | |
| 1707 | if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) { |
| 1708 | uv_so_far = uv; |
| 1709 | uv = UNICODE_REPLACEMENT; |
| 1710 | } |
| 1711 | |
| 1712 | /* Check for overflow. The algorithm requires us to not look past the end |
| 1713 | * of the current character, even if partial, so the upper limit is 's' */ |
| 1714 | if (UNLIKELY(0 < does_utf8_overflow(s0, s, |
| 1715 | 1 /* Do consider overlongs */ |
| 1716 | ))) |
| 1717 | { |
| 1718 | possible_problems |= UTF8_GOT_OVERFLOW; |
| 1719 | uv = UNICODE_REPLACEMENT; |
| 1720 | } |
| 1721 | |
| 1722 | /* Check for overlong. If no problems so far, 'uv' is the correct code |
| 1723 | * point value. Simply see if it is expressible in fewer bytes. Otherwise |
| 1724 | * we must look at the UTF-8 byte sequence itself to see if it is for an |
| 1725 | * overlong */ |
| 1726 | if ( ( LIKELY(! possible_problems) |
| 1727 | && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv))) |
| 1728 | || ( UNLIKELY(possible_problems) |
| 1729 | && ( UNLIKELY(! UTF8_IS_START(*s0)) |
| 1730 | || ( curlen > 1 |
| 1731 | && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0, |
| 1732 | s - s0)))))) |
| 1733 | { |
| 1734 | possible_problems |= UTF8_GOT_LONG; |
| 1735 | |
| 1736 | if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT) |
| 1737 | |
| 1738 | /* The calculation in the 'true' branch of this 'if' |
| 1739 | * below won't work if overflows, and isn't needed |
| 1740 | * anyway. Further below we handle all overflow |
| 1741 | * cases */ |
| 1742 | && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))) |
| 1743 | { |
| 1744 | UV min_uv = uv_so_far; |
| 1745 | STRLEN i; |
| 1746 | |
| 1747 | /* Here, the input is both overlong and is missing some trailing |
| 1748 | * bytes. There is no single code point it could be for, but there |
| 1749 | * may be enough information present to determine if what we have |
| 1750 | * so far is for an unallowed code point, such as for a surrogate. |
| 1751 | * The code further below has the intelligence to determine this, |
| 1752 | * but just for non-overlong UTF-8 sequences. What we do here is |
| 1753 | * calculate the smallest code point the input could represent if |
| 1754 | * there were no too short malformation. Then we compute and save |
| 1755 | * the UTF-8 for that, which is what the code below looks at |
| 1756 | * instead of the raw input. It turns out that the smallest such |
| 1757 | * code point is all we need. */ |
| 1758 | for (i = curlen; i < expectlen; i++) { |
| 1759 | min_uv = UTF8_ACCUMULATE(min_uv, |
| 1760 | I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK)); |
| 1761 | } |
| 1762 | |
| 1763 | adjusted_s0 = temp_char_buf; |
| 1764 | (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0); |
| 1765 | } |
| 1766 | } |
| 1767 | |
| 1768 | /* Here, we have found all the possible problems, except for when the input |
| 1769 | * is for a problematic code point not allowed by the input parameters. */ |
| 1770 | |
| 1771 | /* uv is valid for overlongs */ |
| 1772 | if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG)) |
| 1773 | |
| 1774 | /* isn't problematic if < this */ |
| 1775 | && uv >= UNICODE_SURROGATE_FIRST) |
| 1776 | || ( UNLIKELY(possible_problems) |
| 1777 | |
| 1778 | /* if overflow, we know without looking further |
| 1779 | * precisely which of the problematic types it is, |
| 1780 | * and we deal with those in the overflow handling |
| 1781 | * code */ |
| 1782 | && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)) |
| 1783 | && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0) |
| 1784 | || UNLIKELY(isUTF8_PERL_EXTENDED(s0))))) |
| 1785 | && ((flags & ( UTF8_DISALLOW_NONCHAR |
| 1786 | |UTF8_DISALLOW_SURROGATE |
| 1787 | |UTF8_DISALLOW_SUPER |
| 1788 | |UTF8_DISALLOW_PERL_EXTENDED |
| 1789 | |UTF8_WARN_NONCHAR |
| 1790 | |UTF8_WARN_SURROGATE |
| 1791 | |UTF8_WARN_SUPER |
| 1792 | |UTF8_WARN_PERL_EXTENDED)))) |
| 1793 | { |
| 1794 | /* If there were no malformations, or the only malformation is an |
| 1795 | * overlong, 'uv' is valid */ |
| 1796 | if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) { |
| 1797 | if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { |
| 1798 | possible_problems |= UTF8_GOT_SURROGATE; |
| 1799 | } |
| 1800 | else if (UNLIKELY(uv > PERL_UNICODE_MAX)) { |
| 1801 | possible_problems |= UTF8_GOT_SUPER; |
| 1802 | } |
| 1803 | else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) { |
| 1804 | possible_problems |= UTF8_GOT_NONCHAR; |
| 1805 | } |
| 1806 | } |
| 1807 | else { /* Otherwise, need to look at the source UTF-8, possibly |
| 1808 | adjusted to be non-overlong */ |
| 1809 | |
| 1810 | if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0) |
| 1811 | >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) |
| 1812 | { |
| 1813 | possible_problems |= UTF8_GOT_SUPER; |
| 1814 | } |
| 1815 | else if (curlen > 1) { |
| 1816 | if (UNLIKELY(IS_UTF8_2_BYTE_SUPER( |
| 1817 | NATIVE_UTF8_TO_I8(*adjusted_s0), |
| 1818 | NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) |
| 1819 | { |
| 1820 | possible_problems |= UTF8_GOT_SUPER; |
| 1821 | } |
| 1822 | else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE( |
| 1823 | NATIVE_UTF8_TO_I8(*adjusted_s0), |
| 1824 | NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) |
| 1825 | { |
| 1826 | possible_problems |= UTF8_GOT_SURROGATE; |
| 1827 | } |
| 1828 | } |
| 1829 | |
| 1830 | /* We need a complete well-formed UTF-8 character to discern |
| 1831 | * non-characters, so can't look for them here */ |
| 1832 | } |
| 1833 | } |
| 1834 | |
| 1835 | ready_to_handle_errors: |
| 1836 | |
| 1837 | /* At this point: |
| 1838 | * curlen contains the number of bytes in the sequence that |
| 1839 | * this call should advance the input by. |
| 1840 | * avail_len gives the available number of bytes passed in, but |
| 1841 | * only if this is less than the expected number of |
| 1842 | * bytes, based on the code point's start byte. |
| 1843 | * possible_problems' is 0 if there weren't any problems; otherwise a bit |
| 1844 | * is set in it for each potential problem found. |
| 1845 | * uv contains the code point the input sequence |
| 1846 | * represents; or if there is a problem that prevents |
| 1847 | * a well-defined value from being computed, it is |
| 1848 | * some subsitute value, typically the REPLACEMENT |
| 1849 | * CHARACTER. |
| 1850 | * s0 points to the first byte of the character |
| 1851 | * s points to just after were we left off processing |
| 1852 | * the character |
| 1853 | * send points to just after where that character should |
| 1854 | * end, based on how many bytes the start byte tells |
| 1855 | * us should be in it, but no further than s0 + |
| 1856 | * avail_len |
| 1857 | */ |
| 1858 | |
| 1859 | if (UNLIKELY(possible_problems)) { |
| 1860 | bool disallowed = FALSE; |
| 1861 | const U32 orig_problems = possible_problems; |
| 1862 | |
| 1863 | if (msgs) { |
| 1864 | *msgs = NULL; |
| 1865 | } |
| 1866 | |
| 1867 | while (possible_problems) { /* Handle each possible problem */ |
| 1868 | U32 pack_warn = 0; |
| 1869 | char * message = NULL; |
| 1870 | U32 this_flag_bit = 0; |
| 1871 | |
| 1872 | /* Each 'if' clause handles one problem. They are ordered so that |
| 1873 | * the first ones' messages will be displayed before the later |
| 1874 | * ones; this is kinda in decreasing severity order. But the |
| 1875 | * overlong must come last, as it changes 'uv' looked at by the |
| 1876 | * others */ |
| 1877 | if (possible_problems & UTF8_GOT_OVERFLOW) { |
| 1878 | |
| 1879 | /* Overflow means also got a super and are using Perl's |
| 1880 | * extended UTF-8, but we handle all three cases here */ |
| 1881 | possible_problems |
| 1882 | &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED); |
| 1883 | *errors |= UTF8_GOT_OVERFLOW; |
| 1884 | |
| 1885 | /* But the API says we flag all errors found */ |
| 1886 | if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) { |
| 1887 | *errors |= UTF8_GOT_SUPER; |
| 1888 | } |
| 1889 | if (flags |
| 1890 | & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED)) |
| 1891 | { |
| 1892 | *errors |= UTF8_GOT_PERL_EXTENDED; |
| 1893 | } |
| 1894 | |
| 1895 | /* Disallow if any of the three categories say to */ |
| 1896 | if ( ! (flags & UTF8_ALLOW_OVERFLOW) |
| 1897 | || (flags & ( UTF8_DISALLOW_SUPER |
| 1898 | |UTF8_DISALLOW_PERL_EXTENDED))) |
| 1899 | { |
| 1900 | disallowed = TRUE; |
| 1901 | } |
| 1902 | |
| 1903 | /* Likewise, warn if any say to */ |
| 1904 | if ( ! (flags & UTF8_ALLOW_OVERFLOW) |
| 1905 | || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED))) |
| 1906 | { |
| 1907 | |
| 1908 | /* The warnings code explicitly says it doesn't handle the |
| 1909 | * case of packWARN2 and two categories which have |
| 1910 | * parent-child relationship. Even if it works now to |
| 1911 | * raise the warning if either is enabled, it wouldn't |
| 1912 | * necessarily do so in the future. We output (only) the |
| 1913 | * most dire warning */ |
| 1914 | if (! (flags & UTF8_CHECK_ONLY)) { |
| 1915 | if (msgs || ckWARN_d(WARN_UTF8)) { |
| 1916 | pack_warn = packWARN(WARN_UTF8); |
| 1917 | } |
| 1918 | else if (msgs || ckWARN_d(WARN_NON_UNICODE)) { |
| 1919 | pack_warn = packWARN(WARN_NON_UNICODE); |
| 1920 | } |
| 1921 | if (pack_warn) { |
| 1922 | message = Perl_form(aTHX_ "%s: %s (overflows)", |
| 1923 | malformed_text, |
| 1924 | _byte_dump_string(s0, curlen, 0)); |
| 1925 | this_flag_bit = UTF8_GOT_OVERFLOW; |
| 1926 | } |
| 1927 | } |
| 1928 | } |
| 1929 | } |
| 1930 | else if (possible_problems & UTF8_GOT_EMPTY) { |
| 1931 | possible_problems &= ~UTF8_GOT_EMPTY; |
| 1932 | *errors |= UTF8_GOT_EMPTY; |
| 1933 | |
| 1934 | if (! (flags & UTF8_ALLOW_EMPTY)) { |
| 1935 | |
| 1936 | /* This so-called malformation is now treated as a bug in |
| 1937 | * the caller. If you have nothing to decode, skip calling |
| 1938 | * this function */ |
| 1939 | assert(0); |
| 1940 | |
| 1941 | disallowed = TRUE; |
| 1942 | if ( (msgs |
| 1943 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) |
| 1944 | { |
| 1945 | pack_warn = packWARN(WARN_UTF8); |
| 1946 | message = Perl_form(aTHX_ "%s (empty string)", |
| 1947 | malformed_text); |
| 1948 | this_flag_bit = UTF8_GOT_EMPTY; |
| 1949 | } |
| 1950 | } |
| 1951 | } |
| 1952 | else if (possible_problems & UTF8_GOT_CONTINUATION) { |
| 1953 | possible_problems &= ~UTF8_GOT_CONTINUATION; |
| 1954 | *errors |= UTF8_GOT_CONTINUATION; |
| 1955 | |
| 1956 | if (! (flags & UTF8_ALLOW_CONTINUATION)) { |
| 1957 | disallowed = TRUE; |
| 1958 | if (( msgs |
| 1959 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) |
| 1960 | { |
| 1961 | pack_warn = packWARN(WARN_UTF8); |
| 1962 | message = Perl_form(aTHX_ |
| 1963 | "%s: %s (unexpected continuation byte 0x%02x," |
| 1964 | " with no preceding start byte)", |
| 1965 | malformed_text, |
| 1966 | _byte_dump_string(s0, 1, 0), *s0); |
| 1967 | this_flag_bit = UTF8_GOT_CONTINUATION; |
| 1968 | } |
| 1969 | } |
| 1970 | } |
| 1971 | else if (possible_problems & UTF8_GOT_SHORT) { |
| 1972 | possible_problems &= ~UTF8_GOT_SHORT; |
| 1973 | *errors |= UTF8_GOT_SHORT; |
| 1974 | |
| 1975 | if (! (flags & UTF8_ALLOW_SHORT)) { |
| 1976 | disallowed = TRUE; |
| 1977 | if (( msgs |
| 1978 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) |
| 1979 | { |
| 1980 | pack_warn = packWARN(WARN_UTF8); |
| 1981 | message = Perl_form(aTHX_ |
| 1982 | "%s: %s (too short; %d byte%s available, need %d)", |
| 1983 | malformed_text, |
| 1984 | _byte_dump_string(s0, send - s0, 0), |
| 1985 | (int)avail_len, |
| 1986 | avail_len == 1 ? "" : "s", |
| 1987 | (int)expectlen); |
| 1988 | this_flag_bit = UTF8_GOT_SHORT; |
| 1989 | } |
| 1990 | } |
| 1991 | |
| 1992 | } |
| 1993 | else if (possible_problems & UTF8_GOT_NON_CONTINUATION) { |
| 1994 | possible_problems &= ~UTF8_GOT_NON_CONTINUATION; |
| 1995 | *errors |= UTF8_GOT_NON_CONTINUATION; |
| 1996 | |
| 1997 | if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) { |
| 1998 | disallowed = TRUE; |
| 1999 | if (( msgs |
| 2000 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) |
| 2001 | { |
| 2002 | |
| 2003 | /* If we don't know for sure that the input length is |
| 2004 | * valid, avoid as much as possible reading past the |
| 2005 | * end of the buffer */ |
| 2006 | int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN) |
| 2007 | ? (int) (s - s0) |
| 2008 | : (int) (send - s0); |
| 2009 | pack_warn = packWARN(WARN_UTF8); |
| 2010 | message = Perl_form(aTHX_ "%s", |
| 2011 | unexpected_non_continuation_text(s0, |
| 2012 | printlen, |
| 2013 | s - s0, |
| 2014 | (int) expectlen)); |
| 2015 | this_flag_bit = UTF8_GOT_NON_CONTINUATION; |
| 2016 | } |
| 2017 | } |
| 2018 | } |
| 2019 | else if (possible_problems & UTF8_GOT_SURROGATE) { |
| 2020 | possible_problems &= ~UTF8_GOT_SURROGATE; |
| 2021 | |
| 2022 | if (flags & UTF8_WARN_SURROGATE) { |
| 2023 | *errors |= UTF8_GOT_SURROGATE; |
| 2024 | |
| 2025 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 2026 | && (msgs || ckWARN_d(WARN_SURROGATE))) |
| 2027 | { |
