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 /* Be sure to synchronize this message with the similar one in regcomp.c */ 41 static const char cp_above_legal_max[] = 42 "Use of code point 0x%" UVXf " is not allowed; the" 43 " permissible max is 0x%" UVXf; 44 45 /* 46 =head1 Unicode Support 47 These are various utility functions for manipulating UTF8-encoded 48 strings. For the uninitiated, this is a method of representing arbitrary 49 Unicode characters as a variable number of bytes, in such a way that 50 characters in the ASCII range are unmodified, and a zero byte never appears 51 within non-zero characters. 52 53 =cut 54 */ 55 56 /* helper for Perl__force_out_malformed_utf8_message(). Like 57 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than 58 * PL_compiling */ 59 60 static void 61 S_restore_cop_warnings(pTHX_ void *p) 62 { 63 if (!specialWARN(PL_curcop->cop_warnings)) 64 PerlMemShared_free(PL_curcop->cop_warnings); 65 PL_curcop->cop_warnings = (STRLEN*)p; 66 } 67 68 69 void 70 Perl__force_out_malformed_utf8_message(pTHX_ 71 const U8 *const p, /* First byte in UTF-8 sequence */ 72 const U8 * const e, /* Final byte in sequence (may include 73 multiple chars */ 74 const U32 flags, /* Flags to pass to utf8n_to_uvchr(), 75 usually 0, or some DISALLOW flags */ 76 const bool die_here) /* If TRUE, this function does not return */ 77 { 78 /* This core-only function is to be called when a malformed UTF-8 character 79 * is found, in order to output the detailed information about the 80 * malformation before dieing. The reason it exists is for the occasions 81 * when such a malformation is fatal, but warnings might be turned off, so 82 * that normally they would not be actually output. This ensures that they 83 * do get output. Because a sequence may be malformed in more than one 84 * way, multiple messages may be generated, so we can't make them fatal, as 85 * that would cause the first one to die. 86 * 87 * Instead we pretend -W was passed to perl, then die afterwards. The 88 * flexibility is here to return to the caller so they can finish up and 89 * die themselves */ 90 U32 errors; 91 92 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE; 93 94 ENTER; 95 SAVEI8(PL_dowarn); 96 SAVESPTR(PL_curcop); 97 98 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON; 99 if (PL_curcop) { 100 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather 101 * than PL_compiling */ 102 SAVEDESTRUCTOR_X(S_restore_cop_warnings, 103 (void*)PL_curcop->cop_warnings); 104 PL_curcop->cop_warnings = pWARN_ALL; 105 } 106 107 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors); 108 109 LEAVE; 110 111 if (! errors) { 112 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should" 113 " be called only when there are errors found"); 114 } 115 116 if (die_here) { 117 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); 118 } 119 } 120 121 STATIC HV * 122 S_new_msg_hv(pTHX_ const char * const message, /* The message text */ 123 U32 categories, /* Packed warning categories */ 124 U32 flag) /* Flag associated with this message */ 125 { 126 /* Creates, populates, and returns an HV* that describes an error message 127 * for the translators between UTF8 and code point */ 128 129 SV* msg_sv = newSVpv(message, 0); 130 SV* category_sv = newSVuv(categories); 131 SV* flag_bit_sv = newSVuv(flag); 132 133 HV* msg_hv = newHV(); 134 135 PERL_ARGS_ASSERT_NEW_MSG_HV; 136 137 (void) hv_stores(msg_hv, "text", msg_sv); 138 (void) hv_stores(msg_hv, "warn_categories", category_sv); 139 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv); 140 141 return msg_hv; 142 } 143 144 /* 145 =for apidoc uvoffuni_to_utf8_flags 146 147 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 148 Instead, B<Almost all code should use L</uvchr_to_utf8> or 149 L</uvchr_to_utf8_flags>>. 150 151 This function is like them, but the input is a strict Unicode 152 (as opposed to native) code point. Only in very rare circumstances should code 153 not be using the native code point. 154 155 For details, see the description for L</uvchr_to_utf8_flags>. 156 157 =cut 158 */ 159 160 U8 * 161 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags) 162 { 163 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS; 164 165 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL); 166 } 167 168 /* All these formats take a single UV code point argument */ 169 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf; 170 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf 171 " is not recommended for open interchange"; 172 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode," 173 " may not be portable"; 174 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \ 175 " Unicode, requires a Perl extension," \ 176 " and so is not portable"; 177 178 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \ 179 STMT_START { \ 180 if (flags & UNICODE_WARN_SURROGATE) { \ 181 U32 category = packWARN(WARN_SURROGATE); \ 182 const char * format = surrogate_cp_format; \ 183 if (msgs) { \ 184 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ 185 category, \ 186 UNICODE_GOT_SURROGATE); \ 187 } \ 188 else { \ 189 Perl_ck_warner_d(aTHX_ category, format, uv); \ 190 } \ 191 } \ 192 if (flags & UNICODE_DISALLOW_SURROGATE) { \ 193 return NULL; \ 194 } \ 195 } STMT_END; 196 197 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \ 198 STMT_START { \ 199 if (flags & UNICODE_WARN_NONCHAR) { \ 200 U32 category = packWARN(WARN_NONCHAR); \ 201 const char * format = nonchar_cp_format; \ 202 if (msgs) { \ 203 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ 204 category, \ 205 UNICODE_GOT_NONCHAR); \ 206 } \ 207 else { \ 208 Perl_ck_warner_d(aTHX_ category, format, uv); \ 209 } \ 210 } \ 211 if (flags & UNICODE_DISALLOW_NONCHAR) { \ 212 return NULL; \ 213 } \ 214 } STMT_END; 215 216 /* Use shorter names internally in this file */ 217 #define SHIFT UTF_ACCUMULATION_SHIFT 218 #undef MARK 219 #define MARK UTF_CONTINUATION_MARK 220 #define MASK UTF_CONTINUATION_MASK 221 222 /* 223 =for apidoc uvchr_to_utf8_flags_msgs 224 225 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 226 227 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly. 228 229 This function is for code that wants any warning and/or error messages to be 230 returned to the caller rather than be displayed. All messages that would have 231 been displayed if all lexical warnings are enabled will be returned. 232 233 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter 234 placed after all the others, C<msgs>. If this parameter is 0, this function 235 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should 236 be a pointer to an C<HV *> variable, in which this function creates a new HV to 237 contain any appropriate messages. The hash has three key-value pairs, as 238 follows: 239 240 =over 4 241 242 =item C<text> 243 244 The text of the message as a C<SVpv>. 245 246 =item C<warn_categories> 247 248 The warning category (or categories) packed into a C<SVuv>. 249 250 =item C<flag> 251 252 A single flag bit associated with this message, in a C<SVuv>. 253 The bit corresponds to some bit in the C<*errors> return value, 254 such as C<UNICODE_GOT_SURROGATE>. 255 256 =back 257 258 It's important to note that specifying this parameter as non-null will cause 259 any warnings this function would otherwise generate to be suppressed, and 260 instead be placed in C<*msgs>. The caller can check the lexical warnings state 261 (or not) when choosing what to do with the returned messages. 262 263 The caller, of course, is responsible for freeing any returned HV. 264 265 =cut 266 */ 267 268 /* Undocumented; we don't want people using this. Instead they should use 269 * uvchr_to_utf8_flags_msgs() */ 270 U8 * 271 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs) 272 { 273 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS; 274 275 if (msgs) { 276 *msgs = NULL; 277 } 278 279 if (OFFUNI_IS_INVARIANT(uv)) { 280 *d++ = LATIN1_TO_NATIVE(uv); 281 return d; 282 } 283 284 if (uv <= MAX_UTF8_TWO_BYTE) { 285 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2)); 286 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK); 287 return d; 288 } 289 290 /* Not 2-byte; test for and handle 3-byte result. In the test immediately 291 * below, the 16 is for start bytes E0-EF (which are all the possible ones 292 * for 3 byte characters). The 2 is for 2 continuation bytes; these each 293 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000 294 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC; 295 * 0x800-0xFFFF on ASCII */ 296 if (uv < (16 * (1U << (2 * SHIFT)))) { 297 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3)); 298 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); 299 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); 300 301 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so 302 aren't tested here */ 303 /* The most likely code points in this range are below the surrogates. 304 * Do an extra test to quickly exclude those. */ 305 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) { 306 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) 307 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) 308 { 309 HANDLE_UNICODE_NONCHAR(uv, flags, msgs); 310 } 311 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 312 HANDLE_UNICODE_SURROGATE(uv, flags, msgs); 313 } 314 } 315 #endif 316 return d; 317 } 318 319 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII 320 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can 321 * happen starting with 4-byte characters on ASCII platforms. We unify the 322 * code for these with EBCDIC, even though some of them require 5-bytes on 323 * those, because khw believes the code saving is worth the very slight 324 * performance hit on these high EBCDIC code points. */ 325 326 if (UNLIKELY(UNICODE_IS_SUPER(uv))) { 327 if (UNLIKELY(uv > MAX_LEGAL_CP)) { 328 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_LEGAL_CP); 329 } 330 if ( (flags & UNICODE_WARN_SUPER) 331 || ( (flags & UNICODE_WARN_PERL_EXTENDED) 332 && UNICODE_IS_PERL_EXTENDED(uv))) 333 { 334 const char * format = super_cp_format; 335 U32 category = packWARN(WARN_NON_UNICODE); 336 U32 flag = UNICODE_GOT_SUPER; 337 338 /* Choose the more dire applicable warning */ 339 if (UNICODE_IS_PERL_EXTENDED(uv)) { 340 format = perl_extended_cp_format; 341 if (flags & (UNICODE_WARN_PERL_EXTENDED 342 |UNICODE_DISALLOW_PERL_EXTENDED)) 343 { 344 flag = UNICODE_GOT_PERL_EXTENDED; 345 } 346 } 347 348 if (msgs) { 349 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), 350 category, flag); 351 } 352 else { 353 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv); 354 } 355 } 356 if ( (flags & UNICODE_DISALLOW_SUPER) 357 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED) 358 && UNICODE_IS_PERL_EXTENDED(uv))) 359 { 360 return NULL; 361 } 362 } 363 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) { 364 HANDLE_UNICODE_NONCHAR(uv, flags, msgs); 365 } 366 367 /* Test for and handle 4-byte result. In the test immediately below, the 368 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte 369 * characters). The 3 is for 3 continuation bytes; these each contribute 370 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on 371 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC; 372 * 0x1_0000-0x1F_FFFF on ASCII */ 373 if (uv < (8 * (1U << (3 * SHIFT)))) { 374 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4)); 375 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK); 376 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); 377 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); 378 379 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte 380 characters. The end-plane non-characters for EBCDIC were 381 handled just above */ 382 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) { 383 HANDLE_UNICODE_NONCHAR(uv, flags, msgs); 384 } 385 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 386 HANDLE_UNICODE_SURROGATE(uv, flags, msgs); 387 } 388 #endif 389 390 return d; 391 } 392 393 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII 394 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop 395 * format. The unrolled version above turns out to not save all that much 396 * time, and at these high code points (well above the legal Unicode range 397 * on ASCII platforms, and well above anything in common use in EBCDIC), 398 * khw believes that less code outweighs slight performance gains. */ 399 400 { 401 STRLEN len = OFFUNISKIP(uv); 402 U8 *p = d+len-1; 403 while (p > d) { 404 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK); 405 uv >>= SHIFT; 406 } 407 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len)); 408 return d+len; 409 } 410 } 411 412 /* 413 =for apidoc uvchr_to_utf8 414 415 Adds the UTF-8 representation of the native code point C<uv> to the end 416 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to 417 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to 418 the byte after the end of the new character. In other words, 419 420 d = uvchr_to_utf8(d, uv); 421 422 is the recommended wide native character-aware way of saying 423 424 *(d++) = uv; 425 426 This function accepts any code point from 0..C<IV_MAX> as input. 427 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. 428 429 It is possible to forbid or warn on non-Unicode code points, or those that may 430 be problematic by using L</uvchr_to_utf8_flags>. 431 432 =cut 433 */ 434 435 /* This is also a macro */ 436 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv); 437 438 U8 * 439 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv) 440 { 441 return uvchr_to_utf8(d, uv); 442 } 443 444 /* 445 =for apidoc uvchr_to_utf8_flags 446 447 Adds the UTF-8 representation of the native code point C<uv> to the end 448 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to 449 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to 450 the byte after the end of the new character. In other words, 451 452 d = uvchr_to_utf8_flags(d, uv, flags); 453 454 or, in most cases, 455 456 d = uvchr_to_utf8_flags(d, uv, 0); 457 458 This is the Unicode-aware way of saying 459 460 *(d++) = uv; 461 462 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as 463 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. 464 465 Specifying C<flags> can further restrict what is allowed and not warned on, as 466 follows: 467 468 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set, 469 the function will raise a warning, provided UTF8 warnings are enabled. If 470 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return 471 NULL. If both flags are set, the function will both warn and return NULL. 472 473 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags 474 affect how the function handles a Unicode non-character. 475 476 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags 477 affect the handling of code points that are above the Unicode maximum of 478 0x10FFFF. Languages other than Perl may not be able to accept files that 479 contain these. 480 481 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of 482 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all 483 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the 484 allowed inputs to the strict UTF-8 traditionally defined by Unicode. 485 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and 486 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the 487 above-Unicode and surrogate flags, but not the non-character ones, as 488 defined in 489 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. 490 See L<perlunicode/Noncharacter code points>. 491 492 Extremely high code points were never specified in any standard, and require an 493 extension to UTF-8 to express, which Perl does. It is likely that programs 494 written in something other than Perl would not be able to read files that 495 contain these; nor would Perl understand files written by something that uses a 496 different extension. For these reasons, there is a separate set of flags that 497 can warn and/or disallow these extremely high code points, even if other 498 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED> 499 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see 500 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will 501 treat all above-Unicode code points, including these, as malformations. (Note 502 that the Unicode standard considers anything above 0x10FFFF to be illegal, but 503 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1)) 504 505 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is 506 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly, 507 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named 508 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC 509 platforms,these flags can apply to code points that actually do fit in 31 bits. 510 The new names accurately describe the situation in all cases. 511 512 =cut 513 */ 514 515 /* This is also a macro */ 516 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags); 517 518 U8 * 519 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) 520 { 521 return uvchr_to_utf8_flags(d, uv, flags); 522 } 523 524 #ifndef UV_IS_QUAD 525 526 STATIC int 527 S_is_utf8_cp_above_31_bits(const U8 * const s, 528 const U8 * const e, 529 const bool consider_overlongs) 530 { 531 /* Returns TRUE if the first code point represented by the Perl-extended- 532 * UTF-8-encoded string starting at 's', and looking no further than 'e - 533 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31. 534 * 535 * The function handles the case where the input bytes do not include all 536 * the ones necessary to represent a full character. That is, they may be 537 * the intial bytes of the representation of a code point, but possibly 538 * the final ones necessary for the complete representation may be beyond 539 * 'e - 1'. 540 * 541 * The function also can handle the case where the input is an overlong 542 * sequence. If 'consider_overlongs' is 0, the function assumes the 543 * input is not overlong, without checking, and will return based on that 544 * assumption. If this parameter is 1, the function will go to the trouble 545 * of figuring out if it actually evaluates to above or below 31 bits. 546 * 547 * The sequence is otherwise assumed to be well-formed, without checking. 548 */ 549 550 const STRLEN len = e - s; 551 int is_overlong; 552 553 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS; 554 555 assert(! UTF8_IS_INVARIANT(*s) && e > s); 556 557 #ifdef EBCDIC 558 559 PERL_UNUSED_ARG(consider_overlongs); 560 561 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can 562 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can 563 * also be the start byte for a 31-bit code point; we need at least 2 564 * bytes, and maybe up through 8 bytes, to determine that. (It can also be 565 * the start byte for an overlong sequence, but for 30-bit or smaller code 566 * points, so we don't have to worry about overlongs on EBCDIC.) */ 567 if (*s != 0xFE) { 568 return 0; 569 } 570 571 if (len == 1) { 572 return -1; 573 } 574 575 #else 576 577 /* On ASCII, FE and FF are the only start bytes that can evaluate to 578 * needing more than 31 bits. */ 579 if (LIKELY(*s < 0xFE)) { 580 return 0; 581 } 582 583 /* What we have left are FE and FF. Both of these require more than 31 584 * bits unless they are for overlongs. */ 585 if (! consider_overlongs) { 586 return 1; 587 } 588 589 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to 590 * above 31 bits. But we need more than one byte to discern this, so if 591 * passed just the start byte, it could be an overlong evaluating to 592 * smaller */ 593 if (len == 1) { 594 return -1; 595 } 596 597 /* Having excluded len==1, and knowing that FE and FF are both valid start 598 * bytes, we can call the function below to see if the sequence is 599 * overlong. (We don't need the full generality of the called function, 600 * but for these huge code points, speed shouldn't be a consideration, and 601 * the compiler does have enough information, since it's static to this 602 * file, to optimize to just the needed parts.) */ 603 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len); 604 605 /* If it isn't overlong, more than 31 bits are required. */ 606 if (is_overlong == 0) { 607 return 1; 608 } 609 610 /* If it is indeterminate if it is overlong, return that */ 611 if (is_overlong < 0) { 612 return -1; 613 } 614 615 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as 616 * the max it can be is 2**31 - 1 */ 617 if (*s == 0xFE) { 618 return 0; 619 } 620 621 #endif 622 623 /* Here, ASCII and EBCDIC rejoin: 624 * On ASCII: We have an overlong sequence starting with FF 625 * On EBCDIC: We have a sequence starting with FE. */ 626 627 { /* For C89, use a block so the declaration can be close to its use */ 628 629 #ifdef EBCDIC 630 631 /* U+7FFFFFFF (2 ** 31 - 1) 632 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13 633 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73 634 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72 635 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75 636 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF 637 * U+80000000 (2 ** 31): 638 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 639 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 640 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 641 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0 642 * 643 * and since we know that *s = \xfe, any continuation sequcence 644 * following it that is gt the below is above 31 bits 645 [0] [1] [2] [3] [4] [5] [6] */ 646 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42"; 647 648 #else 649 650 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1) 651 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF 652 * FF overlong for U+80000000 (2 ** 31): 653 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80 654 * and since we know that *s = \xff, any continuation sequcence 655 * following it that is gt the below is above 30 bits 656 [0] [1] [2] [3] [4] [5] [6] */ 657 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81"; 658 659 660 #endif 661 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1; 662 const STRLEN cmp_len = MIN(conts_len, len - 1); 663 664 /* Now compare the continuation bytes in s with the ones we have 665 * compiled in that are for the largest 30 bit code point. If we have 666 * enough bytes available to determine the answer, or the bytes we do 667 * have differ from them, we can compare the two to get a definitive 668 * answer (Note that in UTF-EBCDIC, the two lowest possible 669 * continuation bytes are \x41 and \x42.) */ 670 if (cmp_len >= conts_len || memNE(s + 1, 671 conts_for_highest_30_bit, 672 cmp_len)) 673 { 674 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len)); 675 } 676 677 /* Here, all the bytes we have are the same as the highest 30-bit code 678 * point, but we are missing so many bytes that we can't make the 679 * determination */ 680 return -1; 681 } 682 } 683 684 #endif 685 686 PERL_STATIC_INLINE int 687 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len) 688 { 689 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to 690 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if 691 * it isn't, and -1 if there isn't enough information to tell. This last 692 * return value can happen if the sequence is incomplete, missing some 693 * trailing bytes that would form a complete character. If there are 694 * enough bytes to make a definitive decision, this function does so. 695 * Usually 2 bytes sufficient. 