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 #include "uni_keywords.h" 36 37 static const char malformed_text[] = "Malformed UTF-8 character"; 38 static const char unees[] = 39 "Malformed UTF-8 character (unexpected end of string)"; 40 41 /* Be sure to synchronize this message with the similar one in regcomp.c */ 42 static const char cp_above_legal_max[] = 43 "Use of code point 0x%" UVXf " is not allowed; the" 44 " permissible max is 0x%" UVXf; 45 46 #define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX)) 47 48 /* 49 =head1 Unicode Support 50 These are various utility functions for manipulating UTF8-encoded 51 strings. For the uninitiated, this is a method of representing arbitrary 52 Unicode characters as a variable number of bytes, in such a way that 53 characters in the ASCII range are unmodified, and a zero byte never appears 54 within non-zero characters. 55 56 =cut 57 */ 58 59 void 60 Perl__force_out_malformed_utf8_message(pTHX_ 61 const U8 *const p, /* First byte in UTF-8 sequence */ 62 const U8 * const e, /* Final byte in sequence (may include 63 multiple chars */ 64 const U32 flags, /* Flags to pass to utf8n_to_uvchr(), 65 usually 0, or some DISALLOW flags */ 66 const bool die_here) /* If TRUE, this function does not return */ 67 { 68 /* This core-only function is to be called when a malformed UTF-8 character 69 * is found, in order to output the detailed information about the 70 * malformation before dieing. The reason it exists is for the occasions 71 * when such a malformation is fatal, but warnings might be turned off, so 72 * that normally they would not be actually output. This ensures that they 73 * do get output. Because a sequence may be malformed in more than one 74 * way, multiple messages may be generated, so we can't make them fatal, as 75 * that would cause the first one to die. 76 * 77 * Instead we pretend -W was passed to perl, then die afterwards. The 78 * flexibility is here to return to the caller so they can finish up and 79 * die themselves */ 80 U32 errors; 81 82 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE; 83 84 ENTER; 85 SAVEI8(PL_dowarn); 86 SAVESPTR(PL_curcop); 87 88 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON; 89 if (PL_curcop) { 90 PL_curcop->cop_warnings = pWARN_ALL; 91 } 92 93 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors); 94 95 LEAVE; 96 97 if (! errors) { 98 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should" 99 " be called only when there are errors found"); 100 } 101 102 if (die_here) { 103 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); 104 } 105 } 106 107 STATIC HV * 108 S_new_msg_hv(pTHX_ const char * const message, /* The message text */ 109 U32 categories, /* Packed warning categories */ 110 U32 flag) /* Flag associated with this message */ 111 { 112 /* Creates, populates, and returns an HV* that describes an error message 113 * for the translators between UTF8 and code point */ 114 115 SV* msg_sv = newSVpv(message, 0); 116 SV* category_sv = newSVuv(categories); 117 SV* flag_bit_sv = newSVuv(flag); 118 119 HV* msg_hv = newHV(); 120 121 PERL_ARGS_ASSERT_NEW_MSG_HV; 122 123 (void) hv_stores(msg_hv, "text", msg_sv); 124 (void) hv_stores(msg_hv, "warn_categories", category_sv); 125 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv); 126 127 return msg_hv; 128 } 129 130 /* 131 =for apidoc uvoffuni_to_utf8_flags 132 133 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 134 Instead, B<Almost all code should use L</uvchr_to_utf8> or 135 L</uvchr_to_utf8_flags>>. 136 137 This function is like them, but the input is a strict Unicode 138 (as opposed to native) code point. Only in very rare circumstances should code 139 not be using the native code point. 140 141 For details, see the description for L</uvchr_to_utf8_flags>. 142 143 =cut 144 */ 145 146 U8 * 147 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags) 148 { 149 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS; 150 151 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL); 152 } 153 154 /* All these formats take a single UV code point argument */ 155 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf; 156 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf 157 " is not recommended for open interchange"; 158 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode," 159 " may not be portable"; 160 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \ 161 " Unicode, requires a Perl extension," \ 162 " and so is not portable"; 163 164 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \ 165 STMT_START { \ 166 if (flags & UNICODE_WARN_SURROGATE) { \ 167 U32 category = packWARN(WARN_SURROGATE); \ 168 const char * format = surrogate_cp_format; \ 169 if (msgs) { \ 170 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ 171 category, \ 172 UNICODE_GOT_SURROGATE); \ 173 } \ 174 else { \ 175 Perl_ck_warner_d(aTHX_ category, format, uv); \ 176 } \ 177 } \ 178 if (flags & UNICODE_DISALLOW_SURROGATE) { \ 179 return NULL; \ 180 } \ 181 } STMT_END; 182 183 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \ 184 STMT_START { \ 185 if (flags & UNICODE_WARN_NONCHAR) { \ 186 U32 category = packWARN(WARN_NONCHAR); \ 187 const char * format = nonchar_cp_format; \ 188 if (msgs) { \ 189 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \ 190 category, \ 191 UNICODE_GOT_NONCHAR); \ 192 } \ 193 else { \ 194 Perl_ck_warner_d(aTHX_ category, format, uv); \ 195 } \ 196 } \ 197 if (flags & UNICODE_DISALLOW_NONCHAR) { \ 198 return NULL; \ 199 } \ 200 } STMT_END; 201 202 /* Use shorter names internally in this file */ 203 #define SHIFT UTF_ACCUMULATION_SHIFT 204 #undef MARK 205 #define MARK UTF_CONTINUATION_MARK 206 #define MASK UTF_CONTINUATION_MASK 207 208 /* 209 =for apidoc uvchr_to_utf8_flags_msgs 210 211 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 212 213 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly. 214 215 This function is for code that wants any warning and/or error messages to be 216 returned to the caller rather than be displayed. All messages that would have 217 been displayed if all lexcial warnings are enabled will be returned. 218 219 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter 220 placed after all the others, C<msgs>. If this parameter is 0, this function 221 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should 222 be a pointer to an C<HV *> variable, in which this function creates a new HV to 223 contain any appropriate messages. The hash has three key-value pairs, as 224 follows: 225 226 =over 4 227 228 =item C<text> 229 230 The text of the message as a C<SVpv>. 231 232 =item C<warn_categories> 233 234 The warning category (or categories) packed into a C<SVuv>. 235 236 =item C<flag> 237 238 A single flag bit associated with this message, in a C<SVuv>. 239 The bit corresponds to some bit in the C<*errors> return value, 240 such as C<UNICODE_GOT_SURROGATE>. 241 242 =back 243 244 It's important to note that specifying this parameter as non-null will cause 245 any warnings this function would otherwise generate to be suppressed, and 246 instead be placed in C<*msgs>. The caller can check the lexical warnings state 247 (or not) when choosing what to do with the returned messages. 248 249 The caller, of course, is responsible for freeing any returned HV. 250 251 =cut 252 */ 253 254 /* Undocumented; we don't want people using this. Instead they should use 255 * uvchr_to_utf8_flags_msgs() */ 256 U8 * 257 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs) 258 { 259 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS; 260 261 if (msgs) { 262 *msgs = NULL; 263 } 264 265 if (OFFUNI_IS_INVARIANT(uv)) { 266 *d++ = LATIN1_TO_NATIVE(uv); 267 return d; 268 } 269 270 if (uv <= MAX_UTF8_TWO_BYTE) { 271 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2)); 272 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK); 273 return d; 274 } 275 276 /* Not 2-byte; test for and handle 3-byte result. In the test immediately 277 * below, the 16 is for start bytes E0-EF (which are all the possible ones 278 * for 3 byte characters). The 2 is for 2 continuation bytes; these each 279 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000 280 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC; 281 * 0x800-0xFFFF on ASCII */ 282 if (uv < (16 * (1U << (2 * SHIFT)))) { 283 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3)); 284 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); 285 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); 286 287 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so 288 aren't tested here */ 289 /* The most likely code points in this range are below the surrogates. 290 * Do an extra test to quickly exclude those. */ 291 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) { 292 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) 293 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) 294 { 295 HANDLE_UNICODE_NONCHAR(uv, flags, msgs); 296 } 297 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 298 HANDLE_UNICODE_SURROGATE(uv, flags, msgs); 299 } 300 } 301 #endif 302 return d; 303 } 304 305 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII 306 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can 307 * happen starting with 4-byte characters on ASCII platforms. We unify the 308 * code for these with EBCDIC, even though some of them require 5-bytes on 309 * those, because khw believes the code saving is worth the very slight 310 * performance hit on these high EBCDIC code points. */ 311 312 if (UNLIKELY(UNICODE_IS_SUPER(uv))) { 313 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) { 314 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP); 315 } 316 if ( (flags & UNICODE_WARN_SUPER) 317 || ( (flags & UNICODE_WARN_PERL_EXTENDED) 318 && UNICODE_IS_PERL_EXTENDED(uv))) 319 { 320 const char * format = super_cp_format; 321 U32 category = packWARN(WARN_NON_UNICODE); 322 U32 flag = UNICODE_GOT_SUPER; 323 324 /* Choose the more dire applicable warning */ 325 if (UNICODE_IS_PERL_EXTENDED(uv)) { 326 format = perl_extended_cp_format; 327 if (flags & (UNICODE_WARN_PERL_EXTENDED 328 |UNICODE_DISALLOW_PERL_EXTENDED)) 329 { 330 flag = UNICODE_GOT_PERL_EXTENDED; 331 } 332 } 333 334 if (msgs) { 335 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), 336 category, flag); 337 } 338 else { 339 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv); 340 } 341 } 342 if ( (flags & UNICODE_DISALLOW_SUPER) 343 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED) 344 && UNICODE_IS_PERL_EXTENDED(uv))) 345 { 346 return NULL; 347 } 348 } 349 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) { 350 HANDLE_UNICODE_NONCHAR(uv, flags, msgs); 351 } 352 353 /* Test for and handle 4-byte result. In the test immediately below, the 354 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte 355 * characters). The 3 is for 3 continuation bytes; these each contribute 356 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on 357 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC; 358 * 0x1_0000-0x1F_FFFF on ASCII */ 359 if (uv < (8 * (1U << (3 * SHIFT)))) { 360 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4)); 361 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK); 362 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); 363 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); 364 365 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte 366 characters. The end-plane non-characters for EBCDIC were 367 handled just above */ 368 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) { 369 HANDLE_UNICODE_NONCHAR(uv, flags, msgs); 370 } 371 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 372 HANDLE_UNICODE_SURROGATE(uv, flags, msgs); 373 } 374 #endif 375 376 return d; 377 } 378 379 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII 380 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop 381 * format. The unrolled version above turns out to not save all that much 382 * time, and at these high code points (well above the legal Unicode range 383 * on ASCII platforms, and well above anything in common use in EBCDIC), 384 * khw believes that less code outweighs slight performance gains. */ 385 386 { 387 STRLEN len = OFFUNISKIP(uv); 388 U8 *p = d+len-1; 389 while (p > d) { 390 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK); 391 uv >>= UTF_ACCUMULATION_SHIFT; 392 } 393 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len)); 394 return d+len; 395 } 396 } 397 398 /* 399 =for apidoc uvchr_to_utf8 400 401 Adds the UTF-8 representation of the native code point C<uv> to the end 402 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to 403 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to 404 the byte after the end of the new character. In other words, 405 406 d = uvchr_to_utf8(d, uv); 407 408 is the recommended wide native character-aware way of saying 409 410 *(d++) = uv; 411 412 This function accepts any code point from 0..C<IV_MAX> as input. 413 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. 414 415 It is possible to forbid or warn on non-Unicode code points, or those that may 416 be problematic by using L</uvchr_to_utf8_flags>. 417 418 =cut 419 */ 420 421 /* This is also a macro */ 422 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv); 423 424 U8 * 425 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv) 426 { 427 return uvchr_to_utf8(d, uv); 428 } 429 430 /* 431 =for apidoc uvchr_to_utf8_flags 432 433 Adds the UTF-8 representation of the native code point C<uv> to the end 434 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to 435 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to 436 the byte after the end of the new character. In other words, 437 438 d = uvchr_to_utf8_flags(d, uv, flags); 439 440 or, in most cases, 441 442 d = uvchr_to_utf8_flags(d, uv, 0); 443 444 This is the Unicode-aware way of saying 445 446 *(d++) = uv; 447 448 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as 449 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word. 450 451 Specifying C<flags> can further restrict what is allowed and not warned on, as 452 follows: 453 454 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set, 455 the function will raise a warning, provided UTF8 warnings are enabled. If 456 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return 457 NULL. If both flags are set, the function will both warn and return NULL. 458 459 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags 460 affect how the function handles a Unicode non-character. 461 462 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags 463 affect the handling of code points that are above the Unicode maximum of 464 0x10FFFF. Languages other than Perl may not be able to accept files that 465 contain these. 466 467 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of 468 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all 469 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the 470 allowed inputs to the strict UTF-8 traditionally defined by Unicode. 471 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and 472 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the 473 above-Unicode and surrogate flags, but not the non-character ones, as 474 defined in 475 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. 476 See L<perlunicode/Noncharacter code points>. 