| 2028 | pack_warn = packWARN(WARN_SURROGATE); |
| 2029 | |
| 2030 | /* These are the only errors that can occur with a |
| 2031 | * surrogate when the 'uv' isn't valid */ |
| 2032 | if (orig_problems & UTF8_GOT_TOO_SHORT) { |
| 2033 | message = Perl_form(aTHX_ |
| 2034 | "UTF-16 surrogate (any UTF-8 sequence that" |
| 2035 | " starts with \"%s\" is for a surrogate)", |
| 2036 | _byte_dump_string(s0, curlen, 0)); |
| 2037 | } |
| 2038 | else { |
| 2039 | message = Perl_form(aTHX_ surrogate_cp_format, uv); |
| 2040 | } |
| 2041 | this_flag_bit = UTF8_GOT_SURROGATE; |
| 2042 | } |
| 2043 | } |
| 2044 | |
| 2045 | if (flags & UTF8_DISALLOW_SURROGATE) { |
| 2046 | disallowed = TRUE; |
| 2047 | *errors |= UTF8_GOT_SURROGATE; |
| 2048 | } |
| 2049 | } |
| 2050 | else if (possible_problems & UTF8_GOT_SUPER) { |
| 2051 | possible_problems &= ~UTF8_GOT_SUPER; |
| 2052 | |
| 2053 | if (flags & UTF8_WARN_SUPER) { |
| 2054 | *errors |= UTF8_GOT_SUPER; |
| 2055 | |
| 2056 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 2057 | && (msgs || ckWARN_d(WARN_NON_UNICODE))) |
| 2058 | { |
| 2059 | pack_warn = packWARN(WARN_NON_UNICODE); |
| 2060 | |
| 2061 | if (orig_problems & UTF8_GOT_TOO_SHORT) { |
| 2062 | message = Perl_form(aTHX_ |
| 2063 | "Any UTF-8 sequence that starts with" |
| 2064 | " \"%s\" is for a non-Unicode code point," |
| 2065 | " may not be portable", |
| 2066 | _byte_dump_string(s0, curlen, 0)); |
| 2067 | } |
| 2068 | else { |
| 2069 | message = Perl_form(aTHX_ super_cp_format, uv); |
| 2070 | } |
| 2071 | this_flag_bit = UTF8_GOT_SUPER; |
| 2072 | } |
| 2073 | } |
| 2074 | |
| 2075 | /* Test for Perl's extended UTF-8 after the regular SUPER ones, |
| 2076 | * and before possibly bailing out, so that the more dire |
| 2077 | * warning will override the regular one. */ |
| 2078 | if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) { |
| 2079 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 2080 | && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER)) |
| 2081 | && (msgs || ( ckWARN_d(WARN_NON_UNICODE) |
| 2082 | || ckWARN(WARN_PORTABLE)))) |
| 2083 | { |
| 2084 | pack_warn = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE); |
| 2085 | |
| 2086 | /* If it is an overlong that evaluates to a code point |
| 2087 | * that doesn't have to use the Perl extended UTF-8, it |
| 2088 | * still used it, and so we output a message that |
| 2089 | * doesn't refer to the code point. The same is true |
| 2090 | * if there was a SHORT malformation where the code |
| 2091 | * point is not valid. In that case, 'uv' will have |
| 2092 | * been set to the REPLACEMENT CHAR, and the message |
| 2093 | * below without the code point in it will be selected |
| 2094 | * */ |
| 2095 | if (UNICODE_IS_PERL_EXTENDED(uv)) { |
| 2096 | message = Perl_form(aTHX_ |
| 2097 | PL_extended_cp_format, uv); |
| 2098 | } |
| 2099 | else { |
| 2100 | message = Perl_form(aTHX_ |
| 2101 | "Any UTF-8 sequence that starts with" |
| 2102 | " \"%s\" is a Perl extension, and" |
| 2103 | " so is not portable", |
| 2104 | _byte_dump_string(s0, curlen, 0)); |
| 2105 | } |
| 2106 | this_flag_bit = UTF8_GOT_PERL_EXTENDED; |
| 2107 | } |
| 2108 | |
| 2109 | if (flags & ( UTF8_WARN_PERL_EXTENDED |
| 2110 | |UTF8_DISALLOW_PERL_EXTENDED)) |
| 2111 | { |
| 2112 | *errors |= UTF8_GOT_PERL_EXTENDED; |
| 2113 | |
| 2114 | if (flags & UTF8_DISALLOW_PERL_EXTENDED) { |
| 2115 | disallowed = TRUE; |
| 2116 | } |
| 2117 | } |
| 2118 | } |
| 2119 | |
| 2120 | if (flags & UTF8_DISALLOW_SUPER) { |
| 2121 | *errors |= UTF8_GOT_SUPER; |
| 2122 | disallowed = TRUE; |
| 2123 | } |
| 2124 | } |
| 2125 | else if (possible_problems & UTF8_GOT_NONCHAR) { |
| 2126 | possible_problems &= ~UTF8_GOT_NONCHAR; |
| 2127 | |
| 2128 | if (flags & UTF8_WARN_NONCHAR) { |
| 2129 | *errors |= UTF8_GOT_NONCHAR; |
| 2130 | |
| 2131 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 2132 | && (msgs || ckWARN_d(WARN_NONCHAR))) |
| 2133 | { |
| 2134 | /* The code above should have guaranteed that we don't |
| 2135 | * get here with errors other than overlong */ |
| 2136 | assert (! (orig_problems |
| 2137 | & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR))); |
| 2138 | |
| 2139 | pack_warn = packWARN(WARN_NONCHAR); |
| 2140 | message = Perl_form(aTHX_ nonchar_cp_format, uv); |
| 2141 | this_flag_bit = UTF8_GOT_NONCHAR; |
| 2142 | } |
| 2143 | } |
| 2144 | |
| 2145 | if (flags & UTF8_DISALLOW_NONCHAR) { |
| 2146 | disallowed = TRUE; |
| 2147 | *errors |= UTF8_GOT_NONCHAR; |
| 2148 | } |
| 2149 | } |
| 2150 | else if (possible_problems & UTF8_GOT_LONG) { |
| 2151 | possible_problems &= ~UTF8_GOT_LONG; |
| 2152 | *errors |= UTF8_GOT_LONG; |
| 2153 | |
| 2154 | if (flags & UTF8_ALLOW_LONG) { |
| 2155 | |
| 2156 | /* We don't allow the actual overlong value, unless the |
| 2157 | * special extra bit is also set */ |
| 2158 | if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE |
| 2159 | & ~UTF8_ALLOW_LONG))) |
| 2160 | { |
| 2161 | uv = UNICODE_REPLACEMENT; |
| 2162 | } |
| 2163 | } |
| 2164 | else { |
| 2165 | disallowed = TRUE; |
| 2166 | |
| 2167 | if (( msgs |
| 2168 | || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) |
| 2169 | { |
| 2170 | pack_warn = packWARN(WARN_UTF8); |
| 2171 | |
| 2172 | /* These error types cause 'uv' to be something that |
| 2173 | * isn't what was intended, so can't use it in the |
| 2174 | * message. The other error types either can't |
| 2175 | * generate an overlong, or else the 'uv' is valid */ |
| 2176 | if (orig_problems & |
| 2177 | (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW)) |
| 2178 | { |
| 2179 | message = Perl_form(aTHX_ |
| 2180 | "%s: %s (any UTF-8 sequence that starts" |
| 2181 | " with \"%s\" is overlong which can and" |
| 2182 | " should be represented with a" |
| 2183 | " different, shorter sequence)", |
| 2184 | malformed_text, |
| 2185 | _byte_dump_string(s0, send - s0, 0), |
| 2186 | _byte_dump_string(s0, curlen, 0)); |
| 2187 | } |
| 2188 | else { |
| 2189 | U8 tmpbuf[UTF8_MAXBYTES+1]; |
| 2190 | const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf, |
| 2191 | uv, 0); |
| 2192 | /* Don't use U+ for non-Unicode code points, which |
| 2193 | * includes those in the Latin1 range */ |
| 2194 | const char * preface = ( uv > PERL_UNICODE_MAX |
| 2195 | #ifdef EBCDIC |
| 2196 | || uv <= 0xFF |
| 2197 | #endif |
| 2198 | ) |
| 2199 | ? "0x" |
| 2200 | : "U+"; |
| 2201 | message = Perl_form(aTHX_ |
| 2202 | "%s: %s (overlong; instead use %s to represent" |
| 2203 | " %s%0*" UVXf ")", |
| 2204 | malformed_text, |
| 2205 | _byte_dump_string(s0, send - s0, 0), |
| 2206 | _byte_dump_string(tmpbuf, e - tmpbuf, 0), |
| 2207 | preface, |
| 2208 | ((uv < 256) ? 2 : 4), /* Field width of 2 for |
| 2209 | small code points */ |
| 2210 | UNI_TO_NATIVE(uv)); |
| 2211 | } |
| 2212 | this_flag_bit = UTF8_GOT_LONG; |
| 2213 | } |
| 2214 | } |
| 2215 | } /* End of looking through the possible flags */ |
| 2216 | |
| 2217 | /* Display the message (if any) for the problem being handled in |
| 2218 | * this iteration of the loop */ |
| 2219 | if (message) { |
| 2220 | if (msgs) { |
| 2221 | assert(this_flag_bit); |
| 2222 | |
| 2223 | if (*msgs == NULL) { |
| 2224 | *msgs = newAV(); |
| 2225 | } |
| 2226 | |
| 2227 | av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message, |
| 2228 | pack_warn, |
| 2229 | this_flag_bit))); |
| 2230 | } |
| 2231 | else if (PL_op) |
| 2232 | Perl_warner(aTHX_ pack_warn, "%s in %s", message, |
| 2233 | OP_DESC(PL_op)); |
| 2234 | else |
| 2235 | Perl_warner(aTHX_ pack_warn, "%s", message); |
| 2236 | } |
| 2237 | } /* End of 'while (possible_problems)' */ |
| 2238 | |
| 2239 | /* Since there was a possible problem, the returned length may need to |
| 2240 | * be changed from the one stored at the beginning of this function. |
| 2241 | * Instead of trying to figure out if that's needed, just do it. */ |
| 2242 | if (retlen) { |
| 2243 | *retlen = curlen; |
| 2244 | } |
| 2245 | |
| 2246 | if (disallowed) { |
| 2247 | if (flags & UTF8_CHECK_ONLY && retlen) { |
| 2248 | *retlen = ((STRLEN) -1); |
| 2249 | } |
| 2250 | return 0; |
| 2251 | } |
| 2252 | } |
| 2253 | |
| 2254 | return UNI_TO_NATIVE(uv); |
| 2255 | } |
| 2256 | |
| 2257 | /* |
| 2258 | =for apidoc utf8_to_uvchr_buf |
| 2259 | |
| 2260 | Returns the native code point of the first character in the string C<s> which |
| 2261 | is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. |
| 2262 | C<*retlen> will be set to the length, in bytes, of that character. |
| 2263 | |
| 2264 | If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are |
| 2265 | enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't |
| 2266 | C<NULL>) to -1. If those warnings are off, the computed value, if well-defined |
| 2267 | (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and |
| 2268 | C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is |
| 2269 | the next possible position in C<s> that could begin a non-malformed character. |
| 2270 | See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is |
| 2271 | returned. |
| 2272 | |
| 2273 | =cut |
| 2274 | |
| 2275 | Also implemented as a macro in utf8.h |
| 2276 | |
| 2277 | */ |
| 2278 | |
| 2279 | |
| 2280 | UV |
| 2281 | Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) |
| 2282 | { |
| 2283 | PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF; |
| 2284 | |
| 2285 | return utf8_to_uvchr_buf_helper(s, send, retlen); |
| 2286 | } |
| 2287 | |
| 2288 | /* This is marked as deprecated |
| 2289 | * |
| 2290 | =for apidoc utf8_to_uvuni_buf |
| 2291 | |
| 2292 | Only in very rare circumstances should code need to be dealing in Unicode |
| 2293 | (as opposed to native) code points. In those few cases, use |
| 2294 | C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead. |
| 2295 | If you are not absolutely sure this is one of those cases, then assume it isn't |
| 2296 | and use plain C<utf8_to_uvchr_buf> instead. |
| 2297 | |
| 2298 | Returns the Unicode (not-native) code point of the first character in the |
| 2299 | string C<s> which |
| 2300 | is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. |
| 2301 | C<retlen> will be set to the length, in bytes, of that character. |
| 2302 | |
| 2303 | If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are |
| 2304 | enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't |
| 2305 | NULL) to -1. If those warnings are off, the computed value if well-defined (or |
| 2306 | the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> |
| 2307 | is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the |
| 2308 | next possible position in C<s> that could begin a non-malformed character. |
| 2309 | See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is |
| 2310 | returned. |
| 2311 | |
| 2312 | =cut |
| 2313 | */ |
| 2314 | |
| 2315 | UV |
| 2316 | Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) |
| 2317 | { |
| 2318 | PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF; |
| 2319 | |
| 2320 | assert(send > s); |
| 2321 | |
| 2322 | return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen)); |
| 2323 | } |
| 2324 | |
| 2325 | /* |
| 2326 | =for apidoc utf8_length |
| 2327 | |
| 2328 | Returns the number of characters in the sequence of UTF-8-encoded bytes starting |
| 2329 | at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the |
| 2330 | same place, it returns 0 with no warning raised. |
| 2331 | |
| 2332 | If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning |
| 2333 | and returns the number of valid characters. |
| 2334 | |
| 2335 | =cut |
| 2336 | */ |
| 2337 | |
| 2338 | STRLEN |
| 2339 | Perl_utf8_length(pTHX_ const U8 *s, const U8 *e) |
| 2340 | { |
| 2341 | STRLEN len = 0; |
| 2342 | |
| 2343 | PERL_ARGS_ASSERT_UTF8_LENGTH; |
| 2344 | |
| 2345 | /* Note: cannot use UTF8_IS_...() too eagerly here since e.g. |
| 2346 | * the bitops (especially ~) can create illegal UTF-8. |
| 2347 | * In other words: in Perl UTF-8 is not just for Unicode. */ |
| 2348 | |
| 2349 | if (UNLIKELY(e < s)) |
| 2350 | goto warn_and_return; |
| 2351 | while (s < e) { |
| 2352 | s += UTF8SKIP(s); |
| 2353 | len++; |
| 2354 | } |
| 2355 | |
| 2356 | if (UNLIKELY(e != s)) { |
| 2357 | len--; |
| 2358 | warn_and_return: |
| 2359 | if (PL_op) |
| 2360 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 2361 | "%s in %s", unees, OP_DESC(PL_op)); |
| 2362 | else |
| 2363 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); |
| 2364 | } |
| 2365 | |
| 2366 | return len; |
| 2367 | } |
| 2368 | |
| 2369 | /* |
| 2370 | =for apidoc bytes_cmp_utf8 |
| 2371 | |
| 2372 | Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the |
| 2373 | sequence of characters (stored as UTF-8) |
| 2374 | in C<u>, C<ulen>. Returns 0 if they are |
| 2375 | equal, -1 or -2 if the first string is less than the second string, +1 or +2 |
| 2376 | if the first string is greater than the second string. |
| 2377 | |
| 2378 | -1 or +1 is returned if the shorter string was identical to the start of the |
| 2379 | longer string. -2 or +2 is returned if |
| 2380 | there was a difference between characters |
| 2381 | within the strings. |