696 * 697 * Overlongs can occur whenever the number of continuation bytes changes. 698 * That means whenever the number of leading 1 bits in a start byte 699 * increases from the next lower start byte. That happens for start bytes 700 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal 701 * start bytes have already been excluded, so don't need to be tested here; 702 * ASCII platforms: C0, C1 703 * EBCDIC platforms C0, C1, C2, C3, C4, E0 704 */ 705 706 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); 707 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); 708 709 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK; 710 assert(len > 1 && UTF8_IS_START(*s)); 711 712 /* Each platform has overlongs after the start bytes given above (expressed 713 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but 714 * the logic is the same, except the E0 overlong has already been excluded 715 * on EBCDIC platforms. The values below were found by manually 716 * inspecting the UTF-8 patterns. See the tables in utf8.h and 717 * utfebcdic.h. */ 718 719 # ifdef EBCDIC 720 # define F0_ABOVE_OVERLONG 0xB0 721 # define F8_ABOVE_OVERLONG 0xA8 722 # define FC_ABOVE_OVERLONG 0xA4 723 # define FE_ABOVE_OVERLONG 0xA2 724 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41" 725 /* I8(0xfe) is FF */ 726 # else 727 728 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) { 729 return 1; 730 } 731 732 # define F0_ABOVE_OVERLONG 0x90 733 # define F8_ABOVE_OVERLONG 0x88 734 # define FC_ABOVE_OVERLONG 0x84 735 # define FE_ABOVE_OVERLONG 0x82 736 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80" 737 # endif 738 739 740 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG)) 741 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG)) 742 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG)) 743 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG))) 744 { 745 return 1; 746 } 747 748 /* Check for the FF overlong */ 749 return isFF_OVERLONG(s, len); 750 } 751 752 PERL_STATIC_INLINE int 753 S_isFF_OVERLONG(const U8 * const s, const STRLEN len) 754 { 755 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to 756 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if 757 * it isn't, and -1 if there isn't enough information to tell. This last 758 * return value can happen if the sequence is incomplete, missing some 759 * trailing bytes that would form a complete character. If there are 760 * enough bytes to make a definitive decision, this function does so. */ 761 762 PERL_ARGS_ASSERT_ISFF_OVERLONG; 763 764 /* To be an FF overlong, all the available bytes must match */ 765 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX, 766 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1)))) 767 { 768 return 0; 769 } 770 771 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must 772 * be there; what comes after them doesn't matter. See tables in utf8.h, 773 * utfebcdic.h. */ 774 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) { 775 return 1; 776 } 777 778 /* The missing bytes could cause the result to go one way or the other, so 779 * the result is indeterminate */ 780 return -1; 781 } 782 783 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */ 784 # ifdef EBCDIC /* Actually is I8 */ 785 # define HIGHEST_REPRESENTABLE_UTF8 \ 786 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 787 # else 788 # define HIGHEST_REPRESENTABLE_UTF8 \ 789 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 790 # endif 791 #endif 792 793 PERL_STATIC_INLINE int 794 S_does_utf8_overflow(const U8 * const s, 795 const U8 * e, 796 const bool consider_overlongs) 797 { 798 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to 799 * 'e' - 1 would overflow an IV on this platform; that is if it represents 800 * a code point larger than the highest representable code point. It 801 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't 802 * enough information to tell. This last return value can happen if the 803 * sequence is incomplete, missing some trailing bytes that would form a 804 * complete character. If there are enough bytes to make a definitive 805 * decision, this function does so. 806 * 807 * If 'consider_overlongs' is TRUE, the function checks for the possibility 808 * that the sequence is an overlong that doesn't overflow. Otherwise, it 809 * assumes the sequence is not an overlong. This can give different 810 * results only on ASCII 32-bit platforms. 811 * 812 * (For ASCII platforms, we could use memcmp() because we don't have to 813 * convert each byte to I8, but it's very rare input indeed that would 814 * approach overflow, so the loop below will likely only get executed once.) 815 * 816 * 'e' - 1 must not be beyond a full character. */ 817 818 819 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW; 820 assert(s <= e && s + UTF8SKIP(s) >= e); 821 822 #if ! defined(UV_IS_QUAD) 823 824 return is_utf8_cp_above_31_bits(s, e, consider_overlongs); 825 826 #else 827 828 PERL_UNUSED_ARG(consider_overlongs); 829 830 { 831 const STRLEN len = e - s; 832 const U8 *x; 833 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8; 834 835 for (x = s; x < e; x++, y++) { 836 837 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) { 838 continue; 839 } 840 841 /* If this byte is larger than the corresponding highest UTF-8 842 * byte, the sequence overflow; otherwise the byte is less than, 843 * and so the sequence doesn't overflow */ 844 return NATIVE_UTF8_TO_I8(*x) > *y; 845 846 } 847 848 /* Got to the end and all bytes are the same. If the input is a whole 849 * character, it doesn't overflow. And if it is a partial character, 850 * there's not enough information to tell */ 851 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) { 852 return -1; 853 } 854 855 return 0; 856 } 857 858 #endif 859 860 } 861 862 #if 0 863 864 /* This is the portions of the above function that deal with UV_MAX instead of 865 * IV_MAX. They are left here in case we want to combine them so that internal 866 * uses can have larger code points. The only logic difference is that the 867 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has 868 * different logic. 869 */ 870 871 /* Anything larger than this will overflow the word if it were converted into a UV */ 872 #if defined(UV_IS_QUAD) 873 # ifdef EBCDIC /* Actually is I8 */ 874 # define HIGHEST_REPRESENTABLE_UTF8 \ 875 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 876 # else 877 # define HIGHEST_REPRESENTABLE_UTF8 \ 878 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 879 # endif 880 #else /* 32-bit */ 881 # ifdef EBCDIC 882 # define HIGHEST_REPRESENTABLE_UTF8 \ 883 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF" 884 # else 885 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF" 886 # endif 887 #endif 888 889 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC) 890 891 /* On 32 bit ASCII machines, many overlongs that start with FF don't 892 * overflow */ 893 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) { 894 895 /* To be such an overlong, the first bytes of 's' must match 896 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we 897 * don't have any additional bytes available, the sequence, when 898 * completed might or might not fit in 32 bits. But if we have that 899 * next byte, we can tell for sure. If it is <= 0x83, then it does 900 * fit. */ 901 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) { 902 return -1; 903 } 904 905 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83; 906 } 907 908 /* Starting with the #else, the rest of the function is identical except 909 * 1. we need to move the 'len' declaration to be global to the function 910 * 2. the endif move to just after the UNUSED_ARG. 911 * An empty endif is given just below to satisfy the preprocessor 912 */ 913 #endif 914 915 #endif 916 917 #undef F0_ABOVE_OVERLONG 918 #undef F8_ABOVE_OVERLONG 919 #undef FC_ABOVE_OVERLONG 920 #undef FE_ABOVE_OVERLONG 921 #undef FF_OVERLONG_PREFIX 922 923 STRLEN 924 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags) 925 { 926 STRLEN len; 927 const U8 *x; 928 929 /* A helper function that should not be called directly. 930 * 931 * This function returns non-zero if the string beginning at 's' and 932 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a 933 * code point; otherwise it returns 0. The examination stops after the 934 * first code point in 's' is validated, not looking at the rest of the 935 * input. If 'e' is such that there are not enough bytes to represent a 936 * complete code point, this function will return non-zero anyway, if the 937 * bytes it does have are well-formed UTF-8 as far as they go, and aren't 938 * excluded by 'flags'. 939 * 940 * A non-zero return gives the number of bytes required to represent the 941 * code point. Be aware that if the input is for a partial character, the 942 * return will be larger than 'e - s'. 943 * 944 * This function assumes that the code point represented is UTF-8 variant. 945 * The caller should have excluded the possibility of it being invariant 946 * before calling this function. 947 * 948 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags 949 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return 950 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but 951 * disallowed by the flags. If the input is only for a partial character, 952 * the function will return non-zero if there is any sequence of 953 * well-formed UTF-8 that, when appended to the input sequence, could 954 * result in an allowed code point; otherwise it returns 0. Non characters 955 * cannot be determined based on partial character input. But many of the 956 * other excluded types can be determined with just the first one or two 957 * bytes. 958 * 959 */ 960 961 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER; 962 963 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE 964 |UTF8_DISALLOW_PERL_EXTENDED))); 965 assert(! UTF8_IS_INVARIANT(*s)); 966 967 /* A variant char must begin with a start byte */ 968 if (UNLIKELY(! UTF8_IS_START(*s))) { 969 return 0; 970 } 971 972 /* Examine a maximum of a single whole code point */ 973 if (e - s > UTF8SKIP(s)) { 974 e = s + UTF8SKIP(s); 975 } 976 977 len = e - s; 978 979 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) { 980 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); 981 982 /* Here, we are disallowing some set of largish code points, and the 983 * first byte indicates the sequence is for a code point that could be 984 * in the excluded set. We generally don't have to look beyond this or 985 * the second byte to see if the sequence is actually for one of the 986 * excluded classes. The code below is derived from this table: 987 * 988 * UTF-8 UTF-EBCDIC I8 989 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate 990 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate 991 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode 992 * 993 * Keep in mind that legal continuation bytes range between \x80..\xBF 994 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't 995 * continuation bytes. Hence, we don't have to test the upper edge 996 * because if any of those is encountered, the sequence is malformed, 997 * and would fail elsewhere in this function. 998 * 999 * The code here likewise assumes that there aren't other 1000 * malformations; again the function should fail elsewhere because of 1001 * these. For example, an overlong beginning with FC doesn't actually 1002 * have to be a super; it could actually represent a small code point, 1003 * even U+0000. But, since overlongs (and other malformations) are 1004 * illegal, the function should return FALSE in either case. 1005 */ 1006 1007 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */ 1008 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA 1009 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2) 1010 1011 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \ 1012 /* B6 and B7 */ \ 1013 && ((s1) & 0xFE ) == 0xB6) 1014 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF)) 1015 #else 1016 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5 1017 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90) 1018 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0) 1019 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE) 1020 #endif 1021 1022 if ( (flags & UTF8_DISALLOW_SUPER) 1023 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) 1024 { 1025 return 0; /* Above Unicode */ 1026 } 1027 1028 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED) 1029 && UNLIKELY(isUTF8_PERL_EXTENDED(s))) 1030 { 1031 return 0; 1032 } 1033 1034 if (len > 1) { 1035 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); 1036 1037 if ( (flags & UTF8_DISALLOW_SUPER) 1038 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1))) 1039 { 1040 return 0; /* Above Unicode */ 1041 } 1042 1043 if ( (flags & UTF8_DISALLOW_SURROGATE) 1044 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1))) 1045 { 1046 return 0; /* Surrogate */ 1047 } 1048 1049 if ( (flags & UTF8_DISALLOW_NONCHAR) 1050 && UNLIKELY(UTF8_IS_NONCHAR(s, e))) 1051 { 1052 return 0; /* Noncharacter code point */ 1053 } 1054 } 1055 } 1056 1057 /* Make sure that all that follows are continuation bytes */ 1058 for (x = s + 1; x < e; x++) { 1059 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) { 1060 return 0; 1061 } 1062 } 1063 1064 /* Here is syntactically valid. Next, make sure this isn't the start of an 1065 * overlong. */ 1066 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) { 1067 return 0; 1068 } 1069 1070 /* And finally, that the code point represented fits in a word on this 1071 * platform */ 1072 if (0 < does_utf8_overflow(s, e, 1073 0 /* Don't consider overlongs */ 1074 )) 1075 { 1076 return 0; 1077 } 1078 1079 return UTF8SKIP(s); 1080 } 1081 1082 char * 1083 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format) 1084 { 1085 /* Returns a mortalized C string that is a displayable copy of the 'len' 1086 * bytes starting at 'start'. 'format' gives how to display each byte. 1087 * Currently, there are only two formats, so it is currently a bool: 1088 * 0 \xab 1089 * 1 ab (that is a space between two hex digit bytes) 1090 */ 1091 1092 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a 1093 trailing NUL */ 1094 const U8 * s = start; 1095 const U8 * const e = start + len; 1096 char * output; 1097 char * d; 1098 1099 PERL_ARGS_ASSERT__BYTE_DUMP_STRING; 1100 1101 Newx(output, output_len, char); 1102 SAVEFREEPV(output); 1103 1104 d = output; 1105 for (s = start; s < e; s++) { 1106 const unsigned high_nibble = (*s & 0xF0) >> 4; 1107 const unsigned low_nibble = (*s & 0x0F); 1108 1109 if (format) { 1110 if (s > start) { 1111 *d++ = ' '; 1112 } 1113 } 1114 else { 1115 *d++ = '\\'; 1116 *d++ = 'x'; 1117 } 1118 1119 if (high_nibble < 10) { 1120 *d++ = high_nibble + '0'; 1121 } 1122 else { 1123 *d++ = high_nibble - 10 + 'a'; 1124 } 1125 1126 if (low_nibble < 10) { 1127 *d++ = low_nibble + '0'; 1128 } 1129 else { 1130 *d++ = low_nibble - 10 + 'a'; 1131 } 1132 } 1133 1134 *d = '\0'; 1135 return output; 1136 } 1137 1138 PERL_STATIC_INLINE char * 1139 S_unexpected_non_continuation_text(pTHX_ const U8 * const s, 1140 1141 /* Max number of bytes to print */ 1142 STRLEN print_len, 1143 1144 /* Which one is the non-continuation */ 1145 const STRLEN non_cont_byte_pos, 1146 1147 /* How many bytes should there be? */ 1148 const STRLEN expect_len) 1149 { 1150 /* Return the malformation warning text for an unexpected continuation 1151 * byte. */ 1152 1153 const char * const where = (non_cont_byte_pos == 1) 1154 ? "immediately" 1155 : Perl_form(aTHX_ "%d bytes", 1156 (int) non_cont_byte_pos); 1157 const U8 * x = s + non_cont_byte_pos; 1158 const U8 * e = s + print_len; 1159 1160 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT; 1161 1162 /* We don't need to pass this parameter, but since it has already been 1163 * calculated, it's likely faster to pass it; verify under DEBUGGING */ 1164 assert(expect_len == UTF8SKIP(s)); 1165 1166 /* As a defensive coding measure, don't output anything past a NUL. Such 1167 * bytes shouldn't be in the middle of a malformation, and could mark the 1168 * end of the allocated string, and what comes after is undefined */ 1169 for (; x < e; x++) { 1170 if (*x == '\0') { 1171 x++; /* Output this particular NUL */ 1172 break; 1173 } 1174 } 1175 1176 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x," 1177 " %s after start byte 0x%02x; need %d bytes, got %d)", 1178 malformed_text, 1179 _byte_dump_string(s, x - s, 0), 1180 *(s + non_cont_byte_pos), 1181 where, 1182 *s, 1183 (int) expect_len, 1184 (int) non_cont_byte_pos); 1185 } 1186 1187 /* 1188 1189 =for apidoc utf8n_to_uvchr 1190 1191 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 1192 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. 1193 1194 Bottom level UTF-8 decode routine. 1195 Returns the native code point value of the first character in the string C<s>, 1196 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than 1197 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to 1198 the length, in bytes, of that character. 1199 1200 The value of C<flags> determines the behavior when C<s> does not point to a 1201 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation 1202 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) 1203 is the next possible position in C<s> that could begin a non-malformed 1204 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning 1205 is raised. Some UTF-8 input sequences may contain multiple malformations. 1206 This function tries to find every possible one in each call, so multiple 1207 warnings can be raised for the same sequence. 1208 1209 Various ALLOW flags can be set in C<flags> to allow (and not warn on) 1210 individual types of malformations, such as the sequence being overlong (that 1211 is, when there is a shorter sequence that can express the same code point; 1212 overlong sequences are expressly forbidden in the UTF-8 standard due to 1213 potential security issues). Another malformation example is the first byte of 1214 a character not being a legal first byte. See F<utf8.h> for the list of such 1215 flags. Even if allowed, this function generally returns the Unicode 1216 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in 1217 F<utf8.h> to override this behavior for the overlong malformations, but don't 1218 do that except for very specialized purposes. 1219 1220 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other 1221 flags) malformation is found. If this flag is set, the routine assumes that 1222 the caller will raise a warning, and this function will silently just set 1223 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero. 1224 1225 Note that this API requires disambiguation between successful decoding a C<NUL> 1226 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as 1227 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may 1228 be set to 1. To disambiguate, upon a zero return, see if the first byte of 1229 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an 1230 error. Or you can use C<L</utf8n_to_uvchr_error>>. 1231 1232 Certain code points are considered problematic. These are Unicode surrogates, 1233 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF. 1234 By default these are considered regular code points, but certain situations 1235 warrant special handling for them, which can be specified using the C<flags> 1236 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all 1237 three classes are treated as malformations and handled as such. The flags 1238 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and 1239 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to 1240 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE> 1241 restricts the allowed inputs to the strict UTF-8 traditionally defined by 1242 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness 1243 definition given by 1244 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. 1245 The difference between traditional strictness and C9 strictness is that the 1246 latter does not forbid non-character code points. (They are still discouraged, 1247 however.) For more discussion see L<perlunicode/Noncharacter code points>. 1248 1249 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, 1250 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>, 1251 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be 1252 raised for their respective categories, but otherwise the code points are 1253 considered valid (not malformations). To get a category to both be treated as 1254 a malformation and raise a warning, specify both the WARN and DISALLOW flags. 1255 (But note that warnings are not raised if lexically disabled nor if 1256 C<UTF8_CHECK_ONLY> is also specified.) 1257 1258 Extremely high code points were never specified in any standard, and require an 1259 extension to UTF-8 to express, which Perl does. It is likely that programs 1260 written in something other than Perl would not be able to read files that 1261 contain these; nor would Perl understand files written by something that uses a 1262 different extension. For these reasons, there is a separate set of flags that 1263 can warn and/or disallow these extremely high code points, even if other 1264 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and 1265 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see 1266 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all 1267 above-Unicode code points, including these, as malformations. 1268 (Note that the Unicode standard considers anything above 0x10FFFF to be 1269 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF 1270 (2**31 -1)) 1271 1272 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is 1273 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly, 1274 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named 1275 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags 1276 can apply to code points that actually do fit in 31 bits. This happens on 1277 EBCDIC platforms, and sometimes when the L<overlong 1278 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately 1279 describe the situation in all cases. 1280 1281 1282 All other code points corresponding to Unicode characters, including private 1283 use and those yet to be assigned, are never considered malformed and never 1284 warn. 1285 1286 =cut 1287 1288 Also implemented as a macro in utf8.h 1289 */ 1290 1291 UV 1292 Perl_utf8n_to_uvchr(const U8 *s, 1293 STRLEN curlen, 1294 STRLEN *retlen, 1295 const U32 flags) 1296 { 1297 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR; 1298 1299 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL); 1300 } 1301 1302 /* 1303 1304 =for apidoc utf8n_to_uvchr_error 1305 1306 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 1307 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. 