477 478 Extremely high code points were never specified in any standard, and require an 479 extension to UTF-8 to express, which Perl does. It is likely that programs 480 written in something other than Perl would not be able to read files that 481 contain these; nor would Perl understand files written by something that uses a 482 different extension. For these reasons, there is a separate set of flags that 483 can warn and/or disallow these extremely high code points, even if other 484 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED> 485 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see 486 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will 487 treat all above-Unicode code points, including these, as malformations. (Note 488 that the Unicode standard considers anything above 0x10FFFF to be illegal, but 489 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1)) 490 491 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is 492 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly, 493 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named 494 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC 495 platforms,these flags can apply to code points that actually do fit in 31 bits. 496 The new names accurately describe the situation in all cases. 497 498 =cut 499 */ 500 501 /* This is also a macro */ 502 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags); 503 504 U8 * 505 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) 506 { 507 return uvchr_to_utf8_flags(d, uv, flags); 508 } 509 510 #ifndef UV_IS_QUAD 511 512 STATIC int 513 S_is_utf8_cp_above_31_bits(const U8 * const s, 514 const U8 * const e, 515 const bool consider_overlongs) 516 { 517 /* Returns TRUE if the first code point represented by the Perl-extended- 518 * UTF-8-encoded string starting at 's', and looking no further than 'e - 519 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31. 520 * 521 * The function handles the case where the input bytes do not include all 522 * the ones necessary to represent a full character. That is, they may be 523 * the intial bytes of the representation of a code point, but possibly 524 * the final ones necessary for the complete representation may be beyond 525 * 'e - 1'. 526 * 527 * The function also can handle the case where the input is an overlong 528 * sequence. If 'consider_overlongs' is 0, the function assumes the 529 * input is not overlong, without checking, and will return based on that 530 * assumption. If this parameter is 1, the function will go to the trouble 531 * of figuring out if it actually evaluates to above or below 31 bits. 532 * 533 * The sequence is otherwise assumed to be well-formed, without checking. 534 */ 535 536 const STRLEN len = e - s; 537 int is_overlong; 538 539 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS; 540 541 assert(! UTF8_IS_INVARIANT(*s) && e > s); 542 543 #ifdef EBCDIC 544 545 PERL_UNUSED_ARG(consider_overlongs); 546 547 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can 548 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can 549 * also be the start byte for a 31-bit code point; we need at least 2 550 * bytes, and maybe up through 8 bytes, to determine that. (It can also be 551 * the start byte for an overlong sequence, but for 30-bit or smaller code 552 * points, so we don't have to worry about overlongs on EBCDIC.) */ 553 if (*s != 0xFE) { 554 return 0; 555 } 556 557 if (len == 1) { 558 return -1; 559 } 560 561 #else 562 563 /* On ASCII, FE and FF are the only start bytes that can evaluate to 564 * needing more than 31 bits. */ 565 if (LIKELY(*s < 0xFE)) { 566 return 0; 567 } 568 569 /* What we have left are FE and FF. Both of these require more than 31 570 * bits unless they are for overlongs. */ 571 if (! consider_overlongs) { 572 return 1; 573 } 574 575 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to 576 * above 31 bits. But we need more than one byte to discern this, so if 577 * passed just the start byte, it could be an overlong evaluating to 578 * smaller */ 579 if (len == 1) { 580 return -1; 581 } 582 583 /* Having excluded len==1, and knowing that FE and FF are both valid start 584 * bytes, we can call the function below to see if the sequence is 585 * overlong. (We don't need the full generality of the called function, 586 * but for these huge code points, speed shouldn't be a consideration, and 587 * the compiler does have enough information, since it's static to this 588 * file, to optimize to just the needed parts.) */ 589 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len); 590 591 /* If it isn't overlong, more than 31 bits are required. */ 592 if (is_overlong == 0) { 593 return 1; 594 } 595 596 /* If it is indeterminate if it is overlong, return that */ 597 if (is_overlong < 0) { 598 return -1; 599 } 600 601 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as 602 * the max it can be is 2**31 - 1 */ 603 if (*s == 0xFE) { 604 return 0; 605 } 606 607 #endif 608 609 /* Here, ASCII and EBCDIC rejoin: 610 * On ASCII: We have an overlong sequence starting with FF 611 * On EBCDIC: We have a sequence starting with FE. */ 612 613 { /* For C89, use a block so the declaration can be close to its use */ 614 615 #ifdef EBCDIC 616 617 /* U+7FFFFFFF (2 ** 31 - 1) 618 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13 619 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73 620 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72 621 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75 622 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF 623 * U+80000000 (2 ** 31): 624 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 625 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 626 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 627 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0 628 * 629 * and since we know that *s = \xfe, any continuation sequcence 630 * following it that is gt the below is above 31 bits 631 [0] [1] [2] [3] [4] [5] [6] */ 632 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42"; 633 634 #else 635 636 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1) 637 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF 638 * FF overlong for U+80000000 (2 ** 31): 639 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80 640 * and since we know that *s = \xff, any continuation sequcence 641 * following it that is gt the below is above 30 bits 642 [0] [1] [2] [3] [4] [5] [6] */ 643 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81"; 644 645 646 #endif 647 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1; 648 const STRLEN cmp_len = MIN(conts_len, len - 1); 649 650 /* Now compare the continuation bytes in s with the ones we have 651 * compiled in that are for the largest 30 bit code point. If we have 652 * enough bytes available to determine the answer, or the bytes we do 653 * have differ from them, we can compare the two to get a definitive 654 * answer (Note that in UTF-EBCDIC, the two lowest possible 655 * continuation bytes are \x41 and \x42.) */ 656 if (cmp_len >= conts_len || memNE(s + 1, 657 conts_for_highest_30_bit, 658 cmp_len)) 659 { 660 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len)); 661 } 662 663 /* Here, all the bytes we have are the same as the highest 30-bit code 664 * point, but we are missing so many bytes that we can't make the 665 * determination */ 666 return -1; 667 } 668 } 669 670 #endif 671 672 PERL_STATIC_INLINE int 673 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len) 674 { 675 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to 676 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if 677 * it isn't, and -1 if there isn't enough information to tell. This last 678 * return value can happen if the sequence is incomplete, missing some 679 * trailing bytes that would form a complete character. If there are 680 * enough bytes to make a definitive decision, this function does so. 681 * Usually 2 bytes sufficient. 682 * 683 * Overlongs can occur whenever the number of continuation bytes changes. 684 * That means whenever the number of leading 1 bits in a start byte 685 * increases from the next lower start byte. That happens for start bytes 686 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal 687 * start bytes have already been excluded, so don't need to be tested here; 688 * ASCII platforms: C0, C1 689 * EBCDIC platforms C0, C1, C2, C3, C4, E0 690 */ 691 692 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); 693 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); 694 695 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK; 696 assert(len > 1 && UTF8_IS_START(*s)); 697 698 /* Each platform has overlongs after the start bytes given above (expressed 699 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but 700 * the logic is the same, except the E0 overlong has already been excluded 701 * on EBCDIC platforms. The values below were found by manually 702 * inspecting the UTF-8 patterns. See the tables in utf8.h and 703 * utfebcdic.h. */ 704 705 # ifdef EBCDIC 706 # define F0_ABOVE_OVERLONG 0xB0 707 # define F8_ABOVE_OVERLONG 0xA8 708 # define FC_ABOVE_OVERLONG 0xA4 709 # define FE_ABOVE_OVERLONG 0xA2 710 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41" 711 /* I8(0xfe) is FF */ 712 # else 713 714 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) { 715 return 1; 716 } 717 718 # define F0_ABOVE_OVERLONG 0x90 719 # define F8_ABOVE_OVERLONG 0x88 720 # define FC_ABOVE_OVERLONG 0x84 721 # define FE_ABOVE_OVERLONG 0x82 722 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80" 723 # endif 724 725 726 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG)) 727 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG)) 728 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG)) 729 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG))) 730 { 731 return 1; 732 } 733 734 /* Check for the FF overlong */ 735 return isFF_OVERLONG(s, len); 736 } 737 738 PERL_STATIC_INLINE int 739 S_isFF_OVERLONG(const U8 * const s, const STRLEN len) 740 { 741 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to 742 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if 743 * it isn't, and -1 if there isn't enough information to tell. This last 744 * return value can happen if the sequence is incomplete, missing some 745 * trailing bytes that would form a complete character. If there are 746 * enough bytes to make a definitive decision, this function does so. */ 747 748 PERL_ARGS_ASSERT_ISFF_OVERLONG; 749 750 /* To be an FF overlong, all the available bytes must match */ 751 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX, 752 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1)))) 753 { 754 return 0; 755 } 756 757 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must 758 * be there; what comes after them doesn't matter. See tables in utf8.h, 759 * utfebcdic.h. */ 760 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) { 761 return 1; 762 } 763 764 /* The missing bytes could cause the result to go one way or the other, so 765 * the result is indeterminate */ 766 return -1; 767 } 768 769 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */ 770 # ifdef EBCDIC /* Actually is I8 */ 771 # define HIGHEST_REPRESENTABLE_UTF8 \ 772 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 773 # else 774 # define HIGHEST_REPRESENTABLE_UTF8 \ 775 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 776 # endif 777 #endif 778 779 PERL_STATIC_INLINE int 780 S_does_utf8_overflow(const U8 * const s, 781 const U8 * e, 782 const bool consider_overlongs) 783 { 784 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to 785 * 'e' - 1 would overflow an IV on this platform; that is if it represents 786 * a code point larger than the highest representable code point. It 787 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't 788 * enough information to tell. This last return value can happen if the 789 * sequence is incomplete, missing some trailing bytes that would form a 790 * complete character. If there are enough bytes to make a definitive 791 * decision, this function does so. 792 * 793 * If 'consider_overlongs' is TRUE, the function checks for the possibility 794 * that the sequence is an overlong that doesn't overflow. Otherwise, it 795 * assumes the sequence is not an overlong. This can give different 796 * results only on ASCII 32-bit platforms. 797 * 798 * (For ASCII platforms, we could use memcmp() because we don't have to 799 * convert each byte to I8, but it's very rare input indeed that would 800 * approach overflow, so the loop below will likely only get executed once.) 801 * 802 * 'e' - 1 must not be beyond a full character. */ 803 804 805 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW; 806 assert(s <= e && s + UTF8SKIP(s) >= e); 807 808 #if ! defined(UV_IS_QUAD) 809 810 return is_utf8_cp_above_31_bits(s, e, consider_overlongs); 811 812 #else 813 814 PERL_UNUSED_ARG(consider_overlongs); 815 816 { 817 const STRLEN len = e - s; 818 const U8 *x; 819 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8; 820 821 for (x = s; x < e; x++, y++) { 822 823 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) { 824 continue; 825 } 826 827 /* If this byte is larger than the corresponding highest UTF-8 828 * byte, the sequence overflow; otherwise the byte is less than, 829 * and so the sequence doesn't overflow */ 830 return NATIVE_UTF8_TO_I8(*x) > *y; 831 832 } 833 834 /* Got to the end and all bytes are the same. If the input is a whole 835 * character, it doesn't overflow. And if it is a partial character, 836 * there's not enough information to tell */ 837 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) { 838 return -1; 839 } 840 841 return 0; 842 } 843 844 #endif 845 846 } 847 848 #if 0 849 850 /* This is the portions of the above function that deal with UV_MAX instead of 851 * IV_MAX. They are left here in case we want to combine them so that internal 852 * uses can have larger code points. The only logic difference is that the 853 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has 854 * different logic. 855 */ 856 857 /* Anything larger than this will overflow the word if it were converted into a UV */ 858 #if defined(UV_IS_QUAD) 859 # ifdef EBCDIC /* Actually is I8 */ 860 # define HIGHEST_REPRESENTABLE_UTF8 \ 861 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 862 # else 863 # define HIGHEST_REPRESENTABLE_UTF8 \ 864 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF" 865 # endif 866 #else /* 32-bit */ 867 # ifdef EBCDIC 868 # define HIGHEST_REPRESENTABLE_UTF8 \ 869 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF" 870 # else 871 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF" 872 # endif 873 #endif 874 875 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC) 876 877 /* On 32 bit ASCII machines, many overlongs that start with FF don't 878 * overflow */ 879 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) { 880 881 /* To be such an overlong, the first bytes of 's' must match 882 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we 883 * don't have any additional bytes available, the sequence, when 884 * completed might or might not fit in 32 bits. But if we have that 885 * next byte, we can tell for sure. If it is <= 0x83, then it does 886 * fit. */ 887 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) { 888 return -1; 889 } 890 891 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83; 892 } 893 894 /* Starting with the #else, the rest of the function is identical except 895 * 1. we need to move the 'len' declaration to be global to the function 896 * 2. the endif move to just after the UNUSED_ARG. 897 * An empty endif is given just below to satisfy the preprocessor 898 */ 899 #endif 900 901 #endif 902 903 #undef F0_ABOVE_OVERLONG 904 #undef F8_ABOVE_OVERLONG 905 #undef FC_ABOVE_OVERLONG 906 #undef FE_ABOVE_OVERLONG 907 #undef FF_OVERLONG_PREFIX 908 909 STRLEN 910 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags) 911 { 912 STRLEN len; 913 const U8 *x; 914 915 /* A helper function that should not be called directly. 