| 2382 | |
| 2383 | =cut |
| 2384 | */ |
| 2385 | |
| 2386 | int |
| 2387 | Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen) |
| 2388 | { |
| 2389 | const U8 *const bend = b + blen; |
| 2390 | const U8 *const uend = u + ulen; |
| 2391 | |
| 2392 | PERL_ARGS_ASSERT_BYTES_CMP_UTF8; |
| 2393 | |
| 2394 | while (b < bend && u < uend) { |
| 2395 | U8 c = *u++; |
| 2396 | if (!UTF8_IS_INVARIANT(c)) { |
| 2397 | if (UTF8_IS_DOWNGRADEABLE_START(c)) { |
| 2398 | if (u < uend) { |
| 2399 | U8 c1 = *u++; |
| 2400 | if (UTF8_IS_CONTINUATION(c1)) { |
| 2401 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1); |
| 2402 | } else { |
| 2403 | /* diag_listed_as: Malformed UTF-8 character%s */ |
| 2404 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 2405 | "%s %s%s", |
| 2406 | unexpected_non_continuation_text(u - 2, 2, 1, 2), |
| 2407 | PL_op ? " in " : "", |
| 2408 | PL_op ? OP_DESC(PL_op) : ""); |
| 2409 | return -2; |
| 2410 | } |
| 2411 | } else { |
| 2412 | if (PL_op) |
| 2413 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 2414 | "%s in %s", unees, OP_DESC(PL_op)); |
| 2415 | else |
| 2416 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); |
| 2417 | return -2; /* Really want to return undef :-) */ |
| 2418 | } |
| 2419 | } else { |
| 2420 | return -2; |
| 2421 | } |
| 2422 | } |
| 2423 | if (*b != c) { |
| 2424 | return *b < c ? -2 : +2; |
| 2425 | } |
| 2426 | ++b; |
| 2427 | } |
| 2428 | |
| 2429 | if (b == bend && u == uend) |
| 2430 | return 0; |
| 2431 | |
| 2432 | return b < bend ? +1 : -1; |
| 2433 | } |
| 2434 | |
| 2435 | /* |
| 2436 | =for apidoc utf8_to_bytes |
| 2437 | |
| 2438 | Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding. |
| 2439 | Unlike L</bytes_to_utf8>, this over-writes the original string, and |
| 2440 | updates C<*lenp> to contain the new length. |
| 2441 | Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1. |
| 2442 | |
| 2443 | Upon successful return, the number of variants in the string can be computed by |
| 2444 | having saved the value of C<*lenp> before the call, and subtracting the |
| 2445 | after-call value of C<*lenp> from it. |
| 2446 | |
| 2447 | If you need a copy of the string, see L</bytes_from_utf8>. |
| 2448 | |
| 2449 | =cut |
| 2450 | */ |
| 2451 | |
| 2452 | U8 * |
| 2453 | Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp) |
| 2454 | { |
| 2455 | U8 * first_variant; |
| 2456 | |
| 2457 | PERL_ARGS_ASSERT_UTF8_TO_BYTES; |
| 2458 | PERL_UNUSED_CONTEXT; |
| 2459 | |
| 2460 | /* This is a no-op if no variants at all in the input */ |
| 2461 | if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) { |
| 2462 | return s; |
| 2463 | } |
| 2464 | |
| 2465 | { |
| 2466 | U8 * const save = s; |
| 2467 | U8 * const send = s + *lenp; |
| 2468 | U8 * d; |
| 2469 | |
| 2470 | /* Nothing before the first variant needs to be changed, so start the real |
| 2471 | * work there */ |
| 2472 | s = first_variant; |
| 2473 | while (s < send) { |
| 2474 | if (! UTF8_IS_INVARIANT(*s)) { |
| 2475 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { |
| 2476 | *lenp = ((STRLEN) -1); |
| 2477 | return 0; |
| 2478 | } |
| 2479 | s++; |
| 2480 | } |
| 2481 | s++; |
| 2482 | } |
| 2483 | |
| 2484 | /* Is downgradable, so do it */ |
| 2485 | d = s = first_variant; |
| 2486 | while (s < send) { |
| 2487 | U8 c = *s++; |
| 2488 | if (! UVCHR_IS_INVARIANT(c)) { |
| 2489 | /* Then it is two-byte encoded */ |
| 2490 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); |
| 2491 | s++; |
| 2492 | } |
| 2493 | *d++ = c; |
| 2494 | } |
| 2495 | *d = '\0'; |
| 2496 | *lenp = d - save; |
| 2497 | |
| 2498 | return save; |
| 2499 | } |
| 2500 | } |
| 2501 | |
| 2502 | /* |
| 2503 | =for apidoc bytes_from_utf8 |
| 2504 | |
| 2505 | Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native |
| 2506 | byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is |
| 2507 | actually encoded in UTF-8. |
| 2508 | |
| 2509 | Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of |
| 2510 | the input string. |
| 2511 | |
| 2512 | Do nothing if C<*is_utf8p> is 0, or if there are code points in the string |
| 2513 | not expressible in native byte encoding. In these cases, C<*is_utf8p> and |
| 2514 | C<*lenp> are unchanged, and the return value is the original C<s>. |
| 2515 | |
| 2516 | Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a |
| 2517 | newly created string containing a downgraded copy of C<s>, and whose length is |
| 2518 | returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The |
| 2519 | caller is responsible for arranging for the memory used by this string to get |
| 2520 | freed. |
| 2521 | |
| 2522 | Upon successful return, the number of variants in the string can be computed by |
| 2523 | having saved the value of C<*lenp> before the call, and subtracting the |
| 2524 | after-call value of C<*lenp> from it. |
| 2525 | |
| 2526 | =cut |
| 2527 | |
| 2528 | There is a macro that avoids this function call, but this is retained for |
| 2529 | anyone who calls it with the Perl_ prefix */ |
| 2530 | |
| 2531 | U8 * |
| 2532 | Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p) |
| 2533 | { |
| 2534 | PERL_ARGS_ASSERT_BYTES_FROM_UTF8; |
| 2535 | PERL_UNUSED_CONTEXT; |
| 2536 | |
| 2537 | return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL); |
| 2538 | } |
| 2539 | |
| 2540 | /* |
| 2541 | =for apidoc bytes_from_utf8_loc |
| 2542 | |
| 2543 | Like C<L<perlapi/bytes_from_utf8>()>, but takes an extra parameter, a pointer |
| 2544 | to where to store the location of the first character in C<"s"> that cannot be |
| 2545 | converted to non-UTF8. |
| 2546 | |
| 2547 | If that parameter is C<NULL>, this function behaves identically to |
| 2548 | C<bytes_from_utf8>. |
| 2549 | |
| 2550 | Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to |
| 2551 | C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>. |
| 2552 | |
| 2553 | Otherwise, the function returns a newly created C<NUL>-terminated string |
| 2554 | containing the non-UTF8 equivalent of the convertible first portion of |
| 2555 | C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>. |
| 2556 | If the entire input string was converted, C<*is_utf8p> is set to a FALSE value, |
| 2557 | and C<*first_non_downgradable> is set to C<NULL>. |
| 2558 | |
| 2559 | Otherwise, C<*first_non_downgradable> is set to point to the first byte of the |
| 2560 | first character in the original string that wasn't converted. C<*is_utf8p> is |
| 2561 | unchanged. Note that the new string may have length 0. |
| 2562 | |
| 2563 | Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and |
| 2564 | C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and |
| 2565 | converts as many characters in it as possible stopping at the first one it |
| 2566 | finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is |
| 2567 | set to point to that. The function returns the portion that could be converted |
| 2568 | in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length, |
| 2569 | not including the terminating C<NUL>. If the very first character in the |
| 2570 | original could not be converted, C<*lenp> will be 0, and the new string will |
| 2571 | contain just a single C<NUL>. If the entire input string was converted, |
| 2572 | C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>. |
| 2573 | |
| 2574 | Upon successful return, the number of variants in the converted portion of the |
| 2575 | string can be computed by having saved the value of C<*lenp> before the call, |
| 2576 | and subtracting the after-call value of C<*lenp> from it. |
| 2577 | |
| 2578 | =cut |
| 2579 | |
| 2580 | |
| 2581 | */ |
| 2582 | |
| 2583 | U8 * |
| 2584 | Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted) |
| 2585 | { |
| 2586 | U8 *d; |
| 2587 | const U8 *original = s; |
| 2588 | U8 *converted_start; |
| 2589 | const U8 *send = s + *lenp; |
| 2590 | |
| 2591 | PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC; |
| 2592 | |
| 2593 | if (! *is_utf8p) { |
| 2594 | if (first_unconverted) { |
| 2595 | *first_unconverted = NULL; |
| 2596 | } |
| 2597 | |
| 2598 | return (U8 *) original; |
| 2599 | } |
| 2600 | |
| 2601 | Newx(d, (*lenp) + 1, U8); |
| 2602 | |
| 2603 | converted_start = d; |
| 2604 | while (s < send) { |
| 2605 | U8 c = *s++; |
| 2606 | if (! UTF8_IS_INVARIANT(c)) { |
| 2607 | |
| 2608 | /* Then it is multi-byte encoded. If the code point is above 0xFF, |
| 2609 | * have to stop now */ |
| 2610 | if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) { |
| 2611 | if (first_unconverted) { |
| 2612 | *first_unconverted = s - 1; |
| 2613 | goto finish_and_return; |
| 2614 | } |
| 2615 | else { |
| 2616 | Safefree(converted_start); |
| 2617 | return (U8 *) original; |
| 2618 | } |
| 2619 | } |
| 2620 | |
| 2621 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); |
| 2622 | s++; |
| 2623 | } |
| 2624 | *d++ = c; |
| 2625 | } |
| 2626 | |
| 2627 | /* Here, converted the whole of the input */ |
| 2628 | *is_utf8p = FALSE; |
| 2629 | if (first_unconverted) { |
| 2630 | *first_unconverted = NULL; |
| 2631 | } |
| 2632 | |
| 2633 | finish_and_return: |
| 2634 | *d = '\0'; |
| 2635 | *lenp = d - converted_start; |
| 2636 | |
| 2637 | /* Trim unused space */ |
| 2638 | Renew(converted_start, *lenp + 1, U8); |
| 2639 | |
| 2640 | return converted_start; |
| 2641 | } |
| 2642 | |
| 2643 | /* |
| 2644 | =for apidoc bytes_to_utf8 |
| 2645 | |
| 2646 | Converts a string C<s> of length C<*lenp> bytes from the native encoding into |
| 2647 | UTF-8. |
| 2648 | Returns a pointer to the newly-created string, and sets C<*lenp> to |
| 2649 | reflect the new length in bytes. The caller is responsible for arranging for |
| 2650 | the memory used by this string to get freed. |
| 2651 | |
| 2652 | Upon successful return, the number of variants in the string can be computed by |
| 2653 | having saved the value of C<*lenp> before the call, and subtracting it from the |
| 2654 | after-call value of C<*lenp>. |
| 2655 | |
| 2656 | A C<NUL> character will be written after the end of the string. |
| 2657 | |
| 2658 | If you want to convert to UTF-8 from encodings other than |
| 2659 | the native (Latin1 or EBCDIC), |
| 2660 | see L</sv_recode_to_utf8>(). |
| 2661 | |
| 2662 | =cut |
| 2663 | */ |
| 2664 | |
| 2665 | U8* |
| 2666 | Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp) |
| 2667 | { |
| 2668 | const U8 * const send = s + (*lenp); |
| 2669 | U8 *d; |
| 2670 | U8 *dst; |
| 2671 | |
| 2672 | PERL_ARGS_ASSERT_BYTES_TO_UTF8; |
| 2673 | PERL_UNUSED_CONTEXT; |
| 2674 | |
| 2675 | /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */ |
| 2676 | Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8); |
| 2677 | dst = d; |
| 2678 | |
| 2679 | while (s < send) { |
| 2680 | append_utf8_from_native_byte(*s, &d); |
| 2681 | s++; |
| 2682 | } |
| 2683 | |
| 2684 | *d = '\0'; |
| 2685 | *lenp = d-dst; |
| 2686 | |
| 2687 | return dst; |
| 2688 | } |
| 2689 | |
| 2690 | /* |
| 2691 | * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian), |
| 2692 | * use utf16_to_utf8_reversed(). |
| 2693 | * |
| 2694 | * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes. |
| 2695 | * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes. |
| 2696 | * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes. |
| 2697 | * |
| 2698 | * These functions don't check for overflow. The worst case is every code |
| 2699 | * point in the input is 2 bytes, and requires 4 bytes on output. (If the code |
| 2700 | * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the |
| 2701 | * destination must be pre-extended to 2 times the source length. |
| 2702 | * |
| 2703 | * Do not use in-place. We optimize for native, for obvious reasons. */ |
| 2704 | |
| 2705 | U8* |
| 2706 | Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen) |
| 2707 | { |
| 2708 | U8* pend; |
| 2709 | U8* dstart = d; |
| 2710 | |
| 2711 | PERL_ARGS_ASSERT_UTF16_TO_UTF8; |
| 2712 | |
| 2713 | if (bytelen & 1) |
| 2714 | Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, |
| 2715 | (UV)bytelen); |
| 2716 | |
| 2717 | pend = p + bytelen; |
| 2718 | |
| 2719 | while (p < pend) { |
| 2720 | UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */ |
| 2721 | p += 2; |
| 2722 | if (OFFUNI_IS_INVARIANT(uv)) { |
| 2723 | *d++ = LATIN1_TO_NATIVE((U8) uv); |
| 2724 | continue; |
| 2725 | } |
| 2726 | if (uv <= MAX_UTF8_TWO_BYTE) { |
| 2727 | *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv)); |
| 2728 | *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv)); |
| 2729 | continue; |
| 2730 | } |
| 2731 | |
| 2732 | #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST |
| 2733 | #define LAST_HIGH_SURROGATE 0xDBFF |
| 2734 | #define FIRST_LOW_SURROGATE 0xDC00 |
| 2735 | #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST |
| 2736 | #define FIRST_IN_PLANE1 0x10000 |
| 2737 | |
| 2738 | /* This assumes that most uses will be in the first Unicode plane, not |
| 2739 | * needing surrogates */ |
| 2740 | if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST, |
| 2741 | UNICODE_SURROGATE_LAST))) |
| 2742 | { |