1308 1309 This function is for code that needs to know what the precise malformation(s) 1310 are when an error is found. If you also need to know the generated warning 1311 messages, use L</utf8n_to_uvchr_msgs>() instead. 1312 1313 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after 1314 all the others, C<errors>. If this parameter is 0, this function behaves 1315 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer 1316 to a C<U32> variable, which this function sets to indicate any errors found. 1317 Upon return, if C<*errors> is 0, there were no errors found. Otherwise, 1318 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some 1319 of these bits will be set if a malformation is found, even if the input 1320 C<flags> parameter indicates that the given malformation is allowed; those 1321 exceptions are noted: 1322 1323 =over 4 1324 1325 =item C<UTF8_GOT_PERL_EXTENDED> 1326 1327 The input sequence is not standard UTF-8, but a Perl extension. This bit is 1328 set only if the input C<flags> parameter contains either the 1329 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags. 1330 1331 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, 1332 and so some extension must be used to express them. Perl uses a natural 1333 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further 1334 extension to represent even higher ones, so that any code point that fits in a 1335 64-bit word can be represented. Text using these extensions is not likely to 1336 be portable to non-Perl code. We lump both of these extensions together and 1337 refer to them as Perl extended UTF-8. There exist other extensions that people 1338 have invented, incompatible with Perl's. 1339 1340 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing 1341 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower 1342 than on ASCII. Prior to that, code points 2**31 and higher were simply 1343 unrepresentable, and a different, incompatible method was used to represent 1344 code points between 2**30 and 2**31 - 1. 1345 1346 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if 1347 Perl extended UTF-8 is used. 1348 1349 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still 1350 may use for backward compatibility. That name is misleading, as this flag may 1351 be set when the code point actually does fit in 31 bits. This happens on 1352 EBCDIC platforms, and sometimes when the L<overlong 1353 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately 1354 describes the situation in all cases. 1355 1356 =item C<UTF8_GOT_CONTINUATION> 1357 1358 The input sequence was malformed in that the first byte was a a UTF-8 1359 continuation byte. 1360 1361 =item C<UTF8_GOT_EMPTY> 1362 1363 The input C<curlen> parameter was 0. 1364 1365 =item C<UTF8_GOT_LONG> 1366 1367 The input sequence was malformed in that there is some other sequence that 1368 evaluates to the same code point, but that sequence is shorter than this one. 1369 1370 Until Unicode 3.1, it was legal for programs to accept this malformation, but 1371 it was discovered that this created security issues. 1372 1373 =item C<UTF8_GOT_NONCHAR> 1374 1375 The code point represented by the input UTF-8 sequence is for a Unicode 1376 non-character code point. 1377 This bit is set only if the input C<flags> parameter contains either the 1378 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags. 1379 1380 =item C<UTF8_GOT_NON_CONTINUATION> 1381 1382 The input sequence was malformed in that a non-continuation type byte was found 1383 in a position where only a continuation type one should be. See also 1384 L</C<UTF8_GOT_SHORT>>. 1385 1386 =item C<UTF8_GOT_OVERFLOW> 1387 1388 The input sequence was malformed in that it is for a code point that is not 1389 representable in the number of bits available in an IV on the current platform. 1390 1391 =item C<UTF8_GOT_SHORT> 1392 1393 The input sequence was malformed in that C<curlen> is smaller than required for 1394 a complete sequence. In other words, the input is for a partial character 1395 sequence. 1396 1397 1398 C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short 1399 sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always 1400 that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete 1401 sequence was looked at. If no other flags are present, it means that the 1402 sequence was valid as far as it went. Depending on the application, this could 1403 mean one of three things: 1404 1405 =over 1406 1407 =item * 1408 1409 The C<curlen> length parameter passed in was too small, and the function was 1410 prevented from examining all the necessary bytes. 1411 1412 =item * 1413 1414 The buffer being looked at is based on reading data, and the data received so 1415 far stopped in the middle of a character, so that the next read will 1416 read the remainder of this character. (It is up to the caller to deal with the 1417 split bytes somehow.) 1418 1419 =item * 1420 1421 This is a real error, and the partial sequence is all we're going to get. 1422 1423 =back 1424 1425 =item C<UTF8_GOT_SUPER> 1426 1427 The input sequence was malformed in that it is for a non-Unicode code point; 1428 that is, one above the legal Unicode maximum. 1429 This bit is set only if the input C<flags> parameter contains either the 1430 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags. 1431 1432 =item C<UTF8_GOT_SURROGATE> 1433 1434 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate 1435 code point. 1436 This bit is set only if the input C<flags> parameter contains either the 1437 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags. 1438 1439 =back 1440 1441 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY> 1442 flag to suppress any warnings, and then examine the C<*errors> return. 1443 1444 =cut 1445 1446 Also implemented as a macro in utf8.h 1447 */ 1448 1449 UV 1450 Perl_utf8n_to_uvchr_error(const U8 *s, 1451 STRLEN curlen, 1452 STRLEN *retlen, 1453 const U32 flags, 1454 U32 * errors) 1455 { 1456 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR; 1457 1458 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL); 1459 } 1460 1461 /* 1462 1463 =for apidoc utf8n_to_uvchr_msgs 1464 1465 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 1466 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. 1467 1468 This function is for code that needs to know what the precise malformation(s) 1469 are when an error is found, and wants the corresponding warning and/or error 1470 messages to be returned to the caller rather than be displayed. All messages 1471 that would have been displayed if all lexcial warnings are enabled will be 1472 returned. 1473 1474 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter 1475 placed after all the others, C<msgs>. If this parameter is 0, this function 1476 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should 1477 be a pointer to an C<AV *> variable, in which this function creates a new AV to 1478 contain any appropriate messages. The elements of the array are ordered so 1479 that the first message that would have been displayed is in the 0th element, 1480 and so on. Each element is a hash with three key-value pairs, as follows: 1481 1482 =over 4 1483 1484 =item C<text> 1485 1486 The text of the message as a C<SVpv>. 1487 1488 =item C<warn_categories> 1489 1490 The warning category (or categories) packed into a C<SVuv>. 1491 1492 =item C<flag> 1493 1494 A single flag bit associated with this message, in a C<SVuv>. 1495 The bit corresponds to some bit in the C<*errors> return value, 1496 such as C<UTF8_GOT_LONG>. 1497 1498 =back 1499 1500 It's important to note that specifying this parameter as non-null will cause 1501 any warnings this function would otherwise generate to be suppressed, and 1502 instead be placed in C<*msgs>. The caller can check the lexical warnings state 1503 (or not) when choosing what to do with the returned messages. 1504 1505 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence 1506 no AV is created. 1507 1508 The caller, of course, is responsible for freeing any returned AV. 1509 1510 =cut 1511 */ 1512 1513 UV 1514 Perl__utf8n_to_uvchr_msgs_helper(const U8 *s, 1515 STRLEN curlen, 1516 STRLEN *retlen, 1517 const U32 flags, 1518 U32 * errors, 1519 AV ** msgs) 1520 { 1521 const U8 * const s0 = s; 1522 const U8 * send = s0 + curlen; 1523 U32 possible_problems; /* A bit is set here for each potential problem 1524 found as we go along */ 1525 UV uv; 1526 STRLEN expectlen; /* How long should this sequence be? */ 1527 STRLEN avail_len; /* When input is too short, gives what that is */ 1528 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this 1529 gets set and discarded */ 1530 1531 /* The below are used only if there is both an overlong malformation and a 1532 * too short one. Otherwise the first two are set to 's0' and 'send', and 1533 * the third not used at all */ 1534 U8 * adjusted_s0; 1535 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this 1536 routine; see [perl #130921] */ 1537 UV uv_so_far; 1538 dTHX; 1539 1540 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER; 1541 1542 /* Here, is one of: a) malformed; b) a problematic code point (surrogate, 1543 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul 1544 * syllables that the dfa doesn't properly handle. Quickly dispose of the 1545 * final case. */ 1546 1547 #ifndef EBCDIC 1548 1549 /* Each of the affected Hanguls starts with \xED */ 1550 1551 if (is_HANGUL_ED_utf8_safe(s0, send)) { 1552 if (retlen) { 1553 *retlen = 3; 1554 } 1555 if (errors) { 1556 *errors = 0; 1557 } 1558 if (msgs) { 1559 *msgs = NULL; 1560 } 1561 1562 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT)) 1563 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT) 1564 | (s0[2] & UTF_CONTINUATION_MASK); 1565 } 1566 1567 #endif 1568 1569 /* In conjunction with the exhaustive tests that can be enabled in 1570 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely 1571 * what it is intended to do, and that no flaws in it are masked by 1572 * dropping down and executing the code below 1573 assert(! isUTF8_CHAR(s0, send) 1574 || UTF8_IS_SURROGATE(s0, send) 1575 || UTF8_IS_SUPER(s0, send) 1576 || UTF8_IS_NONCHAR(s0,send)); 1577 */ 1578 1579 s = s0; 1580 uv = *s0; 1581 possible_problems = 0; 1582 expectlen = 0; 1583 avail_len = 0; 1584 discard_errors = 0; 1585 adjusted_s0 = (U8 *) s0; 1586 uv_so_far = 0; 1587 1588 if (errors) { 1589 *errors = 0; 1590 } 1591 else { 1592 errors = &discard_errors; 1593 } 1594 1595 /* The order of malformation tests here is important. We should consume as 1596 * few bytes as possible in order to not skip any valid character. This is 1597 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also 1598 * http://unicode.org/reports/tr36 for more discussion as to why. For 1599 * example, once we've done a UTF8SKIP, we can tell the expected number of 1600 * bytes, and could fail right off the bat if the input parameters indicate 1601 * that there are too few available. But it could be that just that first 1602 * byte is garbled, and the intended character occupies fewer bytes. If we 1603 * blindly assumed that the first byte is correct, and skipped based on 1604 * that number, we could skip over a valid input character. So instead, we 1605 * always examine the sequence byte-by-byte. 1606 * 1607 * We also should not consume too few bytes, otherwise someone could inject 1608 * things. For example, an input could be deliberately designed to 1609 * overflow, and if this code bailed out immediately upon discovering that, 1610 * returning to the caller C<*retlen> pointing to the very next byte (one 1611 * which is actually part of of the overflowing sequence), that could look 1612 * legitimate to the caller, which could discard the initial partial 1613 * sequence and process the rest, inappropriately. 1614 * 1615 * Some possible input sequences are malformed in more than one way. This 1616 * function goes to lengths to try to find all of them. This is necessary 1617 * for correctness, as the inputs may allow one malformation but not 1618 * another, and if we abandon searching for others after finding the 1619 * allowed one, we could allow in something that shouldn't have been. 1620 */ 1621 1622 if (UNLIKELY(curlen == 0)) { 1623 possible_problems |= UTF8_GOT_EMPTY; 1624 curlen = 0; 1625 uv = UNICODE_REPLACEMENT; 1626 goto ready_to_handle_errors; 1627 } 1628 1629 expectlen = UTF8SKIP(s); 1630 1631 /* A well-formed UTF-8 character, as the vast majority of calls to this 1632 * function will be for, has this expected length. For efficiency, set 1633 * things up here to return it. It will be overriden only in those rare 1634 * cases where a malformation is found */ 1635 if (retlen) { 1636 *retlen = expectlen; 1637 } 1638 1639 /* A continuation character can't start a valid sequence */ 1640 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) { 1641 possible_problems |= UTF8_GOT_CONTINUATION; 1642 curlen = 1; 1643 uv = UNICODE_REPLACEMENT; 1644 goto ready_to_handle_errors; 1645 } 1646 1647 /* Here is not a continuation byte, nor an invariant. The only thing left 1648 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START 1649 * because it excludes start bytes like \xC0 that always lead to 1650 * overlongs.) */ 1651 1652 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits 1653 * that indicate the number of bytes in the character's whole UTF-8 1654 * sequence, leaving just the bits that are part of the value. */ 1655 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen); 1656 1657 /* Setup the loop end point, making sure to not look past the end of the 1658 * input string, and flag it as too short if the size isn't big enough. */ 1659 if (UNLIKELY(curlen < expectlen)) { 1660 possible_problems |= UTF8_GOT_SHORT; 1661 avail_len = curlen; 1662 } 1663 else { 1664 send = (U8*) s0 + expectlen; 1665 } 1666 1667 /* Now, loop through the remaining bytes in the character's sequence, 1668 * accumulating each into the working value as we go. */ 1669 for (s = s0 + 1; s < send; s++) { 1670 if (LIKELY(UTF8_IS_CONTINUATION(*s))) { 1671 uv = UTF8_ACCUMULATE(uv, *s); 1672 continue; 1673 } 1674 1675 /* Here, found a non-continuation before processing all expected bytes. 1676 * This byte indicates the beginning of a new character, so quit, even 1677 * if allowing this malformation. */ 1678 possible_problems |= UTF8_GOT_NON_CONTINUATION; 1679 break; 1680 } /* End of loop through the character's bytes */ 1681 1682 /* Save how many bytes were actually in the character */ 1683 curlen = s - s0; 1684 1685 /* Note that there are two types of too-short malformation. One is when 1686 * there is actual wrong data before the normal termination of the 1687 * sequence. The other is that the sequence wasn't complete before the end 1688 * of the data we are allowed to look at, based on the input 'curlen'. 1689 * This means that we were passed data for a partial character, but it is 1690 * valid as far as we saw. The other is definitely invalid. This 1691 * distinction could be important to a caller, so the two types are kept 1692 * separate. 1693 * 1694 * A convenience macro that matches either of the too-short conditions. */ 1695 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION) 1696 1697 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) { 1698 uv_so_far = uv; 1699 uv = UNICODE_REPLACEMENT; 1700 } 1701 1702 /* Check for overflow. The algorithm requires us to not look past the end 1703 * of the current character, even if partial, so the upper limit is 's' */ 1704 if (UNLIKELY(0 < does_utf8_overflow(s0, s, 1705 1 /* Do consider overlongs */ 1706 ))) 1707 { 1708 possible_problems |= UTF8_GOT_OVERFLOW; 1709 uv = UNICODE_REPLACEMENT; 1710 } 1711 1712 /* Check for overlong. If no problems so far, 'uv' is the correct code 1713 * point value. Simply see if it is expressible in fewer bytes. Otherwise 1714 * we must look at the UTF-8 byte sequence itself to see if it is for an 1715 * overlong */ 1716 if ( ( LIKELY(! possible_problems) 1717 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv))) 1718 || ( UNLIKELY(possible_problems) 1719 && ( UNLIKELY(! UTF8_IS_START(*s0)) 1720 || ( curlen > 1 1721 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0, 1722 s - s0)))))) 1723 { 1724 possible_problems |= UTF8_GOT_LONG; 1725 1726 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT) 1727 1728 /* The calculation in the 'true' branch of this 'if' 1729 * below won't work if overflows, and isn't needed 1730 * anyway. Further below we handle all overflow 1731 * cases */ 1732 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))) 1733 { 1734 UV min_uv = uv_so_far; 1735 STRLEN i; 1736 1737 /* Here, the input is both overlong and is missing some trailing 1738 * bytes. There is no single code point it could be for, but there 1739 * may be enough information present to determine if what we have 1740 * so far is for an unallowed code point, such as for a surrogate. 1741 * The code further below has the intelligence to determine this, 1742 * but just for non-overlong UTF-8 sequences. What we do here is 1743 * calculate the smallest code point the input could represent if 1744 * there were no too short malformation. Then we compute and save 1745 * the UTF-8 for that, which is what the code below looks at 1746 * instead of the raw input. It turns out that the smallest such 1747 * code point is all we need. */ 1748 for (i = curlen; i < expectlen; i++) { 1749 min_uv = UTF8_ACCUMULATE(min_uv, 1750 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK)); 1751 } 1752 1753 adjusted_s0 = temp_char_buf; 1754 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0); 1755 } 1756 } 1757 1758 /* Here, we have found all the possible problems, except for when the input 1759 * is for a problematic code point not allowed by the input parameters. */ 1760 1761 /* uv is valid for overlongs */ 1762 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG)) 1763 1764 /* isn't problematic if < this */ 1765 && uv >= UNICODE_SURROGATE_FIRST) 1766 || ( UNLIKELY(possible_problems) 1767 1768 /* if overflow, we know without looking further 1769 * precisely which of the problematic types it is, 1770 * and we deal with those in the overflow handling 1771 * code */ 1772 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)) 1773 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0) 1774 || UNLIKELY(isUTF8_PERL_EXTENDED(s0))))) 1775 && ((flags & ( UTF8_DISALLOW_NONCHAR 1776 |UTF8_DISALLOW_SURROGATE 1777 |UTF8_DISALLOW_SUPER 1778 |UTF8_DISALLOW_PERL_EXTENDED 1779 |UTF8_WARN_NONCHAR 1780 |UTF8_WARN_SURROGATE 1781 |UTF8_WARN_SUPER 1782 |UTF8_WARN_PERL_EXTENDED)))) 1783 { 1784 /* If there were no malformations, or the only malformation is an 1785 * overlong, 'uv' is valid */ 1786 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) { 1787 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 1788 possible_problems |= UTF8_GOT_SURROGATE; 1789 } 1790 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) { 1791 possible_problems |= UTF8_GOT_SUPER; 1792 } 1793 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) { 1794 possible_problems |= UTF8_GOT_NONCHAR; 1795 } 1796 } 1797 else { /* Otherwise, need to look at the source UTF-8, possibly 1798 adjusted to be non-overlong */ 1799 1800 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0) 1801 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) 1802 { 1803 possible_problems |= UTF8_GOT_SUPER; 1804 } 1805 else if (curlen > 1) { 1806 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER( 1807 NATIVE_UTF8_TO_I8(*adjusted_s0), 1808 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) 1809 { 1810 possible_problems |= UTF8_GOT_SUPER; 1811 } 1812 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE( 1813 NATIVE_UTF8_TO_I8(*adjusted_s0), 1814 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) 1815 { 1816 possible_problems |= UTF8_GOT_SURROGATE; 1817 } 1818 } 1819 1820 /* We need a complete well-formed UTF-8 character to discern 1821 * non-characters, so can't look for them here */ 1822 } 1823 } 1824 1825 ready_to_handle_errors: 1826 1827 /* At this point: 1828 * curlen contains the number of bytes in the sequence that 1829 * this call should advance the input by. 1830 * avail_len gives the available number of bytes passed in, but 1831 * only if this is less than the expected number of 1832 * bytes, based on the code point's start byte. 1833 * possible_problems' is 0 if there weren't any problems; otherwise a bit 1834 * is set in it for each potential problem found. 1835 * uv contains the code point the input sequence 1836 * represents; or if there is a problem that prevents 1837 * a well-defined value from being computed, it is 1838 * some subsitute value, typically the REPLACEMENT 1839 * CHARACTER. 