916 * 917 * This function returns non-zero if the string beginning at 's' and 918 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a 919 * code point; otherwise it returns 0. The examination stops after the 920 * first code point in 's' is validated, not looking at the rest of the 921 * input. If 'e' is such that there are not enough bytes to represent a 922 * complete code point, this function will return non-zero anyway, if the 923 * bytes it does have are well-formed UTF-8 as far as they go, and aren't 924 * excluded by 'flags'. 925 * 926 * A non-zero return gives the number of bytes required to represent the 927 * code point. Be aware that if the input is for a partial character, the 928 * return will be larger than 'e - s'. 929 * 930 * This function assumes that the code point represented is UTF-8 variant. 931 * The caller should have excluded the possibility of it being invariant 932 * before calling this function. 933 * 934 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags 935 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return 936 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but 937 * disallowed by the flags. If the input is only for a partial character, 938 * the function will return non-zero if there is any sequence of 939 * well-formed UTF-8 that, when appended to the input sequence, could 940 * result in an allowed code point; otherwise it returns 0. Non characters 941 * cannot be determined based on partial character input. But many of the 942 * other excluded types can be determined with just the first one or two 943 * bytes. 944 * 945 */ 946 947 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER; 948 949 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE 950 |UTF8_DISALLOW_PERL_EXTENDED))); 951 assert(! UTF8_IS_INVARIANT(*s)); 952 953 /* A variant char must begin with a start byte */ 954 if (UNLIKELY(! UTF8_IS_START(*s))) { 955 return 0; 956 } 957 958 /* Examine a maximum of a single whole code point */ 959 if (e - s > UTF8SKIP(s)) { 960 e = s + UTF8SKIP(s); 961 } 962 963 len = e - s; 964 965 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) { 966 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); 967 968 /* Here, we are disallowing some set of largish code points, and the 969 * first byte indicates the sequence is for a code point that could be 970 * in the excluded set. We generally don't have to look beyond this or 971 * the second byte to see if the sequence is actually for one of the 972 * excluded classes. The code below is derived from this table: 973 * 974 * UTF-8 UTF-EBCDIC I8 975 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate 976 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate 977 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode 978 * 979 * Keep in mind that legal continuation bytes range between \x80..\xBF 980 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't 981 * continuation bytes. Hence, we don't have to test the upper edge 982 * because if any of those is encountered, the sequence is malformed, 983 * and would fail elsewhere in this function. 984 * 985 * The code here likewise assumes that there aren't other 986 * malformations; again the function should fail elsewhere because of 987 * these. For example, an overlong beginning with FC doesn't actually 988 * have to be a super; it could actually represent a small code point, 989 * even U+0000. But, since overlongs (and other malformations) are 990 * illegal, the function should return FALSE in either case. 991 */ 992 993 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */ 994 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA 995 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2) 996 997 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \ 998 /* B6 and B7 */ \ 999 && ((s1) & 0xFE ) == 0xB6) 1000 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF)) 1001 #else 1002 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5 1003 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90) 1004 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0) 1005 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE) 1006 #endif 1007 1008 if ( (flags & UTF8_DISALLOW_SUPER) 1009 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) 1010 { 1011 return 0; /* Above Unicode */ 1012 } 1013 1014 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED) 1015 && UNLIKELY(isUTF8_PERL_EXTENDED(s))) 1016 { 1017 return 0; 1018 } 1019 1020 if (len > 1) { 1021 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); 1022 1023 if ( (flags & UTF8_DISALLOW_SUPER) 1024 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1))) 1025 { 1026 return 0; /* Above Unicode */ 1027 } 1028 1029 if ( (flags & UTF8_DISALLOW_SURROGATE) 1030 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1))) 1031 { 1032 return 0; /* Surrogate */ 1033 } 1034 1035 if ( (flags & UTF8_DISALLOW_NONCHAR) 1036 && UNLIKELY(UTF8_IS_NONCHAR(s, e))) 1037 { 1038 return 0; /* Noncharacter code point */ 1039 } 1040 } 1041 } 1042 1043 /* Make sure that all that follows are continuation bytes */ 1044 for (x = s + 1; x < e; x++) { 1045 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) { 1046 return 0; 1047 } 1048 } 1049 1050 /* Here is syntactically valid. Next, make sure this isn't the start of an 1051 * overlong. */ 1052 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) { 1053 return 0; 1054 } 1055 1056 /* And finally, that the code point represented fits in a word on this 1057 * platform */ 1058 if (0 < does_utf8_overflow(s, e, 1059 0 /* Don't consider overlongs */ 1060 )) 1061 { 1062 return 0; 1063 } 1064 1065 return UTF8SKIP(s); 1066 } 1067 1068 char * 1069 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format) 1070 { 1071 /* Returns a mortalized C string that is a displayable copy of the 'len' 1072 * bytes starting at 'start'. 'format' gives how to display each byte. 1073 * Currently, there are only two formats, so it is currently a bool: 1074 * 0 \xab 1075 * 1 ab (that is a space between two hex digit bytes) 1076 */ 1077 1078 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a 1079 trailing NUL */ 1080 const U8 * s = start; 1081 const U8 * const e = start + len; 1082 char * output; 1083 char * d; 1084 1085 PERL_ARGS_ASSERT__BYTE_DUMP_STRING; 1086 1087 Newx(output, output_len, char); 1088 SAVEFREEPV(output); 1089 1090 d = output; 1091 for (s = start; s < e; s++) { 1092 const unsigned high_nibble = (*s & 0xF0) >> 4; 1093 const unsigned low_nibble = (*s & 0x0F); 1094 1095 if (format) { 1096 if (s > start) { 1097 *d++ = ' '; 1098 } 1099 } 1100 else { 1101 *d++ = '\\'; 1102 *d++ = 'x'; 1103 } 1104 1105 if (high_nibble < 10) { 1106 *d++ = high_nibble + '0'; 1107 } 1108 else { 1109 *d++ = high_nibble - 10 + 'a'; 1110 } 1111 1112 if (low_nibble < 10) { 1113 *d++ = low_nibble + '0'; 1114 } 1115 else { 1116 *d++ = low_nibble - 10 + 'a'; 1117 } 1118 } 1119 1120 *d = '\0'; 1121 return output; 1122 } 1123 1124 PERL_STATIC_INLINE char * 1125 S_unexpected_non_continuation_text(pTHX_ const U8 * const s, 1126 1127 /* Max number of bytes to print */ 1128 STRLEN print_len, 1129 1130 /* Which one is the non-continuation */ 1131 const STRLEN non_cont_byte_pos, 1132 1133 /* How many bytes should there be? */ 1134 const STRLEN expect_len) 1135 { 1136 /* Return the malformation warning text for an unexpected continuation 1137 * byte. */ 1138 1139 const char * const where = (non_cont_byte_pos == 1) 1140 ? "immediately" 1141 : Perl_form(aTHX_ "%d bytes", 1142 (int) non_cont_byte_pos); 1143 const U8 * x = s + non_cont_byte_pos; 1144 const U8 * e = s + print_len; 1145 1146 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT; 1147 1148 /* We don't need to pass this parameter, but since it has already been 1149 * calculated, it's likely faster to pass it; verify under DEBUGGING */ 1150 assert(expect_len == UTF8SKIP(s)); 1151 1152 /* As a defensive coding measure, don't output anything past a NUL. Such 1153 * bytes shouldn't be in the middle of a malformation, and could mark the 1154 * end of the allocated string, and what comes after is undefined */ 1155 for (; x < e; x++) { 1156 if (*x == '\0') { 1157 x++; /* Output this particular NUL */ 1158 break; 1159 } 1160 } 1161 1162 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x," 1163 " %s after start byte 0x%02x; need %d bytes, got %d)", 1164 malformed_text, 1165 _byte_dump_string(s, x - s, 0), 1166 *(s + non_cont_byte_pos), 1167 where, 1168 *s, 1169 (int) expect_len, 1170 (int) non_cont_byte_pos); 1171 } 1172 1173 /* 1174 1175 =for apidoc utf8n_to_uvchr 1176 1177 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 1178 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. 1179 1180 Bottom level UTF-8 decode routine. 1181 Returns the native code point value of the first character in the string C<s>, 1182 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than 1183 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to 1184 the length, in bytes, of that character. 1185 1186 The value of C<flags> determines the behavior when C<s> does not point to a 1187 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation 1188 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) 1189 is the next possible position in C<s> that could begin a non-malformed 1190 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning 1191 is raised. Some UTF-8 input sequences may contain multiple malformations. 1192 This function tries to find every possible one in each call, so multiple 1193 warnings can be raised for the same sequence. 1194 1195 Various ALLOW flags can be set in C<flags> to allow (and not warn on) 1196 individual types of malformations, such as the sequence being overlong (that 1197 is, when there is a shorter sequence that can express the same code point; 1198 overlong sequences are expressly forbidden in the UTF-8 standard due to 1199 potential security issues). Another malformation example is the first byte of 1200 a character not being a legal first byte. See F<utf8.h> for the list of such 1201 flags. Even if allowed, this function generally returns the Unicode 1202 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in 1203 F<utf8.h> to override this behavior for the overlong malformations, but don't 1204 do that except for very specialized purposes. 1205 1206 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other 1207 flags) malformation is found. If this flag is set, the routine assumes that 1208 the caller will raise a warning, and this function will silently just set 1209 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero. 1210 1211 Note that this API requires disambiguation between successful decoding a C<NUL> 1212 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as 1213 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may 1214 be set to 1. To disambiguate, upon a zero return, see if the first byte of 1215 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an 1216 error. Or you can use C<L</utf8n_to_uvchr_error>>. 1217 1218 Certain code points are considered problematic. These are Unicode surrogates, 1219 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF. 1220 By default these are considered regular code points, but certain situations 1221 warrant special handling for them, which can be specified using the C<flags> 1222 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all 1223 three classes are treated as malformations and handled as such. The flags 1224 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and 1225 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to 1226 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE> 1227 restricts the allowed inputs to the strict UTF-8 traditionally defined by 1228 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness 1229 definition given by 1230 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. 1231 The difference between traditional strictness and C9 strictness is that the 1232 latter does not forbid non-character code points. (They are still discouraged, 1233 however.) For more discussion see L<perlunicode/Noncharacter code points>. 1234 1235 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, 1236 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>, 1237 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be 1238 raised for their respective categories, but otherwise the code points are 1239 considered valid (not malformations). To get a category to both be treated as 1240 a malformation and raise a warning, specify both the WARN and DISALLOW flags. 1241 (But note that warnings are not raised if lexically disabled nor if 1242 C<UTF8_CHECK_ONLY> is also specified.) 1243 1244 Extremely high code points were never specified in any standard, and require an 1245 extension to UTF-8 to express, which Perl does. It is likely that programs 1246 written in something other than Perl would not be able to read files that 1247 contain these; nor would Perl understand files written by something that uses a 1248 different extension. For these reasons, there is a separate set of flags that 1249 can warn and/or disallow these extremely high code points, even if other 1250 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and 1251 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see 1252 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all 1253 above-Unicode code points, including these, as malformations. 1254 (Note that the Unicode standard considers anything above 0x10FFFF to be 1255 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF 1256 (2**31 -1)) 1257 1258 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is 1259 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly, 1260 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named 1261 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags 1262 can apply to code points that actually do fit in 31 bits. This happens on 1263 EBCDIC platforms, and sometimes when the L<overlong 1264 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately 1265 describe the situation in all cases. 1266 1267 1268 All other code points corresponding to Unicode characters, including private 1269 use and those yet to be assigned, are never considered malformed and never 1270 warn. 1271 1272 =cut 1273 1274 Also implemented as a macro in utf8.h 1275 */ 1276 1277 UV 1278 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, 1279 STRLEN curlen, 1280 STRLEN *retlen, 1281 const U32 flags) 1282 { 1283 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR; 1284 1285 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL); 1286 } 1287 1288 /* The tables below come from http://bjoern.hoehrmann.de/utf-8/decoder/dfa/, 1289 * which requires this copyright notice */ 1290 1291 /* Copyright (c) 2008-2009 Bjoern Hoehrmann <bjoern@hoehrmann.de> 1292 1293 Permission is hereby granted, free of charge, to any person obtaining a copy of 1294 this software and associated documentation files (the "Software"), to deal in 1295 the Software without restriction, including without limitation the rights to 1296 use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies 1297 of the Software, and to permit persons to whom the Software is furnished to do 1298 so, subject to the following conditions: 1299 1300 The above copyright notice and this permission notice shall be included in all 1301 copies or substantial portions of the Software. 