| 2743 | if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) { |
| 2744 | Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); |
| 2745 | } |
| 2746 | else { |
| 2747 | UV low = (p[0] << 8) + p[1]; |
| 2748 | if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE, |
| 2749 | LAST_LOW_SURROGATE))) |
| 2750 | { |
| 2751 | Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); |
| 2752 | } |
| 2753 | p += 2; |
| 2754 | uv = ((uv - FIRST_HIGH_SURROGATE) << 10) |
| 2755 | + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1; |
| 2756 | } |
| 2757 | } |
| 2758 | #ifdef EBCDIC |
| 2759 | d = uvoffuni_to_utf8_flags(d, uv, 0); |
| 2760 | #else |
| 2761 | if (uv < FIRST_IN_PLANE1) { |
| 2762 | *d++ = (U8)(( uv >> 12) | 0xe0); |
| 2763 | *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); |
| 2764 | *d++ = (U8)(( uv & 0x3f) | 0x80); |
| 2765 | continue; |
| 2766 | } |
| 2767 | else { |
| 2768 | *d++ = (U8)(( uv >> 18) | 0xf0); |
| 2769 | *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80); |
| 2770 | *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); |
| 2771 | *d++ = (U8)(( uv & 0x3f) | 0x80); |
| 2772 | continue; |
| 2773 | } |
| 2774 | #endif |
| 2775 | } |
| 2776 | *newlen = d - dstart; |
| 2777 | return d; |
| 2778 | } |
| 2779 | |
| 2780 | /* Note: this one is slightly destructive of the source. */ |
| 2781 | |
| 2782 | U8* |
| 2783 | Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen) |
| 2784 | { |
| 2785 | U8* s = (U8*)p; |
| 2786 | U8* const send = s + bytelen; |
| 2787 | |
| 2788 | PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED; |
| 2789 | |
| 2790 | if (bytelen & 1) |
| 2791 | Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf, |
| 2792 | (UV)bytelen); |
| 2793 | |
| 2794 | while (s < send) { |
| 2795 | const U8 tmp = s[0]; |
| 2796 | s[0] = s[1]; |
| 2797 | s[1] = tmp; |
| 2798 | s += 2; |
| 2799 | } |
| 2800 | return utf16_to_utf8(p, d, bytelen, newlen); |
| 2801 | } |
| 2802 | |
| 2803 | bool |
| 2804 | Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c) |
| 2805 | { |
| 2806 | return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c); |
| 2807 | } |
| 2808 | |
| 2809 | bool |
| 2810 | Perl__is_uni_perl_idcont(pTHX_ UV c) |
| 2811 | { |
| 2812 | return _invlist_contains_cp(PL_utf8_perl_idcont, c); |
| 2813 | } |
| 2814 | |
| 2815 | bool |
| 2816 | Perl__is_uni_perl_idstart(pTHX_ UV c) |
| 2817 | { |
| 2818 | return _invlist_contains_cp(PL_utf8_perl_idstart, c); |
| 2819 | } |
| 2820 | |
| 2821 | UV |
| 2822 | Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, |
| 2823 | const char S_or_s) |
| 2824 | { |
| 2825 | /* We have the latin1-range values compiled into the core, so just use |
| 2826 | * those, converting the result to UTF-8. The only difference between upper |
| 2827 | * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is |
| 2828 | * either "SS" or "Ss". Which one to use is passed into the routine in |
| 2829 | * 'S_or_s' to avoid a test */ |
| 2830 | |
| 2831 | UV converted = toUPPER_LATIN1_MOD(c); |
| 2832 | |
| 2833 | PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1; |
| 2834 | |
| 2835 | assert(S_or_s == 'S' || S_or_s == 's'); |
| 2836 | |
| 2837 | if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for |
| 2838 | characters in this range */ |
| 2839 | *p = (U8) converted; |
| 2840 | *lenp = 1; |
| 2841 | return converted; |
| 2842 | } |
| 2843 | |
| 2844 | /* toUPPER_LATIN1_MOD gives the correct results except for three outliers, |
| 2845 | * which it maps to one of them, so as to only have to have one check for |
| 2846 | * it in the main case */ |
| 2847 | if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { |
| 2848 | switch (c) { |
| 2849 | case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: |
| 2850 | converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; |
| 2851 | break; |
| 2852 | case MICRO_SIGN: |
| 2853 | converted = GREEK_CAPITAL_LETTER_MU; |
| 2854 | break; |
| 2855 | #if UNICODE_MAJOR_VERSION > 2 \ |
| 2856 | || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \ |
| 2857 | && UNICODE_DOT_DOT_VERSION >= 8) |
| 2858 | case LATIN_SMALL_LETTER_SHARP_S: |
| 2859 | *(p)++ = 'S'; |
| 2860 | *p = S_or_s; |
| 2861 | *lenp = 2; |
| 2862 | return 'S'; |
| 2863 | #endif |
| 2864 | default: |
| 2865 | Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect" |
| 2866 | " '%c' to map to '%c'", |
| 2867 | c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS); |
| 2868 | NOT_REACHED; /* NOTREACHED */ |
| 2869 | } |
| 2870 | } |
| 2871 | |
| 2872 | *(p)++ = UTF8_TWO_BYTE_HI(converted); |
| 2873 | *p = UTF8_TWO_BYTE_LO(converted); |
| 2874 | *lenp = 2; |
| 2875 | |
| 2876 | return converted; |
| 2877 | } |
| 2878 | |
| 2879 | /* If compiled on an early Unicode version, there may not be auxiliary tables |
| 2880 | * */ |
| 2881 | #ifndef HAS_UC_AUX_TABLES |
| 2882 | # define UC_AUX_TABLE_ptrs NULL |
| 2883 | # define UC_AUX_TABLE_lengths NULL |
| 2884 | #endif |
| 2885 | #ifndef HAS_TC_AUX_TABLES |
| 2886 | # define TC_AUX_TABLE_ptrs NULL |
| 2887 | # define TC_AUX_TABLE_lengths NULL |
| 2888 | #endif |
| 2889 | #ifndef HAS_LC_AUX_TABLES |
| 2890 | # define LC_AUX_TABLE_ptrs NULL |
| 2891 | # define LC_AUX_TABLE_lengths NULL |
| 2892 | #endif |
| 2893 | #ifndef HAS_CF_AUX_TABLES |
| 2894 | # define CF_AUX_TABLE_ptrs NULL |
| 2895 | # define CF_AUX_TABLE_lengths NULL |
| 2896 | #endif |
| 2897 | #ifndef HAS_UC_AUX_TABLES |
| 2898 | # define UC_AUX_TABLE_ptrs NULL |
| 2899 | # define UC_AUX_TABLE_lengths NULL |
| 2900 | #endif |
| 2901 | |
| 2902 | /* Call the function to convert a UTF-8 encoded character to the specified case. |
| 2903 | * Note that there may be more than one character in the result. |
| 2904 | * 's' is a pointer to the first byte of the input character |
| 2905 | * 'd' will be set to the first byte of the string of changed characters. It |
| 2906 | * needs to have space for UTF8_MAXBYTES_CASE+1 bytes |
| 2907 | * 'lenp' will be set to the length in bytes of the string of changed characters |
| 2908 | * |
| 2909 | * The functions return the ordinal of the first character in the string of |
| 2910 | * 'd' */ |
| 2911 | #define CALL_UPPER_CASE(uv, s, d, lenp) \ |
| 2912 | _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \ |
| 2913 | Uppercase_Mapping_invmap, \ |
| 2914 | UC_AUX_TABLE_ptrs, \ |
| 2915 | UC_AUX_TABLE_lengths, \ |
| 2916 | "uppercase") |
| 2917 | #define CALL_TITLE_CASE(uv, s, d, lenp) \ |
| 2918 | _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \ |
| 2919 | Titlecase_Mapping_invmap, \ |
| 2920 | TC_AUX_TABLE_ptrs, \ |
| 2921 | TC_AUX_TABLE_lengths, \ |
| 2922 | "titlecase") |
| 2923 | #define CALL_LOWER_CASE(uv, s, d, lenp) \ |
| 2924 | _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \ |
| 2925 | Lowercase_Mapping_invmap, \ |
| 2926 | LC_AUX_TABLE_ptrs, \ |
| 2927 | LC_AUX_TABLE_lengths, \ |
| 2928 | "lowercase") |
| 2929 | |
| 2930 | |
| 2931 | /* This additionally has the input parameter 'specials', which if non-zero will |
| 2932 | * cause this to use the specials hash for folding (meaning get full case |
| 2933 | * folding); otherwise, when zero, this implies a simple case fold */ |
| 2934 | #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \ |
| 2935 | (specials) \ |
| 2936 | ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \ |
| 2937 | Case_Folding_invmap, \ |
| 2938 | CF_AUX_TABLE_ptrs, \ |
| 2939 | CF_AUX_TABLE_lengths, \ |
| 2940 | "foldcase") \ |
| 2941 | : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \ |
| 2942 | Simple_Case_Folding_invmap, \ |
| 2943 | NULL, NULL, \ |
| 2944 | "foldcase") |
| 2945 | |
| 2946 | UV |
| 2947 | Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp) |
| 2948 | { |
| 2949 | /* Convert the Unicode character whose ordinal is <c> to its uppercase |
| 2950 | * version and store that in UTF-8 in <p> and its length in bytes in <lenp>. |
| 2951 | * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since |
| 2952 | * the changed version may be longer than the original character. |
| 2953 | * |
| 2954 | * The ordinal of the first character of the changed version is returned |
| 2955 | * (but note, as explained above, that there may be more.) */ |
| 2956 | |
| 2957 | PERL_ARGS_ASSERT_TO_UNI_UPPER; |
| 2958 | |
| 2959 | if (c < 256) { |
| 2960 | return _to_upper_title_latin1((U8) c, p, lenp, 'S'); |
| 2961 | } |
| 2962 | |
| 2963 | return CALL_UPPER_CASE(c, NULL, p, lenp); |
| 2964 | } |
| 2965 | |
| 2966 | UV |
| 2967 | Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp) |
| 2968 | { |
| 2969 | PERL_ARGS_ASSERT_TO_UNI_TITLE; |
| 2970 | |
| 2971 | if (c < 256) { |
| 2972 | return _to_upper_title_latin1((U8) c, p, lenp, 's'); |
| 2973 | } |
| 2974 | |
| 2975 | return CALL_TITLE_CASE(c, NULL, p, lenp); |
| 2976 | } |
| 2977 | |
| 2978 | STATIC U8 |
| 2979 | S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy) |
| 2980 | { |
| 2981 | /* We have the latin1-range values compiled into the core, so just use |
| 2982 | * those, converting the result to UTF-8. Since the result is always just |
| 2983 | * one character, we allow <p> to be NULL */ |
| 2984 | |
| 2985 | U8 converted = toLOWER_LATIN1(c); |
| 2986 | |
| 2987 | PERL_UNUSED_ARG(dummy); |
| 2988 | |
| 2989 | if (p != NULL) { |
| 2990 | if (NATIVE_BYTE_IS_INVARIANT(converted)) { |
| 2991 | *p = converted; |
| 2992 | *lenp = 1; |
| 2993 | } |
| 2994 | else { |
| 2995 | /* Result is known to always be < 256, so can use the EIGHT_BIT |
| 2996 | * macros */ |
| 2997 | *p = UTF8_EIGHT_BIT_HI(converted); |
| 2998 | *(p+1) = UTF8_EIGHT_BIT_LO(converted); |
| 2999 | *lenp = 2; |
| 3000 | } |
| 3001 | } |
| 3002 | return converted; |
| 3003 | } |
| 3004 | |
| 3005 | UV |
| 3006 | Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp) |
| 3007 | { |
| 3008 | PERL_ARGS_ASSERT_TO_UNI_LOWER; |
| 3009 | |
| 3010 | if (c < 256) { |
| 3011 | return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ ); |
| 3012 | } |
| 3013 | |
| 3014 | return CALL_LOWER_CASE(c, NULL, p, lenp); |
| 3015 | } |
| 3016 | |
| 3017 | UV |
| 3018 | Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags) |
| 3019 | { |
| 3020 | /* Corresponds to to_lower_latin1(); <flags> bits meanings: |
| 3021 | * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited |
| 3022 | * FOLD_FLAGS_FULL iff full folding is to be used; |
| 3023 | * |
| 3024 | * Not to be used for locale folds |
| 3025 | */ |
| 3026 | |
| 3027 | UV converted; |
| 3028 | |
| 3029 | PERL_ARGS_ASSERT__TO_FOLD_LATIN1; |
| 3030 | |
| 3031 | assert (! (flags & FOLD_FLAGS_LOCALE)); |
| 3032 | |
| 3033 | if (UNLIKELY(c == MICRO_SIGN)) { |
| 3034 | converted = GREEK_SMALL_LETTER_MU; |
| 3035 | } |
| 3036 | #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ |
| 3037 | || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ |
| 3038 | || UNICODE_DOT_DOT_VERSION > 0) |
| 3039 | else if ( (flags & FOLD_FLAGS_FULL) |
| 3040 | && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S)) |
| 3041 | { |
| 3042 | /* If can't cross 127/128 boundary, can't return "ss"; instead return |
| 3043 | * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}") |
| 3044 | * under those circumstances. */ |
| 3045 | if (flags & FOLD_FLAGS_NOMIX_ASCII) { |
| 3046 | *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; |
| 3047 | Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, |
| 3048 | p, *lenp, U8); |
| 3049 | return LATIN_SMALL_LETTER_LONG_S; |
| 3050 | } |
| 3051 | else { |
| 3052 | *(p)++ = 's'; |
| 3053 | *p = 's'; |
| 3054 | *lenp = 2; |
| 3055 | return 's'; |
| 3056 | } |
| 3057 | } |
| 3058 | #endif |
| 3059 | else { /* In this range the fold of all other characters is their lower |
| 3060 | case */ |
| 3061 | converted = toLOWER_LATIN1(c); |
| 3062 | } |
| 3063 | |
| 3064 | if (UVCHR_IS_INVARIANT(converted)) { |
| 3065 | *p = (U8) converted; |
| 3066 | *lenp = 1; |
| 3067 | } |
| 3068 | else { |
| 3069 | *(p)++ = UTF8_TWO_BYTE_HI(converted); |
| 3070 | *p = UTF8_TWO_BYTE_LO(converted); |
| 3071 | *lenp = 2; |
| 3072 | } |
| 3073 | |
| 3074 | return converted; |
| 3075 | } |
| 3076 | |
| 3077 | UV |
| 3078 | Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags) |
| 3079 | { |
| 3080 | |
| 3081 | /* Not currently externally documented, and subject to change |
| 3082 | * <flags> bits meanings: |
| 3083 | * FOLD_FLAGS_FULL iff full folding is to be used; |
| 3084 | * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying |
| 3085 | * locale are to be used. |
| 3086 | * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited |
| 3087 | */ |
| 3088 | |
| 3089 | PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS; |
| 3090 | |
| 3091 | if (flags & FOLD_FLAGS_LOCALE) { |
| 3092 | /* Treat a non-Turkic UTF-8 locale as not being in locale at all, |
| 3093 | * except for potentially warning */ |
| 3094 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 3095 | if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) { |
| 3096 | flags &= ~FOLD_FLAGS_LOCALE; |
| 3097 | } |
| 3098 | else { |
| 3099 | goto needs_full_generality; |
| 3100 | } |
| 3101 | } |
| 3102 | |
| 3103 | if (c < 256) { |