1840 * s0 points to the first byte of the character 1841 * s points to just after were we left off processing 1842 * the character 1843 * send points to just after where that character should 1844 * end, based on how many bytes the start byte tells 1845 * us should be in it, but no further than s0 + 1846 * avail_len 1847 */ 1848 1849 if (UNLIKELY(possible_problems)) { 1850 bool disallowed = FALSE; 1851 const U32 orig_problems = possible_problems; 1852 1853 if (msgs) { 1854 *msgs = NULL; 1855 } 1856 1857 while (possible_problems) { /* Handle each possible problem */ 1858 UV pack_warn = 0; 1859 char * message = NULL; 1860 U32 this_flag_bit = 0; 1861 1862 /* Each 'if' clause handles one problem. They are ordered so that 1863 * the first ones' messages will be displayed before the later 1864 * ones; this is kinda in decreasing severity order. But the 1865 * overlong must come last, as it changes 'uv' looked at by the 1866 * others */ 1867 if (possible_problems & UTF8_GOT_OVERFLOW) { 1868 1869 /* Overflow means also got a super and are using Perl's 1870 * extended UTF-8, but we handle all three cases here */ 1871 possible_problems 1872 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED); 1873 *errors |= UTF8_GOT_OVERFLOW; 1874 1875 /* But the API says we flag all errors found */ 1876 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) { 1877 *errors |= UTF8_GOT_SUPER; 1878 } 1879 if (flags 1880 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED)) 1881 { 1882 *errors |= UTF8_GOT_PERL_EXTENDED; 1883 } 1884 1885 /* Disallow if any of the three categories say to */ 1886 if ( ! (flags & UTF8_ALLOW_OVERFLOW) 1887 || (flags & ( UTF8_DISALLOW_SUPER 1888 |UTF8_DISALLOW_PERL_EXTENDED))) 1889 { 1890 disallowed = TRUE; 1891 } 1892 1893 /* Likewise, warn if any say to */ 1894 if ( ! (flags & UTF8_ALLOW_OVERFLOW) 1895 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED))) 1896 { 1897 1898 /* The warnings code explicitly says it doesn't handle the 1899 * case of packWARN2 and two categories which have 1900 * parent-child relationship. Even if it works now to 1901 * raise the warning if either is enabled, it wouldn't 1902 * necessarily do so in the future. We output (only) the 1903 * most dire warning */ 1904 if (! (flags & UTF8_CHECK_ONLY)) { 1905 if (msgs || ckWARN_d(WARN_UTF8)) { 1906 pack_warn = packWARN(WARN_UTF8); 1907 } 1908 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) { 1909 pack_warn = packWARN(WARN_NON_UNICODE); 1910 } 1911 if (pack_warn) { 1912 message = Perl_form(aTHX_ "%s: %s (overflows)", 1913 malformed_text, 1914 _byte_dump_string(s0, curlen, 0)); 1915 this_flag_bit = UTF8_GOT_OVERFLOW; 1916 } 1917 } 1918 } 1919 } 1920 else if (possible_problems & UTF8_GOT_EMPTY) { 1921 possible_problems &= ~UTF8_GOT_EMPTY; 1922 *errors |= UTF8_GOT_EMPTY; 1923 1924 if (! (flags & UTF8_ALLOW_EMPTY)) { 1925 1926 /* This so-called malformation is now treated as a bug in 1927 * the caller. If you have nothing to decode, skip calling 1928 * this function */ 1929 assert(0); 1930 1931 disallowed = TRUE; 1932 if ( (msgs 1933 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 1934 { 1935 pack_warn = packWARN(WARN_UTF8); 1936 message = Perl_form(aTHX_ "%s (empty string)", 1937 malformed_text); 1938 this_flag_bit = UTF8_GOT_EMPTY; 1939 } 1940 } 1941 } 1942 else if (possible_problems & UTF8_GOT_CONTINUATION) { 1943 possible_problems &= ~UTF8_GOT_CONTINUATION; 1944 *errors |= UTF8_GOT_CONTINUATION; 1945 1946 if (! (flags & UTF8_ALLOW_CONTINUATION)) { 1947 disallowed = TRUE; 1948 if (( msgs 1949 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 1950 { 1951 pack_warn = packWARN(WARN_UTF8); 1952 message = Perl_form(aTHX_ 1953 "%s: %s (unexpected continuation byte 0x%02x," 1954 " with no preceding start byte)", 1955 malformed_text, 1956 _byte_dump_string(s0, 1, 0), *s0); 1957 this_flag_bit = UTF8_GOT_CONTINUATION; 1958 } 1959 } 1960 } 1961 else if (possible_problems & UTF8_GOT_SHORT) { 1962 possible_problems &= ~UTF8_GOT_SHORT; 1963 *errors |= UTF8_GOT_SHORT; 1964 1965 if (! (flags & UTF8_ALLOW_SHORT)) { 1966 disallowed = TRUE; 1967 if (( msgs 1968 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 1969 { 1970 pack_warn = packWARN(WARN_UTF8); 1971 message = Perl_form(aTHX_ 1972 "%s: %s (too short; %d byte%s available, need %d)", 1973 malformed_text, 1974 _byte_dump_string(s0, send - s0, 0), 1975 (int)avail_len, 1976 avail_len == 1 ? "" : "s", 1977 (int)expectlen); 1978 this_flag_bit = UTF8_GOT_SHORT; 1979 } 1980 } 1981 1982 } 1983 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) { 1984 possible_problems &= ~UTF8_GOT_NON_CONTINUATION; 1985 *errors |= UTF8_GOT_NON_CONTINUATION; 1986 1987 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) { 1988 disallowed = TRUE; 1989 if (( msgs 1990 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 1991 { 1992 1993 /* If we don't know for sure that the input length is 1994 * valid, avoid as much as possible reading past the 1995 * end of the buffer */ 1996 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN) 1997 ? s - s0 1998 : send - s0; 1999 pack_warn = packWARN(WARN_UTF8); 2000 message = Perl_form(aTHX_ "%s", 2001 unexpected_non_continuation_text(s0, 2002 printlen, 2003 s - s0, 2004 (int) expectlen)); 2005 this_flag_bit = UTF8_GOT_NON_CONTINUATION; 2006 } 2007 } 2008 } 2009 else if (possible_problems & UTF8_GOT_SURROGATE) { 2010 possible_problems &= ~UTF8_GOT_SURROGATE; 2011 2012 if (flags & UTF8_WARN_SURROGATE) { 2013 *errors |= UTF8_GOT_SURROGATE; 2014 2015 if ( ! (flags & UTF8_CHECK_ONLY) 2016 && (msgs || ckWARN_d(WARN_SURROGATE))) 2017 { 2018 pack_warn = packWARN(WARN_SURROGATE); 2019 2020 /* These are the only errors that can occur with a 2021 * surrogate when the 'uv' isn't valid */ 2022 if (orig_problems & UTF8_GOT_TOO_SHORT) { 2023 message = Perl_form(aTHX_ 2024 "UTF-16 surrogate (any UTF-8 sequence that" 2025 " starts with \"%s\" is for a surrogate)", 2026 _byte_dump_string(s0, curlen, 0)); 2027 } 2028 else { 2029 message = Perl_form(aTHX_ surrogate_cp_format, uv); 2030 } 2031 this_flag_bit = UTF8_GOT_SURROGATE; 2032 } 2033 } 2034 2035 if (flags & UTF8_DISALLOW_SURROGATE) { 2036 disallowed = TRUE; 2037 *errors |= UTF8_GOT_SURROGATE; 2038 } 2039 } 2040 else if (possible_problems & UTF8_GOT_SUPER) { 2041 possible_problems &= ~UTF8_GOT_SUPER; 2042 2043 if (flags & UTF8_WARN_SUPER) { 2044 *errors |= UTF8_GOT_SUPER; 2045 2046 if ( ! (flags & UTF8_CHECK_ONLY) 2047 && (msgs || ckWARN_d(WARN_NON_UNICODE))) 2048 { 2049 pack_warn = packWARN(WARN_NON_UNICODE); 2050 2051 if (orig_problems & UTF8_GOT_TOO_SHORT) { 2052 message = Perl_form(aTHX_ 2053 "Any UTF-8 sequence that starts with" 2054 " \"%s\" is for a non-Unicode code point," 2055 " may not be portable", 2056 _byte_dump_string(s0, curlen, 0)); 2057 } 2058 else { 2059 message = Perl_form(aTHX_ super_cp_format, uv); 2060 } 2061 this_flag_bit = UTF8_GOT_SUPER; 2062 } 2063 } 2064 2065 /* Test for Perl's extended UTF-8 after the regular SUPER ones, 2066 * and before possibly bailing out, so that the more dire 2067 * warning will override the regular one. */ 2068 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) { 2069 if ( ! (flags & UTF8_CHECK_ONLY) 2070 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER)) 2071 && (msgs || ckWARN_d(WARN_NON_UNICODE))) 2072 { 2073 pack_warn = packWARN(WARN_NON_UNICODE); 2074 2075 /* If it is an overlong that evaluates to a code point 2076 * that doesn't have to use the Perl extended UTF-8, it 2077 * still used it, and so we output a message that 2078 * doesn't refer to the code point. The same is true 2079 * if there was a SHORT malformation where the code 2080 * point is not valid. In that case, 'uv' will have 2081 * been set to the REPLACEMENT CHAR, and the message 2082 * below without the code point in it will be selected 2083 * */ 2084 if (UNICODE_IS_PERL_EXTENDED(uv)) { 2085 message = Perl_form(aTHX_ 2086 perl_extended_cp_format, uv); 2087 } 2088 else { 2089 message = Perl_form(aTHX_ 2090 "Any UTF-8 sequence that starts with" 2091 " \"%s\" is a Perl extension, and" 2092 " so is not portable", 2093 _byte_dump_string(s0, curlen, 0)); 2094 } 2095 this_flag_bit = UTF8_GOT_PERL_EXTENDED; 2096 } 2097 2098 if (flags & ( UTF8_WARN_PERL_EXTENDED 2099 |UTF8_DISALLOW_PERL_EXTENDED)) 2100 { 2101 *errors |= UTF8_GOT_PERL_EXTENDED; 2102 2103 if (flags & UTF8_DISALLOW_PERL_EXTENDED) { 2104 disallowed = TRUE; 2105 } 2106 } 2107 } 2108 2109 if (flags & UTF8_DISALLOW_SUPER) { 2110 *errors |= UTF8_GOT_SUPER; 2111 disallowed = TRUE; 2112 } 2113 } 2114 else if (possible_problems & UTF8_GOT_NONCHAR) { 2115 possible_problems &= ~UTF8_GOT_NONCHAR; 2116 2117 if (flags & UTF8_WARN_NONCHAR) { 2118 *errors |= UTF8_GOT_NONCHAR; 2119 2120 if ( ! (flags & UTF8_CHECK_ONLY) 2121 && (msgs || ckWARN_d(WARN_NONCHAR))) 2122 { 2123 /* The code above should have guaranteed that we don't 2124 * get here with errors other than overlong */ 2125 assert (! (orig_problems 2126 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR))); 2127 2128 pack_warn = packWARN(WARN_NONCHAR); 2129 message = Perl_form(aTHX_ nonchar_cp_format, uv); 2130 this_flag_bit = UTF8_GOT_NONCHAR; 2131 } 2132 } 2133 2134 if (flags & UTF8_DISALLOW_NONCHAR) { 2135 disallowed = TRUE; 2136 *errors |= UTF8_GOT_NONCHAR; 2137 } 2138 } 2139 else if (possible_problems & UTF8_GOT_LONG) { 2140 possible_problems &= ~UTF8_GOT_LONG; 2141 *errors |= UTF8_GOT_LONG; 2142 2143 if (flags & UTF8_ALLOW_LONG) { 2144 2145 /* We don't allow the actual overlong value, unless the 2146 * special extra bit is also set */ 2147 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE 2148 & ~UTF8_ALLOW_LONG))) 2149 { 2150 uv = UNICODE_REPLACEMENT; 2151 } 2152 } 2153 else { 2154 disallowed = TRUE; 2155 2156 if (( msgs 2157 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 2158 { 2159 pack_warn = packWARN(WARN_UTF8); 2160 2161 /* These error types cause 'uv' to be something that 2162 * isn't what was intended, so can't use it in the 2163 * message. The other error types either can't 2164 * generate an overlong, or else the 'uv' is valid */ 2165 if (orig_problems & 2166 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW)) 2167 { 2168 message = Perl_form(aTHX_ 2169 "%s: %s (any UTF-8 sequence that starts" 2170 " with \"%s\" is overlong which can and" 2171 " should be represented with a" 2172 " different, shorter sequence)", 2173 malformed_text, 2174 _byte_dump_string(s0, send - s0, 0), 2175 _byte_dump_string(s0, curlen, 0)); 2176 } 2177 else { 2178 U8 tmpbuf[UTF8_MAXBYTES+1]; 2179 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf, 2180 uv, 0); 2181 /* Don't use U+ for non-Unicode code points, which 2182 * includes those in the Latin1 range */ 2183 const char * preface = ( uv > PERL_UNICODE_MAX 2184 #ifdef EBCDIC 2185 || uv <= 0xFF 2186 #endif 2187 ) 2188 ? "0x" 2189 : "U+"; 2190 message = Perl_form(aTHX_ 2191 "%s: %s (overlong; instead use %s to represent" 2192 " %s%0*" UVXf ")", 2193 malformed_text, 2194 _byte_dump_string(s0, send - s0, 0), 2195 _byte_dump_string(tmpbuf, e - tmpbuf, 0), 2196 preface, 2197 ((uv < 256) ? 2 : 4), /* Field width of 2 for 2198 small code points */ 2199 UNI_TO_NATIVE(uv)); 2200 } 2201 this_flag_bit = UTF8_GOT_LONG; 2202 } 2203 } 2204 } /* End of looking through the possible flags */ 2205 2206 /* Display the message (if any) for the problem being handled in 2207 * this iteration of the loop */ 2208 if (message) { 2209 if (msgs) { 2210 assert(this_flag_bit); 2211 2212 if (*msgs == NULL) { 2213 *msgs = newAV(); 2214 } 2215 2216 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message, 2217 pack_warn, 2218 this_flag_bit))); 2219 } 2220 else if (PL_op) 2221 Perl_warner(aTHX_ pack_warn, "%s in %s", message, 2222 OP_DESC(PL_op)); 2223 else 2224 Perl_warner(aTHX_ pack_warn, "%s", message); 2225 } 2226 } /* End of 'while (possible_problems)' */ 2227 2228 /* Since there was a possible problem, the returned length may need to 2229 * be changed from the one stored at the beginning of this function. 2230 * Instead of trying to figure out if that's needed, just do it. */ 2231 if (retlen) { 2232 *retlen = curlen; 2233 } 2234 2235 if (disallowed) { 2236 if (flags & UTF8_CHECK_ONLY && retlen) { 2237 *retlen = ((STRLEN) -1); 2238 } 2239 return 0; 2240 } 2241 } 2242 2243 return UNI_TO_NATIVE(uv); 2244 } 2245 2246 /* 2247 =for apidoc utf8_to_uvchr_buf 2248 2249 Returns the native code point of the first character in the string C<s> which 2250 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. 2251 C<*retlen> will be set to the length, in bytes, of that character. 2252 2253 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are 2254 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 2255 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined 2256 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and 2257 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is 2258 the next possible position in C<s> that could begin a non-malformed character. 2259 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is 2260 returned. 2261 2262 =cut 2263 2264 Also implemented as a macro in utf8.h 2265 2266 */ 2267 2268 2269 UV 2270 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) 2271 { 2272 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF; 2273 2274 assert(s < send); 2275 2276 return utf8n_to_uvchr(s, send - s, retlen, 2277 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY); 2278 } 2279 2280 /* This is marked as deprecated 2281 * 2282 =for apidoc utf8_to_uvuni_buf 2283 2284 Only in very rare circumstances should code need to be dealing in Unicode 2285 (as opposed to native) code points. In those few cases, use 2286 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you 2287 are not absolutely sure this is one of those cases, then assume it isn't and 2288 use plain C<utf8_to_uvchr_buf> instead. 2289 2290 Returns the Unicode (not-native) code point of the first character in the 2291 string C<s> which 2292 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. 2293 C<retlen> will be set to the length, in bytes, of that character. 2294 2295 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are 2296 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 2297 NULL) to -1. If those warnings are off, the computed value if well-defined (or 2298 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> 2299 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the 2300 next possible position in C<s> that could begin a non-malformed character. 2301 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. 2302 2303 =cut 2304 */ 2305 2306 UV 2307 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) 2308 { 2309 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF; 2310 2311 assert(send > s); 2312 2313 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen)); 2314 } 2315 2316 /* 2317 =for apidoc utf8_length 2318 2319 Returns the number of characters in the sequence of UTF-8-encoded bytes starting 2320 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the 2321 same place, it returns 0 with no warning raised. 2322 2323 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning 2324 and returns the number of valid characters. 2325 2326 =cut 2327 */ 2328 2329 STRLEN 2330 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e) 2331 { 2332 STRLEN len = 0; 2333 2334 PERL_ARGS_ASSERT_UTF8_LENGTH; 2335 2336 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g. 2337 * the bitops (especially ~) can create illegal UTF-8. 2338 * In other words: in Perl UTF-8 is not just for Unicode. */ 2339 2340 if (UNLIKELY(e < s)) 2341 goto warn_and_return; 2342 while (s < e) { 2343 s += UTF8SKIP(s); 2344 len++; 2345 } 2346 2347 if (UNLIKELY(e != s)) { 2348 len--; 2349 warn_and_return: 2350 if (PL_op) 2351 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 2352 "%s in %s", unees, OP_DESC(PL_op)); 2353 else 2354 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); 2355 } 2356 2357 return len; 2358 } 2359 2360 /* 2361 =for apidoc bytes_cmp_utf8 2362 2363 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the 2364 sequence of characters (stored as UTF-8) 2365 in C<u>, C<ulen>. Returns 0 if they are 2366 equal, -1 or -2 if the first string is less than the second string, +1 or +2 2367 if the first string is greater than the second string. 2368 2369 -1 or +1 is returned if the shorter string was identical to the start of the 2370 longer string. -2 or +2 is returned if 2371 there was a difference between characters 2372 within the strings. 2373 2374 =cut 2375 */ 2376 2377 int 2378 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen) 2379 { 2380 const U8 *const bend = b + blen; 2381 const U8 *const uend = u + ulen; 2382 2383 PERL_ARGS_ASSERT_BYTES_CMP_UTF8; 2384 2385 while (b < bend && u < uend) { 2386 U8 c = *u++; 2387 if (!UTF8_IS_INVARIANT(c)) { 2388 if (UTF8_IS_DOWNGRADEABLE_START(c)) { 2389 if (u < uend) { 2390 U8 c1 = *u++; 2391 if (UTF8_IS_CONTINUATION(c1)) { 2392 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1); 2393 } else { 2394 /* diag_listed_as: Malformed UTF-8 character%s */ 2395 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 2396 "%s %s%s", 2397 unexpected_non_continuation_text(u - 2, 2, 1, 2), 2398 PL_op ? " in " : "", 2399 PL_op ? OP_DESC(PL_op) : ""); 2400 return -2; 2401 } 2402 } else { 2403 if (PL_op) 2404 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 2405 "%s in %s", unees, OP_DESC(PL_op)); 2406 else 2407 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); 2408 return -2; /* Really want to return undef :-) */ 2409 } 2410 } else { 2411 return -2; 2412 } 2413 } 2414 if (*b != c) { 2415 return *b < c ? -2 : +2; 2416 } 2417 ++b; 2418 } 2419 2420 if (b == bend && u == uend) 2421 return 0; 2422 2423 return b < bend ? +1 : -1; 2424 } 2425 2426 /* 2427 =for apidoc utf8_to_bytes 2428 2429 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding. 2430 Unlike L</bytes_to_utf8>, this over-writes the original string, and 2431 updates C<*lenp> to contain the new length. 2432 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1. 2433 2434 Upon successful return, the number of variants in the string can be computed by 2435 having saved the value of C<*lenp> before the call, and subtracting the 2436 after-call value of C<*lenp> from it. 2437 2438 If you need a copy of the string, see L</bytes_from_utf8>. 2439 2440 =cut 2441 */ 2442 2443 U8 * 2444 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp) 2445 { 2446 U8 * first_variant; 2447 2448 PERL_ARGS_ASSERT_UTF8_TO_BYTES; 2449 PERL_UNUSED_CONTEXT; 2450 2451 /* This is a no-op if no variants at all in the input */ 2452 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) { 2453 return s; 2454 } 2455 2456 { 2457 U8 * const save = s; 2458 U8 * const send = s + *lenp; 2459 U8 * d; 2460 2461 /* Nothing before the first variant needs to be changed, so start the real 2462 * work there */ 2463 s = first_variant; 2464 while (s < send) { 2465 if (! UTF8_IS_INVARIANT(*s)) { 2466 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { 2467 *lenp = ((STRLEN) -1); 2468 return 0; 2469 } 2470 s++; 2471 } 2472 s++; 2473 } 2474 2475 /* Is downgradable, so do it */ 2476 d = s = first_variant; 2477 while (s < send) { 2478 U8 c = *s++; 2479 if (! UVCHR_IS_INVARIANT(c)) { 2480 /* Then it is two-byte encoded */ 2481 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); 2482 s++; 2483 } 2484 *d++ = c; 2485 } 2486 *d = '\0'; 2487 *lenp = d - save; 2488 2489 return save; 2490 } 2491 } 2492 2493 /* 2494 =for apidoc bytes_from_utf8 2495 2496 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native 2497 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is 2498 actually encoded in UTF-8. 2499 2500 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of 2501 the input string. 2502 2503 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string 2504 not expressible in native byte encoding. In these cases, C<*is_utf8p> and 2505 C<*lenp> are unchanged, and the return value is the original C<s>. 2506 2507 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a 2508 newly created string containing a downgraded copy of C<s>, and whose length is 2509 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The 2510 caller is responsible for arranging for the memory used by this string to get 2511 freed. 2512 2513 Upon successful return, the number of variants in the string can be computed by 2514 having saved the value of C<*lenp> before the call, and subtracting the 2515 after-call value of C<*lenp> from it. 2516 2517 =cut 2518 2519 There is a macro that avoids this function call, but this is retained for 2520 anyone who calls it with the Perl_ prefix */ 2521 2522 U8 * 2523 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p) 2524 { 2525 PERL_ARGS_ASSERT_BYTES_FROM_UTF8; 2526 PERL_UNUSED_CONTEXT; 2527 2528 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL); 2529 } 2530 2531 /* 2532 No = here because currently externally undocumented 2533 for apidoc bytes_from_utf8_loc 2534 2535 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where 2536 to store the location of the first character in C<"s"> that cannot be 2537 converted to non-UTF8. 2538 2539 If that parameter is C<NULL>, this function behaves identically to 2540 C<bytes_from_utf8>. 2541 2542 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to 2543 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>. 2544 2545 Otherwise, the function returns a newly created C<NUL>-terminated string 2546 containing the non-UTF8 equivalent of the convertible first portion of 2547 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>. 2548 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value, 2549 and C<*first_non_downgradable> is set to C<NULL>. 2550 2551 Otherwise, C<*first_non_downgradable> set to point to the first byte of the 2552 first character in the original string that wasn't converted. C<*is_utf8p> is 2553 unchanged. Note that the new string may have length 0. 2554 2555 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and 2556 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and 2557 converts as many characters in it as possible stopping at the first one it 2558 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is 2559 set to point to that. The function returns the portion that could be converted 2560 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length, 2561 not including the terminating C<NUL>. If the very first character in the 2562 original could not be converted, C<*lenp> will be 0, and the new string will 2563 contain just a single C<NUL>. If the entire input string was converted, 2564 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>. 2565 2566 Upon successful return, the number of variants in the converted portion of the 2567 string can be computed by having saved the value of C<*lenp> before the call, 2568 and subtracting the after-call value of C<*lenp> from it. 2569 2570 =cut 2571 2572 2573 */ 2574 2575 U8 * 2576 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted) 2577 { 2578 U8 *d; 2579 const U8 *original = s; 2580 U8 *converted_start; 2581 const U8 *send = s + *lenp; 2582 2583 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC; 2584 2585 if (! *is_utf8p) { 2586 if (first_unconverted) { 2587 *first_unconverted = NULL; 2588 } 2589 2590 return (U8 *) original; 2591 } 2592 2593 Newx(d, (*lenp) + 1, U8); 2594 2595 converted_start = d; 2596 while (s < send) { 2597 U8 c = *s++; 2598 if (! UTF8_IS_INVARIANT(c)) { 2599 2600 /* Then it is multi-byte encoded. If the code point is above 0xFF, 2601 * have to stop now */ 2602 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) { 2603 if (first_unconverted) { 2604 *first_unconverted = s - 1; 2605 goto finish_and_return; 2606 } 2607 else { 2608 Safefree(converted_start); 2609 return (U8 *) original; 2610 } 2611 } 2612 2613 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); 2614 s++; 2615 } 2616 *d++ = c; 2617 } 2618 2619 /* Here, converted the whole of the input */ 2620 *is_utf8p = FALSE; 2621 if (first_unconverted) { 2622 *first_unconverted = NULL; 2623 } 2624 2625 finish_and_return: 2626 *d = '\0'; 2627 *lenp = d - converted_start; 2628 2629 /* Trim unused space */ 2630 Renew(converted_start, *lenp + 1, U8); 2631 2632 return converted_start; 2633 } 2634 2635 /* 2636 =for apidoc bytes_to_utf8 2637 2638 Converts a string C<s> of length C<*lenp> bytes from the native encoding into 2639 UTF-8. 2640 Returns a pointer to the newly-created string, and sets C<*lenp> to 2641 reflect the new length in bytes. The caller is responsible for arranging for 2642 the memory used by this string to get freed. 2643 2644 Upon successful return, the number of variants in the string can be computed by 2645 having saved the value of C<*lenp> before the call, and subtracting it from the 2646 after-call value of C<*lenp>. 2647 2648 A C<NUL> character will be written after the end of the string. 2649 2650 If you want to convert to UTF-8 from encodings other than 2651 the native (Latin1 or EBCDIC), 2652 see L</sv_recode_to_utf8>(). 2653 2654 =cut 2655 */ 2656 2657 U8* 2658 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp) 2659 { 2660 const U8 * const send = s + (*lenp); 2661 U8 *d; 2662 U8 *dst; 2663 2664 PERL_ARGS_ASSERT_BYTES_TO_UTF8; 2665 PERL_UNUSED_CONTEXT; 2666 2667 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */ 2668 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8); 2669 dst = d; 2670 2671 while (s < send) { 2672 append_utf8_from_native_byte(*s, &d); 2673 s++; 2674 } 2675 2676 *d = '\0'; 2677 *lenp = d-dst; 2678 2679 return dst; 2680 } 2681 2682 /* 2683 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian), 2684 * use utf16_to_utf8_reversed(). 