1302 1303 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 1304 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 1305 FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 1306 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 1307 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 1308 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 1309 SOFTWARE. 1310 1311 */ 1312 1313 #if 0 1314 static U8 utf8d_C9[] = { 1315 /* The first part of the table maps bytes to character classes that 1316 * to reduce the size of the transition table and create bitmasks. */ 1317 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/ 1318 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/ 1319 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/ 1320 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/ 1321 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/ 1322 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/ 1323 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/ 1324 10,3,3,3,3,3,3,3,3,3,3,3,3,4,3,3, 11,6,6,6,5,8,8,8,8,8,8,8,8,8,8,8, /*-FF*/ 1325 1326 /* The second part is a transition table that maps a combination 1327 * of a state of the automaton and a character class to a state. */ 1328 0,12,24,36,60,96,84,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12, 1329 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12, 1330 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12, 1331 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12, 1332 12,36,12,12,12,12,12,12,12,12,12,12 1333 }; 1334 1335 #endif 1336 1337 #ifndef EBCDIC 1338 1339 /* This is a version of the above table customized for Perl that doesn't 1340 * exclude surrogates and accepts start bytes up through F7 (representing 1341 * 2**21 - 1). */ 1342 static U8 dfa_tab_for_perl[] = { 1343 /* The first part of the table maps bytes to character classes to reduce 1344 * the size of the transition table and create bitmasks. */ 1345 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/ 1346 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/ 1347 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/ 1348 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/ 1349 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/ 1350 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/ 1351 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/ 1352 10,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, 11,4,4,4,4,4,4,4,8,8,8,8,8,8,8,8, /*-FF*/ 1353 1354 /* The second part is a transition table that maps a combination 1355 * of a state of the automaton and a character class to a state. */ 1356 0,12,24,36,96,12,12,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,/*23*/ 1357 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,/*47*/ 1358 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,/*71*/ 1359 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,/*95*/ 1360 12,36,12,12,12,12,12,36,12,36,12,12 /* 96- 107 */ 1361 1362 /* The customization was to repurpose the surrogates type '4' to instead be 1363 * for start bytes F1-F7. Types 5 and 6 are now unused, and their entries in 1364 * the transition part of the table are set to 12, so are illegal. 1365 * 1366 * To do higher code points would require expansion and some rearrangement of 1367 * the table. The type '1' entries for continuation bytes 80-8f would have to 1368 * be split into several types, because they aren't treated uniformly for 1369 * higher start bytes, since overlongs for F8 are 80-87; FC: 80-83; and FE: 1370 * 80-81. We start needing to worry about overflow if FE is included. 1371 * Ignoring, FE and FF, we could use type 5 for F9-FB, and 6 for FD (remember 1372 * from the web site that these are used to right shift). FE would 1373 * necessarily be type 7; and FF, type 8. And new states would have to be 1374 * created for F8 and FC (and FE and FF if used), so quite a bit of work would 1375 * be involved. 1376 * 1377 * XXX Better would be to customize the table so that the noncharacters are 1378 * excluded. This again is non trivial, but doing so would simplify the code 1379 * that uses this, and might make it small enough to make it inlinable */ 1380 }; 1381 1382 #endif 1383 1384 /* 1385 1386 =for apidoc utf8n_to_uvchr_error 1387 1388 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 1389 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. 1390 1391 This function is for code that needs to know what the precise malformation(s) 1392 are when an error is found. If you also need to know the generated warning 1393 messages, use L</utf8n_to_uvchr_msgs>() instead. 1394 1395 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after 1396 all the others, C<errors>. If this parameter is 0, this function behaves 1397 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer 1398 to a C<U32> variable, which this function sets to indicate any errors found. 1399 Upon return, if C<*errors> is 0, there were no errors found. Otherwise, 1400 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some 1401 of these bits will be set if a malformation is found, even if the input 1402 C<flags> parameter indicates that the given malformation is allowed; those 1403 exceptions are noted: 1404 1405 =over 4 1406 1407 =item C<UTF8_GOT_PERL_EXTENDED> 1408 1409 The input sequence is not standard UTF-8, but a Perl extension. This bit is 1410 set only if the input C<flags> parameter contains either the 1411 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags. 1412 1413 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, 1414 and so some extension must be used to express them. Perl uses a natural 1415 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further 1416 extension to represent even higher ones, so that any code point that fits in a 1417 64-bit word can be represented. Text using these extensions is not likely to 1418 be portable to non-Perl code. We lump both of these extensions together and 1419 refer to them as Perl extended UTF-8. There exist other extensions that people 1420 have invented, incompatible with Perl's. 1421 1422 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing 1423 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower 1424 than on ASCII. Prior to that, code points 2**31 and higher were simply 1425 unrepresentable, and a different, incompatible method was used to represent 1426 code points between 2**30 and 2**31 - 1. 1427 1428 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if 1429 Perl extended UTF-8 is used. 1430 1431 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still 1432 may use for backward compatibility. That name is misleading, as this flag may 1433 be set when the code point actually does fit in 31 bits. This happens on 1434 EBCDIC platforms, and sometimes when the L<overlong 1435 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately 1436 describes the situation in all cases. 1437 1438 =item C<UTF8_GOT_CONTINUATION> 1439 1440 The input sequence was malformed in that the first byte was a a UTF-8 1441 continuation byte. 1442 1443 =item C<UTF8_GOT_EMPTY> 1444 1445 The input C<curlen> parameter was 0. 1446 1447 =item C<UTF8_GOT_LONG> 1448 1449 The input sequence was malformed in that there is some other sequence that 1450 evaluates to the same code point, but that sequence is shorter than this one. 1451 1452 Until Unicode 3.1, it was legal for programs to accept this malformation, but 1453 it was discovered that this created security issues. 1454 1455 =item C<UTF8_GOT_NONCHAR> 1456 1457 The code point represented by the input UTF-8 sequence is for a Unicode 1458 non-character code point. 1459 This bit is set only if the input C<flags> parameter contains either the 1460 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags. 1461 1462 =item C<UTF8_GOT_NON_CONTINUATION> 1463 1464 The input sequence was malformed in that a non-continuation type byte was found 1465 in a position where only a continuation type one should be. 1466 1467 =item C<UTF8_GOT_OVERFLOW> 1468 1469 The input sequence was malformed in that it is for a code point that is not 1470 representable in the number of bits available in an IV on the current platform. 1471 1472 =item C<UTF8_GOT_SHORT> 1473 1474 The input sequence was malformed in that C<curlen> is smaller than required for 1475 a complete sequence. In other words, the input is for a partial character 1476 sequence. 1477 1478 =item C<UTF8_GOT_SUPER> 1479 1480 The input sequence was malformed in that it is for a non-Unicode code point; 1481 that is, one above the legal Unicode maximum. 1482 This bit is set only if the input C<flags> parameter contains either the 1483 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags. 1484 1485 =item C<UTF8_GOT_SURROGATE> 1486 1487 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate 1488 code point. 1489 This bit is set only if the input C<flags> parameter contains either the 1490 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags. 1491 1492 =back 1493 1494 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY> 1495 flag to suppress any warnings, and then examine the C<*errors> return. 1496 1497 =cut 1498 1499 Also implemented as a macro in utf8.h 1500 */ 1501 1502 UV 1503 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s, 1504 STRLEN curlen, 1505 STRLEN *retlen, 1506 const U32 flags, 1507 U32 * errors) 1508 { 1509 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR; 1510 1511 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL); 1512 } 1513 1514 /* 1515 1516 =for apidoc utf8n_to_uvchr_msgs 1517 1518 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 1519 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. 1520 1521 This function is for code that needs to know what the precise malformation(s) 1522 are when an error is found, and wants the corresponding warning and/or error 1523 messages to be returned to the caller rather than be displayed. All messages 1524 that would have been displayed if all lexcial warnings are enabled will be 1525 returned. 1526 1527 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter 1528 placed after all the others, C<msgs>. If this parameter is 0, this function 1529 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should 1530 be a pointer to an C<AV *> variable, in which this function creates a new AV to 1531 contain any appropriate messages. The elements of the array are ordered so 1532 that the first message that would have been displayed is in the 0th element, 1533 and so on. Each element is a hash with three key-value pairs, as follows: 1534 1535 =over 4 1536 1537 =item C<text> 1538 1539 The text of the message as a C<SVpv>. 1540 1541 =item C<warn_categories> 1542 1543 The warning category (or categories) packed into a C<SVuv>. 1544 1545 =item C<flag> 1546 1547 A single flag bit associated with this message, in a C<SVuv>. 1548 The bit corresponds to some bit in the C<*errors> return value, 1549 such as C<UTF8_GOT_LONG>. 1550 1551 =back 1552 1553 It's important to note that specifying this parameter as non-null will cause 1554 any warnings this function would otherwise generate to be suppressed, and 1555 instead be placed in C<*msgs>. The caller can check the lexical warnings state 1556 (or not) when choosing what to do with the returned messages. 1557 1558 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence 1559 no AV is created. 1560 1561 The caller, of course, is responsible for freeing any returned AV. 1562 1563 =cut 1564 */ 1565 1566 UV 1567 Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s, 1568 STRLEN curlen, 1569 STRLEN *retlen, 1570 const U32 flags, 1571 U32 * errors, 1572 AV ** msgs) 1573 { 1574 const U8 * const s0 = s; 1575 const U8 * send = s0 + curlen; 1576 U32 possible_problems = 0; /* A bit is set here for each potential problem 1577 found as we go along */ 1578 UV uv = (UV) -1; 1579 STRLEN expectlen = 0; /* How long should this sequence be? 1580 (initialized to silence compilers' wrong 1581 warning) */ 1582 STRLEN avail_len = 0; /* When input is too short, gives what that is */ 1583 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL; 1584 this gets set and discarded */ 1585 1586 /* The below are used only if there is both an overlong malformation and a 1587 * too short one. Otherwise the first two are set to 's0' and 'send', and 1588 * the third not used at all */ 1589 U8 * adjusted_s0 = (U8 *) s0; 1590 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this 1591 routine; see [perl #130921] */ 1592 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */ 1593 1594 UV state = 0; 1595 1596 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS; 1597 1598 if (errors) { 1599 *errors = 0; 1600 } 1601 else { 1602 errors = &discard_errors; 1603 } 1604 1605 /* The order of malformation tests here is important. We should consume as 1606 * few bytes as possible in order to not skip any valid character. This is 1607 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also 1608 * http://unicode.org/reports/tr36 for more discussion as to why. For 1609 * example, once we've done a UTF8SKIP, we can tell the expected number of 1610 * bytes, and could fail right off the bat if the input parameters indicate 1611 * that there are too few available. But it could be that just that first 1612 * byte is garbled, and the intended character occupies fewer bytes. If we 1613 * blindly assumed that the first byte is correct, and skipped based on 1614 * that number, we could skip over a valid input character. So instead, we 1615 * always examine the sequence byte-by-byte. 1616 * 1617 * We also should not consume too few bytes, otherwise someone could inject 1618 * things. For example, an input could be deliberately designed to 1619 * overflow, and if this code bailed out immediately upon discovering that, 1620 * returning to the caller C<*retlen> pointing to the very next byte (one 1621 * which is actually part of of the overflowing sequence), that could look 1622 * legitimate to the caller, which could discard the initial partial 1623 * sequence and process the rest, inappropriately. 