| 3104 | return _to_fold_latin1((U8) c, p, lenp, |
| 3105 | flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); |
| 3106 | } |
| 3107 | |
| 3108 | /* Here, above 255. If no special needs, just use the macro */ |
| 3109 | if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) { |
| 3110 | return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL); |
| 3111 | } |
| 3112 | else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with |
| 3113 | the special flags. */ |
| 3114 | U8 utf8_c[UTF8_MAXBYTES + 1]; |
| 3115 | |
| 3116 | needs_full_generality: |
| 3117 | uvchr_to_utf8(utf8_c, c); |
| 3118 | return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), |
| 3119 | p, lenp, flags); |
| 3120 | } |
| 3121 | } |
| 3122 | |
| 3123 | PERL_STATIC_INLINE bool |
| 3124 | S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e, |
| 3125 | SV* const invlist) |
| 3126 | { |
| 3127 | /* returns a boolean giving whether or not the UTF8-encoded character that |
| 3128 | * starts at <p>, and extending no further than <e - 1> is in the inversion |
| 3129 | * list <invlist>. */ |
| 3130 | |
| 3131 | UV cp = utf8n_to_uvchr(p, e - p, NULL, 0); |
| 3132 | |
| 3133 | PERL_ARGS_ASSERT_IS_UTF8_COMMON; |
| 3134 | |
| 3135 | if (cp == 0 && (p >= e || *p != '\0')) { |
| 3136 | _force_out_malformed_utf8_message(p, e, 0, 1); |
| 3137 | NOT_REACHED; /* NOTREACHED */ |
| 3138 | } |
| 3139 | |
| 3140 | assert(invlist); |
| 3141 | return _invlist_contains_cp(invlist, cp); |
| 3142 | } |
| 3143 | |
| 3144 | #if 0 /* Not currently used, but may be needed in the future */ |
| 3145 | PERLVAR(I, seen_deprecated_macro, HV *) |
| 3146 | |
| 3147 | STATIC void |
| 3148 | S_warn_on_first_deprecated_use(pTHX_ const char * const name, |
| 3149 | const char * const alternative, |
| 3150 | const bool use_locale, |
| 3151 | const char * const file, |
| 3152 | const unsigned line) |
| 3153 | { |
| 3154 | const char * key; |
| 3155 | |
| 3156 | PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE; |
| 3157 | |
| 3158 | if (ckWARN_d(WARN_DEPRECATED)) { |
| 3159 | |
| 3160 | key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line); |
| 3161 | if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) { |
| 3162 | if (! PL_seen_deprecated_macro) { |
| 3163 | PL_seen_deprecated_macro = newHV(); |
| 3164 | } |
| 3165 | if (! hv_store(PL_seen_deprecated_macro, key, |
| 3166 | strlen(key), &PL_sv_undef, 0)) |
| 3167 | { |
| 3168 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); |
| 3169 | } |
| 3170 | |
| 3171 | if (instr(file, "mathoms.c")) { |
| 3172 | Perl_warner(aTHX_ WARN_DEPRECATED, |
| 3173 | "In %s, line %d, starting in Perl v5.32, %s()" |
| 3174 | " will be removed. Avoid this message by" |
| 3175 | " converting to use %s().\n", |
| 3176 | file, line, name, alternative); |
| 3177 | } |
| 3178 | else { |
| 3179 | Perl_warner(aTHX_ WARN_DEPRECATED, |
| 3180 | "In %s, line %d, starting in Perl v5.32, %s() will" |
| 3181 | " require an additional parameter. Avoid this" |
| 3182 | " message by converting to use %s().\n", |
| 3183 | file, line, name, alternative); |
| 3184 | } |
| 3185 | } |
| 3186 | } |
| 3187 | } |
| 3188 | #endif |
| 3189 | |
| 3190 | bool |
| 3191 | Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e) |
| 3192 | { |
| 3193 | PERL_ARGS_ASSERT__IS_UTF8_FOO; |
| 3194 | |
| 3195 | return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]); |
| 3196 | } |
| 3197 | |
| 3198 | bool |
| 3199 | Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e) |
| 3200 | { |
| 3201 | PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART; |
| 3202 | |
| 3203 | return is_utf8_common(p, e, PL_utf8_perl_idstart); |
| 3204 | } |
| 3205 | |
| 3206 | bool |
| 3207 | Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e) |
| 3208 | { |
| 3209 | PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT; |
| 3210 | |
| 3211 | return is_utf8_common(p, e, PL_utf8_perl_idcont); |
| 3212 | } |
| 3213 | |
| 3214 | STATIC UV |
| 3215 | S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, |
| 3216 | U8* ustrp, STRLEN *lenp, |
| 3217 | SV *invlist, const I32 * const invmap, |
| 3218 | const U32 * const * const aux_tables, |
| 3219 | const U8 * const aux_table_lengths, |
| 3220 | const char * const normal) |
| 3221 | { |
| 3222 | STRLEN len = 0; |
| 3223 | |
| 3224 | /* Change the case of code point 'uv1' whose UTF-8 representation (assumed |
| 3225 | * by this routine to be valid) begins at 'p'. 'normal' is a string to use |
| 3226 | * to name the new case in any generated messages, as a fallback if the |
| 3227 | * operation being used is not available. The new case is given by the |
| 3228 | * data structures in the remaining arguments. |
| 3229 | * |
| 3230 | * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the |
| 3231 | * entire changed case string, and the return value is the first code point |
| 3232 | * in that string */ |
| 3233 | |
| 3234 | PERL_ARGS_ASSERT__TO_UTF8_CASE; |
| 3235 | |
| 3236 | /* For code points that don't change case, we already know that the output |
| 3237 | * of this function is the unchanged input, so we can skip doing look-ups |
| 3238 | * for them. Unfortunately the case-changing code points are scattered |
| 3239 | * around. But there are some long consecutive ranges where there are no |
| 3240 | * case changing code points. By adding tests, we can eliminate the lookup |
| 3241 | * for all the ones in such ranges. This is currently done here only for |
| 3242 | * just a few cases where the scripts are in common use in modern commerce |
| 3243 | * (and scripts adjacent to those which can be included without additional |
| 3244 | * tests). */ |
| 3245 | |
| 3246 | if (uv1 >= 0x0590) { |
| 3247 | /* This keeps from needing further processing the code points most |
| 3248 | * likely to be used in the following non-cased scripts: Hebrew, |
| 3249 | * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari, |
| 3250 | * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada, |
| 3251 | * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */ |
| 3252 | if (uv1 < 0x10A0) { |
| 3253 | goto cases_to_self; |
| 3254 | } |
| 3255 | |
| 3256 | /* The following largish code point ranges also don't have case |
| 3257 | * changes, but khw didn't think they warranted extra tests to speed |
| 3258 | * them up (which would slightly slow down everything else above them): |
| 3259 | * 1100..139F Hangul Jamo, Ethiopic |
| 3260 | * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic, |
| 3261 | * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian, |
| 3262 | * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham, |
| 3263 | * Combining Diacritical Marks Extended, Balinese, |
| 3264 | * Sundanese, Batak, Lepcha, Ol Chiki |
| 3265 | * 2000..206F General Punctuation |
| 3266 | */ |
| 3267 | |
| 3268 | if (uv1 >= 0x2D30) { |
| 3269 | |
| 3270 | /* This keeps the from needing further processing the code points |
| 3271 | * most likely to be used in the following non-cased major scripts: |
| 3272 | * CJK, Katakana, Hiragana, plus some less-likely scripts. |
| 3273 | * |
| 3274 | * (0x2D30 above might have to be changed to 2F00 in the unlikely |
| 3275 | * event that Unicode eventually allocates the unused block as of |
| 3276 | * v8.0 2FE0..2FEF to code points that are cased. khw has verified |
| 3277 | * that the test suite will start having failures to alert you |
| 3278 | * should that happen) */ |
| 3279 | if (uv1 < 0xA640) { |
| 3280 | goto cases_to_self; |
| 3281 | } |
| 3282 | |
| 3283 | if (uv1 >= 0xAC00) { |
| 3284 | if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) { |
| 3285 | if (ckWARN_d(WARN_SURROGATE)) { |
| 3286 | const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; |
| 3287 | Perl_warner(aTHX_ packWARN(WARN_SURROGATE), |
| 3288 | "Operation \"%s\" returns its argument for" |
| 3289 | " UTF-16 surrogate U+%04" UVXf, desc, uv1); |
| 3290 | } |
| 3291 | goto cases_to_self; |
| 3292 | } |
| 3293 | |
| 3294 | /* AC00..FAFF Catches Hangul syllables and private use, plus |
| 3295 | * some others */ |
| 3296 | if (uv1 < 0xFB00) { |
| 3297 | goto cases_to_self; |
| 3298 | } |
| 3299 | |
| 3300 | if (UNLIKELY(UNICODE_IS_SUPER(uv1))) { |
| 3301 | if (UNLIKELY(uv1 > MAX_LEGAL_CP)) { |
| 3302 | Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv1)); |
| 3303 | } |
| 3304 | if (ckWARN_d(WARN_NON_UNICODE)) { |
| 3305 | const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; |
| 3306 | Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), |
| 3307 | "Operation \"%s\" returns its argument for" |
| 3308 | " non-Unicode code point 0x%04" UVXf, desc, uv1); |
| 3309 | } |
| 3310 | goto cases_to_self; |
| 3311 | } |
| 3312 | #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C |
| 3313 | if (UNLIKELY(uv1 |
| 3314 | > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C)) |
| 3315 | { |
| 3316 | |
| 3317 | goto cases_to_self; |
| 3318 | } |
| 3319 | #endif |
| 3320 | } |
| 3321 | } |
| 3322 | |
| 3323 | /* Note that non-characters are perfectly legal, so no warning should |
| 3324 | * be given. */ |
| 3325 | } |
| 3326 | |
| 3327 | { |
| 3328 | unsigned int i; |
| 3329 | const U32 * cp_list; |
| 3330 | U8 * d; |
| 3331 | |
| 3332 | /* 'index' is guaranteed to be non-negative, as this is an inversion |
| 3333 | * map that covers all possible inputs. See [perl #133365] */ |
| 3334 | SSize_t index = _invlist_search(invlist, uv1); |
| 3335 | I32 base = invmap[index]; |
| 3336 | |
| 3337 | /* The data structures are set up so that if 'base' is non-negative, |
| 3338 | * the case change is 1-to-1; and if 0, the change is to itself */ |
| 3339 | if (base >= 0) { |
| 3340 | IV lc; |
| 3341 | |
| 3342 | if (base == 0) { |
| 3343 | goto cases_to_self; |
| 3344 | } |
| 3345 | |
| 3346 | /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */ |
| 3347 | lc = base + uv1 - invlist_array(invlist)[index]; |
| 3348 | *lenp = uvchr_to_utf8(ustrp, lc) - ustrp; |
| 3349 | return lc; |
| 3350 | } |
| 3351 | |
| 3352 | /* Here 'base' is negative. That means the mapping is 1-to-many, and |
| 3353 | * requires an auxiliary table look up. abs(base) gives the index into |
| 3354 | * a list of such tables which points to the proper aux table. And a |
| 3355 | * parallel list gives the length of each corresponding aux table. */ |
| 3356 | cp_list = aux_tables[-base]; |
| 3357 | |
| 3358 | /* Create the string of UTF-8 from the mapped-to code points */ |
| 3359 | d = ustrp; |
| 3360 | for (i = 0; i < aux_table_lengths[-base]; i++) { |
| 3361 | d = uvchr_to_utf8(d, cp_list[i]); |
| 3362 | } |
| 3363 | *d = '\0'; |
| 3364 | *lenp = d - ustrp; |
| 3365 | |
| 3366 | return cp_list[0]; |
| 3367 | } |
| 3368 | |
| 3369 | /* Here, there was no mapping defined, which means that the code point maps |
| 3370 | * to itself. Return the inputs */ |
| 3371 | cases_to_self: |
| 3372 | if (p) { |
| 3373 | len = UTF8SKIP(p); |
| 3374 | if (p != ustrp) { /* Don't copy onto itself */ |
| 3375 | Copy(p, ustrp, len, U8); |
| 3376 | } |
| 3377 | *lenp = len; |
| 3378 | } |
| 3379 | else { |
| 3380 | *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp; |
| 3381 | } |
| 3382 | |
| 3383 | return uv1; |
| 3384 | |
| 3385 | } |
| 3386 | |
| 3387 | Size_t |
| 3388 | Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to, |
| 3389 | const U32 ** remaining_folds_to) |
| 3390 | { |
| 3391 | /* Returns the count of the number of code points that fold to the input |
| 3392 | * 'cp' (besides itself). |
| 3393 | * |
| 3394 | * If the return is 0, there is nothing else that folds to it, and |
| 3395 | * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL. |
| 3396 | * |
| 3397 | * If the return is 1, '*first_folds_to' is set to the single code point, |
| 3398 | * and '*remaining_folds_to' is set to NULL. |
| 3399 | * |
| 3400 | * Otherwise, '*first_folds_to' is set to a code point, and |
| 3401 | * '*remaining_fold_to' is set to an array that contains the others. The |
| 3402 | * length of this array is the returned count minus 1. |
| 3403 | * |
| 3404 | * The reason for this convolution is to avoid having to deal with |
| 3405 | * allocating and freeing memory. The lists are already constructed, so |
| 3406 | * the return can point to them, but single code points aren't, so would |
| 3407 | * need to be constructed if we didn't employ something like this API |
| 3408 | * |
| 3409 | * The code points returned by this function are all legal Unicode, which |
| 3410 | * occupy at most 21 bits, and so a U32 is sufficient, and the lists are |
| 3411 | * constructed with this size (to save space and memory), and we return |
| 3412 | * pointers, so they must be this size */ |
| 3413 | |
| 3414 | /* 'index' is guaranteed to be non-negative, as this is an inversion map |
| 3415 | * that covers all possible inputs. See [perl #133365] */ |
| 3416 | SSize_t index = _invlist_search(PL_utf8_foldclosures, cp); |
| 3417 | I32 base = _Perl_IVCF_invmap[index]; |
| 3418 | |
| 3419 | PERL_ARGS_ASSERT__INVERSE_FOLDS; |
| 3420 | |
| 3421 | if (base == 0) { /* No fold */ |
| 3422 | *first_folds_to = 0; |
| 3423 | *remaining_folds_to = NULL; |