2685 * 2686 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes. 2687 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes. 2688 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes. 2689 * 2690 * These functions don't check for overflow. The worst case is every code 2691 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code 2692 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the 2693 * destination must be pre-extended to 2 times the source length. 2694 * 2695 * Do not use in-place. We optimize for native, for obvious reasons. */ 2696 2697 U8* 2698 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) 2699 { 2700 U8* pend; 2701 U8* dstart = d; 2702 2703 PERL_ARGS_ASSERT_UTF16_TO_UTF8; 2704 2705 if (bytelen & 1) 2706 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, 2707 (UV)bytelen); 2708 2709 pend = p + bytelen; 2710 2711 while (p < pend) { 2712 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */ 2713 p += 2; 2714 if (OFFUNI_IS_INVARIANT(uv)) { 2715 *d++ = LATIN1_TO_NATIVE((U8) uv); 2716 continue; 2717 } 2718 if (uv <= MAX_UTF8_TWO_BYTE) { 2719 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv)); 2720 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv)); 2721 continue; 2722 } 2723 2724 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST 2725 #define LAST_HIGH_SURROGATE 0xDBFF 2726 #define FIRST_LOW_SURROGATE 0xDC00 2727 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST 2728 #define FIRST_IN_PLANE1 0x10000 2729 2730 /* This assumes that most uses will be in the first Unicode plane, not 2731 * needing surrogates */ 2732 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST 2733 && uv <= UNICODE_SURROGATE_LAST)) 2734 { 2735 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) { 2736 Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); 2737 } 2738 else { 2739 UV low = (p[0] << 8) + p[1]; 2740 if ( UNLIKELY(low < FIRST_LOW_SURROGATE) 2741 || UNLIKELY(low > LAST_LOW_SURROGATE)) 2742 { 2743 Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); 2744 } 2745 p += 2; 2746 uv = ((uv - FIRST_HIGH_SURROGATE) << 10) 2747 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1; 2748 } 2749 } 2750 #ifdef EBCDIC 2751 d = uvoffuni_to_utf8_flags(d, uv, 0); 2752 #else 2753 if (uv < FIRST_IN_PLANE1) { 2754 *d++ = (U8)(( uv >> 12) | 0xe0); 2755 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); 2756 *d++ = (U8)(( uv & 0x3f) | 0x80); 2757 continue; 2758 } 2759 else { 2760 *d++ = (U8)(( uv >> 18) | 0xf0); 2761 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80); 2762 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); 2763 *d++ = (U8)(( uv & 0x3f) | 0x80); 2764 continue; 2765 } 2766 #endif 2767 } 2768 *newlen = d - dstart; 2769 return d; 2770 } 2771 2772 /* Note: this one is slightly destructive of the source. */ 2773 2774 U8* 2775 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) 2776 { 2777 U8* s = (U8*)p; 2778 U8* const send = s + bytelen; 2779 2780 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED; 2781 2782 if (bytelen & 1) 2783 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf, 2784 (UV)bytelen); 2785 2786 while (s < send) { 2787 const U8 tmp = s[0]; 2788 s[0] = s[1]; 2789 s[1] = tmp; 2790 s += 2; 2791 } 2792 return utf16_to_utf8(p, d, bytelen, newlen); 2793 } 2794 2795 bool 2796 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c) 2797 { 2798 dVAR; 2799 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c); 2800 } 2801 2802 /* Internal function so we can deprecate the external one, and call 2803 this one from other deprecated functions in this file */ 2804 2805 bool 2806 Perl__is_utf8_idstart(pTHX_ const U8 *p) 2807 { 2808 dVAR; 2809 2810 PERL_ARGS_ASSERT__IS_UTF8_IDSTART; 2811 2812 if (*p == '_') 2813 return TRUE; 2814 return is_utf8_common(p, PL_utf8_idstart); 2815 } 2816 2817 bool 2818 Perl__is_uni_perl_idcont(pTHX_ UV c) 2819 { 2820 dVAR; 2821 return _invlist_contains_cp(PL_utf8_perl_idcont, c); 2822 } 2823 2824 bool 2825 Perl__is_uni_perl_idstart(pTHX_ UV c) 2826 { 2827 dVAR; 2828 return _invlist_contains_cp(PL_utf8_perl_idstart, c); 2829 } 2830 2831 UV 2832 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, 2833 const char S_or_s) 2834 { 2835 /* We have the latin1-range values compiled into the core, so just use 2836 * those, converting the result to UTF-8. The only difference between upper 2837 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is 2838 * either "SS" or "Ss". Which one to use is passed into the routine in 2839 * 'S_or_s' to avoid a test */ 2840 2841 UV converted = toUPPER_LATIN1_MOD(c); 2842 2843 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1; 2844 2845 assert(S_or_s == 'S' || S_or_s == 's'); 2846 2847 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for 2848 characters in this range */ 2849 *p = (U8) converted; 2850 *lenp = 1; 2851 return converted; 2852 } 2853 2854 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers, 2855 * which it maps to one of them, so as to only have to have one check for 2856 * it in the main case */ 2857 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { 2858 switch (c) { 2859 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: 2860 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; 2861 break; 2862 case MICRO_SIGN: 2863 converted = GREEK_CAPITAL_LETTER_MU; 2864 break; 2865 #if UNICODE_MAJOR_VERSION > 2 \ 2866 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \ 2867 && UNICODE_DOT_DOT_VERSION >= 8) 2868 case LATIN_SMALL_LETTER_SHARP_S: 2869 *(p)++ = 'S'; 2870 *p = S_or_s; 2871 *lenp = 2; 2872 return 'S'; 2873 #endif 2874 default: 2875 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect" 2876 " '%c' to map to '%c'", 2877 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS); 2878 NOT_REACHED; /* NOTREACHED */ 2879 } 2880 } 2881 2882 *(p)++ = UTF8_TWO_BYTE_HI(converted); 2883 *p = UTF8_TWO_BYTE_LO(converted); 2884 *lenp = 2; 2885 2886 return converted; 2887 } 2888 2889 /* If compiled on an early Unicode version, there may not be auxiliary tables 2890 * */ 2891 #ifndef HAS_UC_AUX_TABLES 2892 # define UC_AUX_TABLE_ptrs NULL 2893 # define UC_AUX_TABLE_lengths NULL 2894 #endif 2895 #ifndef HAS_TC_AUX_TABLES 2896 # define TC_AUX_TABLE_ptrs NULL 2897 # define TC_AUX_TABLE_lengths NULL 2898 #endif 2899 #ifndef HAS_LC_AUX_TABLES 2900 # define LC_AUX_TABLE_ptrs NULL 2901 # define LC_AUX_TABLE_lengths NULL 2902 #endif 2903 #ifndef HAS_CF_AUX_TABLES 2904 # define CF_AUX_TABLE_ptrs NULL 2905 # define CF_AUX_TABLE_lengths NULL 2906 #endif 2907 #ifndef HAS_UC_AUX_TABLES 2908 # define UC_AUX_TABLE_ptrs NULL 2909 # define UC_AUX_TABLE_lengths NULL 2910 #endif 2911 2912 /* Call the function to convert a UTF-8 encoded character to the specified case. 2913 * Note that there may be more than one character in the result. 2914 * 's' is a pointer to the first byte of the input character 2915 * 'd' will be set to the first byte of the string of changed characters. It 2916 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes 2917 * 'lenp' will be set to the length in bytes of the string of changed characters 2918 * 2919 * The functions return the ordinal of the first character in the string of 2920 * 'd' */ 2921 #define CALL_UPPER_CASE(uv, s, d, lenp) \ 2922 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \ 2923 Uppercase_Mapping_invmap, \ 2924 UC_AUX_TABLE_ptrs, \ 2925 UC_AUX_TABLE_lengths, \ 2926 "uppercase") 2927 #define CALL_TITLE_CASE(uv, s, d, lenp) \ 2928 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \ 2929 Titlecase_Mapping_invmap, \ 2930 TC_AUX_TABLE_ptrs, \ 2931 TC_AUX_TABLE_lengths, \ 2932 "titlecase") 2933 #define CALL_LOWER_CASE(uv, s, d, lenp) \ 2934 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \ 2935 Lowercase_Mapping_invmap, \ 2936 LC_AUX_TABLE_ptrs, \ 2937 LC_AUX_TABLE_lengths, \ 2938 "lowercase") 2939 2940 2941 /* This additionally has the input parameter 'specials', which if non-zero will 2942 * cause this to use the specials hash for folding (meaning get full case 2943 * folding); otherwise, when zero, this implies a simple case fold */ 2944 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \ 2945 (specials) \ 2946 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \ 2947 Case_Folding_invmap, \ 2948 CF_AUX_TABLE_ptrs, \ 2949 CF_AUX_TABLE_lengths, \ 2950 "foldcase") \ 2951 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \ 2952 Simple_Case_Folding_invmap, \ 2953 NULL, NULL, \ 2954 "foldcase") 2955 2956 UV 2957 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp) 2958 { 2959 /* Convert the Unicode character whose ordinal is <c> to its uppercase 2960 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>. 2961 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since 2962 * the changed version may be longer than the original character. 2963 * 2964 * The ordinal of the first character of the changed version is returned 2965 * (but note, as explained above, that there may be more.) */ 2966 2967 dVAR; 2968 PERL_ARGS_ASSERT_TO_UNI_UPPER; 2969 2970 if (c < 256) { 2971 return _to_upper_title_latin1((U8) c, p, lenp, 'S'); 2972 } 2973 2974 return CALL_UPPER_CASE(c, NULL, p, lenp); 2975 } 2976 2977 UV 2978 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp) 2979 { 2980 dVAR; 2981 PERL_ARGS_ASSERT_TO_UNI_TITLE; 2982 2983 if (c < 256) { 2984 return _to_upper_title_latin1((U8) c, p, lenp, 's'); 2985 } 2986 2987 return CALL_TITLE_CASE(c, NULL, p, lenp); 2988 } 2989 2990 STATIC U8 2991 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy) 2992 { 2993 /* We have the latin1-range values compiled into the core, so just use 2994 * those, converting the result to UTF-8. Since the result is always just 2995 * one character, we allow <p> to be NULL */ 2996 2997 U8 converted = toLOWER_LATIN1(c); 2998 2999 PERL_UNUSED_ARG(dummy); 3000 3001 if (p != NULL) { 3002 if (NATIVE_BYTE_IS_INVARIANT(converted)) { 3003 *p = converted; 3004 *lenp = 1; 3005 } 3006 else { 3007 /* Result is known to always be < 256, so can use the EIGHT_BIT 3008 * macros */ 3009 *p = UTF8_EIGHT_BIT_HI(converted); 3010 *(p+1) = UTF8_EIGHT_BIT_LO(converted); 3011 *lenp = 2; 3012 } 3013 } 3014 return converted; 3015 } 3016 3017 UV 3018 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp) 3019 { 3020 dVAR; 3021 PERL_ARGS_ASSERT_TO_UNI_LOWER; 3022 3023 if (c < 256) { 3024 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ ); 3025 } 3026 3027 return CALL_LOWER_CASE(c, NULL, p, lenp); 3028 } 3029 3030 UV 3031 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags) 3032 { 3033 /* Corresponds to to_lower_latin1(); <flags> bits meanings: 3034 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited 3035 * FOLD_FLAGS_FULL iff full folding is to be used; 3036 * 3037 * Not to be used for locale folds 3038 */ 3039 3040 UV converted; 3041 3042 PERL_ARGS_ASSERT__TO_FOLD_LATIN1; 3043 3044 assert (! (flags & FOLD_FLAGS_LOCALE)); 3045 3046 if (UNLIKELY(c == MICRO_SIGN)) { 3047 converted = GREEK_SMALL_LETTER_MU; 3048 } 3049 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ 3050 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ 3051 || UNICODE_DOT_DOT_VERSION > 0) 3052 else if ( (flags & FOLD_FLAGS_FULL) 3053 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S)) 3054 { 3055 /* If can't cross 127/128 boundary, can't return "ss"; instead return 3056 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}") 3057 * under those circumstances. */ 3058 if (flags & FOLD_FLAGS_NOMIX_ASCII) { 3059 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; 3060 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, 3061 p, *lenp, U8); 3062 return LATIN_SMALL_LETTER_LONG_S; 3063 } 3064 else { 3065 *(p)++ = 's'; 3066 *p = 's'; 3067 *lenp = 2; 3068 return 's'; 3069 } 3070 } 3071 #endif 3072 else { /* In this range the fold of all other characters is their lower 3073 case */ 3074 converted = toLOWER_LATIN1(c); 3075 } 3076 3077 if (UVCHR_IS_INVARIANT(converted)) { 3078 *p = (U8) converted; 3079 *lenp = 1; 3080 } 3081 else { 3082 *(p)++ = UTF8_TWO_BYTE_HI(converted); 3083 *p = UTF8_TWO_BYTE_LO(converted); 3084 *lenp = 2; 3085 } 3086 3087 return converted; 3088 } 3089 3090 UV 3091 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags) 3092 { 3093 3094 /* Not currently externally documented, and subject to change 3095 * <flags> bits meanings: 3096 * FOLD_FLAGS_FULL iff full folding is to be used; 3097 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying 3098 * locale are to be used. 3099 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited 3100 */ 3101 3102 dVAR; 3103 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS; 3104 3105 if (flags & FOLD_FLAGS_LOCALE) { 3106 /* Treat a non-Turkic UTF-8 locale as not being in locale at all, 3107 * except for potentially warning */ 3108 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 3109 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) { 3110 flags &= ~FOLD_FLAGS_LOCALE; 3111 } 3112 else { 3113 goto needs_full_generality; 3114 } 3115 } 3116 3117 if (c < 256) { 3118 return _to_fold_latin1((U8) c, p, lenp, 3119 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); 3120 } 3121 3122 /* Here, above 255. If no special needs, just use the macro */ 3123 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) { 3124 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL); 3125 } 3126 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with 3127 the special flags. */ 3128 U8 utf8_c[UTF8_MAXBYTES + 1]; 3129 3130 needs_full_generality: 3131 uvchr_to_utf8(utf8_c, c); 3132 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), 3133 p, lenp, flags); 3134 } 3135 } 3136 3137 PERL_STATIC_INLINE bool 3138 S_is_utf8_common(pTHX_ const U8 *const p, SV* const invlist) 3139 { 3140 /* returns a boolean giving whether or not the UTF8-encoded character that 3141 * starts at <p> is in the inversion list indicated by <invlist>. 3142 * 3143 * Note that it is assumed that the buffer length of <p> is enough to 3144 * contain all the bytes that comprise the character. Thus, <*p> should 3145 * have been checked before this call for mal-formedness enough to assure 3146 * that. This function, does make sure to not look past any NUL, so it is 3147 * safe to use on C, NUL-terminated, strings */ 3148 STRLEN len = my_strnlen((char *) p, UTF8SKIP(p)); 3149 3150 PERL_ARGS_ASSERT_IS_UTF8_COMMON; 3151 3152 /* The API should have included a length for the UTF-8 character in <p>, 3153 * but it doesn't. We therefore assume that p has been validated at least 3154 * as far as there being enough bytes available in it to accommodate the 3155 * character without reading beyond the end, and pass that number on to the 3156 * validating routine */ 3157 if (! isUTF8_CHAR(p, p + len)) { 3158 _force_out_malformed_utf8_message(p, p + len, _UTF8_NO_CONFIDENCE_IN_CURLEN, 3159 1 /* Die */ ); 3160 NOT_REACHED; /* NOTREACHED */ 3161 } 3162 3163 return is_utf8_common_with_len(p, p + len, invlist); 3164 } 3165 3166 PERL_STATIC_INLINE bool 3167 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, 3168 SV* const invlist) 3169 { 3170 /* returns a boolean giving whether or not the UTF8-encoded character that 3171 * starts at <p>, and extending no further than <e - 1> is in the inversion 3172 * list <invlist>. */ 3173 3174 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0); 3175 3176 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN; 3177 3178 if (cp == 0 && (p >= e || *p != '\0')) { 3179 _force_out_malformed_utf8_message(p, e, 0, 1); 3180 NOT_REACHED; /* NOTREACHED */ 3181 } 3182 3183 assert(invlist); 3184 return _invlist_contains_cp(invlist, cp); 3185 } 3186 3187 STATIC void 3188 S_warn_on_first_deprecated_use(pTHX_ const char * const name, 3189 const char * const alternative, 3190 const bool use_locale, 3191 const char * const file, 3192 const unsigned line) 3193 { 3194 const char * key; 3195 3196 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE; 3197 3198 if (ckWARN_d(WARN_DEPRECATED)) { 3199 3200 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line); 3201 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) { 3202 if (! PL_seen_deprecated_macro) { 3203 PL_seen_deprecated_macro = newHV(); 3204 } 3205 if (! hv_store(PL_seen_deprecated_macro, key, 3206 strlen(key), &PL_sv_undef, 0)) 3207 { 3208 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); 3209 } 3210 3211 if (instr(file, "mathoms.c")) { 3212 Perl_warner(aTHX_ WARN_DEPRECATED, 3213 "In %s, line %d, starting in Perl v5.32, %s()" 3214 " will be removed. Avoid this message by" 3215 " converting to use %s().\n", 3216 file, line, name, alternative); 3217 } 3218 else { 3219 Perl_warner(aTHX_ WARN_DEPRECATED, 3220 "In %s, line %d, starting in Perl v5.32, %s() will" 3221 " require an additional parameter. Avoid this" 3222 " message by converting to use %s().\n", 3223 file, line, name, alternative); 3224 } 3225 } 3226 } 3227 } 3228 3229 bool 3230 Perl__is_utf8_FOO(pTHX_ U8 classnum, 3231 const U8 * const p, 3232 const char * const name, 3233 const char * const alternative, 3234 const bool use_utf8, 3235 const bool use_locale, 3236 const char * const file, 3237 const unsigned line) 3238 { 3239 dVAR; 3240 PERL_ARGS_ASSERT__IS_UTF8_FOO; 3241 3242 warn_on_first_deprecated_use(name, alternative, use_locale, file, line); 3243 3244 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) { 3245 3246 switch (classnum) { 3247 case _CC_WORDCHAR: 3248 case _CC_DIGIT: 3249 case _CC_ALPHA: 3250 case _CC_LOWER: 3251 case _CC_UPPER: 3252 case _CC_PUNCT: 3253 case _CC_PRINT: 3254 case _CC_ALPHANUMERIC: 3255 case _CC_GRAPH: 3256 case _CC_CASED: 3257 3258 return is_utf8_common(p, PL_XPosix_ptrs[classnum]); 3259 3260 case _CC_SPACE: 3261 return is_XPERLSPACE_high(p); 3262 case _CC_BLANK: 3263 return is_HORIZWS_high(p); 3264 case _CC_XDIGIT: 3265 return is_XDIGIT_high(p); 3266 case _CC_CNTRL: 3267 return 0; 3268 case _CC_ASCII: 3269 return 0; 3270 case _CC_VERTSPACE: 3271 return is_VERTWS_high(p); 3272 case _CC_IDFIRST: 3273 return is_utf8_common(p, PL_utf8_perl_idstart); 3274 case _CC_IDCONT: 3275 return is_utf8_common(p, PL_utf8_perl_idcont); 3276 } 3277 } 3278 3279 /* idcont is the same as wordchar below 256 */ 3280 if (classnum == _CC_IDCONT) { 3281 classnum = _CC_WORDCHAR; 3282 } 3283 else if (classnum == _CC_IDFIRST) { 3284 if (*p == '_') { 3285 return TRUE; 3286 } 3287 classnum = _CC_ALPHA; 3288 } 3289 3290 if (! use_locale) { 3291 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) { 3292 return _generic_isCC(*p, classnum); 3293 } 3294 3295 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum); 3296 } 3297 else { 3298 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) { 3299 return isFOO_lc(classnum, *p); 3300 } 3301 3302 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 ))); 3303 } 3304 3305 NOT_REACHED; /* NOTREACHED */ 3306 } 3307 3308 bool 3309 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p, 3310 const U8 * const e) 3311 { 3312 dVAR; 3313 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN; 3314 3315 return is_utf8_common_with_len(p, e, PL_XPosix_ptrs[classnum]); 3316 } 3317 3318 bool 3319 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e) 3320 { 3321 dVAR; 3322 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN; 3323 3324 return is_utf8_common_with_len(p, e, PL_utf8_perl_idstart); 3325 } 3326 3327 bool 3328 Perl__is_utf8_xidstart(pTHX_ const U8 *p) 3329 { 3330 dVAR; 3331 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART; 3332 3333 if (*p == '_') 3334 return TRUE; 3335 return is_utf8_common(p, PL_utf8_xidstart); 3336 } 3337 3338 bool 3339 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e) 3340 { 3341 dVAR; 3342 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN; 3343 3344 return is_utf8_common_with_len(p, e, PL_utf8_perl_idcont); 3345 } 3346 3347 bool 3348 Perl__is_utf8_idcont(pTHX_ const U8 *p) 3349 { 3350 dVAR; 3351 PERL_ARGS_ASSERT__IS_UTF8_IDCONT; 3352 3353 return is_utf8_common(p, PL_utf8_idcont); 3354 } 3355 3356 bool 3357 Perl__is_utf8_xidcont(pTHX_ const U8 *p) 3358 { 3359 dVAR; 3360 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT; 3361 3362 return is_utf8_common(p, PL_utf8_xidcont); 3363 } 3364 3365 bool 3366 Perl__is_utf8_mark(pTHX_ const U8 *p) 3367 { 3368 dVAR; 3369 PERL_ARGS_ASSERT__IS_UTF8_MARK; 3370 3371 return is_utf8_common(p, PL_utf8_mark); 3372 } 3373 3374 STATIC UV 3375 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, 3376 U8* ustrp, STRLEN *lenp, 3377 SV *invlist, const int * const invmap, 3378 const unsigned int * const * const aux_tables, 3379 const U8 * const aux_table_lengths, 3380 const char * const normal) 3381 { 3382 STRLEN len = 0; 3383 3384 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed 3385 * by this routine to be valid) begins at 'p'. 'normal' is a string to use 3386 * to name the new case in any generated messages, as a fallback if the 3387 * operation being used is not available. The new case is given by the 3388 * data structures in the remaining arguments. 3389 * 3390 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the 3391 * entire changed case string, and the return value is the first code point 3392 * in that string */ 3393 3394 PERL_ARGS_ASSERT__TO_UTF8_CASE; 3395 3396 /* For code points that don't change case, we already know that the output 3397 * of this function is the unchanged input, so we can skip doing look-ups 3398 * for them. Unfortunately the case-changing code points are scattered 3399 * around. But there are some long consecutive ranges where there are no 3400 * case changing code points. By adding tests, we can eliminate the lookup 3401 * for all the ones in such ranges. This is currently done here only for 3402 * just a few cases where the scripts are in common use in modern commerce 3403 * (and scripts adjacent to those which can be included without additional 3404 * tests). */ 3405 3406 if (uv1 >= 0x0590) { 3407 /* This keeps from needing further processing the code points most 3408 * likely to be used in the following non-cased scripts: Hebrew, 3409 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari, 3410 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada, 3411 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */ 3412 if (uv1 < 0x10A0) { 3413 goto cases_to_self; 3414 } 3415 3416 /* The following largish code point ranges also don't have case 3417 * changes, but khw didn't think they warranted extra tests to speed 3418 * them up (which would slightly slow down everything else above them): 3419 * 1100..139F Hangul Jamo, Ethiopic 3420 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic, 3421 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian, 3422 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham, 3423 * Combining Diacritical Marks Extended, Balinese, 3424 * Sundanese, Batak, Lepcha, Ol Chiki 3425 * 2000..206F General Punctuation 3426 */ 3427 3428 if (uv1 >= 0x2D30) { 3429 3430 /* This keeps the from needing further processing the code points 3431 * most likely to be used in the following non-cased major scripts: 3432 * CJK, Katakana, Hiragana, plus some less-likely scripts. 