1624 * 1625 * Some possible input sequences are malformed in more than one way. This 1626 * function goes to lengths to try to find all of them. This is necessary 1627 * for correctness, as the inputs may allow one malformation but not 1628 * another, and if we abandon searching for others after finding the 1629 * allowed one, we could allow in something that shouldn't have been. 1630 */ 1631 1632 if (UNLIKELY(curlen == 0)) { 1633 possible_problems |= UTF8_GOT_EMPTY; 1634 curlen = 0; 1635 uv = UNICODE_REPLACEMENT; 1636 goto ready_to_handle_errors; 1637 } 1638 1639 expectlen = UTF8SKIP(s); 1640 1641 /* A well-formed UTF-8 character, as the vast majority of calls to this 1642 * function will be for, has this expected length. For efficiency, set 1643 * things up here to return it. It will be overriden only in those rare 1644 * cases where a malformation is found */ 1645 if (retlen) { 1646 *retlen = expectlen; 1647 } 1648 1649 /* An invariant is trivially well-formed */ 1650 if (UTF8_IS_INVARIANT(*s0)) { 1651 return *s0; 1652 } 1653 1654 #ifndef EBCDIC 1655 1656 /* Measurements show that this dfa is somewhat faster than the regular code 1657 * below, so use it first, dropping down for the non-normal cases. */ 1658 1659 # define PERL_UTF8_DECODE_REJECT 12 1660 1661 while (s < send && LIKELY(state != PERL_UTF8_DECODE_REJECT)) { 1662 UV type = dfa_tab_for_perl[*s]; 1663 1664 if (state != 0) { 1665 uv = (*s & 0x3fu) | (uv << UTF_ACCUMULATION_SHIFT); 1666 state = dfa_tab_for_perl[256 + state + type]; 1667 } 1668 else { 1669 uv = (0xff >> type) & (*s); 1670 state = dfa_tab_for_perl[256 + type]; 1671 } 1672 1673 if (state == 0) { 1674 1675 /* If this could be a code point that the flags don't allow (the first 1676 * surrogate is the first such possible one), delve further, but we already 1677 * have calculated 'uv' */ 1678 if ( (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE 1679 |UTF8_WARN_ILLEGAL_INTERCHANGE)) 1680 && uv >= UNICODE_SURROGATE_FIRST) 1681 { 1682 curlen = s + 1 - s0; 1683 goto got_uv; 1684 } 1685 1686 return uv; 1687 } 1688 1689 s++; 1690 } 1691 1692 /* Here, is some sort of failure. Use the full mechanism */ 1693 1694 uv = *s0; 1695 1696 #endif 1697 1698 /* A continuation character can't start a valid sequence */ 1699 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) { 1700 possible_problems |= UTF8_GOT_CONTINUATION; 1701 curlen = 1; 1702 uv = UNICODE_REPLACEMENT; 1703 goto ready_to_handle_errors; 1704 } 1705 1706 /* Here is not a continuation byte, nor an invariant. The only thing left 1707 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START 1708 * because it excludes start bytes like \xC0 that always lead to 1709 * overlongs.) */ 1710 1711 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits 1712 * that indicate the number of bytes in the character's whole UTF-8 1713 * sequence, leaving just the bits that are part of the value. */ 1714 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen); 1715 1716 /* Setup the loop end point, making sure to not look past the end of the 1717 * input string, and flag it as too short if the size isn't big enough. */ 1718 if (UNLIKELY(curlen < expectlen)) { 1719 possible_problems |= UTF8_GOT_SHORT; 1720 avail_len = curlen; 1721 } 1722 else { 1723 send = (U8*) s0 + expectlen; 1724 } 1725 1726 /* Now, loop through the remaining bytes in the character's sequence, 1727 * accumulating each into the working value as we go. */ 1728 for (s = s0 + 1; s < send; s++) { 1729 if (LIKELY(UTF8_IS_CONTINUATION(*s))) { 1730 uv = UTF8_ACCUMULATE(uv, *s); 1731 continue; 1732 } 1733 1734 /* Here, found a non-continuation before processing all expected bytes. 1735 * This byte indicates the beginning of a new character, so quit, even 1736 * if allowing this malformation. */ 1737 possible_problems |= UTF8_GOT_NON_CONTINUATION; 1738 break; 1739 } /* End of loop through the character's bytes */ 1740 1741 /* Save how many bytes were actually in the character */ 1742 curlen = s - s0; 1743 1744 /* Note that there are two types of too-short malformation. One is when 1745 * there is actual wrong data before the normal termination of the 1746 * sequence. The other is that the sequence wasn't complete before the end 1747 * of the data we are allowed to look at, based on the input 'curlen'. 1748 * This means that we were passed data for a partial character, but it is 1749 * valid as far as we saw. The other is definitely invalid. This 1750 * distinction could be important to a caller, so the two types are kept 1751 * separate. 1752 * 1753 * A convenience macro that matches either of the too-short conditions. */ 1754 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION) 1755 1756 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) { 1757 uv_so_far = uv; 1758 uv = UNICODE_REPLACEMENT; 1759 } 1760 1761 /* Check for overflow. The algorithm requires us to not look past the end 1762 * of the current character, even if partial, so the upper limit is 's' */ 1763 if (UNLIKELY(0 < does_utf8_overflow(s0, s, 1764 1 /* Do consider overlongs */ 1765 ))) 1766 { 1767 possible_problems |= UTF8_GOT_OVERFLOW; 1768 uv = UNICODE_REPLACEMENT; 1769 } 1770 1771 /* Check for overlong. If no problems so far, 'uv' is the correct code 1772 * point value. Simply see if it is expressible in fewer bytes. Otherwise 1773 * we must look at the UTF-8 byte sequence itself to see if it is for an 1774 * overlong */ 1775 if ( ( LIKELY(! possible_problems) 1776 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv))) 1777 || ( UNLIKELY(possible_problems) 1778 && ( UNLIKELY(! UTF8_IS_START(*s0)) 1779 || ( curlen > 1 1780 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0, 1781 s - s0)))))) 1782 { 1783 possible_problems |= UTF8_GOT_LONG; 1784 1785 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT) 1786 1787 /* The calculation in the 'true' branch of this 'if' 1788 * below won't work if overflows, and isn't needed 1789 * anyway. Further below we handle all overflow 1790 * cases */ 1791 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))) 1792 { 1793 UV min_uv = uv_so_far; 1794 STRLEN i; 1795 1796 /* Here, the input is both overlong and is missing some trailing 1797 * bytes. There is no single code point it could be for, but there 1798 * may be enough information present to determine if what we have 1799 * so far is for an unallowed code point, such as for a surrogate. 1800 * The code further below has the intelligence to determine this, 1801 * but just for non-overlong UTF-8 sequences. What we do here is 1802 * calculate the smallest code point the input could represent if 1803 * there were no too short malformation. Then we compute and save 1804 * the UTF-8 for that, which is what the code below looks at 1805 * instead of the raw input. It turns out that the smallest such 1806 * code point is all we need. */ 1807 for (i = curlen; i < expectlen; i++) { 1808 min_uv = UTF8_ACCUMULATE(min_uv, 1809 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK)); 1810 } 1811 1812 adjusted_s0 = temp_char_buf; 1813 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0); 1814 } 1815 } 1816 1817 got_uv: 1818 1819 /* Here, we have found all the possible problems, except for when the input 1820 * is for a problematic code point not allowed by the input parameters. */ 1821 1822 /* uv is valid for overlongs */ 1823 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG)) 1824 1825 /* isn't problematic if < this */ 1826 && uv >= UNICODE_SURROGATE_FIRST) 1827 || ( UNLIKELY(possible_problems) 1828 1829 /* if overflow, we know without looking further 1830 * precisely which of the problematic types it is, 1831 * and we deal with those in the overflow handling 1832 * code */ 1833 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)) 1834 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0) 1835 || UNLIKELY(isUTF8_PERL_EXTENDED(s0))))) 1836 && ((flags & ( UTF8_DISALLOW_NONCHAR 1837 |UTF8_DISALLOW_SURROGATE 1838 |UTF8_DISALLOW_SUPER 1839 |UTF8_DISALLOW_PERL_EXTENDED 1840 |UTF8_WARN_NONCHAR 1841 |UTF8_WARN_SURROGATE 1842 |UTF8_WARN_SUPER 1843 |UTF8_WARN_PERL_EXTENDED)))) 1844 { 1845 /* If there were no malformations, or the only malformation is an 1846 * overlong, 'uv' is valid */ 1847 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) { 1848 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 1849 possible_problems |= UTF8_GOT_SURROGATE; 1850 } 1851 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) { 1852 possible_problems |= UTF8_GOT_SUPER; 1853 } 1854 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) { 1855 possible_problems |= UTF8_GOT_NONCHAR; 1856 } 1857 } 1858 else { /* Otherwise, need to look at the source UTF-8, possibly 1859 adjusted to be non-overlong */ 1860 1861 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0) 1862 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) 1863 { 1864 possible_problems |= UTF8_GOT_SUPER; 1865 } 1866 else if (curlen > 1) { 1867 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER( 1868 NATIVE_UTF8_TO_I8(*adjusted_s0), 1869 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) 1870 { 1871 possible_problems |= UTF8_GOT_SUPER; 1872 } 1873 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE( 1874 NATIVE_UTF8_TO_I8(*adjusted_s0), 1875 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) 1876 { 1877 possible_problems |= UTF8_GOT_SURROGATE; 1878 } 1879 } 1880 1881 /* We need a complete well-formed UTF-8 character to discern 1882 * non-characters, so can't look for them here */ 1883 } 1884 } 1885 1886 ready_to_handle_errors: 1887 1888 /* At this point: 1889 * curlen contains the number of bytes in the sequence that 1890 * this call should advance the input by. 1891 * avail_len gives the available number of bytes passed in, but 1892 * only if this is less than the expected number of 1893 * bytes, based on the code point's start byte. 1894 * possible_problems' is 0 if there weren't any problems; otherwise a bit 1895 * is set in it for each potential problem found. 1896 * uv contains the code point the input sequence 1897 * represents; or if there is a problem that prevents 1898 * a well-defined value from being computed, it is 1899 * some subsitute value, typically the REPLACEMENT 1900 * CHARACTER. 1901 * s0 points to the first byte of the character 1902 * s points to just after were we left off processing 1903 * the character 1904 * send points to just after where that character should 1905 * end, based on how many bytes the start byte tells 1906 * us should be in it, but no further than s0 + 1907 * avail_len 1908 */ 1909 1910 if (UNLIKELY(possible_problems)) { 1911 bool disallowed = FALSE; 1912 const U32 orig_problems = possible_problems; 1913 1914 if (msgs) { 1915 *msgs = NULL; 1916 } 1917 1918 while (possible_problems) { /* Handle each possible problem */ 1919 UV pack_warn = 0; 1920 char * message = NULL; 1921 U32 this_flag_bit = 0; 1922 1923 /* Each 'if' clause handles one problem. They are ordered so that 1924 * the first ones' messages will be displayed before the later 1925 * ones; this is kinda in decreasing severity order. But the 1926 * overlong must come last, as it changes 'uv' looked at by the 1927 * others */ 1928 if (possible_problems & UTF8_GOT_OVERFLOW) { 1929 1930 /* Overflow means also got a super and are using Perl's 1931 * extended UTF-8, but we handle all three cases here */ 1932 possible_problems 1933 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED); 1934 *errors |= UTF8_GOT_OVERFLOW; 1935 1936 /* But the API says we flag all errors found */ 1937 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) { 1938 *errors |= UTF8_GOT_SUPER; 1939 } 1940 if (flags 1941 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED)) 1942 { 1943 *errors |= UTF8_GOT_PERL_EXTENDED; 1944 } 1945 1946 /* Disallow if any of the three categories say to */ 1947 if ( ! (flags & UTF8_ALLOW_OVERFLOW) 1948 || (flags & ( UTF8_DISALLOW_SUPER 1949 |UTF8_DISALLOW_PERL_EXTENDED))) 1950 { 1951 disallowed = TRUE; 1952 } 1953 1954 /* Likewise, warn if any say to */ 1955 if ( ! (flags & UTF8_ALLOW_OVERFLOW) 1956 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED))) 1957 { 1958 1959 /* The warnings code explicitly says it doesn't handle the 1960 * case of packWARN2 and two categories which have 1961 * parent-child relationship. Even if it works now to 1962 * raise the warning if either is enabled, it wouldn't 1963 * necessarily do so in the future. We output (only) the 1964 * most dire warning */ 1965 if (! (flags & UTF8_CHECK_ONLY)) { 1966 if (msgs || ckWARN_d(WARN_UTF8)) { 1967 pack_warn = packWARN(WARN_UTF8); 1968 } 1969 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) { 1970 pack_warn = packWARN(WARN_NON_UNICODE); 1971 } 1972 if (pack_warn) { 1973 message = Perl_form(aTHX_ "%s: %s (overflows)", 1974 malformed_text, 1975 _byte_dump_string(s0, curlen, 0)); 1976 this_flag_bit = UTF8_GOT_OVERFLOW; 1977 } 1978 } 1979 } 1980 } 1981 else if (possible_problems & UTF8_GOT_EMPTY) { 1982 possible_problems &= ~UTF8_GOT_EMPTY; 1983 *errors |= UTF8_GOT_EMPTY; 1984 1985 if (! (flags & UTF8_ALLOW_EMPTY)) { 1986 1987 /* This so-called malformation is now treated as a bug in 1988 * the caller. If you have nothing to decode, skip calling 1989 * this function */ 1990 assert(0); 1991 1992 disallowed = TRUE; 1993 if ( (msgs 1994 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 1995 { 1996 pack_warn = packWARN(WARN_UTF8); 1997 message = Perl_form(aTHX_ "%s (empty string)", 1998 malformed_text); 1999 this_flag_bit = UTF8_GOT_EMPTY; 2000 } 2001 } 2002 } 2003 else if (possible_problems & UTF8_GOT_CONTINUATION) { 2004 possible_problems &= ~UTF8_GOT_CONTINUATION; 2005 *errors |= UTF8_GOT_CONTINUATION; 2006 2007 if (! (flags & UTF8_ALLOW_CONTINUATION)) { 2008 disallowed = TRUE; 2009 if (( msgs 2010 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 2011 { 2012 pack_warn = packWARN(WARN_UTF8); 2013 message = Perl_form(aTHX_ 2014 "%s: %s (unexpected continuation byte 0x%02x," 2015 " with no preceding start byte)", 2016 malformed_text, 2017 _byte_dump_string(s0, 1, 0), *s0); 2018 this_flag_bit = UTF8_GOT_CONTINUATION; 2019 } 2020 } 2021 } 2022 else if (possible_problems & UTF8_GOT_SHORT) { 2023 possible_problems &= ~UTF8_GOT_SHORT; 2024 *errors |= UTF8_GOT_SHORT; 2025 2026 if (! (flags & UTF8_ALLOW_SHORT)) { 2027 disallowed = TRUE; 2028 if (( msgs 2029 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 2030 { 2031 pack_warn = packWARN(WARN_UTF8); 2032 message = Perl_form(aTHX_ 2033 "%s: %s (too short; %d byte%s available, need %d)", 2034 malformed_text, 2035 _byte_dump_string(s0, send - s0, 0), 2036 (int)avail_len, 2037 avail_len == 1 ? "" : "s", 2038 (int)expectlen); 2039 this_flag_bit = UTF8_GOT_SHORT; 2040 } 2041 } 2042 2043 } 2044 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) { 2045 possible_problems &= ~UTF8_GOT_NON_CONTINUATION; 2046 *errors |= UTF8_GOT_NON_CONTINUATION; 2047 2048 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) { 2049 disallowed = TRUE; 2050 if (( msgs 2051 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 2052 { 2053 2054 /* If we don't know for sure that the input length is 2055 * valid, avoid as much as possible reading past the 2056 * end of the buffer */ 2057 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN) 2058 ? s - s0 2059 : send - s0; 2060 pack_warn = packWARN(WARN_UTF8); 2061 message = Perl_form(aTHX_ "%s", 2062 unexpected_non_continuation_text(s0, 2063 printlen, 2064 s - s0, 2065 (int) expectlen)); 2066 this_flag_bit = UTF8_GOT_NON_CONTINUATION; 2067 } 2068 } 2069 } 2070 else if (possible_problems & UTF8_GOT_SURROGATE) { 2071 possible_problems &= ~UTF8_GOT_SURROGATE; 2072 2073 if (flags & UTF8_WARN_SURROGATE) { 2074 *errors |= UTF8_GOT_SURROGATE; 2075 2076 if ( ! (flags & UTF8_CHECK_ONLY) 2077 && (msgs || ckWARN_d(WARN_SURROGATE))) 2078 { 2079 pack_warn = packWARN(WARN_SURROGATE); 2080 2081 /* These are the only errors that can occur with a 2082 * surrogate when the 'uv' isn't valid */ 2083 if (orig_problems & UTF8_GOT_TOO_SHORT) { 2084 message = Perl_form(aTHX_ 2085 "UTF-16 surrogate (any UTF-8 sequence that" 2086 " starts with \"%s\" is for a surrogate)", 2087 _byte_dump_string(s0, curlen, 0)); 2088 } 2089 else { 2090 message = Perl_form(aTHX_ surrogate_cp_format, uv); 2091 } 2092 this_flag_bit = UTF8_GOT_SURROGATE; 2093 } 2094 } 2095 2096 