| 3424 | return 0; |
| 3425 | } |
| 3426 | |
| 3427 | #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */ |
| 3428 | |
| 3429 | assert(base > 0); |
| 3430 | |
| 3431 | #else |
| 3432 | |
| 3433 | if (UNLIKELY(base < 0)) { /* Folds to more than one character */ |
| 3434 | |
| 3435 | /* The data structure is set up so that the absolute value of 'base' is |
| 3436 | * an index into a table of pointers to arrays, with the array |
| 3437 | * corresponding to the index being the list of code points that fold |
| 3438 | * to 'cp', and the parallel array containing the length of the list |
| 3439 | * array */ |
| 3440 | *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0]; |
| 3441 | *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; |
| 3442 | /* +1 excludes first_folds_to */ |
| 3443 | return IVCF_AUX_TABLE_lengths[-base]; |
| 3444 | } |
| 3445 | |
| 3446 | #endif |
| 3447 | |
| 3448 | /* Only the single code point. This works like 'fc(G) = G - A + a' */ |
| 3449 | *first_folds_to = (U32) (base + cp |
| 3450 | - invlist_array(PL_utf8_foldclosures)[index]); |
| 3451 | *remaining_folds_to = NULL; |
| 3452 | return 1; |
| 3453 | } |
| 3454 | |
| 3455 | STATIC UV |
| 3456 | S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, |
| 3457 | U8* const ustrp, STRLEN *lenp) |
| 3458 | { |
| 3459 | /* This is called when changing the case of a UTF-8-encoded character above |
| 3460 | * the Latin1 range, and the operation is in a non-UTF-8 locale. If the |
| 3461 | * result contains a character that crosses the 255/256 boundary, disallow |
| 3462 | * the change, and return the original code point. See L<perlfunc/lc> for |
| 3463 | * why; |
| 3464 | * |
| 3465 | * p points to the original string whose case was changed; assumed |
| 3466 | * by this routine to be well-formed |
| 3467 | * result the code point of the first character in the changed-case string |
| 3468 | * ustrp points to the changed-case string (<result> represents its |
| 3469 | * first char) |
| 3470 | * lenp points to the length of <ustrp> */ |
| 3471 | |
| 3472 | UV original; /* To store the first code point of <p> */ |
| 3473 | |
| 3474 | PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING; |
| 3475 | |
| 3476 | assert(UTF8_IS_ABOVE_LATIN1(*p)); |
| 3477 | |
| 3478 | /* We know immediately if the first character in the string crosses the |
| 3479 | * boundary, so can skip testing */ |
| 3480 | if (result > 255) { |
| 3481 | |
| 3482 | /* Look at every character in the result; if any cross the |
| 3483 | * boundary, the whole thing is disallowed */ |
| 3484 | U8* s = ustrp + UTF8SKIP(ustrp); |
| 3485 | U8* e = ustrp + *lenp; |
| 3486 | while (s < e) { |
| 3487 | if (! UTF8_IS_ABOVE_LATIN1(*s)) { |
| 3488 | goto bad_crossing; |
| 3489 | } |
| 3490 | s += UTF8SKIP(s); |
| 3491 | } |
| 3492 | |
| 3493 | /* Here, no characters crossed, result is ok as-is, but we warn. */ |
| 3494 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p)); |
| 3495 | return result; |
| 3496 | } |
| 3497 | |
| 3498 | bad_crossing: |
| 3499 | |
| 3500 | /* Failed, have to return the original */ |
| 3501 | original = valid_utf8_to_uvchr(p, lenp); |
| 3502 | |
| 3503 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 3504 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 3505 | "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8" |
| 3506 | " locale; resolved to \"\\x{%" UVXf "}\".", |
| 3507 | OP_DESC(PL_op), |
| 3508 | original, |
| 3509 | original); |
| 3510 | Copy(p, ustrp, *lenp, char); |
| 3511 | return original; |
| 3512 | } |
| 3513 | |
| 3514 | STATIC UV |
| 3515 | S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e, |
| 3516 | U8 * ustrp, STRLEN *lenp) |
| 3517 | { |
| 3518 | /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from |
| 3519 | * p0..e-1 according to Turkic rules is the same as for non-Turkic. |
| 3520 | * Otherwise, it returns the first code point of the Turkic foldcased |
| 3521 | * sequence, and the entire sequence will be stored in *ustrp. ustrp will |
| 3522 | * contain *lenp bytes |
| 3523 | * |
| 3524 | * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER |
| 3525 | * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER |
| 3526 | * DOTLESS I */ |
| 3527 | |
| 3528 | PERL_ARGS_ASSERT_TURKIC_FC; |
| 3529 | assert(e > p); |
| 3530 | |
| 3531 | if (UNLIKELY(*p == 'I')) { |
| 3532 | *lenp = 2; |
| 3533 | ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I); |
| 3534 | ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I); |
| 3535 | return LATIN_SMALL_LETTER_DOTLESS_I; |
| 3536 | } |
| 3537 | |
| 3538 | if (UNLIKELY(memBEGINs(p, e - p, |
| 3539 | LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8))) |
| 3540 | { |
| 3541 | *lenp = 1; |
| 3542 | *ustrp = 'i'; |
| 3543 | return 'i'; |
| 3544 | } |
| 3545 | |
| 3546 | return 0; |
| 3547 | } |
| 3548 | |
| 3549 | STATIC UV |
| 3550 | S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e, |
| 3551 | U8 * ustrp, STRLEN *lenp) |
| 3552 | { |
| 3553 | /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from |
| 3554 | * p0..e-1 according to Turkic rules is the same as for non-Turkic. |
| 3555 | * Otherwise, it returns the first code point of the Turkic lowercased |
| 3556 | * sequence, and the entire sequence will be stored in *ustrp. ustrp will |
| 3557 | * contain *lenp bytes */ |
| 3558 | |
| 3559 | PERL_ARGS_ASSERT_TURKIC_LC; |
| 3560 | assert(e > p0); |
| 3561 | |
| 3562 | /* A 'I' requires context as to what to do */ |
| 3563 | if (UNLIKELY(*p0 == 'I')) { |
| 3564 | const U8 * p = p0 + 1; |
| 3565 | |
| 3566 | /* According to the Unicode SpecialCasing.txt file, a capital 'I' |
| 3567 | * modified by a dot above lowercases to 'i' even in turkic locales. */ |
| 3568 | while (p < e) { |
| 3569 | UV cp; |
| 3570 | |
| 3571 | if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) { |
| 3572 | ustrp[0] = 'i'; |
| 3573 | *lenp = 1; |
| 3574 | return 'i'; |
| 3575 | } |
| 3576 | |
| 3577 | /* For the dot above to modify the 'I', it must be part of a |
| 3578 | * combining sequence immediately following the 'I', and no other |
| 3579 | * modifier with a ccc of 230 may intervene */ |
| 3580 | cp = utf8_to_uvchr_buf(p, e, NULL); |
| 3581 | if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) { |
| 3582 | break; |
| 3583 | } |
| 3584 | |
| 3585 | /* Here the combining sequence continues */ |
| 3586 | p += UTF8SKIP(p); |
| 3587 | } |
| 3588 | } |
| 3589 | |
| 3590 | /* In all other cases the lc is the same as the fold */ |
| 3591 | return turkic_fc(p0, e, ustrp, lenp); |
| 3592 | } |
| 3593 | |
| 3594 | STATIC UV |
| 3595 | S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e, |
| 3596 | U8 * ustrp, STRLEN *lenp) |
| 3597 | { |
| 3598 | /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence |
| 3599 | * from p0..e-1 according to Turkic rules is the same as for non-Turkic. |
| 3600 | * Otherwise, it returns the first code point of the Turkic upper or |
| 3601 | * title-cased sequence, and the entire sequence will be stored in *ustrp. |
| 3602 | * ustrp will contain *lenp bytes |
| 3603 | * |
| 3604 | * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER |
| 3605 | * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER |
| 3606 | * DOTLESS I */ |
| 3607 | |
| 3608 | PERL_ARGS_ASSERT_TURKIC_UC; |
| 3609 | assert(e > p); |
| 3610 | |
| 3611 | if (*p == 'i') { |
| 3612 | *lenp = 2; |
| 3613 | ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE); |
| 3614 | ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE); |
| 3615 | return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE; |
| 3616 | } |
| 3617 | |
| 3618 | if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) { |
| 3619 | *lenp = 1; |
| 3620 | *ustrp = 'I'; |
| 3621 | return 'I'; |
| 3622 | } |
| 3623 | |
| 3624 | return 0; |
| 3625 | } |
| 3626 | |
| 3627 | /* The process for changing the case is essentially the same for the four case |
| 3628 | * change types, except there are complications for folding. Otherwise the |
| 3629 | * difference is only which case to change to. To make sure that they all do |
| 3630 | * the same thing, the bodies of the functions are extracted out into the |
| 3631 | * following two macros. The functions are written with the same variable |
| 3632 | * names, and these are known and used inside these macros. It would be |
| 3633 | * better, of course, to have inline functions to do it, but since different |
| 3634 | * macros are called, depending on which case is being changed to, this is not |
| 3635 | * feasible in C (to khw's knowledge). Two macros are created so that the fold |
| 3636 | * function can start with the common start macro, then finish with its special |
| 3637 | * handling; while the other three cases can just use the common end macro. |
| 3638 | * |
| 3639 | * The algorithm is to use the proper (passed in) macro or function to change |
| 3640 | * the case for code points that are below 256. The macro is used if using |
| 3641 | * locale rules for the case change; the function if not. If the code point is |
| 3642 | * above 255, it is computed from the input UTF-8, and another macro is called |
| 3643 | * to do the conversion. If necessary, the output is converted to UTF-8. If |
| 3644 | * using a locale, we have to check that the change did not cross the 255/256 |
| 3645 | * boundary, see check_locale_boundary_crossing() for further details. |
| 3646 | * |
| 3647 | * The macros are split with the correct case change for the below-256 case |
| 3648 | * stored into 'result', and in the middle of an else clause for the above-255 |
| 3649 | * case. At that point in the 'else', 'result' is not the final result, but is |
| 3650 | * the input code point calculated from the UTF-8. The fold code needs to |
| 3651 | * realize all this and take it from there. |
| 3652 | * |
| 3653 | * To deal with Turkic locales, the function specified by the parameter |
| 3654 | * 'turkic' is called when appropriate. |
| 3655 | * |
| 3656 | * If you read the two macros as sequential, it's easier to understand what's |
| 3657 | * going on. */ |
| 3658 | #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \ |
| 3659 | L1_func_extra_param, turkic) \ |
| 3660 | \ |
| 3661 | if (flags & (locale_flags)) { \ |
| 3662 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ |
| 3663 | if (IN_UTF8_CTYPE_LOCALE) { \ |
| 3664 | if (UNLIKELY(PL_in_utf8_turkic_locale)) { \ |
| 3665 | UV ret = turkic(p, e, ustrp, lenp); \ |
| 3666 | if (ret) return ret; \ |
| 3667 | } \ |
| 3668 | \ |
| 3669 | /* Otherwise, treat a UTF-8 locale as not being in locale at \ |
| 3670 | * all */ \ |
| 3671 | flags &= ~(locale_flags); \ |
| 3672 | } \ |
| 3673 | } \ |
| 3674 | \ |
| 3675 | if (UTF8_IS_INVARIANT(*p)) { \ |
| 3676 | if (flags & (locale_flags)) { \ |
| 3677 | result = LC_L1_change_macro(*p); \ |
| 3678 | } \ |
| 3679 | else { \ |
| 3680 | return L1_func(*p, ustrp, lenp, L1_func_extra_param); \ |
| 3681 | } \ |
| 3682 | } \ |
| 3683 | else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \ |
| 3684 | U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \ |
| 3685 | if (flags & (locale_flags)) { \ |
| 3686 | result = LC_L1_change_macro(c); \ |
| 3687 | } \ |
| 3688 | else { \ |
| 3689 | return L1_func(c, ustrp, lenp, L1_func_extra_param); \ |
| 3690 | } \ |
| 3691 | } \ |
| 3692 | else { /* malformed UTF-8 or ord above 255 */ \ |
| 3693 | STRLEN len_result; \ |
| 3694 | result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \ |
| 3695 | if (len_result == (STRLEN) -1) { \ |
| 3696 | _force_out_malformed_utf8_message(p, e, 0, 1 /* Die */ ); \ |
| 3697 | } |
| 3698 | |
| 3699 | #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \ |
| 3700 | result = change_macro(result, p, ustrp, lenp); \ |
| 3701 | \ |
| 3702 | if (flags & (locale_flags)) { \ |
| 3703 | result = check_locale_boundary_crossing(p, result, ustrp, lenp); \ |
| 3704 | } \ |
| 3705 | return result; \ |
| 3706 | } \ |
| 3707 | \ |
| 3708 | /* Here, used locale rules. Convert back to UTF-8 */ \ |
| 3709 | if (UTF8_IS_INVARIANT(result)) { \ |
| 3710 | *ustrp = (U8) result; \ |
| 3711 | *lenp = 1; \ |
| 3712 | } \ |
| 3713 | else { \ |
| 3714 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \ |
| 3715 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \ |
| 3716 | *lenp = 2; \ |
| 3717 | } \ |
| 3718 | \ |
| 3719 | return result; |
| 3720 | |
| 3721 | /* Not currently externally documented, and subject to change: |
| 3722 | * <flags> is set iff the rules from the current underlying locale are to |
| 3723 | * be used. */ |
| 3724 | |
| 3725 | UV |
| 3726 | Perl__to_utf8_upper_flags(pTHX_ const U8 *p, |
| 3727 | const U8 *e, |
| 3728 | U8* ustrp, |
| 3729 | STRLEN *lenp, |
| 3730 | bool flags) |
| 3731 | { |
| 3732 | UV result; |
| 3733 | |
| 3734 | PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS; |
| 3735 | |
| 3736 | /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */ |
| 3737 | /* 2nd char of uc(U+DF) is 'S' */ |
| 3738 | CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S', |
| 3739 | turkic_uc); |
| 3740 | CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE); |