3433 * 3434 * (0x2D30 above might have to be changed to 2F00 in the unlikely 3435 * event that Unicode eventually allocates the unused block as of 3436 * v8.0 2FE0..2FEF to code points that are cased. khw has verified 3437 * that the test suite will start having failures to alert you 3438 * should that happen) */ 3439 if (uv1 < 0xA640) { 3440 goto cases_to_self; 3441 } 3442 3443 if (uv1 >= 0xAC00) { 3444 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) { 3445 if (ckWARN_d(WARN_SURROGATE)) { 3446 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; 3447 Perl_warner(aTHX_ packWARN(WARN_SURROGATE), 3448 "Operation \"%s\" returns its argument for" 3449 " UTF-16 surrogate U+%04" UVXf, desc, uv1); 3450 } 3451 goto cases_to_self; 3452 } 3453 3454 /* AC00..FAFF Catches Hangul syllables and private use, plus 3455 * some others */ 3456 if (uv1 < 0xFB00) { 3457 goto cases_to_self; 3458 } 3459 3460 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) { 3461 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) { 3462 Perl_croak(aTHX_ cp_above_legal_max, uv1, 3463 MAX_LEGAL_CP); 3464 } 3465 if (ckWARN_d(WARN_NON_UNICODE)) { 3466 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; 3467 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), 3468 "Operation \"%s\" returns its argument for" 3469 " non-Unicode code point 0x%04" UVXf, desc, uv1); 3470 } 3471 goto cases_to_self; 3472 } 3473 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C 3474 if (UNLIKELY(uv1 3475 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C)) 3476 { 3477 3478 /* As of Unicode 10.0, this means we avoid swash creation 3479 * for anything beyond high Plane 1 (below emojis) */ 3480 goto cases_to_self; 3481 } 3482 #endif 3483 } 3484 } 3485 3486 /* Note that non-characters are perfectly legal, so no warning should 3487 * be given. */ 3488 } 3489 3490 { 3491 unsigned int i; 3492 const unsigned int * cp_list; 3493 U8 * d; 3494 3495 /* 'index' is guaranteed to be non-negative, as this is an inversion 3496 * map that covers all possible inputs. See [perl #133365] */ 3497 SSize_t index = _invlist_search(invlist, uv1); 3498 IV base = invmap[index]; 3499 3500 /* The data structures are set up so that if 'base' is non-negative, 3501 * the case change is 1-to-1; and if 0, the change is to itself */ 3502 if (base >= 0) { 3503 IV lc; 3504 3505 if (base == 0) { 3506 goto cases_to_self; 3507 } 3508 3509 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */ 3510 lc = base + uv1 - invlist_array(invlist)[index]; 3511 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp; 3512 return lc; 3513 } 3514 3515 /* Here 'base' is negative. That means the mapping is 1-to-many, and 3516 * requires an auxiliary table look up. abs(base) gives the index into 3517 * a list of such tables which points to the proper aux table. And a 3518 * parallel list gives the length of each corresponding aux table. */ 3519 cp_list = aux_tables[-base]; 3520 3521 /* Create the string of UTF-8 from the mapped-to code points */ 3522 d = ustrp; 3523 for (i = 0; i < aux_table_lengths[-base]; i++) { 3524 d = uvchr_to_utf8(d, cp_list[i]); 3525 } 3526 *d = '\0'; 3527 *lenp = d - ustrp; 3528 3529 return cp_list[0]; 3530 } 3531 3532 /* Here, there was no mapping defined, which means that the code point maps 3533 * to itself. Return the inputs */ 3534 cases_to_self: 3535 if (p) { 3536 len = UTF8SKIP(p); 3537 if (p != ustrp) { /* Don't copy onto itself */ 3538 Copy(p, ustrp, len, U8); 3539 } 3540 *lenp = len; 3541 } 3542 else { 3543 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp; 3544 } 3545 3546 return uv1; 3547 3548 } 3549 3550 Size_t 3551 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to, 3552 const unsigned int ** remaining_folds_to) 3553 { 3554 /* Returns the count of the number of code points that fold to the input 3555 * 'cp' (besides itself). 3556 * 3557 * If the return is 0, there is nothing else that folds to it, and 3558 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL. 3559 * 3560 * If the return is 1, '*first_folds_to' is set to the single code point, 3561 * and '*remaining_folds_to' is set to NULL. 3562 * 3563 * Otherwise, '*first_folds_to' is set to a code point, and 3564 * '*remaining_fold_to' is set to an array that contains the others. The 3565 * length of this array is the returned count minus 1. 3566 * 3567 * The reason for this convolution is to avoid having to deal with 3568 * allocating and freeing memory. The lists are already constructed, so 3569 * the return can point to them, but single code points aren't, so would 3570 * need to be constructed if we didn't employ something like this API */ 3571 3572 dVAR; 3573 /* 'index' is guaranteed to be non-negative, as this is an inversion map 3574 * that covers all possible inputs. See [perl #133365] */ 3575 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp); 3576 int base = _Perl_IVCF_invmap[index]; 3577 3578 PERL_ARGS_ASSERT__INVERSE_FOLDS; 3579 3580 if (base == 0) { /* No fold */ 3581 *first_folds_to = 0; 3582 *remaining_folds_to = NULL; 3583 return 0; 3584 } 3585 3586 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */ 3587 3588 assert(base > 0); 3589 3590 #else 3591 3592 if (UNLIKELY(base < 0)) { /* Folds to more than one character */ 3593 3594 /* The data structure is set up so that the absolute value of 'base' is 3595 * an index into a table of pointers to arrays, with the array 3596 * corresponding to the index being the list of code points that fold 3597 * to 'cp', and the parallel array containing the length of the list 3598 * array */ 3599 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0]; 3600 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes 3601 *first_folds_to 3602 */ 3603 return IVCF_AUX_TABLE_lengths[-base]; 3604 } 3605 3606 #endif 3607 3608 /* Only the single code point. This works like 'fc(G) = G - A + a' */ 3609 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index]; 3610 *remaining_folds_to = NULL; 3611 return 1; 3612 } 3613 3614 STATIC UV 3615 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, 3616 U8* const ustrp, STRLEN *lenp) 3617 { 3618 /* This is called when changing the case of a UTF-8-encoded character above 3619 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the 3620 * result contains a character that crosses the 255/256 boundary, disallow 3621 * the change, and return the original code point. See L<perlfunc/lc> for 3622 * why; 3623 * 3624 * p points to the original string whose case was changed; assumed 3625 * by this routine to be well-formed 3626 * result the code point of the first character in the changed-case string 3627 * ustrp points to the changed-case string (<result> represents its 3628 * first char) 3629 * lenp points to the length of <ustrp> */ 3630 3631 UV original; /* To store the first code point of <p> */ 3632 3633 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING; 3634 3635 assert(UTF8_IS_ABOVE_LATIN1(*p)); 3636 3637 /* We know immediately if the first character in the string crosses the 3638 * boundary, so can skip testing */ 3639 if (result > 255) { 3640 3641 /* Look at every character in the result; if any cross the 3642 * boundary, the whole thing is disallowed */ 3643 U8* s = ustrp + UTF8SKIP(ustrp); 3644 U8* e = ustrp + *lenp; 3645 while (s < e) { 3646 if (! UTF8_IS_ABOVE_LATIN1(*s)) { 3647 goto bad_crossing; 3648 } 3649 s += UTF8SKIP(s); 3650 } 3651 3652 /* Here, no characters crossed, result is ok as-is, but we warn. */ 3653 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p)); 3654 return result; 3655 } 3656 3657 bad_crossing: 3658 3659 /* Failed, have to return the original */ 3660 original = valid_utf8_to_uvchr(p, lenp); 3661 3662 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 3663 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 3664 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8" 3665 " locale; resolved to \"\\x{%" UVXf "}\".", 3666 OP_DESC(PL_op), 3667 original, 3668 original); 3669 Copy(p, ustrp, *lenp, char); 3670 return original; 3671 } 3672 3673 STATIC U32 3674 S_check_and_deprecate(pTHX_ const U8 *p, 3675 const U8 **e, 3676 const unsigned int type, /* See below */ 3677 const bool use_locale, /* Is this a 'LC_' 3678 macro call? */ 3679 const char * const file, 3680 const unsigned line) 3681 { 3682 /* This is a temporary function to deprecate the unsafe calls to the case 3683 * changing macros and functions. It keeps all the special stuff in just 3684 * one place. 3685 * 3686 * It updates *e with the pointer to the end of the input string. If using 3687 * the old-style macros, *e is NULL on input, and so this function assumes 3688 * the input string is long enough to hold the entire UTF-8 sequence, and 3689 * sets *e accordingly, but it then returns a flag to pass the 3690 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid 3691 * using the full length if possible. 3692 * 3693 * It also does the assert that *e > p when *e is not NULL. This should be 3694 * migrated to the callers when this function gets deleted. 3695 * 3696 * The 'type' parameter is used for the caller to specify which case 3697 * changing function this is called from: */ 3698 3699 # define DEPRECATE_TO_UPPER 0 3700 # define DEPRECATE_TO_TITLE 1 3701 # define DEPRECATE_TO_LOWER 2 3702 # define DEPRECATE_TO_FOLD 3 3703 3704 U32 utf8n_flags = 0; 3705 const char * name; 3706 const char * alternative; 3707 3708 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE; 3709 3710 if (*e == NULL) { 3711 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN; 3712 3713 /* strnlen() makes this function safe for the common case of 3714 * NUL-terminated strings */ 3715 *e = p + my_strnlen((char *) p, UTF8SKIP(p)); 3716 3717 /* For mathoms.c calls, we use the function name we know is stored 3718 * there. It could be part of a larger path */ 3719 if (type == DEPRECATE_TO_UPPER) { 3720 name = instr(file, "mathoms.c") 3721 ? "to_utf8_upper" 3722 : "toUPPER_utf8"; 3723 alternative = "toUPPER_utf8_safe"; 3724 } 3725 else if (type == DEPRECATE_TO_TITLE) { 3726 name = instr(file, "mathoms.c") 3727 ? "to_utf8_title" 3728 : "toTITLE_utf8"; 3729 alternative = "toTITLE_utf8_safe"; 3730 } 3731 else if (type == DEPRECATE_TO_LOWER) { 3732 name = instr(file, "mathoms.c") 3733 ? "to_utf8_lower" 3734 : "toLOWER_utf8"; 3735 alternative = "toLOWER_utf8_safe"; 3736 } 3737 else if (type == DEPRECATE_TO_FOLD) { 3738 name = instr(file, "mathoms.c") 3739 ? "to_utf8_fold" 3740 : "toFOLD_utf8"; 3741 alternative = "toFOLD_utf8_safe"; 3742 } 3743 else Perl_croak(aTHX_ "panic: Unexpected case change type"); 3744 3745 warn_on_first_deprecated_use(name, alternative, use_locale, file, line); 3746 } 3747 else { 3748 assert (p < *e); 3749 } 3750 3751 return utf8n_flags; 3752 } 3753 3754 STATIC UV 3755 S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e, 3756 U8 * ustrp, STRLEN *lenp) 3757 { 3758 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from 3759 * p0..e-1 according to Turkic rules is the same as for non-Turkic. 3760 * Otherwise, it returns the first code point of the Turkic foldcased 3761 * sequence, and the entire sequence will be stored in *ustrp. ustrp will 3762 * contain *lenp bytes 3763 * 3764 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER 3765 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER 3766 * DOTLESS I */ 3767 3768 PERL_ARGS_ASSERT_TURKIC_FC; 3769 assert(e > p); 3770 3771 if (UNLIKELY(*p == 'I')) { 3772 *lenp = 2; 3773 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I); 3774 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I); 3775 return LATIN_SMALL_LETTER_DOTLESS_I; 3776 } 3777 3778 if (UNLIKELY(memBEGINs(p, e - p, 3779 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8))) 3780 { 3781 *lenp = 1; 3782 *ustrp = 'i'; 3783 return 'i'; 3784 } 3785 3786 return 0; 3787 } 3788 3789 STATIC UV 3790 S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e, 3791 U8 * ustrp, STRLEN *lenp) 3792 { 3793 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from 3794 * p0..e-1 according to Turkic rules is the same as for non-Turkic. 3795 * Otherwise, it returns the first code point of the Turkic lowercased 3796 * sequence, and the entire sequence will be stored in *ustrp. ustrp will 3797 * contain *lenp bytes */ 3798 3799 dVAR; 3800 PERL_ARGS_ASSERT_TURKIC_LC; 3801 assert(e > p0); 3802 3803 /* A 'I' requires context as to what to do */ 3804 if (UNLIKELY(*p0 == 'I')) { 3805 const U8 * p = p0 + 1; 3806 3807 /* According to the Unicode SpecialCasing.txt file, a capital 'I' 3808 * modified by a dot above lowercases to 'i' even in turkic locales. */ 3809 while (p < e) { 3810 UV cp; 3811 3812 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) { 3813 ustrp[0] = 'i'; 3814 *lenp = 1; 3815 return 'i'; 3816 } 3817 3818 /* For the dot above to modify the 'I', it must be part of a 3819 * combining sequence immediately following the 'I', and no other 3820 * modifier with a ccc of 230 may intervene */ 3821 cp = utf8_to_uvchr_buf(p, e, NULL); 3822 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) { 3823 break; 3824 } 3825 3826 /* Here the combining sequence continues */ 3827 p += UTF8SKIP(p); 3828 } 3829 } 3830 3831 /* In all other cases the lc is the same as the fold */ 3832 return turkic_fc(p0, e, ustrp, lenp); 3833 } 3834 3835 STATIC UV 3836 S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e, 3837 U8 * ustrp, STRLEN *lenp) 3838 { 3839 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence 3840 * from p0..e-1 according to Turkic rules is the same as for non-Turkic. 3841 * Otherwise, it returns the first code point of the Turkic upper or 3842 * title-cased sequence, and the entire sequence will be stored in *ustrp. 3843 * ustrp will contain *lenp bytes 3844 * 3845 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER 3846 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER 3847 * DOTLESS I */ 3848 3849 PERL_ARGS_ASSERT_TURKIC_UC; 3850 assert(e > p); 3851 3852 if (*p == 'i') { 3853 *lenp = 2; 3854 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE); 3855 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE); 3856 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE; 3857 } 3858 3859 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) { 3860 *lenp = 1; 3861 *ustrp = 'I'; 3862 return 'I'; 3863 } 3864 3865 return 0; 3866 } 3867 3868 /* The process for changing the case is essentially the same for the four case 3869 * change types, except there are complications for folding. Otherwise the 3870 * difference is only which case to change to. To make sure that they all do 3871 * the same thing, the bodies of the functions are extracted out into the 3872 * following two macros. The functions are written with the same variable 3873 * names, and these are known and used inside these macros. It would be 3874 * better, of course, to have inline functions to do it, but since different 3875 * macros are called, depending on which case is being changed to, this is not 3876 * feasible in C (to khw's knowledge). Two macros are created so that the fold 3877 * function can start with the common start macro, then finish with its special 3878 * handling; while the other three cases can just use the common end macro. 3879 * 3880 * The algorithm is to use the proper (passed in) macro or function to change 3881 * the case for code points that are below 256. The macro is used if using 3882 * locale rules for the case change; the function if not. If the code point is 3883 * above 255, it is computed from the input UTF-8, and another macro is called 3884 * to do the conversion. If necessary, the output is converted to UTF-8. If 3885 * using a locale, we have to check that the change did not cross the 255/256 3886 * boundary, see check_locale_boundary_crossing() for further details. 3887 * 3888 * The macros are split with the correct case change for the below-256 case 3889 * stored into 'result', and in the middle of an else clause for the above-255 3890 * case. At that point in the 'else', 'result' is not the final result, but is 3891 * the input code point calculated from the UTF-8. The fold code needs to 3892 * realize all this and take it from there. 3893 * 3894 * To deal with Turkic locales, the function specified by the parameter 3895 * 'turkic' is called when appropriate. 3896 * 3897 * If you read the two macros as sequential, it's easier to understand what's 3898 * going on. */ 3899 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \ 3900 L1_func_extra_param, turkic) \ 3901 \ 3902 if (flags & (locale_flags)) { \ 3903 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ 3904 if (IN_UTF8_CTYPE_LOCALE) { \ 3905 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \ 3906 UV ret = turkic(p, e, ustrp, lenp); \ 3907 if (ret) return ret; \ 3908 } \ 3909 \ 3910 /* Otherwise, treat a UTF-8 locale as not being in locale at \ 3911 * all */ \ 3912 flags &= ~(locale_flags); \ 3913 } \ 3914 } \ 3915 \ 3916 if (UTF8_IS_INVARIANT(*p)) { \ 3917 if (flags & (locale_flags)) { \ 3918 result = LC_L1_change_macro(*p); \ 3919 } \ 3920 else { \ 3921 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \ 3922 } \ 3923 } \ 3924 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \ 3925 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \ 3926 if (flags & (locale_flags)) { \ 3927 result = LC_L1_change_macro(c); \ 3928 } \ 3929 else { \ 3930 return L1_func(c, ustrp, lenp, L1_func_extra_param); \ 3931 } \ 3932 } \ 3933 else { /* malformed UTF-8 or ord above 255 */ \ 3934 STRLEN len_result; \ 3935 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \ 3936 if (len_result == (STRLEN) -1) { \ 3937 _force_out_malformed_utf8_message(p, e, utf8n_flags, \ 3938 1 /* Die */ ); \ 3939 } 3940 3941 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \ 3942 result = change_macro(result, p, ustrp, lenp); \ 3943 \ 3944 if (flags & (locale_flags)) { \ 3945 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \ 3946 } \ 3947 return result; \ 3948 } \ 3949 \ 3950 /* Here, used locale rules. Convert back to UTF-8 */ \ 3951 if (UTF8_IS_INVARIANT(result)) { \ 3952 *ustrp = (U8) result; \ 3953 *lenp = 1; \ 3954 } \ 3955 else { \ 3956 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \ 3957 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \ 3958 *lenp = 2; \ 3959 } \ 3960 \ 3961 return result; 3962 3963 /* 3964 =for apidoc to_utf8_upper 3965 3966 Instead use L</toUPPER_utf8_safe>. 3967 3968 =cut */ 3969 3970 /* Not currently externally documented, and subject to change: 3971 * <flags> is set iff iff the rules from the current underlying locale are to 3972 * be used. */ 3973 3974 UV 3975 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, 3976 const U8 *e, 3977 U8* ustrp, 3978 STRLEN *lenp, 3979 bool flags, 3980 const char * const file, 3981 const int line) 3982 { 3983 dVAR; 3984 UV result; 3985 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER, 3986 cBOOL(flags), file, line); 3987 3988 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS; 3989 3990 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */ 3991 /* 2nd char of uc(U+DF) is 'S' */ 3992 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S', 3993 turkic_uc); 3994 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE); 3995 } 3996 3997 /* 3998 =for apidoc to_utf8_title 3999 4000 Instead use L</toTITLE_utf8_safe>. 4001 4002 =cut */ 4003 4004 /* Not currently externally documented, and subject to change: 4005 * <flags> is set iff the rules from the current underlying locale are to be 4006 * used. Since titlecase is not defined in POSIX, for other than a 4007 * UTF-8 locale, uppercase is used instead for code points < 256. 4008 */ 4009 4010 UV 4011 Perl__to_utf8_title_flags(pTHX_ const U8 *p, 4012 const U8 *e, 4013 U8* ustrp, 4014 STRLEN *lenp, 4015 bool flags, 4016 const char * const file, 4017 const int line) 4018 { 4019 dVAR; 4020 UV result; 4021 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE, 4022 cBOOL(flags), file, line); 4023 4024 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS; 4025 4026 /* 2nd char of ucfirst(U+DF) is 's' */ 4027 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's', 4028 turkic_uc); 4029 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE); 4030 } 4031 4032 /* 4033 =for apidoc to_utf8_lower 4034 4035 Instead use L</toLOWER_utf8_safe>. 4036 4037 =cut */ 4038 4039 /* Not currently externally documented, and subject to change: 4040 * <flags> is set iff iff the rules from the current underlying locale are to 4041 * be used. 4042 */ 4043 4044 UV 4045 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, 4046 const U8 *e, 4047 U8* ustrp, 4048 STRLEN *lenp, 4049 bool flags, 4050 const char * const file, 4051 const int line) 4052 { 4053 dVAR; 4054 UV result; 4055 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER, 4056 cBOOL(flags), file, line); 4057 4058 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS; 4059 4060 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */, 4061 turkic_lc); 4062 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE) 4063 } 4064 4065 /* 4066 =for apidoc to_utf8_fold 4067 4068 Instead use L</toFOLD_utf8_safe>. 