if (flags & UTF8_DISALLOW_SURROGATE) { 2097 disallowed = TRUE; 2098 *errors |= UTF8_GOT_SURROGATE; 2099 } 2100 } 2101 else if (possible_problems & UTF8_GOT_SUPER) { 2102 possible_problems &= ~UTF8_GOT_SUPER; 2103 2104 if (flags & UTF8_WARN_SUPER) { 2105 *errors |= UTF8_GOT_SUPER; 2106 2107 if ( ! (flags & UTF8_CHECK_ONLY) 2108 && (msgs || ckWARN_d(WARN_NON_UNICODE))) 2109 { 2110 pack_warn = packWARN(WARN_NON_UNICODE); 2111 2112 if (orig_problems & UTF8_GOT_TOO_SHORT) { 2113 message = Perl_form(aTHX_ 2114 "Any UTF-8 sequence that starts with" 2115 " \"%s\" is for a non-Unicode code point," 2116 " may not be portable", 2117 _byte_dump_string(s0, curlen, 0)); 2118 } 2119 else { 2120 message = Perl_form(aTHX_ super_cp_format, uv); 2121 } 2122 this_flag_bit = UTF8_GOT_SUPER; 2123 } 2124 } 2125 2126 /* Test for Perl's extended UTF-8 after the regular SUPER ones, 2127 * and before possibly bailing out, so that the more dire 2128 * warning will override the regular one. */ 2129 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) { 2130 if ( ! (flags & UTF8_CHECK_ONLY) 2131 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER)) 2132 && (msgs || ckWARN_d(WARN_NON_UNICODE))) 2133 { 2134 pack_warn = packWARN(WARN_NON_UNICODE); 2135 2136 /* If it is an overlong that evaluates to a code point 2137 * that doesn't have to use the Perl extended UTF-8, it 2138 * still used it, and so we output a message that 2139 * doesn't refer to the code point. The same is true 2140 * if there was a SHORT malformation where the code 2141 * point is not valid. In that case, 'uv' will have 2142 * been set to the REPLACEMENT CHAR, and the message 2143 * below without the code point in it will be selected 2144 * */ 2145 if (UNICODE_IS_PERL_EXTENDED(uv)) { 2146 message = Perl_form(aTHX_ 2147 perl_extended_cp_format, uv); 2148 } 2149 else { 2150 message = Perl_form(aTHX_ 2151 "Any UTF-8 sequence that starts with" 2152 " \"%s\" is a Perl extension, and" 2153 " so is not portable", 2154 _byte_dump_string(s0, curlen, 0)); 2155 } 2156 this_flag_bit = UTF8_GOT_PERL_EXTENDED; 2157 } 2158 2159 if (flags & ( UTF8_WARN_PERL_EXTENDED 2160 |UTF8_DISALLOW_PERL_EXTENDED)) 2161 { 2162 *errors |= UTF8_GOT_PERL_EXTENDED; 2163 2164 if (flags & UTF8_DISALLOW_PERL_EXTENDED) { 2165 disallowed = TRUE; 2166 } 2167 } 2168 } 2169 2170 if (flags & UTF8_DISALLOW_SUPER) { 2171 *errors |= UTF8_GOT_SUPER; 2172 disallowed = TRUE; 2173 } 2174 } 2175 else if (possible_problems & UTF8_GOT_NONCHAR) { 2176 possible_problems &= ~UTF8_GOT_NONCHAR; 2177 2178 if (flags & UTF8_WARN_NONCHAR) { 2179 *errors |= UTF8_GOT_NONCHAR; 2180 2181 if ( ! (flags & UTF8_CHECK_ONLY) 2182 && (msgs || ckWARN_d(WARN_NONCHAR))) 2183 { 2184 /* The code above should have guaranteed that we don't 2185 * get here with errors other than overlong */ 2186 assert (! (orig_problems 2187 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR))); 2188 2189 pack_warn = packWARN(WARN_NONCHAR); 2190 message = Perl_form(aTHX_ nonchar_cp_format, uv); 2191 this_flag_bit = UTF8_GOT_NONCHAR; 2192 } 2193 } 2194 2195 if (flags & UTF8_DISALLOW_NONCHAR) { 2196 disallowed = TRUE; 2197 *errors |= UTF8_GOT_NONCHAR; 2198 } 2199 } 2200 else if (possible_problems & UTF8_GOT_LONG) { 2201 possible_problems &= ~UTF8_GOT_LONG; 2202 *errors |= UTF8_GOT_LONG; 2203 2204 if (flags & UTF8_ALLOW_LONG) { 2205 2206 /* We don't allow the actual overlong value, unless the 2207 * special extra bit is also set */ 2208 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE 2209 & ~UTF8_ALLOW_LONG))) 2210 { 2211 uv = UNICODE_REPLACEMENT; 2212 } 2213 } 2214 else { 2215 disallowed = TRUE; 2216 2217 if (( msgs 2218 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY)) 2219 { 2220 pack_warn = packWARN(WARN_UTF8); 2221 2222 /* These error types cause 'uv' to be something that 2223 * isn't what was intended, so can't use it in the 2224 * message. The other error types either can't 2225 * generate an overlong, or else the 'uv' is valid */ 2226 if (orig_problems & 2227 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW)) 2228 { 2229 message = Perl_form(aTHX_ 2230 "%s: %s (any UTF-8 sequence that starts" 2231 " with \"%s\" is overlong which can and" 2232 " should be represented with a" 2233 " different, shorter sequence)", 2234 malformed_text, 2235 _byte_dump_string(s0, send - s0, 0), 2236 _byte_dump_string(s0, curlen, 0)); 2237 } 2238 else { 2239 U8 tmpbuf[UTF8_MAXBYTES+1]; 2240 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf, 2241 uv, 0); 2242 /* Don't use U+ for non-Unicode code points, which 2243 * includes those in the Latin1 range */ 2244 const char * preface = ( uv > PERL_UNICODE_MAX 2245 #ifdef EBCDIC 2246 || uv <= 0xFF 2247 #endif 2248 ) 2249 ? "0x" 2250 : "U+"; 2251 message = Perl_form(aTHX_ 2252 "%s: %s (overlong; instead use %s to represent" 2253 " %s%0*" UVXf ")", 2254 malformed_text, 2255 _byte_dump_string(s0, send - s0, 0), 2256 _byte_dump_string(tmpbuf, e - tmpbuf, 0), 2257 preface, 2258 ((uv < 256) ? 2 : 4), /* Field width of 2 for 2259 small code points */ 2260 UNI_TO_NATIVE(uv)); 2261 } 2262 this_flag_bit = UTF8_GOT_LONG; 2263 } 2264 } 2265 } /* End of looking through the possible flags */ 2266 2267 /* Display the message (if any) for the problem being handled in 2268 * this iteration of the loop */ 2269 if (message) { 2270 if (msgs) { 2271 assert(this_flag_bit); 2272 2273 if (*msgs == NULL) { 2274 *msgs = newAV(); 2275 } 2276 2277 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message, 2278 pack_warn, 2279 this_flag_bit))); 2280 } 2281 else if (PL_op) 2282 Perl_warner(aTHX_ pack_warn, "%s in %s", message, 2283 OP_DESC(PL_op)); 2284 else 2285 Perl_warner(aTHX_ pack_warn, "%s", message); 2286 } 2287 } /* End of 'while (possible_problems)' */ 2288 2289 /* Since there was a possible problem, the returned length may need to 2290 * be changed from the one stored at the beginning of this function. 2291 * Instead of trying to figure out if that's needed, just do it. */ 2292 if (retlen) { 2293 *retlen = curlen; 2294 } 2295 2296 if (disallowed) { 2297 if (flags & UTF8_CHECK_ONLY && retlen) { 2298 *retlen = ((STRLEN) -1); 2299 } 2300 return 0; 2301 } 2302 } 2303 2304 return UNI_TO_NATIVE(uv); 2305 } 2306 2307 /* 2308 =for apidoc utf8_to_uvchr_buf 2309 2310 Returns the native code point of the first character in the string C<s> which 2311 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. 2312 C<*retlen> will be set to the length, in bytes, of that character. 2313 2314 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are 2315 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 2316 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined 2317 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and 2318 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is 2319 the next possible position in C<s> that could begin a non-malformed character. 2320 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is 2321 returned. 2322 2323 =cut 2324 2325 Also implemented as a macro in utf8.h 2326 2327 */ 2328 2329 2330 UV 2331 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) 2332 { 2333 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF; 2334 2335 assert(s < send); 2336 2337 return utf8n_to_uvchr(s, send - s, retlen, 2338 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY); 2339 } 2340 2341 /* This is marked as deprecated 2342 * 2343 =for apidoc utf8_to_uvuni_buf 2344 2345 Only in very rare circumstances should code need to be dealing in Unicode 2346 (as opposed to native) code points. In those few cases, use 2347 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you 2348 are not absolutely sure this is one of those cases, then assume it isn't and 2349 use plain C<utf8_to_uvchr_buf> instead. 2350 2351 Returns the Unicode (not-native) code point of the first character in the 2352 string C<s> which 2353 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. 2354 C<retlen> will be set to the length, in bytes, of that character. 2355 2356 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are 2357 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 2358 NULL) to -1. If those warnings are off, the computed value if well-defined (or 2359 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> 2360 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the 2361 next possible position in C<s> that could begin a non-malformed character. 2362 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. 2363 2364 =cut 2365 */ 2366 2367 UV 2368 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) 2369 { 2370 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF; 2371 2372 assert(send > s); 2373 2374 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen)); 2375 } 2376 2377 /* 2378 =for apidoc utf8_length 2379 2380 Returns the number of characters in the sequence of UTF-8-encoded bytes starting 2381 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the 2382 same place, it returns 0 with no warning raised. 2383 2384 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning 2385 and returns the number of valid characters. 2386 2387 =cut 2388 */ 2389 2390 STRLEN 2391 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e) 2392 { 2393 STRLEN len = 0; 2394 2395 PERL_ARGS_ASSERT_UTF8_LENGTH; 2396 2397 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g. 2398 * the bitops (especially ~) can create illegal UTF-8. 2399 * In other words: in Perl UTF-8 is not just for Unicode. */ 2400 2401 if (e < s) 2402 goto warn_and_return; 2403 while (s < e) { 2404 s += UTF8SKIP(s); 2405 len++; 2406 } 2407 2408 if (e != s) { 2409 len--; 2410 warn_and_return: 2411 if (PL_op) 2412 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 2413 "%s in %s", unees, OP_DESC(PL_op)); 2414 else 2415 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); 2416 } 2417 2418 return len; 2419 } 2420 2421 /* 2422 =for apidoc bytes_cmp_utf8 2423 2424 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the 2425 sequence of characters (stored as UTF-8) 2426 in C<u>, C<ulen>. Returns 0 if they are 2427 equal, -1 or -2 if the first string is less than the second string, +1 or +2 2428 if the first string is greater than the second string. 2429 2430 -1 or +1 is returned if the shorter string was identical to the start of the 2431 longer string. -2 or +2 is returned if 2432 there was a difference between characters 2433 within the strings. 2434 2435 =cut 2436 */ 2437 2438 int 2439 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen) 2440 { 2441 const U8 *const bend = b + blen; 2442 const U8 *const uend = u + ulen; 2443 2444 PERL_ARGS_ASSERT_BYTES_CMP_UTF8; 2445 2446 while (b < bend && u < uend) { 2447 U8 c = *u++; 2448 if (!UTF8_IS_INVARIANT(c)) { 2449 if (UTF8_IS_DOWNGRADEABLE_START(c)) { 2450 if (u < uend) { 2451 U8 c1 = *u++; 2452 if (UTF8_IS_CONTINUATION(c1)) { 2453 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1); 2454 } else { 2455 /* diag_listed_as: Malformed UTF-8 character%s */ 2456 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 2457 "%s %s%s", 2458 unexpected_non_continuation_text(u - 2, 2, 1, 2), 2459 PL_op ? " in " : "", 2460 PL_op ? OP_DESC(PL_op) : ""); 2461 return -2; 2462 } 2463 } else { 2464 if (PL_op) 2465 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 2466 "%s in %s", unees, OP_DESC(PL_op)); 2467 else 2468 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); 2469 return -2; /* Really want to return undef :-) */ 2470 } 2471 } else { 2472 return -2; 2473 } 2474 } 2475 if (*b != c) { 2476 return *b < c ? -2 : +2; 2477 } 2478 ++b; 2479 } 2480 2481 if (b == bend && u == uend) 2482 return 0; 2483 2484 return b < bend ? +1 : -1; 2485 } 2486 2487 /* 2488 =for apidoc utf8_to_bytes 2489 2490 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding. 2491 Unlike L</bytes_to_utf8>, this over-writes the original string, and 2492 updates C<*lenp> to contain the new length. 2493 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1. 2494 2495 Upon successful return, the number of variants in the string can be computed by 2496 having saved the value of C<*lenp> before the call, and subtracting the 2497 after-call value of C<*lenp> from it. 2498 2499 If you need a copy of the string, see L</bytes_from_utf8>. 2500 2501 =cut 2502 */ 2503 2504 U8 * 2505 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp) 2506 { 2507 U8 * first_variant; 2508 2509 PERL_ARGS_ASSERT_UTF8_TO_BYTES; 2510 PERL_UNUSED_CONTEXT; 2511 2512 /* This is a no-op if no variants at all in the input */ 2513 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) { 2514 return s; 2515 } 2516 2517 { 2518 U8 * const save = s; 2519 U8 * const send = s + *lenp; 2520 U8 * d; 2521 2522 /* Nothing before the first variant needs to be changed, so start the real 2523 * work there */ 2524 s = first_variant; 2525 while (s < send) { 2526 if (! UTF8_IS_INVARIANT(*s)) { 2527 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { 2528 *lenp = ((STRLEN) -1); 2529 return 0; 2530 } 2531 s++; 2532 } 2533 s++; 2534 } 2535 2536 /* Is downgradable, so do it */ 2537 d = s = first_variant; 2538 while (s < send) { 2539 U8 c = *s++; 2540 if (! UVCHR_IS_INVARIANT(c)) { 2541 /* Then it is two-byte encoded */ 2542 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); 2543 s++; 2544 } 2545 *d++ = c; 2546 } 2547 *d = '\0'; 2548 *lenp = d - save; 2549 2550 return save; 2551 } 2552 } 2553 2554 /* 2555 =for apidoc bytes_from_utf8 2556 2557 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native 2558 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is 2559 actually encoded in UTF-8. 2560 2561 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of 2562 the input string. 2563 2564 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string 2565 not expressible in native byte encoding. In these cases, C<*is_utf8p> and 2566 C<*lenp> are unchanged, and the return value is the original C<s>. 2567 2568 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a 2569 newly created string containing a downgraded copy of C<s>, and whose length is 2570 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The 2571 caller is responsible for arranging for the memory used by this string to get 2572 freed. 2573 2574 Upon successful return, the number of variants in the string can be computed by 2575 having saved the value of C<*lenp> before the call, and subtracting the 2576 after-call value of C<*lenp> from it. 2577 2578 =cut 2579 2580 There is a macro that avoids this function call, but this is retained for 2581 anyone who calls it with the Perl_ prefix */ 2582 2583 U8 * 2584 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p) 2585 { 2586 PERL_ARGS_ASSERT_BYTES_FROM_UTF8; 2587 PERL_UNUSED_CONTEXT; 2588 2589 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL); 2590 } 2591 2592 /* 2593 No = here because currently externally undocumented 2594 for apidoc bytes_from_utf8_loc 2595 2596 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where 2597 to store the location of the first character in C<"s"> that cannot be 2598 converted to non-UTF8. 2599 2600 If that parameter is C<NULL>, this function behaves identically to 2601 C<bytes_from_utf8>. 2602 2603 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to 2604 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>. 2605 2606 Otherwise, the function returns a newly created C<NUL>-terminated string 2607 containing the non-UTF8 equivalent of the convertible first portion of 2608 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>. 