| 3741 | } |
| 3742 | |
| 3743 | /* Not currently externally documented, and subject to change: |
| 3744 | * <flags> is set iff the rules from the current underlying locale are to be |
| 3745 | * used. Since titlecase is not defined in POSIX, for other than a |
| 3746 | * UTF-8 locale, uppercase is used instead for code points < 256. |
| 3747 | */ |
| 3748 | |
| 3749 | UV |
| 3750 | Perl__to_utf8_title_flags(pTHX_ const U8 *p, |
| 3751 | const U8 *e, |
| 3752 | U8* ustrp, |
| 3753 | STRLEN *lenp, |
| 3754 | bool flags) |
| 3755 | { |
| 3756 | UV result; |
| 3757 | |
| 3758 | PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS; |
| 3759 | |
| 3760 | /* 2nd char of ucfirst(U+DF) is 's' */ |
| 3761 | CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's', |
| 3762 | turkic_uc); |
| 3763 | CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE); |
| 3764 | } |
| 3765 | |
| 3766 | /* Not currently externally documented, and subject to change: |
| 3767 | * <flags> is set iff the rules from the current underlying locale are to |
| 3768 | * be used. |
| 3769 | */ |
| 3770 | |
| 3771 | UV |
| 3772 | Perl__to_utf8_lower_flags(pTHX_ const U8 *p, |
| 3773 | const U8 *e, |
| 3774 | U8* ustrp, |
| 3775 | STRLEN *lenp, |
| 3776 | bool flags) |
| 3777 | { |
| 3778 | UV result; |
| 3779 | |
| 3780 | PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS; |
| 3781 | |
| 3782 | CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */, |
| 3783 | turkic_lc); |
| 3784 | CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE) |
| 3785 | } |
| 3786 | |
| 3787 | /* Not currently externally documented, and subject to change, |
| 3788 | * in <flags> |
| 3789 | * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying |
| 3790 | * locale are to be used. |
| 3791 | * bit FOLD_FLAGS_FULL is set iff full case folds are to be used; |
| 3792 | * otherwise simple folds |
| 3793 | * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are |
| 3794 | * prohibited |
| 3795 | */ |
| 3796 | |
| 3797 | UV |
| 3798 | Perl__to_utf8_fold_flags(pTHX_ const U8 *p, |
| 3799 | const U8 *e, |
| 3800 | U8* ustrp, |
| 3801 | STRLEN *lenp, |
| 3802 | U8 flags) |
| 3803 | { |
| 3804 | UV result; |
| 3805 | |
| 3806 | PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS; |
| 3807 | |
| 3808 | /* These are mutually exclusive */ |
| 3809 | assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII))); |
| 3810 | |
| 3811 | assert(p != ustrp); /* Otherwise overwrites */ |
| 3812 | |
| 3813 | CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1, |
| 3814 | ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)), |
| 3815 | turkic_fc); |
| 3816 | |
| 3817 | result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL); |
| 3818 | |
| 3819 | if (flags & FOLD_FLAGS_LOCALE) { |
| 3820 | |
| 3821 | # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8 |
| 3822 | # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8 |
| 3823 | # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8 |
| 3824 | |
| 3825 | /* Special case these two characters, as what normally gets |
| 3826 | * returned under locale doesn't work */ |
| 3827 | if (memBEGINs((char *) p, e - p, CAP_SHARP_S)) |
| 3828 | { |
| 3829 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 3830 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 3831 | "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; " |
| 3832 | "resolved to \"\\x{17F}\\x{17F}\"."); |
| 3833 | goto return_long_s; |
| 3834 | } |
| 3835 | else |
| 3836 | #endif |
| 3837 | if (memBEGINs((char *) p, e - p, LONG_S_T)) |
| 3838 | { |
| 3839 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 3840 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 3841 | "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; " |
| 3842 | "resolved to \"\\x{FB06}\"."); |
| 3843 | goto return_ligature_st; |
| 3844 | } |
| 3845 | |
| 3846 | #if UNICODE_MAJOR_VERSION == 3 \ |
| 3847 | && UNICODE_DOT_VERSION == 0 \ |
| 3848 | && UNICODE_DOT_DOT_VERSION == 1 |
| 3849 | # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8 |
| 3850 | |
| 3851 | /* And special case this on this Unicode version only, for the same |
| 3852 | * reaons the other two are special cased. They would cross the |
| 3853 | * 255/256 boundary which is forbidden under /l, and so the code |
| 3854 | * wouldn't catch that they are equivalent (which they are only in |
| 3855 | * this release) */ |
| 3856 | else if (memBEGINs((char *) p, e - p, DOTTED_I)) { |
| 3857 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 3858 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 3859 | "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; " |
| 3860 | "resolved to \"\\x{0131}\"."); |
| 3861 | goto return_dotless_i; |
| 3862 | } |
| 3863 | #endif |
| 3864 | |
| 3865 | return check_locale_boundary_crossing(p, result, ustrp, lenp); |
| 3866 | } |
| 3867 | else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) { |
| 3868 | return result; |
| 3869 | } |
| 3870 | else { |
| 3871 | /* This is called when changing the case of a UTF-8-encoded |
| 3872 | * character above the ASCII range, and the result should not |
| 3873 | * contain an ASCII character. */ |
| 3874 | |
| 3875 | UV original; /* To store the first code point of <p> */ |
| 3876 | |
| 3877 | /* Look at every character in the result; if any cross the |
| 3878 | * boundary, the whole thing is disallowed */ |
| 3879 | U8* s = ustrp; |
| 3880 | U8* send = ustrp + *lenp; |
| 3881 | while (s < send) { |
| 3882 | if (isASCII(*s)) { |
| 3883 | /* Crossed, have to return the original */ |
| 3884 | original = valid_utf8_to_uvchr(p, lenp); |
| 3885 | |
| 3886 | /* But in these instances, there is an alternative we can |
| 3887 | * return that is valid */ |
| 3888 | if (original == LATIN_SMALL_LETTER_SHARP_S |
| 3889 | #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */ |
| 3890 | || original == LATIN_CAPITAL_LETTER_SHARP_S |
| 3891 | #endif |
| 3892 | ) { |
| 3893 | goto return_long_s; |
| 3894 | } |
| 3895 | else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) { |
| 3896 | goto return_ligature_st; |
| 3897 | } |
| 3898 | #if UNICODE_MAJOR_VERSION == 3 \ |
| 3899 | && UNICODE_DOT_VERSION == 0 \ |
| 3900 | && UNICODE_DOT_DOT_VERSION == 1 |
| 3901 | |
| 3902 | else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) { |
| 3903 | goto return_dotless_i; |
| 3904 | } |
| 3905 | #endif |
| 3906 | Copy(p, ustrp, *lenp, char); |
| 3907 | return original; |
| 3908 | } |
| 3909 | s += UTF8SKIP(s); |
| 3910 | } |
| 3911 | |
| 3912 | /* Here, no characters crossed, result is ok as-is */ |
| 3913 | return result; |
| 3914 | } |
| 3915 | } |
| 3916 | |
| 3917 | /* Here, used locale rules. Convert back to UTF-8 */ |
| 3918 | if (UTF8_IS_INVARIANT(result)) { |
| 3919 | *ustrp = (U8) result; |
| 3920 | *lenp = 1; |
| 3921 | } |
| 3922 | else { |
| 3923 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); |
| 3924 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); |
| 3925 | *lenp = 2; |
| 3926 | } |
| 3927 | |
| 3928 | return result; |
| 3929 | |
| 3930 | return_long_s: |
| 3931 | /* Certain folds to 'ss' are prohibited by the options, but they do allow |
| 3932 | * folds to a string of two of these characters. By returning this |
| 3933 | * instead, then, e.g., |
| 3934 | * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}") |
| 3935 | * works. */ |
| 3936 | |
| 3937 | *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; |
| 3938 | Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, |
| 3939 | ustrp, *lenp, U8); |
| 3940 | return LATIN_SMALL_LETTER_LONG_S; |
| 3941 | |
| 3942 | return_ligature_st: |
| 3943 | /* Two folds to 'st' are prohibited by the options; instead we pick one and |
| 3944 | * have the other one fold to it */ |
| 3945 | |
| 3946 | *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1; |
| 3947 | Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8); |
| 3948 | return LATIN_SMALL_LIGATURE_ST; |
| 3949 | |
| 3950 | #if UNICODE_MAJOR_VERSION == 3 \ |
| 3951 | && UNICODE_DOT_VERSION == 0 \ |
| 3952 | && UNICODE_DOT_DOT_VERSION == 1 |
| 3953 | |
| 3954 | return_dotless_i: |
| 3955 | *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1; |
| 3956 | Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8); |
| 3957 | return LATIN_SMALL_LETTER_DOTLESS_I; |
| 3958 | |
| 3959 | #endif |
| 3960 | |
| 3961 | } |
| 3962 | |
| 3963 | bool |
| 3964 | Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len) |
| 3965 | { |
| 3966 | /* May change: warns if surrogates, non-character code points, or |
| 3967 | * non-Unicode code points are in 's' which has length 'len' bytes. |
| 3968 | * Returns TRUE if none found; FALSE otherwise. The only other validity |
| 3969 | * check is to make sure that this won't exceed the string's length nor |
| 3970 | * overflow */ |
| 3971 | |
| 3972 | const U8* const e = s + len; |
| 3973 | bool ok = TRUE; |
| 3974 | |
| 3975 | PERL_ARGS_ASSERT_CHECK_UTF8_PRINT; |
| 3976 | |
| 3977 | while (s < e) { |
| 3978 | if (UTF8SKIP(s) > len) { |
| 3979 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 3980 | "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print"); |
| 3981 | return FALSE; |
| 3982 | } |
| 3983 | if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) { |
| 3984 | if (UNLIKELY(UTF8_IS_SUPER(s, e))) { |
| 3985 | if ( ckWARN_d(WARN_NON_UNICODE) |
| 3986 | || UNLIKELY(0 < does_utf8_overflow(s, s + len, |
| 3987 | 0 /* Don't consider overlongs */ |
| 3988 | ))) |
| 3989 | { |
| 3990 | /* A side effect of this function will be to warn */ |
| 3991 | (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER); |
| 3992 | ok = FALSE; |
| 3993 | } |
| 3994 | } |
| 3995 | else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) { |
| 3996 | if (ckWARN_d(WARN_SURROGATE)) { |
| 3997 | /* This has a different warning than the one the called |
| 3998 | * function would output, so can't just call it, unlike we |
| 3999 | * do for the non-chars and above-unicodes */ |
| 4000 | UV uv = utf8_to_uvchr_buf(s, e, NULL); |
| 4001 | Perl_warner(aTHX_ packWARN(WARN_SURROGATE), |
| 4002 | "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", |
| 4003 | uv); |
| 4004 | ok = FALSE; |
| 4005 | } |
| 4006 | } |
| 4007 | else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e)) |
| 4008 | && (ckWARN_d(WARN_NONCHAR))) |
| 4009 | { |
| 4010 | /* A side effect of this function will be to warn */ |
| 4011 | (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR); |
| 4012 | ok = FALSE; |
| 4013 | } |
| 4014 | } |
| 4015 | s += UTF8SKIP(s); |
| 4016 | } |
| 4017 | |
| 4018 | return ok; |
| 4019 | } |
| 4020 | |
| 4021 | /* |
| 4022 | =for apidoc pv_uni_display |
| 4023 | |
| 4024 | Build to the scalar C<dsv> a displayable version of the UTF-8 encoded string |
| 4025 | C<spv>, length C<len>, the displayable version being at most C<pvlim> bytes |
| 4026 | long (if longer, the rest is truncated and C<"..."> will be appended). |
| 4027 | |
| 4028 | The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display |
| 4029 | C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH> |
| 4030 | to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">) |
| 4031 | (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">). |
| 4032 | C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both |
| 4033 | C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on. |
| 4034 | |
| 4035 | Additionally, there is now C<UNI_DISPLAY_BACKSPACE> which allows C<\b> for a |
| 4036 | backspace, but only when C<UNI_DISPLAY_BACKSLASH> also is set. |
| 4037 | |
| 4038 | The pointer to the PV of the C<dsv> is returned. |
| 4039 | |
| 4040 | See also L</sv_uni_display>. |
| 4041 | |
| 4042 | =cut */ |
| 4043 | char * |
| 4044 | Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, |
| 4045 | UV flags) |
| 4046 | { |
| 4047 | int truncated = 0; |
| 4048 | const char *s, *e; |
| 4049 | |
| 4050 | PERL_ARGS_ASSERT_PV_UNI_DISPLAY; |
| 4051 | |
| 4052 | SvPVCLEAR(dsv); |
| 4053 | SvUTF8_off(dsv); |
| 4054 | for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) { |
| 4055 | UV u; |
| 4056 | bool ok = 0; |
| 4057 | |
| 4058 | if (pvlim && SvCUR(dsv) >= pvlim) { |
| 4059 | truncated++; |
| 4060 | break; |
| 4061 | } |
| 4062 | u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0); |
| 4063 | if (u < 256) { |
| 4064 | const unsigned char c = (unsigned char)u & 0xFF; |
| 4065 | if (flags & UNI_DISPLAY_BACKSLASH) { |
| 4066 | if ( isMNEMONIC_CNTRL(c) |
| 4067 | && ( c != '\b' |
| 4068 | || (flags & UNI_DISPLAY_BACKSPACE))) |
| 4069 | { |
| 4070 | const char * mnemonic = cntrl_to_mnemonic(c); |
| 4071 | sv_catpvn(dsv, mnemonic, strlen(mnemonic)); |
| 4072 | ok = 1; |
| 4073 | } |
| 4074 | else if (c == '\\') { |
| 4075 | sv_catpvs(dsv, "\\\\"); |
| 4076 | ok = 1; |
| 4077 | } |
| 4078 | } |
| 4079 | /* isPRINT() is the locale-blind version. */ |
| 4080 | if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) { |
| 4081 | const char string = c; |
| 4082 | sv_catpvn(dsv, &string, 1); |
| 4083 | ok = 1; |
| 4084 | } |
| 4085 | } |
| 4086 | if (!ok) |