4069 4070 =cut */ 4071 4072 /* Not currently externally documented, and subject to change, 4073 * in <flags> 4074 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying 4075 * locale are to be used. 4076 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used; 4077 * otherwise simple folds 4078 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are 4079 * prohibited 4080 */ 4081 4082 UV 4083 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, 4084 const U8 *e, 4085 U8* ustrp, 4086 STRLEN *lenp, 4087 U8 flags, 4088 const char * const file, 4089 const int line) 4090 { 4091 dVAR; 4092 UV result; 4093 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD, 4094 cBOOL(flags), file, line); 4095 4096 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS; 4097 4098 /* These are mutually exclusive */ 4099 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII))); 4100 4101 assert(p != ustrp); /* Otherwise overwrites */ 4102 4103 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1, 4104 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)), 4105 turkic_fc); 4106 4107 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL); 4108 4109 if (flags & FOLD_FLAGS_LOCALE) { 4110 4111 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8 4112 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8 4113 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8 4114 4115 /* Special case these two characters, as what normally gets 4116 * returned under locale doesn't work */ 4117 if (memBEGINs((char *) p, e - p, CAP_SHARP_S)) 4118 { 4119 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 4120 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 4121 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; " 4122 "resolved to \"\\x{17F}\\x{17F}\"."); 4123 goto return_long_s; 4124 } 4125 else 4126 #endif 4127 if (memBEGINs((char *) p, e - p, LONG_S_T)) 4128 { 4129 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 4130 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 4131 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; " 4132 "resolved to \"\\x{FB06}\"."); 4133 goto return_ligature_st; 4134 } 4135 4136 #if UNICODE_MAJOR_VERSION == 3 \ 4137 && UNICODE_DOT_VERSION == 0 \ 4138 && UNICODE_DOT_DOT_VERSION == 1 4139 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8 4140 4141 /* And special case this on this Unicode version only, for the same 4142 * reaons the other two are special cased. They would cross the 4143 * 255/256 boundary which is forbidden under /l, and so the code 4144 * wouldn't catch that they are equivalent (which they are only in 4145 * this release) */ 4146 else if (memBEGINs((char *) p, e - p, DOTTED_I)) { 4147 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 4148 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 4149 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; " 4150 "resolved to \"\\x{0131}\"."); 4151 goto return_dotless_i; 4152 } 4153 #endif 4154 4155 return check_locale_boundary_crossing(p, result, ustrp, lenp); 4156 } 4157 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) { 4158 return result; 4159 } 4160 else { 4161 /* This is called when changing the case of a UTF-8-encoded 4162 * character above the ASCII range, and the result should not 4163 * contain an ASCII character. */ 4164 4165 UV original; /* To store the first code point of <p> */ 4166 4167 /* Look at every character in the result; if any cross the 4168 * boundary, the whole thing is disallowed */ 4169 U8* s = ustrp; 4170 U8* e = ustrp + *lenp; 4171 while (s < e) { 4172 if (isASCII(*s)) { 4173 /* Crossed, have to return the original */ 4174 original = valid_utf8_to_uvchr(p, lenp); 4175 4176 /* But in these instances, there is an alternative we can 4177 * return that is valid */ 4178 if (original == LATIN_SMALL_LETTER_SHARP_S 4179 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */ 4180 || original == LATIN_CAPITAL_LETTER_SHARP_S 4181 #endif 4182 ) { 4183 goto return_long_s; 4184 } 4185 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) { 4186 goto return_ligature_st; 4187 } 4188 #if UNICODE_MAJOR_VERSION == 3 \ 4189 && UNICODE_DOT_VERSION == 0 \ 4190 && UNICODE_DOT_DOT_VERSION == 1 4191 4192 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) { 4193 goto return_dotless_i; 4194 } 4195 #endif 4196 Copy(p, ustrp, *lenp, char); 4197 return original; 4198 } 4199 s += UTF8SKIP(s); 4200 } 4201 4202 /* Here, no characters crossed, result is ok as-is */ 4203 return result; 4204 } 4205 } 4206 4207 /* Here, used locale rules. Convert back to UTF-8 */ 4208 if (UTF8_IS_INVARIANT(result)) { 4209 *ustrp = (U8) result; 4210 *lenp = 1; 4211 } 4212 else { 4213 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); 4214 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); 4215 *lenp = 2; 4216 } 4217 4218 return result; 4219 4220 return_long_s: 4221 /* Certain folds to 'ss' are prohibited by the options, but they do allow 4222 * folds to a string of two of these characters. By returning this 4223 * instead, then, e.g., 4224 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}") 4225 * works. */ 4226 4227 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; 4228 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, 4229 ustrp, *lenp, U8); 4230 return LATIN_SMALL_LETTER_LONG_S; 4231 4232 return_ligature_st: 4233 /* Two folds to 'st' are prohibited by the options; instead we pick one and 4234 * have the other one fold to it */ 4235 4236 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1; 4237 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8); 4238 return LATIN_SMALL_LIGATURE_ST; 4239 4240 #if UNICODE_MAJOR_VERSION == 3 \ 4241 && UNICODE_DOT_VERSION == 0 \ 4242 && UNICODE_DOT_DOT_VERSION == 1 4243 4244 return_dotless_i: 4245 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1; 4246 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8); 4247 return LATIN_SMALL_LETTER_DOTLESS_I; 4248 4249 #endif 4250 4251 } 4252 4253 /* Note: 4254 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch(). 4255 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8". 4256 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl. 4257 */ 4258 4259 SV* 4260 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, 4261 I32 minbits, I32 none) 4262 { 4263 /* Returns a copy of a swash initiated by the called function. This is the 4264 * public interface, and returning a copy prevents others from doing 4265 * mischief on the original. The only remaining use of this is in tr/// */ 4266 4267 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST 4268 * use the following define */ 4269 4270 #define SWASH_INIT_RETURN(x) \ 4271 PL_curpm= old_PL_curpm; \ 4272 return newSVsv(x) 4273 4274 /* Initialize and return a swash, creating it if necessary. It does this 4275 * by calling utf8_heavy.pl in the general case. 4276 * 4277 * pkg is the name of the package that <name> should be in. 4278 * name is the name of the swash to find. 4279 * listsv is a string to initialize the swash with. It must be of the form 4280 * documented as the subroutine return value in 4281 * L<perlunicode/User-Defined Character Properties> 4282 * minbits is the number of bits required to represent each data element. 4283 * none I (khw) do not understand this one, but it is used only in tr///. 4284 * 4285 * Thus there are two possible inputs to find the swash: <name> and 4286 * <listsv>. At least one must be specified. The result 4287 * will be the union of the specified ones, although <listsv>'s various 4288 * actions can intersect, etc. what <name> gives. To avoid going out to 4289 * disk at all, <invlist> should specify completely what the swash should 4290 * have, and <listsv> should be &PL_sv_undef and <name> should be "". 4291 */ 4292 4293 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */ 4294 4295 SV* retval = &PL_sv_undef; 4296 4297 PERL_ARGS_ASSERT_SWASH_INIT; 4298 4299 assert(listsv != &PL_sv_undef || strNE(name, "")); 4300 4301 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the 4302 regex that triggered the swash init and the swash init 4303 perl logic itself. See perl #122747 */ 4304 4305 /* If data was passed in to go out to utf8_heavy to find the swash of, do 4306 * so */ 4307 if (listsv != &PL_sv_undef || strNE(name, "")) { 4308 dSP; 4309 const size_t pkg_len = strlen(pkg); 4310 const size_t name_len = strlen(name); 4311 HV * const stash = gv_stashpvn(pkg, pkg_len, 0); 4312 SV* errsv_save; 4313 GV *method; 4314 4315 4316 PUSHSTACKi(PERLSI_MAGIC); 4317 ENTER; 4318 SAVEHINTS(); 4319 save_re_context(); 4320 /* We might get here via a subroutine signature which uses a utf8 4321 * parameter name, at which point PL_subname will have been set 4322 * but not yet used. */ 4323 save_item(PL_subname); 4324 if (PL_parser && PL_parser->error_count) 4325 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0; 4326 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1); 4327 if (!method) { /* demand load UTF-8 */ 4328 ENTER; 4329 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); 4330 GvSV(PL_errgv) = NULL; 4331 #ifndef NO_TAINT_SUPPORT 4332 /* It is assumed that callers of this routine are not passing in 4333 * any user derived data. */ 4334 /* Need to do this after save_re_context() as it will set 4335 * PL_tainted to 1 while saving $1 etc (see the code after getrx: 4336 * in Perl_magic_get). Even line to create errsv_save can turn on 4337 * PL_tainted. */ 4338 SAVEBOOL(TAINT_get); 4339 TAINT_NOT; 4340 #endif 4341 require_pv("utf8_heavy.pl"); 4342 { 4343 /* Not ERRSV, as there is no need to vivify a scalar we are 4344 about to discard. */ 4345 SV * const errsv = GvSV(PL_errgv); 4346 if (!SvTRUE(errsv)) { 4347 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); 4348 SvREFCNT_dec(errsv); 4349 } 4350 } 4351 LEAVE; 4352 } 4353 SPAGAIN; 4354 PUSHMARK(SP); 4355 EXTEND(SP,5); 4356 mPUSHp(pkg, pkg_len); 4357 mPUSHp(name, name_len); 4358 PUSHs(listsv); 4359 mPUSHi(minbits); 4360 mPUSHi(none); 4361 PUTBACK; 4362 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); 4363 GvSV(PL_errgv) = NULL; 4364 /* If we already have a pointer to the method, no need to use 4365 * call_method() to repeat the lookup. */ 4366 if (method 4367 ? call_sv(MUTABLE_SV(method), G_SCALAR) 4368 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD)) 4369 { 4370 retval = *PL_stack_sp--; 4371 SvREFCNT_inc(retval); 4372 } 4373 { 4374 /* Not ERRSV. See above. */ 4375 SV * const errsv = GvSV(PL_errgv); 4376 if (!SvTRUE(errsv)) { 4377 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); 4378 SvREFCNT_dec(errsv); 4379 } 4380 } 4381 LEAVE; 4382 POPSTACK; 4383 if (IN_PERL_COMPILETIME) { 4384 CopHINTS_set(PL_curcop, PL_hints); 4385 } 4386 } /* End of calling the module to find the swash */ 4387 4388 SWASH_INIT_RETURN(retval); 4389 #undef SWASH_INIT_RETURN 4390 } 4391 4392 4393 /* This API is wrong for special case conversions since we may need to 4394 * return several Unicode characters for a single Unicode character 4395 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is 4396 * the lower-level routine, and it is similarly broken for returning 4397 * multiple values. --jhi 4398 * For those, you should use S__to_utf8_case() instead */ 4399 /* Now SWASHGET is recasted into S_swatch_get in this file. */ 4400 4401 /* Note: 4402 * Returns the value of property/mapping C<swash> for the first character 4403 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is 4404 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr> 4405 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>. 4406 * 4407 * A "swash" is a hash which contains initially the keys/values set up by 4408 * SWASHNEW. The purpose is to be able to completely represent a Unicode 4409 * property for all possible code points. Things are stored in a compact form 4410 * (see utf8_heavy.pl) so that calculation is required to find the actual 4411 * property value for a given code point. As code points are looked up, new 4412 * key/value pairs are added to the hash, so that the calculation doesn't have 4413 * to ever be re-done. Further, each calculation is done, not just for the 4414 * desired one, but for a whole block of code points adjacent to that one. 4415 * For binary properties on ASCII machines, the block is usually for 64 code 4416 * points, starting with a code point evenly divisible by 64. Thus if the 4417 * property value for code point 257 is requested, the code goes out and 4418 * calculates the property values for all 64 code points between 256 and 319, 4419 * and stores these as a single 64-bit long bit vector, called a "swatch", 4420 * under the key for code point 256. The key is the UTF-8 encoding for code 4421 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding 4422 * for a code point is 13 bytes, the key will be 12 bytes long. If the value 4423 * for code point 258 is then requested, this code realizes that it would be 4424 * stored under the key for 256, and would find that value and extract the 4425 * relevant bit, offset from 256. 4426 * 4427 * Non-binary properties are stored in as many bits as necessary to represent 4428 * their values (32 currently, though the code is more general than that), not 4429 * as single bits, but the principle is the same: the value for each key is a 4430 * vector that encompasses the property values for all code points whose UTF-8 4431 * representations are represented by the key. That is, for all code points 4432 * whose UTF-8 representations are length N bytes, and the key is the first N-1 4433 * bytes of that. 4434 */ 4435 UV 4436 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8) 4437 { 4438 HV *const hv = MUTABLE_HV(SvRV(swash)); 4439 U32 klen; 4440 U32 off; 4441 STRLEN slen = 0; 4442 STRLEN needents; 4443 const U8 *tmps = NULL; 4444 SV *swatch; 4445 const U8 c = *ptr; 4446 4447 PERL_ARGS_ASSERT_SWASH_FETCH; 4448 4449 /* If it really isn't a hash, it isn't really swash; must be an inversion 4450 * list */ 4451 if (SvTYPE(hv) != SVt_PVHV) { 4452 return _invlist_contains_cp((SV*)hv, 4453 (do_utf8) 4454 ? valid_utf8_to_uvchr(ptr, NULL) 4455 : c); 4456 } 4457 4458 /* We store the values in a "swatch" which is a vec() value in a swash 4459 * hash. Code points 0-255 are a single vec() stored with key length 4460 * (klen) 0. All other code points have a UTF-8 representation 4461 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which 4462 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key 4463 * length for them is the length of the encoded char - 1. ptr[klen] is the 4464 * final byte in the sequence representing the character */ 4465 if (!do_utf8 || UTF8_IS_INVARIANT(c)) { 4466 klen = 0; 4467 needents = 256; 4468 off = c; 4469 } 4470 else if (UTF8_IS_DOWNGRADEABLE_START(c)) { 4471 klen = 0; 4472 needents = 256; 4473 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1)); 4474 } 4475 else { 4476 klen = UTF8SKIP(ptr) - 1; 4477 4478 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into 4479 * the vec is the final byte in the sequence. (In EBCDIC this is 4480 * converted to I8 to get consecutive values.) To help you visualize 4481 * all this: 4482 * Straight 1047 After final byte 4483 * UTF-8 UTF-EBCDIC I8 transform 4484 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0 4485 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1 4486 * ... 4487 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9 4488 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA 4489 * ... 4490 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2 4491 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3 4492 * ... 4493 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB 4494 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC 4495 * ... 4496 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF 4497 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41 4498 * 4499 * (There are no discontinuities in the elided (...) entries.) 4500 * The UTF-8 key for these 33 code points is '\xD0' (which also is the 4501 * key for the next 31, up through U+043F, whose UTF-8 final byte is 4502 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points. 4503 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an 4504 * index into the vec() swatch (after subtracting 0x80, which we 4505 * actually do with an '&'). 4506 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32 4507 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has 4508 * dicontinuities which go away by transforming it into I8, and we 4509 * effectively subtract 0xA0 to get the index. */ 4510 needents = (1 << UTF_ACCUMULATION_SHIFT); 4511 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK; 4512 } 4513 4514 /* 4515 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test 4516 * suite. (That is, only 7-8% overall over just a hash cache. Still, 4517 * it's nothing to sniff at.) Pity we usually come through at least 4518 * two function calls to get here... 4519 * 4520 * NB: this code assumes that swatches are never modified, once generated! 4521 */ 4522 4523 if (hv == PL_last_swash_hv && 4524 klen == PL_last_swash_klen && 4525 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) ) 4526 { 4527 tmps = PL_last_swash_tmps; 4528 slen = PL_last_swash_slen; 4529 } 4530 else { 4531 /* Try our second-level swatch cache, kept in a hash. */ 4532 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE); 4533 4534 /* If not cached, generate it via swatch_get */ 4535 if (!svp || !SvPOK(*svp) 4536 || !(tmps = (const U8*)SvPV_const(*svp, slen))) 4537 { 4538 if (klen) { 4539 const UV code_point = valid_utf8_to_uvchr(ptr, NULL); 4540 swatch = swatch_get(swash, 4541 code_point & ~((UV)needents - 1), 4542 needents); 4543 } 4544 else { /* For the first 256 code points, the swatch has a key of 4545 length 0 */ 4546 swatch = swatch_get(swash, 0, needents); 4547 } 4548 4549 if (IN_PERL_COMPILETIME) 4550 CopHINTS_set(PL_curcop, PL_hints); 4551 4552 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0); 4553 4554 if (!svp || !(tmps = (U8*)SvPV(*svp, slen)) 4555 || (slen << 3) < needents) 4556 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, " 4557 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf, 4558 svp, tmps, (UV)slen, (UV)needents); 4559 } 4560 4561 PL_last_swash_hv = hv; 4562 assert(klen <= sizeof(PL_last_swash_key)); 4563 PL_last_swash_klen = (U8)klen; 4564 /* FIXME change interpvar.h? */ 4565 PL_last_swash_tmps = (U8 *) tmps; 4566 PL_last_swash_slen = slen; 4567 if (klen) 4568 Copy(ptr, PL_last_swash_key, klen, U8); 4569 } 4570 4571 switch ((int)((slen << 3) / needents)) { 4572 case 1: 4573 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0; 4574 case 8: 4575 return ((UV) tmps[off]); 4576 case 16: 4577 off <<= 1; 4578 return 4579 ((UV) tmps[off ] << 8) + 4580 ((UV) tmps[off + 1]); 4581 case 32: 4582 off <<= 2; 4583 return 4584 ((UV) tmps[off ] << 24) + 4585 ((UV) tmps[off + 1] << 16) + 4586 ((UV) tmps[off + 2] << 8) + 4587 ((UV) tmps[off + 3]); 4588 } 4589 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, " 4590 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents); 4591 NORETURN_FUNCTION_END; 4592 } 4593 4594 /* Read a single line of the main body of the swash input text. These are of 4595 * the form: 4596 * 0053 0056 0073 4597 * where each number is hex. The first two numbers form the minimum and 4598 * maximum of a range, and the third is the value associated with the range. 4599 * Not all swashes should have a third number 4600 * 4601 * On input: l points to the beginning of the line to be examined; it points 4602 * to somewhere in the string of the whole input text, and is 4603 * terminated by a \n or the null string terminator. 4604 * lend points to the null terminator of that string 4605 * wants_value is non-zero if the swash expects a third number 4606 * typestr is the name of the swash's mapping, like 'ToLower' 4607 * On output: *min, *max, and *val are set to the values read from the line. 4608 * returns a pointer just beyond the line examined. If there was no 4609 * valid min number on the line, returns lend+1 4610 */ 4611 4612 STATIC U8* 4613 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val, 4614 const bool wants_value, const U8* const typestr) 4615 { 4616 const int typeto = typestr[0] == 'T' && typestr[1] == 'o'; 4617 STRLEN numlen; /* Length of the number */ 4618 I32 flags = PERL_SCAN_SILENT_ILLDIGIT 4619 | PERL_SCAN_DISALLOW_PREFIX 4620 | PERL_SCAN_SILENT_NON_PORTABLE; 4621 4622 /* nl points to the next \n in the scan */ 4623 U8* const nl = (U8*)memchr(l, '\n', lend - l); 4624 4625 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE; 4626 4627 /* Get the first number on the line: the range minimum */ 4628 numlen = lend - l; 4629 *min = grok_hex((char *)l, &numlen, &flags, NULL); 4630 *max = *min; /* So can never return without setting max */ 4631 if (numlen) /* If found a hex number, position past it */ 4632 l += numlen; 4633 else if (nl) { /* Else, go handle next line, if any */ 4634 return nl + 1; /* 1 is length of "\n" */ 4635 } 4636 else { /* Else, no next line */ 4637 return lend + 1; /* to LIST's end at which \n is not found */ 4638 } 4639 4640 /* The max range value follows, separated by a BLANK */ 4641 if (isBLANK(*l)) { 4642 ++l; 4643 flags = PERL_SCAN_SILENT_ILLDIGIT 4644 | PERL_SCAN_DISALLOW_PREFIX 4645 | PERL_SCAN_SILENT_NON_PORTABLE; 4646 numlen = lend - l; 4647 *max = grok_hex((char *)l, &numlen, &flags, NULL); 4648 if (numlen) 4649 l += numlen; 4650 else /* If no value here, it is a single element range */ 4651 *max = *min; 4652 4653 /* Non-binary tables have a third entry: what the first element of the 4654 * range maps to. The map for those currently read here is in hex */ 4655 if (wants_value) { 4656 if (isBLANK(*l)) { 4657 ++l; 4658 flags = PERL_SCAN_SILENT_ILLDIGIT 4659 | PERL_SCAN_DISALLOW_PREFIX 4660 | PERL_SCAN_SILENT_NON_PORTABLE; 4661 numlen = lend - l; 4662 *val = grok_hex((char *)l, &numlen, &flags, NULL); 4663 if (numlen) 4664 l += numlen; 4665 else 4666 *val = 0; 4667 } 4668 else { 4669 *val = 0; 4670 if (typeto) { 4671 /* diag_listed_as: To%s: illegal mapping '%s' */ 4672 Perl_croak(aTHX_ "%s: illegal mapping '%s'", 4673 typestr, l); 4674 } 4675 } 4676 } 4677 else 4678 *val = 0; /* bits == 1, then any val should be ignored */ 4679 } 4680 else { /* Nothing following range min, should be single element with no 4681 mapping expected */ 4682 if (wants_value) { 4683 *val = 0; 4684 if (typeto) { 4685 /* diag_listed_as: To%s: illegal mapping '%s' */ 4686 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l); 4687 } 4688 } 4689 else 4690 *val = 0; /* bits == 1, then val should be ignored */ 4691 } 4692 4693 /* Position to next line if any, or EOF */ 4694 if (nl) 4695 l = nl + 1; 4696 else 4697 l = lend; 4698 4699 return l; 4700 } 4701 4702 /* Note: 4703 * Returns a swatch (a bit vector string) for a code point sequence 4704 * that starts from the value C<start> and comprises the number C<span>. 4705 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl). 4706 * Should be used via swash_fetch, which will cache the swatch in C<swash>. 