2609 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value, 2610 and C<*first_non_downgradable> is set to C<NULL>. 2611 2612 Otherwise, C<*first_non_downgradable> set to point to the first byte of the 2613 first character in the original string that wasn't converted. C<*is_utf8p> is 2614 unchanged. Note that the new string may have length 0. 2615 2616 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and 2617 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and 2618 converts as many characters in it as possible stopping at the first one it 2619 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is 2620 set to point to that. The function returns the portion that could be converted 2621 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length, 2622 not including the terminating C<NUL>. If the very first character in the 2623 original could not be converted, C<*lenp> will be 0, and the new string will 2624 contain just a single C<NUL>. If the entire input string was converted, 2625 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>. 2626 2627 Upon successful return, the number of variants in the converted portion of the 2628 string can be computed by having saved the value of C<*lenp> before the call, 2629 and subtracting the after-call value of C<*lenp> from it. 2630 2631 =cut 2632 2633 2634 */ 2635 2636 U8 * 2637 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted) 2638 { 2639 U8 *d; 2640 const U8 *original = s; 2641 U8 *converted_start; 2642 const U8 *send = s + *lenp; 2643 2644 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC; 2645 2646 if (! *is_utf8p) { 2647 if (first_unconverted) { 2648 *first_unconverted = NULL; 2649 } 2650 2651 return (U8 *) original; 2652 } 2653 2654 Newx(d, (*lenp) + 1, U8); 2655 2656 converted_start = d; 2657 while (s < send) { 2658 U8 c = *s++; 2659 if (! UTF8_IS_INVARIANT(c)) { 2660 2661 /* Then it is multi-byte encoded. If the code point is above 0xFF, 2662 * have to stop now */ 2663 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) { 2664 if (first_unconverted) { 2665 *first_unconverted = s - 1; 2666 goto finish_and_return; 2667 } 2668 else { 2669 Safefree(converted_start); 2670 return (U8 *) original; 2671 } 2672 } 2673 2674 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); 2675 s++; 2676 } 2677 *d++ = c; 2678 } 2679 2680 /* Here, converted the whole of the input */ 2681 *is_utf8p = FALSE; 2682 if (first_unconverted) { 2683 *first_unconverted = NULL; 2684 } 2685 2686 finish_and_return: 2687 *d = '\0'; 2688 *lenp = d - converted_start; 2689 2690 /* Trim unused space */ 2691 Renew(converted_start, *lenp + 1, U8); 2692 2693 return converted_start; 2694 } 2695 2696 /* 2697 =for apidoc bytes_to_utf8 2698 2699 Converts a string C<s> of length C<*lenp> bytes from the native encoding into 2700 UTF-8. 2701 Returns a pointer to the newly-created string, and sets C<*lenp> to 2702 reflect the new length in bytes. The caller is responsible for arranging for 2703 the memory used by this string to get freed. 2704 2705 Upon successful return, the number of variants in the string can be computed by 2706 having saved the value of C<*lenp> before the call, and subtracting it from the 2707 after-call value of C<*lenp>. 2708 2709 A C<NUL> character will be written after the end of the string. 2710 2711 If you want to convert to UTF-8 from encodings other than 2712 the native (Latin1 or EBCDIC), 2713 see L</sv_recode_to_utf8>(). 2714 2715 =cut 2716 */ 2717 2718 U8* 2719 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp) 2720 { 2721 const U8 * const send = s + (*lenp); 2722 U8 *d; 2723 U8 *dst; 2724 2725 PERL_ARGS_ASSERT_BYTES_TO_UTF8; 2726 PERL_UNUSED_CONTEXT; 2727 2728 Newx(d, (*lenp) * 2 + 1, U8); 2729 dst = d; 2730 2731 while (s < send) { 2732 append_utf8_from_native_byte(*s, &d); 2733 s++; 2734 } 2735 2736 *d = '\0'; 2737 *lenp = d-dst; 2738 2739 /* Trim unused space */ 2740 Renew(dst, *lenp + 1, U8); 2741 2742 return dst; 2743 } 2744 2745 /* 2746 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian), 2747 * use utf16_to_utf8_reversed(). 2748 * 2749 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes. 2750 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes. 2751 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes. 2752 * 2753 * These functions don't check for overflow. The worst case is every code 2754 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code 2755 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the 2756 * destination must be pre-extended to 2 times the source length. 2757 * 2758 * Do not use in-place. We optimize for native, for obvious reasons. */ 2759 2760 U8* 2761 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) 2762 { 2763 U8* pend; 2764 U8* dstart = d; 2765 2766 PERL_ARGS_ASSERT_UTF16_TO_UTF8; 2767 2768 if (bytelen & 1) 2769 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, 2770 (UV)bytelen); 2771 2772 pend = p + bytelen; 2773 2774 while (p < pend) { 2775 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */ 2776 p += 2; 2777 if (OFFUNI_IS_INVARIANT(uv)) { 2778 *d++ = LATIN1_TO_NATIVE((U8) uv); 2779 continue; 2780 } 2781 if (uv <= MAX_UTF8_TWO_BYTE) { 2782 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv)); 2783 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv)); 2784 continue; 2785 } 2786 2787 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST 2788 #define LAST_HIGH_SURROGATE 0xDBFF 2789 #define FIRST_LOW_SURROGATE 0xDC00 2790 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST 2791 #define FIRST_IN_PLANE1 0x10000 2792 2793 /* This assumes that most uses will be in the first Unicode plane, not 2794 * needing surrogates */ 2795 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST 2796 && uv <= UNICODE_SURROGATE_LAST)) 2797 { 2798 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) { 2799 Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); 2800 } 2801 else { 2802 UV low = (p[0] << 8) + p[1]; 2803 if ( UNLIKELY(low < FIRST_LOW_SURROGATE) 2804 || UNLIKELY(low > LAST_LOW_SURROGATE)) 2805 { 2806 Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); 2807 } 2808 p += 2; 2809 uv = ((uv - FIRST_HIGH_SURROGATE) << 10) 2810 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1; 2811 } 2812 } 2813 #ifdef EBCDIC 2814 d = uvoffuni_to_utf8_flags(d, uv, 0); 2815 #else 2816 if (uv < FIRST_IN_PLANE1) { 2817 *d++ = (U8)(( uv >> 12) | 0xe0); 2818 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); 2819 *d++ = (U8)(( uv & 0x3f) | 0x80); 2820 continue; 2821 } 2822 else { 2823 *d++ = (U8)(( uv >> 18) | 0xf0); 2824 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80); 2825 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); 2826 *d++ = (U8)(( uv & 0x3f) | 0x80); 2827 continue; 2828 } 2829 #endif 2830 } 2831 *newlen = d - dstart; 2832 return d; 2833 } 2834 2835 /* Note: this one is slightly destructive of the source. */ 2836 2837 U8* 2838 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) 2839 { 2840 U8* s = (U8*)p; 2841 U8* const send = s + bytelen; 2842 2843 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED; 2844 2845 if (bytelen & 1) 2846 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf, 2847 (UV)bytelen); 2848 2849 while (s < send) { 2850 const U8 tmp = s[0]; 2851 s[0] = s[1]; 2852 s[1] = tmp; 2853 s += 2; 2854 } 2855 return utf16_to_utf8(p, d, bytelen, newlen); 2856 } 2857 2858 bool 2859 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c) 2860 { 2861 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c); 2862 } 2863 2864 /* Internal function so we can deprecate the external one, and call 2865 this one from other deprecated functions in this file */ 2866 2867 bool 2868 Perl__is_utf8_idstart(pTHX_ const U8 *p) 2869 { 2870 PERL_ARGS_ASSERT__IS_UTF8_IDSTART; 2871 2872 if (*p == '_') 2873 return TRUE; 2874 return is_utf8_common(p, NULL, 2875 "This is buggy if this gets used", 2876 PL_utf8_idstart); 2877 } 2878 2879 bool 2880 Perl__is_uni_perl_idcont(pTHX_ UV c) 2881 { 2882 return _invlist_contains_cp(PL_utf8_perl_idcont, c); 2883 } 2884 2885 bool 2886 Perl__is_uni_perl_idstart(pTHX_ UV c) 2887 { 2888 return _invlist_contains_cp(PL_utf8_perl_idstart, c); 2889 } 2890 2891 UV 2892 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, 2893 const char S_or_s) 2894 { 2895 /* We have the latin1-range values compiled into the core, so just use 2896 * those, converting the result to UTF-8. The only difference between upper 2897 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is 2898 * either "SS" or "Ss". Which one to use is passed into the routine in 2899 * 'S_or_s' to avoid a test */ 2900 2901 UV converted = toUPPER_LATIN1_MOD(c); 2902 2903 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1; 2904 2905 assert(S_or_s == 'S' || S_or_s == 's'); 2906 2907 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for 2908 characters in this range */ 2909 *p = (U8) converted; 2910 *lenp = 1; 2911 return converted; 2912 } 2913 2914 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers, 2915 * which it maps to one of them, so as to only have to have one check for 2916 * it in the main case */ 2917 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { 2918 switch (c) { 2919 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: 2920 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; 2921 break; 2922 case MICRO_SIGN: 2923 converted = GREEK_CAPITAL_LETTER_MU; 2924 break; 2925 #if UNICODE_MAJOR_VERSION > 2 \ 2926 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \ 2927 && UNICODE_DOT_DOT_VERSION >= 8) 2928 case LATIN_SMALL_LETTER_SHARP_S: 2929 *(p)++ = 'S'; 2930 *p = S_or_s; 2931 *lenp = 2; 2932 return 'S'; 2933 #endif 2934 default: 2935 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect" 2936 " '%c' to map to '%c'", 2937 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS); 2938 NOT_REACHED; /* NOTREACHED */ 2939 } 2940 } 2941 2942 *(p)++ = UTF8_TWO_BYTE_HI(converted); 2943 *p = UTF8_TWO_BYTE_LO(converted); 2944 *lenp = 2; 2945 2946 return converted; 2947 } 2948 2949 /* If compiled on an early Unicode version, there may not be auxiliary tables 2950 * */ 2951 #ifndef HAS_UC_AUX_TABLES 2952 # define UC_AUX_TABLE_ptrs NULL 2953 # define UC_AUX_TABLE_lengths NULL 2954 #endif 2955 #ifndef HAS_TC_AUX_TABLES 2956 # define TC_AUX_TABLE_ptrs NULL 2957 # define TC_AUX_TABLE_lengths NULL 2958 #endif 2959 #ifndef HAS_LC_AUX_TABLES 2960 # define LC_AUX_TABLE_ptrs NULL 2961 # define LC_AUX_TABLE_lengths NULL 2962 #endif 2963 #ifndef HAS_CF_AUX_TABLES 2964 # define CF_AUX_TABLE_ptrs NULL 2965 # define CF_AUX_TABLE_lengths NULL 2966 #endif 2967 #ifndef HAS_UC_AUX_TABLES 2968 # define UC_AUX_TABLE_ptrs NULL 2969 # define UC_AUX_TABLE_lengths NULL 2970 #endif 2971 2972 /* Call the function to convert a UTF-8 encoded character to the specified case. 2973 * Note that there may be more than one character in the result. 2974 * 's' is a pointer to the first byte of the input character 2975 * 'd' will be set to the first byte of the string of changed characters. It 2976 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes 2977 * 'lenp' will be set to the length in bytes of the string of changed characters 2978 * 2979 * The functions return the ordinal of the first character in the string of 2980 * 'd' */ 2981 #define CALL_UPPER_CASE(uv, s, d, lenp) \ 2982 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \ 2983 Uppercase_Mapping_invmap, \ 2984 UC_AUX_TABLE_ptrs, \ 2985 UC_AUX_TABLE_lengths, \ 2986 "uppercase") 2987 #define CALL_TITLE_CASE(uv, s, d, lenp) \ 2988 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \ 2989 Titlecase_Mapping_invmap, \ 2990 TC_AUX_TABLE_ptrs, \ 2991 TC_AUX_TABLE_lengths, \ 2992 "titlecase") 2993 #define CALL_LOWER_CASE(uv, s, d, lenp) \ 2994 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \ 2995 Lowercase_Mapping_invmap, \ 2996 LC_AUX_TABLE_ptrs, \ 2997 LC_AUX_TABLE_lengths, \ 2998 "lowercase") 2999 3000 3001 /* This additionally has the input parameter 'specials', which if non-zero will 3002 * cause this to use the specials hash for folding (meaning get full case 3003 * folding); otherwise, when zero, this implies a simple case fold */ 3004 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \ 3005 (specials) \ 3006 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \ 3007 Case_Folding_invmap, \ 3008 CF_AUX_TABLE_ptrs, \ 3009 CF_AUX_TABLE_lengths, \ 3010 "foldcase") \ 3011 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \ 3012 Simple_Case_Folding_invmap, \ 3013 NULL, NULL, \ 3014 "foldcase") 3015 3016 UV 3017 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp) 3018 { 3019 /* Convert the Unicode character whose ordinal is <c> to its uppercase 3020 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>. 3021 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since 3022 * the changed version may be longer than the original character. 3023 * 3024 * The ordinal of the first character of the changed version is returned 3025 * (but note, as explained above, that there may be more.) */ 3026 3027 PERL_ARGS_ASSERT_TO_UNI_UPPER; 3028 3029 if (c < 256) { 3030 return _to_upper_title_latin1((U8) c, p, lenp, 'S'); 3031 } 3032 3033 uvchr_to_utf8(p, c); 3034 return CALL_UPPER_CASE(c, p, p, lenp); 3035 } 3036 3037 UV 3038 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp) 3039 { 3040 PERL_ARGS_ASSERT_TO_UNI_TITLE; 3041 3042 if (c < 256) { 3043 return _to_upper_title_latin1((U8) c, p, lenp, 's'); 3044 } 3045 3046 uvchr_to_utf8(p, c); 3047 return CALL_TITLE_CASE(c, p, p, lenp); 3048 } 3049 3050 STATIC U8 3051 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy) 3052 { 3053 /* We have the latin1-range values compiled into the core, so just use 3054 * those, converting the result to UTF-8. Since the result is always just 3055 * one character, we allow <p> to be NULL */ 3056 3057 U8 converted = toLOWER_LATIN1(c); 3058 3059 PERL_UNUSED_ARG(dummy); 3060 3061 if (p != NULL) { 3062 if (NATIVE_BYTE_IS_INVARIANT(converted)) { 3063 *p = converted; 3064 *lenp = 1; 3065 } 3066 else { 3067 /* Result is known to always be < 256, so can use the EIGHT_BIT 3068 * macros */ 3069 *p = UTF8_EIGHT_BIT_HI(converted); 3070 *(p+1) = UTF8_EIGHT_BIT_LO(converted); 3071 *lenp = 2; 3072 } 3073 } 3074 return converted; 3075 } 3076 3077 UV 3078 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp) 3079 { 3080 PERL_ARGS_ASSERT_TO_UNI_LOWER; 3081 3082 if (c < 256) { 3083 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ ); 3084 } 3085 3086 uvchr_to_utf8(p, c); 3087 return CALL_LOWER_CASE(c, p, p, lenp); 3088 } 3089 3090 UV 3091 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags) 3092 { 3093 /* Corresponds to to_lower_latin1(); <flags> bits meanings: 3094 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited 3095 * FOLD_FLAGS_FULL iff full folding is to be used; 3096 * 3097 * Not to be used for locale folds 3098 */ 3099 3100 UV converted; 3101 3102 PERL_ARGS_ASSERT__TO_FOLD_LATIN1; 3103 3104 assert (! (flags & FOLD_FLAGS_LOCALE)); 3105 3106 if (UNLIKELY(c == MICRO_SIGN)) { 3107 converted = GREEK_SMALL_LETTER_MU; 3108 } 3109 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ 3110 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ 3111 || UNICODE_DOT_DOT_VERSION > 0) 3112 else if ( (flags & FOLD_FLAGS_FULL) 3113 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S)) 3114 { 3115 /* If can't cross 127/128 boundary, can't return "ss"; instead return 3116 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}") 