| 4087 | Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u); |
| 4088 | } |
| 4089 | if (truncated) |
| 4090 | sv_catpvs(dsv, "..."); |
| 4091 | |
| 4092 | return SvPVX(dsv); |
| 4093 | } |
| 4094 | |
| 4095 | /* |
| 4096 | =for apidoc sv_uni_display |
| 4097 | |
| 4098 | Build to the scalar C<dsv> a displayable version of the scalar C<sv>, |
| 4099 | the displayable version being at most C<pvlim> bytes long |
| 4100 | (if longer, the rest is truncated and "..." will be appended). |
| 4101 | |
| 4102 | The C<flags> argument is as in L</pv_uni_display>(). |
| 4103 | |
| 4104 | The pointer to the PV of the C<dsv> is returned. |
| 4105 | |
| 4106 | =cut |
| 4107 | */ |
| 4108 | char * |
| 4109 | Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags) |
| 4110 | { |
| 4111 | const char * const ptr = |
| 4112 | isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv); |
| 4113 | |
| 4114 | PERL_ARGS_ASSERT_SV_UNI_DISPLAY; |
| 4115 | |
| 4116 | return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr, |
| 4117 | SvCUR(ssv), pvlim, flags); |
| 4118 | } |
| 4119 | |
| 4120 | /* |
| 4121 | =for apidoc foldEQ_utf8 |
| 4122 | |
| 4123 | Returns true if the leading portions of the strings C<s1> and C<s2> (either or |
| 4124 | both of which may be in UTF-8) are the same case-insensitively; false |
| 4125 | otherwise. How far into the strings to compare is determined by other input |
| 4126 | parameters. |
| 4127 | |
| 4128 | If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode; |
| 4129 | otherwise it is assumed to be in native 8-bit encoding. Correspondingly for |
| 4130 | C<u2> with respect to C<s2>. |
| 4131 | |
| 4132 | If the byte length C<l1> is non-zero, it says how far into C<s1> to check for |
| 4133 | fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach. |
| 4134 | The scan will not be considered to be a match unless the goal is reached, and |
| 4135 | scanning won't continue past that goal. Correspondingly for C<l2> with respect |
| 4136 | to C<s2>. |
| 4137 | |
| 4138 | If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that |
| 4139 | pointer is considered an end pointer to the position 1 byte past the maximum |
| 4140 | point in C<s1> beyond which scanning will not continue under any circumstances. |
| 4141 | (This routine assumes that UTF-8 encoded input strings are not malformed; |
| 4142 | malformed input can cause it to read past C<pe1>). This means that if both |
| 4143 | C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match |
| 4144 | will never be successful because it can never |
| 4145 | get as far as its goal (and in fact is asserted against). Correspondingly for |
| 4146 | C<pe2> with respect to C<s2>. |
| 4147 | |
| 4148 | At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and |
| 4149 | C<l2> must be non-zero), and if both do, both have to be |
| 4150 | reached for a successful match. Also, if the fold of a character is multiple |
| 4151 | characters, all of them must be matched (see tr21 reference below for |
| 4152 | 'folding'). |
| 4153 | |
| 4154 | Upon a successful match, if C<pe1> is non-C<NULL>, |
| 4155 | it will be set to point to the beginning of the I<next> character of C<s1> |
| 4156 | beyond what was matched. Correspondingly for C<pe2> and C<s2>. |
| 4157 | |
| 4158 | For case-insensitiveness, the "casefolding" of Unicode is used |
| 4159 | instead of upper/lowercasing both the characters, see |
| 4160 | L<https://www.unicode.org/unicode/reports/tr21/> (Case Mappings). |
| 4161 | |
| 4162 | =cut */ |
| 4163 | |
| 4164 | /* A flags parameter has been added which may change, and hence isn't |
| 4165 | * externally documented. Currently it is: |
| 4166 | * 0 for as-documented above |
| 4167 | * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an |
| 4168 | ASCII one, to not match |
| 4169 | * FOLDEQ_LOCALE is set iff the rules from the current underlying |
| 4170 | * locale are to be used. |
| 4171 | * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this |
| 4172 | * routine. This allows that step to be skipped. |
| 4173 | * Currently, this requires s1 to be encoded as UTF-8 |
| 4174 | * (u1 must be true), which is asserted for. |
| 4175 | * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may |
| 4176 | * cross certain boundaries. Hence, the caller should |
| 4177 | * let this function do the folding instead of |
| 4178 | * pre-folding. This code contains an assertion to |
| 4179 | * that effect. However, if the caller knows what |
| 4180 | * it's doing, it can pass this flag to indicate that, |
| 4181 | * and the assertion is skipped. |
| 4182 | * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies |
| 4183 | * to s2, and s2 doesn't have to be UTF-8 encoded. |
| 4184 | * This introduces an asymmetry to save a few branches |
| 4185 | * in a loop. Currently, this is not a problem, as |
| 4186 | * never are both inputs pre-folded. Simply call this |
| 4187 | * function with the pre-folded one as the second |
| 4188 | * string. |
| 4189 | * FOLDEQ_S2_FOLDS_SANE |
| 4190 | */ |
| 4191 | |
| 4192 | I32 |
| 4193 | Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, |
| 4194 | const char *s2, char **pe2, UV l2, bool u2, |
| 4195 | U32 flags) |
| 4196 | { |
| 4197 | const U8 *p1 = (const U8*)s1; /* Point to current char */ |
| 4198 | const U8 *p2 = (const U8*)s2; |
| 4199 | const U8 *g1 = NULL; /* goal for s1 */ |
| 4200 | const U8 *g2 = NULL; |
| 4201 | const U8 *e1 = NULL; /* Don't scan s1 past this */ |
| 4202 | U8 *f1 = NULL; /* Point to current folded */ |
| 4203 | const U8 *e2 = NULL; |
| 4204 | U8 *f2 = NULL; |
| 4205 | STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */ |
| 4206 | U8 foldbuf1[UTF8_MAXBYTES_CASE+1]; |
| 4207 | U8 foldbuf2[UTF8_MAXBYTES_CASE+1]; |
| 4208 | U8 flags_for_folder = FOLD_FLAGS_FULL; |
| 4209 | |
| 4210 | PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS; |
| 4211 | |
| 4212 | assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE)) |
| 4213 | && (( (flags & FOLDEQ_S1_ALREADY_FOLDED) |
| 4214 | && !(flags & FOLDEQ_S1_FOLDS_SANE)) |
| 4215 | || ( (flags & FOLDEQ_S2_ALREADY_FOLDED) |
| 4216 | && !(flags & FOLDEQ_S2_FOLDS_SANE))))); |
| 4217 | /* The algorithm is to trial the folds without regard to the flags on |
| 4218 | * the first line of the above assert(), and then see if the result |
| 4219 | * violates them. This means that the inputs can't be pre-folded to a |
| 4220 | * violating result, hence the assert. This could be changed, with the |
| 4221 | * addition of extra tests here for the already-folded case, which would |
| 4222 | * slow it down. That cost is more than any possible gain for when these |
| 4223 | * flags are specified, as the flags indicate /il or /iaa matching which |
| 4224 | * is less common than /iu, and I (khw) also believe that real-world /il |
| 4225 | * and /iaa matches are most likely to involve code points 0-255, and this |
| 4226 | * function only under rare conditions gets called for 0-255. */ |
| 4227 | |
| 4228 | if (flags & FOLDEQ_LOCALE) { |
| 4229 | if (IN_UTF8_CTYPE_LOCALE) { |
| 4230 | if (UNLIKELY(PL_in_utf8_turkic_locale)) { |
| 4231 | flags_for_folder |= FOLD_FLAGS_LOCALE; |
| 4232 | } |
| 4233 | else { |
| 4234 | flags &= ~FOLDEQ_LOCALE; |
| 4235 | } |
| 4236 | } |
| 4237 | else { |
| 4238 | flags_for_folder |= FOLD_FLAGS_LOCALE; |
| 4239 | } |
| 4240 | } |
| 4241 | if (flags & FOLDEQ_UTF8_NOMIX_ASCII) { |
| 4242 | flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII; |
| 4243 | } |
| 4244 | |
| 4245 | if (pe1) { |
| 4246 | e1 = *(U8**)pe1; |
| 4247 | } |
| 4248 | |
| 4249 | if (l1) { |
| 4250 | g1 = (const U8*)s1 + l1; |
| 4251 | } |
| 4252 | |
| 4253 | if (pe2) { |
| 4254 | e2 = *(U8**)pe2; |
| 4255 | } |
| 4256 | |
| 4257 | if (l2) { |
| 4258 | g2 = (const U8*)s2 + l2; |
| 4259 | } |
| 4260 | |
| 4261 | /* Must have at least one goal */ |
| 4262 | assert(g1 || g2); |
| 4263 | |
| 4264 | if (g1) { |
| 4265 | |
| 4266 | /* Will never match if goal is out-of-bounds */ |
| 4267 | assert(! e1 || e1 >= g1); |
| 4268 | |
| 4269 | /* Here, there isn't an end pointer, or it is beyond the goal. We |
| 4270 | * only go as far as the goal */ |
| 4271 | e1 = g1; |
| 4272 | } |
| 4273 | else { |
| 4274 | assert(e1); /* Must have an end for looking at s1 */ |
| 4275 | } |
| 4276 | |
| 4277 | /* Same for goal for s2 */ |
| 4278 | if (g2) { |
| 4279 | assert(! e2 || e2 >= g2); |
| 4280 | e2 = g2; |
| 4281 | } |
| 4282 | else { |
| 4283 | assert(e2); |
| 4284 | } |
| 4285 | |
| 4286 | /* If both operands are already folded, we could just do a memEQ on the |
| 4287 | * whole strings at once, but it would be better if the caller realized |
| 4288 | * this and didn't even call us */ |
| 4289 | |
| 4290 | /* Look through both strings, a character at a time */ |
| 4291 | while (p1 < e1 && p2 < e2) { |
| 4292 | |
| 4293 | /* If at the beginning of a new character in s1, get its fold to use |
| 4294 | * and the length of the fold. */ |
| 4295 | if (n1 == 0) { |
| 4296 | if (flags & FOLDEQ_S1_ALREADY_FOLDED) { |
| 4297 | f1 = (U8 *) p1; |
| 4298 | assert(u1); |
| 4299 | n1 = UTF8SKIP(f1); |
| 4300 | } |
| 4301 | else { |
| 4302 | if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) { |
| 4303 | |
| 4304 | /* We have to forbid mixing ASCII with non-ASCII if the |
| 4305 | * flags so indicate. And, we can short circuit having to |
| 4306 | * call the general functions for this common ASCII case, |
| 4307 | * all of whose non-locale folds are also ASCII, and hence |
| 4308 | * UTF-8 invariants, so the UTF8ness of the strings is not |
| 4309 | * relevant. */ |
| 4310 | if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) { |
| 4311 | return 0; |
| 4312 | } |
| 4313 | n1 = 1; |
| 4314 | *foldbuf1 = toFOLD(*p1); |
| 4315 | } |
| 4316 | else if (u1) { |
| 4317 | _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder); |
| 4318 | } |
| 4319 | else { /* Not UTF-8, get UTF-8 fold */ |
| 4320 | _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder); |
| 4321 | } |
| 4322 | f1 = foldbuf1; |
| 4323 | } |
| 4324 | } |
| 4325 | |
| 4326 | if (n2 == 0) { /* Same for s2 */ |
| 4327 | if (flags & FOLDEQ_S2_ALREADY_FOLDED) { |
| 4328 | |
| 4329 | /* Point to the already-folded character. But for non-UTF-8 |
| 4330 | * variants, convert to UTF-8 for the algorithm below */ |
| 4331 | if (UTF8_IS_INVARIANT(*p2)) { |
| 4332 | f2 = (U8 *) p2; |
| 4333 | n2 = 1; |
| 4334 | } |
| 4335 | else if (u2) { |
| 4336 | f2 = (U8 *) p2; |
| 4337 | n2 = UTF8SKIP(f2); |
| 4338 | } |
| 4339 | else { |
| 4340 | foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2); |
| 4341 | foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2); |
| 4342 | f2 = foldbuf2; |
| 4343 | n2 = 2; |
| 4344 | } |
| 4345 | } |
| 4346 | else { |
| 4347 | if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) { |
| 4348 | if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) { |
| 4349 | return 0; |
| 4350 | } |
| 4351 | n2 = 1; |
| 4352 | *foldbuf2 = toFOLD(*p2); |
| 4353 | } |
| 4354 | else if (u2) { |
| 4355 | _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder); |
| 4356 | } |
| 4357 | else { |
| 4358 | _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder); |
| 4359 | } |
| 4360 | f2 = foldbuf2; |
| 4361 | } |
| 4362 | } |
| 4363 | |
| 4364 | /* Here f1 and f2 point to the beginning of the strings to compare. |
| 4365 | * These strings are the folds of the next character from each input |
| 4366 | * string, stored in UTF-8. */ |
| 4367 | |
| 4368 | /* While there is more to look for in both folds, see if they |
| 4369 | * continue to match */ |
| 4370 | while (n1 && n2) { |
| 4371 | U8 fold_length = UTF8SKIP(f1); |
| 4372 | if (fold_length != UTF8SKIP(f2) |
| 4373 | || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE |
| 4374 | function call for single |
| 4375 | byte */ |
| 4376 | || memNE((char*)f1, (char*)f2, fold_length)) |
| 4377 | { |
| 4378 | return 0; /* mismatch */ |
| 4379 | } |
| 4380 | |
| 4381 | /* Here, they matched, advance past them */ |
| 4382 | n1 -= fold_length; |
| 4383 | f1 += fold_length; |
| 4384 | n2 -= fold_length; |
| 4385 | f2 += fold_length; |
| 4386 | } |
| 4387 | |
| 4388 | /* When reach the end of any fold, advance the input past it */ |
| 4389 | if (n1 == 0) { |
| 4390 | p1 += u1 ? UTF8SKIP(p1) : 1; |
| 4391 | } |
| 4392 | if (n2 == 0) { |
| 4393 | p2 += u2 ? UTF8SKIP(p2) : 1; |
| 4394 | } |
| 4395 | } /* End of loop through both strings */ |
| 4396 | |
| 4397 | /* A match is defined by each scan that specified an explicit length |
| 4398 | * reaching its final goal, and the other not having matched a partial |
| 4399 | * character (which can happen when the fold of a character is more than one |
| 4400 | * character). */ |
| 4401 | if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) { |
| 4402 | return 0; |
| 4403 | } |
| 4404 | |
| 4405 | /* Successful match. Set output pointers */ |
| 4406 | if (pe1) { |
| 4407 | *pe1 = (char*)p1; |
| 4408 | } |
| 4409 | if (pe2) { |
| 4410 | *pe2 = (char*)p2; |
| 4411 | } |
| 4412 | return 1; |
| 4413 | } |
| 4414 | |
| 4415 | /* |
| 4416 | * ex: set ts=8 sts=4 sw=4 et: |
| 4417 | */ |