4707 */ 4708 STATIC SV* 4709 S_swatch_get(pTHX_ SV* swash, UV start, UV span) 4710 { 4711 SV *swatch; 4712 U8 *l, *lend, *x, *xend, *s; 4713 STRLEN lcur, xcur, scur; 4714 HV *const hv = MUTABLE_HV(SvRV(swash)); 4715 4716 SV** listsvp = NULL; /* The string containing the main body of the table */ 4717 SV** extssvp = NULL; 4718 U8* typestr = NULL; 4719 STRLEN bits = 0; 4720 STRLEN octets; /* if bits == 1, then octets == 0 */ 4721 UV none; 4722 UV end = start + span; 4723 4724 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE); 4725 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE); 4726 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE); 4727 extssvp = hv_fetchs(hv, "EXTRAS", FALSE); 4728 listsvp = hv_fetchs(hv, "LIST", FALSE); 4729 4730 bits = SvUV(*bitssvp); 4731 none = SvUV(*nonesvp); 4732 typestr = (U8*)SvPV_nolen(*typesvp); 4733 octets = bits >> 3; /* if bits == 1, then octets == 0 */ 4734 4735 PERL_ARGS_ASSERT_SWATCH_GET; 4736 4737 if (bits != 8 && bits != 16 && bits != 32) { 4738 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf, 4739 (UV)bits); 4740 } 4741 4742 /* If overflowed, use the max possible */ 4743 if (end < start) { 4744 end = UV_MAX; 4745 span = end - start; 4746 } 4747 4748 /* create and initialize $swatch */ 4749 scur = octets ? (span * octets) : (span + 7) / 8; 4750 swatch = newSV(scur); 4751 SvPOK_on(swatch); 4752 s = (U8*)SvPVX(swatch); 4753 if (octets && none) { 4754 const U8* const e = s + scur; 4755 while (s < e) { 4756 if (bits == 8) 4757 *s++ = (U8)(none & 0xff); 4758 else if (bits == 16) { 4759 *s++ = (U8)((none >> 8) & 0xff); 4760 *s++ = (U8)( none & 0xff); 4761 } 4762 else if (bits == 32) { 4763 *s++ = (U8)((none >> 24) & 0xff); 4764 *s++ = (U8)((none >> 16) & 0xff); 4765 *s++ = (U8)((none >> 8) & 0xff); 4766 *s++ = (U8)( none & 0xff); 4767 } 4768 } 4769 *s = '\0'; 4770 } 4771 else { 4772 (void)memzero((U8*)s, scur + 1); 4773 } 4774 SvCUR_set(swatch, scur); 4775 s = (U8*)SvPVX(swatch); 4776 4777 /* read $swash->{LIST} */ 4778 l = (U8*)SvPV(*listsvp, lcur); 4779 lend = l + lcur; 4780 while (l < lend) { 4781 UV min = 0, max = 0, val = 0, upper; 4782 l = swash_scan_list_line(l, lend, &min, &max, &val, 4783 cBOOL(octets), typestr); 4784 if (l > lend) { 4785 break; 4786 } 4787 4788 /* If looking for something beyond this range, go try the next one */ 4789 if (max < start) 4790 continue; 4791 4792 /* <end> is generally 1 beyond where we want to set things, but at the 4793 * platform's infinity, where we can't go any higher, we want to 4794 * include the code point at <end> */ 4795 upper = (max < end) 4796 ? max 4797 : (max != UV_MAX || end != UV_MAX) 4798 ? end - 1 4799 : end; 4800 4801 if (octets) { 4802 UV key; 4803 if (min < start) { 4804 if (!none || val < none) { 4805 val += start - min; 4806 } 4807 min = start; 4808 } 4809 for (key = min; key <= upper; key++) { 4810 STRLEN offset; 4811 /* offset must be non-negative (start <= min <= key < end) */ 4812 offset = octets * (key - start); 4813 if (bits == 8) 4814 s[offset] = (U8)(val & 0xff); 4815 else if (bits == 16) { 4816 s[offset ] = (U8)((val >> 8) & 0xff); 4817 s[offset + 1] = (U8)( val & 0xff); 4818 } 4819 else if (bits == 32) { 4820 s[offset ] = (U8)((val >> 24) & 0xff); 4821 s[offset + 1] = (U8)((val >> 16) & 0xff); 4822 s[offset + 2] = (U8)((val >> 8) & 0xff); 4823 s[offset + 3] = (U8)( val & 0xff); 4824 } 4825 4826 if (!none || val < none) 4827 ++val; 4828 } 4829 } 4830 } /* while */ 4831 4832 /* read $swash->{EXTRAS} */ 4833 x = (U8*)SvPV(*extssvp, xcur); 4834 xend = x + xcur; 4835 while (x < xend) { 4836 STRLEN namelen; 4837 U8 *namestr; 4838 SV** othersvp; 4839 HV* otherhv; 4840 STRLEN otherbits; 4841 SV **otherbitssvp, *other; 4842 U8 *s, *o, *nl; 4843 STRLEN slen, olen; 4844 4845 const U8 opc = *x++; 4846 if (opc == '\n') 4847 continue; 4848 4849 nl = (U8*)memchr(x, '\n', xend - x); 4850 4851 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { 4852 if (nl) { 4853 x = nl + 1; /* 1 is length of "\n" */ 4854 continue; 4855 } 4856 else { 4857 x = xend; /* to EXTRAS' end at which \n is not found */ 4858 break; 4859 } 4860 } 4861 4862 namestr = x; 4863 if (nl) { 4864 namelen = nl - namestr; 4865 x = nl + 1; 4866 } 4867 else { 4868 namelen = xend - namestr; 4869 x = xend; 4870 } 4871 4872 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); 4873 otherhv = MUTABLE_HV(SvRV(*othersvp)); 4874 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); 4875 otherbits = (STRLEN)SvUV(*otherbitssvp); 4876 if (bits < otherbits) 4877 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, " 4878 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits); 4879 4880 /* The "other" swatch must be destroyed after. */ 4881 other = swatch_get(*othersvp, start, span); 4882 o = (U8*)SvPV(other, olen); 4883 4884 if (!olen) 4885 Perl_croak(aTHX_ "panic: swatch_get got improper swatch"); 4886 4887 s = (U8*)SvPV(swatch, slen); 4888 { 4889 STRLEN otheroctets = otherbits >> 3; 4890 STRLEN offset = 0; 4891 U8* const send = s + slen; 4892 4893 while (s < send) { 4894 UV otherval = 0; 4895 4896 if (otherbits == 1) { 4897 otherval = (o[offset >> 3] >> (offset & 7)) & 1; 4898 ++offset; 4899 } 4900 else { 4901 STRLEN vlen = otheroctets; 4902 otherval = *o++; 4903 while (--vlen) { 4904 otherval <<= 8; 4905 otherval |= *o++; 4906 } 4907 } 4908 4909 if (opc == '+' && otherval) 4910 NOOP; /* replace with otherval */ 4911 else if (opc == '!' && !otherval) 4912 otherval = 1; 4913 else if (opc == '-' && otherval) 4914 otherval = 0; 4915 else if (opc == '&' && !otherval) 4916 otherval = 0; 4917 else { 4918 s += octets; /* no replacement */ 4919 continue; 4920 } 4921 4922 if (bits == 8) 4923 *s++ = (U8)( otherval & 0xff); 4924 else if (bits == 16) { 4925 *s++ = (U8)((otherval >> 8) & 0xff); 4926 *s++ = (U8)( otherval & 0xff); 4927 } 4928 else if (bits == 32) { 4929 *s++ = (U8)((otherval >> 24) & 0xff); 4930 *s++ = (U8)((otherval >> 16) & 0xff); 4931 *s++ = (U8)((otherval >> 8) & 0xff); 4932 *s++ = (U8)( otherval & 0xff); 4933 } 4934 } 4935 } 4936 sv_free(other); /* through with it! */ 4937 } /* while */ 4938 return swatch; 4939 } 4940 4941 bool 4942 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len) 4943 { 4944 /* May change: warns if surrogates, non-character code points, or 4945 * non-Unicode code points are in 's' which has length 'len' bytes. 4946 * Returns TRUE if none found; FALSE otherwise. The only other validity 4947 * check is to make sure that this won't exceed the string's length nor 4948 * overflow */ 4949 4950 const U8* const e = s + len; 4951 bool ok = TRUE; 4952 4953 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT; 4954 4955 while (s < e) { 4956 if (UTF8SKIP(s) > len) { 4957 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 4958 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print"); 4959 return FALSE; 4960 } 4961 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) { 4962 if (UNLIKELY(UTF8_IS_SUPER(s, e))) { 4963 if ( ckWARN_d(WARN_NON_UNICODE) 4964 || UNLIKELY(0 < does_utf8_overflow(s, s + len, 4965 0 /* Don't consider overlongs */ 4966 ))) 4967 { 4968 /* A side effect of this function will be to warn */ 4969 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER); 4970 ok = FALSE; 4971 } 4972 } 4973 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) { 4974 if (ckWARN_d(WARN_SURROGATE)) { 4975 /* This has a different warning than the one the called 4976 * function would output, so can't just call it, unlike we 4977 * do for the non-chars and above-unicodes */ 4978 UV uv = utf8_to_uvchr_buf(s, e, NULL); 4979 Perl_warner(aTHX_ packWARN(WARN_SURROGATE), 4980 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", 4981 uv); 4982 ok = FALSE; 4983 } 4984 } 4985 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e)) 4986 && (ckWARN_d(WARN_NONCHAR))) 4987 { 4988 /* A side effect of this function will be to warn */ 4989 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR); 4990 ok = FALSE; 4991 } 4992 } 4993 s += UTF8SKIP(s); 4994 } 4995 4996 return ok; 4997 } 4998 4999 /* 5000 =for apidoc pv_uni_display 5001 5002 Build to the scalar C<dsv> a displayable version of the string C<spv>, 5003 length C<len>, the displayable version being at most C<pvlim> bytes long 5004 (if longer, the rest is truncated and C<"..."> will be appended). 5005 5006 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display 5007 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH> 5008 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">) 5009 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">). 5010 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both 5011 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on. 5012 5013 The pointer to the PV of the C<dsv> is returned. 5014 5015 See also L</sv_uni_display>. 5016 5017 =cut */ 5018 char * 5019 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, 5020 UV flags) 5021 { 5022 int truncated = 0; 5023 const char *s, *e; 5024 5025 PERL_ARGS_ASSERT_PV_UNI_DISPLAY; 5026 5027 SvPVCLEAR(dsv); 5028 SvUTF8_off(dsv); 5029 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) { 5030 UV u; 5031 /* This serves double duty as a flag and a character to print after 5032 a \ when flags & UNI_DISPLAY_BACKSLASH is true. 5033 */ 5034 char ok = 0; 5035 5036 if (pvlim && SvCUR(dsv) >= pvlim) { 5037 truncated++; 5038 break; 5039 } 5040 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0); 5041 if (u < 256) { 5042 const unsigned char c = (unsigned char)u & 0xFF; 5043 if (flags & UNI_DISPLAY_BACKSLASH) { 5044 switch (c) { 5045 case '\n': 5046 ok = 'n'; break; 5047 case '\r': 5048 ok = 'r'; break; 5049 case '\t': 5050 ok = 't'; break; 5051 case '\f': 5052 ok = 'f'; break; 5053 case '\a': 5054 ok = 'a'; break; 5055 case '\\': 5056 ok = '\\'; break; 5057 default: break; 5058 } 5059 if (ok) { 5060 const char string = ok; 5061 sv_catpvs(dsv, "\\"); 5062 sv_catpvn(dsv, &string, 1); 5063 } 5064 } 5065 /* isPRINT() is the locale-blind version. */ 5066 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) { 5067 const char string = c; 5068 sv_catpvn(dsv, &string, 1); 5069 ok = 1; 5070 } 5071 } 5072 if (!ok) 5073 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u); 5074 } 5075 if (truncated) 5076 sv_catpvs(dsv, "..."); 5077 5078 return SvPVX(dsv); 5079 } 5080 5081 /* 5082 =for apidoc sv_uni_display 5083 5084 Build to the scalar C<dsv> a displayable version of the scalar C<sv>, 5085 the displayable version being at most C<pvlim> bytes long 5086 (if longer, the rest is truncated and "..." will be appended). 5087 5088 The C<flags> argument is as in L</pv_uni_display>(). 5089 5090 The pointer to the PV of the C<dsv> is returned. 5091 5092 =cut 5093 */ 5094 char * 5095 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags) 5096 { 5097 const char * const ptr = 5098 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv); 5099 5100 PERL_ARGS_ASSERT_SV_UNI_DISPLAY; 5101 5102 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr, 5103 SvCUR(ssv), pvlim, flags); 5104 } 5105 5106 /* 5107 =for apidoc foldEQ_utf8 5108 5109 Returns true if the leading portions of the strings C<s1> and C<s2> (either or 5110 both of which may be in UTF-8) are the same case-insensitively; false 5111 otherwise. How far into the strings to compare is determined by other input 5112 parameters. 5113 5114 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode; 5115 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for 5116 C<u2> with respect to C<s2>. 5117 5118 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for 5119 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach. 5120 The scan will not be considered to be a match unless the goal is reached, and 5121 scanning won't continue past that goal. Correspondingly for C<l2> with respect 5122 to C<s2>. 5123 5124 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that 5125 pointer is considered an end pointer to the position 1 byte past the maximum 5126 point in C<s1> beyond which scanning will not continue under any circumstances. 5127 (This routine assumes that UTF-8 encoded input strings are not malformed; 5128 malformed input can cause it to read past C<pe1>). This means that if both 5129 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match 5130 will never be successful because it can never 5131 get as far as its goal (and in fact is asserted against). Correspondingly for 5132 C<pe2> with respect to C<s2>. 5133 5134 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and 5135 C<l2> must be non-zero), and if both do, both have to be 5136 reached for a successful match. Also, if the fold of a character is multiple 5137 characters, all of them must be matched (see tr21 reference below for 5138 'folding'). 5139 5140 Upon a successful match, if C<pe1> is non-C<NULL>, 5141 it will be set to point to the beginning of the I<next> character of C<s1> 5142 beyond what was matched. Correspondingly for C<pe2> and C<s2>. 5143 5144 For case-insensitiveness, the "casefolding" of Unicode is used 5145 instead of upper/lowercasing both the characters, see 5146 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings). 5147 5148 =cut */ 5149 5150 /* A flags parameter has been added which may change, and hence isn't 5151 * externally documented. Currently it is: 5152 * 0 for as-documented above 5153 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an 5154 ASCII one, to not match 5155 * FOLDEQ_LOCALE is set iff the rules from the current underlying 5156 * locale are to be used. 5157 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this 5158 * routine. This allows that step to be skipped. 5159 * Currently, this requires s1 to be encoded as UTF-8 5160 * (u1 must be true), which is asserted for. 5161 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may 5162 * cross certain boundaries. Hence, the caller should 5163 * let this function do the folding instead of 5164 * pre-folding. This code contains an assertion to 5165 * that effect. However, if the caller knows what 5166 * it's doing, it can pass this flag to indicate that, 5167 * and the assertion is skipped. 5168 * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies 5169 * to s2, and s2 doesn't have to be UTF-8 encoded. 5170 * This introduces an asymmetry to save a few branches 5171 * in a loop. Currently, this is not a problem, as 5172 * never are both inputs pre-folded. Simply call this 5173 * function with the pre-folded one as the second 5174 * string. 5175 * FOLDEQ_S2_FOLDS_SANE 5176 */ 5177 I32 5178 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, 5179 const char *s2, char **pe2, UV l2, bool u2, 5180 U32 flags) 5181 { 5182 const U8 *p1 = (const U8*)s1; /* Point to current char */ 5183 const U8 *p2 = (const U8*)s2; 5184 const U8 *g1 = NULL; /* goal for s1 */ 5185 const U8 *g2 = NULL; 5186 const U8 *e1 = NULL; /* Don't scan s1 past this */ 5187 U8 *f1 = NULL; /* Point to current folded */ 5188 const U8 *e2 = NULL; 5189 U8 *f2 = NULL; 5190 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */ 5191 U8 foldbuf1[UTF8_MAXBYTES_CASE+1]; 5192 U8 foldbuf2[UTF8_MAXBYTES_CASE+1]; 5193 U8 flags_for_folder = FOLD_FLAGS_FULL; 5194 5195 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS; 5196 5197 assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE)) 5198 && (( (flags & FOLDEQ_S1_ALREADY_FOLDED) 5199 && !(flags & FOLDEQ_S1_FOLDS_SANE)) 5200 || ( (flags & FOLDEQ_S2_ALREADY_FOLDED) 5201 && !(flags & FOLDEQ_S2_FOLDS_SANE))))); 5202 /* The algorithm is to trial the folds without regard to the flags on 5203 * the first line of the above assert(), and then see if the result 5204 * violates them. This means that the inputs can't be pre-folded to a 5205 * violating result, hence the assert. This could be changed, with the 5206 * addition of extra tests here for the already-folded case, which would 5207 * slow it down. That cost is more than any possible gain for when these 5208 * flags are specified, as the flags indicate /il or /iaa matching which 5209 * is less common than /iu, and I (khw) also believe that real-world /il 5210 * and /iaa matches are most likely to involve code points 0-255, and this 5211 * function only under rare conditions gets called for 0-255. */ 5212 5213 if (flags & FOLDEQ_LOCALE) { 5214 if (IN_UTF8_CTYPE_LOCALE) { 5215 if (UNLIKELY(PL_in_utf8_turkic_locale)) { 5216 flags_for_folder |= FOLD_FLAGS_LOCALE; 5217 } 5218 else { 5219 flags &= ~FOLDEQ_LOCALE; 5220 } 5221 } 5222 else { 5223 flags_for_folder |= FOLD_FLAGS_LOCALE; 5224 } 5225 } 5226 if (flags & FOLDEQ_UTF8_NOMIX_ASCII) { 5227 flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII; 5228 } 5229 5230 if (pe1) { 5231 e1 = *(U8**)pe1; 5232 } 5233 5234 if (l1) { 5235 g1 = (const U8*)s1 + l1; 5236 } 5237 5238 if (pe2) { 5239 e2 = *(U8**)pe2; 5240 } 5241 5242 if (l2) { 5243 g2 = (const U8*)s2 + l2; 5244 } 5245 5246 /* Must have at least one goal */ 5247 assert(g1 || g2); 5248 5249 if (g1) { 5250 5251 /* Will never match if goal is out-of-bounds */ 5252 assert(! e1 || e1 >= g1); 5253 5254 /* Here, there isn't an end pointer, or it is beyond the goal. We 5255 * only go as far as the goal */ 5256 e1 = g1; 5257 } 5258 else { 5259 assert(e1); /* Must have an end for looking at s1 */ 5260 } 5261 5262 /* Same for goal for s2 */ 5263 if (g2) { 5264 assert(! e2 || e2 >= g2); 5265 e2 = g2; 5266 } 5267 else { 5268 assert(e2); 5269 } 5270 5271 /* If both operands are already folded, we could just do a memEQ on the 5272 * whole strings at once, but it would be better if the caller realized 5273 * this and didn't even call us */ 5274 5275 /* Look through both strings, a character at a time */ 5276 while (p1 < e1 && p2 < e2) { 5277 5278 /* If at the beginning of a new character in s1, get its fold to use 5279 * and the length of the fold. */ 5280 if (n1 == 0) { 5281 if (flags & FOLDEQ_S1_ALREADY_FOLDED) { 5282 f1 = (U8 *) p1; 5283 assert(u1); 5284 n1 = UTF8SKIP(f1); 5285 } 5286 else { 5287 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) { 5288 5289 /* We have to forbid mixing ASCII with non-ASCII if the 5290 * flags so indicate. And, we can short circuit having to 5291 * call the general functions for this common ASCII case, 5292 * all of whose non-locale folds are also ASCII, and hence 5293 * UTF-8 invariants, so the UTF8ness of the strings is not 5294 * relevant. */ 5295 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) { 5296 return 0; 5297 } 5298 n1 = 1; 5299 *foldbuf1 = toFOLD(*p1); 5300 } 5301 else if (u1) { 5302 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder); 5303 } 5304 else { /* Not UTF-8, get UTF-8 fold */ 5305 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder); 5306 } 5307 f1 = foldbuf1; 5308 } 5309 } 5310 5311 if (n2 == 0) { /* Same for s2 */ 5312 if (flags & FOLDEQ_S2_ALREADY_FOLDED) { 5313 5314 /* Point to the already-folded character. But for non-UTF-8 5315 * variants, convert to UTF-8 for the algorithm below */ 5316 if (UTF8_IS_INVARIANT(*p2)) { 5317 f2 = (U8 *) p2; 5318 n2 = 1; 5319 } 5320 else if (u2) { 5321 f2 = (U8 *) p2; 5322 n2 = UTF8SKIP(f2); 5323 } 5324 else { 5325 foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2); 5326 foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2); 5327 f2 = foldbuf2; 5328 n2 = 2; 5329 } 5330 } 5331 else { 5332 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) { 5333 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) { 5334 return 0; 5335 } 5336 n2 = 1; 5337 *foldbuf2 = toFOLD(*p2); 5338 } 5339 else if (u2) { 5340 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder); 5341 } 5342 else { 5343 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder); 5344 } 5345 f2 = foldbuf2; 5346 } 5347 } 5348 5349 /* Here f1 and f2 point to the beginning of the strings to compare. 5350 * These strings are the folds of the next character from each input 5351 * string, stored in UTF-8. */ 5352 5353 /* While there is more to look for in both folds, see if they 5354 * continue to match */ 5355 while (n1 && n2) { 5356 U8 fold_length = UTF8SKIP(f1); 5357 if (fold_length != UTF8SKIP(f2) 5358 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE 5359 function call for single 5360 byte */ 5361 || memNE((char*)f1, (char*)f2, fold_length)) 5362 { 5363 return 0; /* mismatch */ 5364 } 5365 5366 /* Here, they matched, advance past them */ 5367 n1 -= fold_length; 5368 f1 += fold_length; 5369 n2 -= fold_length; 5370 f2 += fold_length; 5371 } 5372 5373 /* When reach the end of any fold, advance the input past it */ 5374 if (n1 == 0) { 5375 p1 += u1 ? UTF8SKIP(p1) : 1; 5376 } 5377 if (n2 == 0) { 5378 p2 += u2 ? UTF8SKIP(p2) : 1; 5379 } 5380 } /* End of loop through both strings */ 5381 5382 /* A match is defined by each scan that specified an explicit length 5383 * reaching its final goal, and the other not having matched a partial 5384 * character (which can happen when the fold of a character is more than one 5385 * character). */ 5386 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) { 5387 return 0; 5388 } 5389 5390 /* Successful match. Set output pointers */ 5391 if (pe1) { 5392 *pe1 = (char*)p1; 5393 } 5394 if (pe2) { 5395 *pe2 = (char*)p2; 5396 } 5397 return 1; 5398 } 5399 5400 /* XXX The next two functions should likely be moved to mathoms.c once all 5401 * occurrences of them are removed from the core; some cpan-upstream modules 5402 * still use them */ 5403 5404 U8 * 5405 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv) 5406 { 5407 PERL_ARGS_ASSERT_UVUNI_TO_UTF8; 5408 5409 return uvoffuni_to_utf8_flags(d, uv, 0); 5410 } 5411 5412 /* 5413 =for apidoc utf8n_to_uvuni 5414 5415 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>. 5416 5417 This function was useful for code that wanted to handle both EBCDIC and 5418 ASCII platforms with Unicode properties, but starting in Perl v5.20, the 5419 distinctions between the platforms have mostly been made invisible to most 5420 code, so this function is quite unlikely to be what you want. If you do need 5421 this precise functionality, use instead 5422 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> 5423 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>. 5424 5425 =cut 5426 */ 5427 5428 UV 5429 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags) 5430 { 5431 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI; 5432 5433 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags)); 5434 } 5435 5436 /* 5437 =for apidoc uvuni_to_utf8_flags 5438 5439 Instead you almost certainly want to use L</uvchr_to_utf8> or 5440 L</uvchr_to_utf8_flags>. 5441 5442 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>, 5443 which itself, while not deprecated, should be used only in isolated 5444 circumstances. These functions were useful for code that wanted to handle 5445 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl 5446 v5.20, the distinctions between the platforms have mostly been made invisible 5447 to most code, so this function is quite unlikely to be what you want. 5448 5449 =cut 5450 */ 5451 5452 U8 * 5453 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) 5454 { 5455 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS; 5456 5457 return uvoffuni_to_utf8_flags(d, uv, flags); 5458 } 5459 5460 /* 5461 =for apidoc utf8_to_uvchr 5462 5463 Returns the native code point of the first character in the string C<s> 5464 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the 5465 length, in bytes, of that character. 5466 5467 Some, but not all, UTF-8 malformations are detected, and in fact, some 5468 malformed input could cause reading beyond the end of the input buffer, which 5469 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead. 5470 5471 If C<s> points to one of the detected malformations, and UTF8 warnings are 5472 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 5473 C<NULL>) to -1. If those warnings are off, the computed value if well-defined (or 5474 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> 5475 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the 5476 next possible position in C<s> that could begin a non-malformed character. 5477 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. 5478 5479 =cut 5480 */ 5481 5482 UV 5483 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen) 5484 { 5485 PERL_ARGS_ASSERT_UTF8_TO_UVCHR; 5486 5487 /* This function is unsafe if malformed UTF-8 input is given it, which is 5488 * why the function is deprecated. If the first byte of the input 5489 * indicates that there are more bytes remaining in the sequence that forms 5490 * the character than there are in the input buffer, it can read past the 5491 * end. But we can make it safe if the input string happens to be 5492 * NUL-terminated, as many strings in Perl are, by refusing to read past a 5493 * NUL. A NUL indicates the start of the next character anyway. If the 5494 * input isn't NUL-terminated, the function remains unsafe, as it always 5495 * has been. 5496 * 5497 * An initial NUL has to be handled separately, but all ASCIIs can be 5498 * handled the same way, speeding up this common case */ 5499 5500 if (UTF8_IS_INVARIANT(*s)) { /* Assumes 's' contains at least 1 byte */ 5501 if (retlen) { 5502 *retlen = 1; 5503 } 5504 return (UV) *s; 5505 } 5506 5507 return utf8_to_uvchr_buf(s, 5508 s + my_strnlen((char *) s, UTF8SKIP(s)), 5509 retlen); 5510 } 5511 5512 /* 5513 * ex: set ts=8 sts=4 sw=4 et: 5514 */ 5515