3117 * under those circumstances. */ 3118 if (flags & FOLD_FLAGS_NOMIX_ASCII) { 3119 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; 3120 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, 3121 p, *lenp, U8); 3122 return LATIN_SMALL_LETTER_LONG_S; 3123 } 3124 else { 3125 *(p)++ = 's'; 3126 *p = 's'; 3127 *lenp = 2; 3128 return 's'; 3129 } 3130 } 3131 #endif 3132 else { /* In this range the fold of all other characters is their lower 3133 case */ 3134 converted = toLOWER_LATIN1(c); 3135 } 3136 3137 if (UVCHR_IS_INVARIANT(converted)) { 3138 *p = (U8) converted; 3139 *lenp = 1; 3140 } 3141 else { 3142 *(p)++ = UTF8_TWO_BYTE_HI(converted); 3143 *p = UTF8_TWO_BYTE_LO(converted); 3144 *lenp = 2; 3145 } 3146 3147 return converted; 3148 } 3149 3150 UV 3151 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags) 3152 { 3153 3154 /* Not currently externally documented, and subject to change 3155 * <flags> bits meanings: 3156 * FOLD_FLAGS_FULL iff full folding is to be used; 3157 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying 3158 * locale are to be used. 3159 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited 3160 */ 3161 3162 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS; 3163 3164 if (flags & FOLD_FLAGS_LOCALE) { 3165 /* Treat a UTF-8 locale as not being in locale at all, except for 3166 * potentially warning */ 3167 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 3168 if (IN_UTF8_CTYPE_LOCALE) { 3169 flags &= ~FOLD_FLAGS_LOCALE; 3170 } 3171 else { 3172 goto needs_full_generality; 3173 } 3174 } 3175 3176 if (c < 256) { 3177 return _to_fold_latin1((U8) c, p, lenp, 3178 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); 3179 } 3180 3181 /* Here, above 255. If no special needs, just use the macro */ 3182 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) { 3183 uvchr_to_utf8(p, c); 3184 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL); 3185 } 3186 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with 3187 the special flags. */ 3188 U8 utf8_c[UTF8_MAXBYTES + 1]; 3189 3190 needs_full_generality: 3191 uvchr_to_utf8(utf8_c, c); 3192 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), 3193 p, lenp, flags); 3194 } 3195 } 3196 3197 PERL_STATIC_INLINE bool 3198 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash, 3199 const char *const swashname, SV* const invlist) 3200 { 3201 /* returns a boolean giving whether or not the UTF8-encoded character that 3202 * starts at <p> is in the swash indicated by <swashname>. <swash> 3203 * contains a pointer to where the swash indicated by <swashname> 3204 * is to be stored; which this routine will do, so that future calls will 3205 * look at <*swash> and only generate a swash if it is not null. <invlist> 3206 * is NULL or an inversion list that defines the swash. If not null, it 3207 * saves time during initialization of the swash. 3208 * 3209 * Note that it is assumed that the buffer length of <p> is enough to 3210 * contain all the bytes that comprise the character. Thus, <*p> should 3211 * have been checked before this call for mal-formedness enough to assure 3212 * that. */ 3213 3214 PERL_ARGS_ASSERT_IS_UTF8_COMMON; 3215 3216 /* The API should have included a length for the UTF-8 character in <p>, 3217 * but it doesn't. We therefore assume that p has been validated at least 3218 * as far as there being enough bytes available in it to accommodate the 3219 * character without reading beyond the end, and pass that number on to the 3220 * validating routine */ 3221 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) { 3222 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p), 3223 _UTF8_NO_CONFIDENCE_IN_CURLEN, 3224 1 /* Die */ ); 3225 NOT_REACHED; /* NOTREACHED */ 3226 } 3227 3228 if (invlist) { 3229 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL)); 3230 } 3231 3232 assert(swash); 3233 3234 if (!*swash) { 3235 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 3236 *swash = _core_swash_init("utf8", 3237 3238 /* Only use the name if there is no inversion 3239 * list; otherwise will go out to disk */ 3240 (invlist) ? "" : swashname, 3241 3242 &PL_sv_undef, 1, 0, invlist, &flags); 3243 } 3244 3245 return swash_fetch(*swash, p, TRUE) != 0; 3246 } 3247 3248 PERL_STATIC_INLINE bool 3249 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, 3250 SV **swash, const char *const swashname, 3251 SV* const invlist) 3252 { 3253 /* returns a boolean giving whether or not the UTF8-encoded character that 3254 * starts at <p>, and extending no further than <e - 1> is in the swash 3255 * indicated by <swashname>. <swash> contains a pointer to where the swash 3256 * indicated by <swashname> is to be stored; which this routine will do, so 3257 * that future calls will look at <*swash> and only generate a swash if it 3258 * is not null. <invlist> is NULL or an inversion list that defines the 3259 * swash. If not null, it saves time during initialization of the swash. 3260 */ 3261 3262 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN; 3263 3264 if (! isUTF8_CHAR(p, e)) { 3265 _force_out_malformed_utf8_message(p, e, 0, 1); 3266 NOT_REACHED; /* NOTREACHED */ 3267 } 3268 3269 if (invlist) { 3270 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL)); 3271 } 3272 3273 assert(swash); 3274 3275 if (!*swash) { 3276 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 3277 *swash = _core_swash_init("utf8", 3278 3279 /* Only use the name if there is no inversion 3280 * list; otherwise will go out to disk */ 3281 (invlist) ? "" : swashname, 3282 3283 &PL_sv_undef, 1, 0, invlist, &flags); 3284 } 3285 3286 return swash_fetch(*swash, p, TRUE) != 0; 3287 } 3288 3289 STATIC void 3290 S_warn_on_first_deprecated_use(pTHX_ const char * const name, 3291 const char * const alternative, 3292 const bool use_locale, 3293 const char * const file, 3294 const unsigned line) 3295 { 3296 const char * key; 3297 3298 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE; 3299 3300 if (ckWARN_d(WARN_DEPRECATED)) { 3301 3302 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line); 3303 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) { 3304 if (! PL_seen_deprecated_macro) { 3305 PL_seen_deprecated_macro = newHV(); 3306 } 3307 if (! hv_store(PL_seen_deprecated_macro, key, 3308 strlen(key), &PL_sv_undef, 0)) 3309 { 3310 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); 3311 } 3312 3313 if (instr(file, "mathoms.c")) { 3314 Perl_warner(aTHX_ WARN_DEPRECATED, 3315 "In %s, line %d, starting in Perl v5.30, %s()" 3316 " will be removed. Avoid this message by" 3317 " converting to use %s().\n", 3318 file, line, name, alternative); 3319 } 3320 else { 3321 Perl_warner(aTHX_ WARN_DEPRECATED, 3322 "In %s, line %d, starting in Perl v5.30, %s() will" 3323 " require an additional parameter. Avoid this" 3324 " message by converting to use %s().\n", 3325 file, line, name, alternative); 3326 } 3327 } 3328 } 3329 } 3330 3331 bool 3332 Perl__is_utf8_FOO(pTHX_ U8 classnum, 3333 const U8 * const p, 3334 const char * const name, 3335 const char * const alternative, 3336 const bool use_utf8, 3337 const bool use_locale, 3338 const char * const file, 3339 const unsigned line) 3340 { 3341 PERL_ARGS_ASSERT__IS_UTF8_FOO; 3342 3343 warn_on_first_deprecated_use(name, alternative, use_locale, file, line); 3344 3345 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) { 3346 3347 switch (classnum) { 3348 case _CC_WORDCHAR: 3349 case _CC_DIGIT: 3350 case _CC_ALPHA: 3351 case _CC_LOWER: 3352 case _CC_UPPER: 3353 case _CC_PUNCT: 3354 case _CC_PRINT: 3355 case _CC_ALPHANUMERIC: 3356 case _CC_GRAPH: 3357 case _CC_CASED: 3358 3359 return is_utf8_common(p, 3360 NULL, 3361 "This is buggy if this gets used", 3362 PL_XPosix_ptrs[classnum]); 3363 3364 case _CC_SPACE: 3365 return is_XPERLSPACE_high(p); 3366 case _CC_BLANK: 3367 return is_HORIZWS_high(p); 3368 case _CC_XDIGIT: 3369 return is_XDIGIT_high(p); 3370 case _CC_CNTRL: 3371 return 0; 3372 case _CC_ASCII: 3373 return 0; 3374 case _CC_VERTSPACE: 3375 return is_VERTWS_high(p); 3376 case _CC_IDFIRST: 3377 return is_utf8_common(p, NULL, 3378 "This is buggy if this gets used", 3379 PL_utf8_perl_idstart); 3380 case _CC_IDCONT: 3381 return is_utf8_common(p, NULL, 3382 "This is buggy if this gets used", 3383 PL_utf8_perl_idcont); 3384 } 3385 } 3386 3387 /* idcont is the same as wordchar below 256 */ 3388 if (classnum == _CC_IDCONT) { 3389 classnum = _CC_WORDCHAR; 3390 } 3391 else if (classnum == _CC_IDFIRST) { 3392 if (*p == '_') { 3393 return TRUE; 3394 } 3395 classnum = _CC_ALPHA; 3396 } 3397 3398 if (! use_locale) { 3399 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) { 3400 return _generic_isCC(*p, classnum); 3401 } 3402 3403 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum); 3404 } 3405 else { 3406 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) { 3407 return isFOO_lc(classnum, *p); 3408 } 3409 3410 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 ))); 3411 } 3412 3413 NOT_REACHED; /* NOTREACHED */ 3414 } 3415 3416 bool 3417 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p, 3418 const U8 * const e) 3419 { 3420 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN; 3421 3422 return is_utf8_common_with_len(p, e, NULL, 3423 "This is buggy if this gets used", 3424 PL_XPosix_ptrs[classnum]); 3425 } 3426 3427 bool 3428 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e) 3429 { 3430 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN; 3431 3432 return is_utf8_common_with_len(p, e, NULL, 3433 "This is buggy if this gets used", 3434 PL_utf8_perl_idstart); 3435 } 3436 3437 bool 3438 Perl__is_utf8_xidstart(pTHX_ const U8 *p) 3439 { 3440 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART; 3441 3442 if (*p == '_') 3443 return TRUE; 3444 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL); 3445 } 3446 3447 bool 3448 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e) 3449 { 3450 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN; 3451 3452 return is_utf8_common_with_len(p, e, NULL, 3453 "This is buggy if this gets used", 3454 PL_utf8_perl_idcont); 3455 } 3456 3457 bool 3458 Perl__is_utf8_idcont(pTHX_ const U8 *p) 3459 { 3460 PERL_ARGS_ASSERT__IS_UTF8_IDCONT; 3461 3462 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL); 3463 } 3464 3465 bool 3466 Perl__is_utf8_xidcont(pTHX_ const U8 *p) 3467 { 3468 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT; 3469 3470 return is_utf8_common(p, &PL_utf8_xidcont, "XIdContinue", NULL); 3471 } 3472 3473 bool 3474 Perl__is_utf8_mark(pTHX_ const U8 *p) 3475 { 3476 PERL_ARGS_ASSERT__IS_UTF8_MARK; 3477 3478 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL); 3479 } 3480 3481 STATIC UV 3482 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, 3483 U8* ustrp, STRLEN *lenp, 3484 SV *invlist, const int * const invmap, 3485 const unsigned int * const * const aux_tables, 3486 const U8 * const aux_table_lengths, 3487 const char * const normal) 3488 { 3489 STRLEN len = 0; 3490 3491 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed 3492 * by this routine to be valid) begins at 'p'. 'normal' is a string to use 3493 * to name the new case in any generated messages, as a fallback if the 3494 * operation being used is not available. The new case is given by the 3495 * data structures in the remaining arguments. 3496 * 3497 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the 3498 * entire changed case string, and the return value is the first code point 3499 * in that string */ 3500 3501 PERL_ARGS_ASSERT__TO_UTF8_CASE; 3502 3503 /* For code points that don't change case, we already know that the output 3504 * of this function is the unchanged input, so we can skip doing look-ups 3505 * for them. Unfortunately the case-changing code points are scattered 3506 * around. But there are some long consecutive ranges where there are no 3507 * case changing code points. By adding tests, we can eliminate the lookup 3508 * for all the ones in such ranges. This is currently done here only for 3509 * just a few cases where the scripts are in common use in modern commerce 3510 * (and scripts adjacent to those which can be included without additional 3511 * tests). */ 3512 3513 if (uv1 >= 0x0590) { 3514 /* This keeps from needing further processing the code points most 3515 * likely to be used in the following non-cased scripts: Hebrew, 3516 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari, 3517 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada, 3518 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */ 3519 if (uv1 < 0x10A0) { 3520 goto cases_to_self; 3521 } 3522 3523 /* The following largish code point ranges also don't have case 3524 * changes, but khw didn't think they warranted extra tests to speed 3525 * them up (which would slightly slow down everything else above them): 3526 * 1100..139F Hangul Jamo, Ethiopic 3527 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic, 3528 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian, 3529 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham, 3530 * Combining Diacritical Marks Extended, Balinese, 3531 * Sundanese, Batak, Lepcha, Ol Chiki 3532 * 2000..206F General Punctuation 3533 */ 3534 3535 if (uv1 >= 0x2D30) { 3536 3537 /* This keeps the from needing further processing the code points 3538 * most likely to be used in the following non-cased major scripts: 3539 * CJK, Katakana, Hiragana, plus some less-likely scripts. 3540 * 3541 * (0x2D30 above might have to be changed to 2F00 in the unlikely 3542 * event that Unicode eventually allocates the unused block as of 3543 * v8.0 2FE0..2FEF to code points that are cased. khw has verified 3544 * that the test suite will start having failures to alert you 3545 * should that happen) */ 3546 if (uv1 < 0xA640) { 3547 goto cases_to_self; 3548 } 3549 3550 if (uv1 >= 0xAC00) { 3551 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) { 3552 if (ckWARN_d(WARN_SURROGATE)) { 3553 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; 3554 Perl_warner(aTHX_ packWARN(WARN_SURROGATE), 3555 "Operation \"%s\" returns its argument for" 3556 " UTF-16 surrogate U+%04" UVXf, desc, uv1); 3557 } 3558 goto cases_to_self; 3559 } 3560 3561 /* AC00..FAFF Catches Hangul syllables and private use, plus 3562 * some others */ 3563 if (uv1 < 0xFB00) { 3564 goto cases_to_self; 3565 } 3566 3567 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) { 3568 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) { 3569 Perl_croak(aTHX_ cp_above_legal_max, uv1, 3570 MAX_EXTERNALLY_LEGAL_CP); 3571 } 3572 if (ckWARN_d(WARN_NON_UNICODE)) { 3573 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; 3574 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), 3575 "Operation \"%s\" returns its argument for" 3576 " non-Unicode code point 0x%04" UVXf, desc, uv1); 3577 } 3578 goto cases_to_self; 3579 } 3580 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C 3581 if (UNLIKELY(uv1 3582 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C)) 3583 { 3584 3585 /* As of Unicode 10.0, this means we avoid swash creation 3586 * for anything beyond high Plane 1 (below emojis) */ 3587 goto cases_to_self; 3588 } 3589 #endif 3590 } 3591 } 3592 3593 /* Note that non-characters are perfectly legal, so no warning should 3594 * be given. */ 3595 } 3596 3597 { 3598 unsigned int i; 3599 const unsigned int * cp_list; 3600 U8 * d; 3601 SSize_t index = _invlist_search(invlist, uv1); 3602 IV base = invmap[index]; 3603 3604 /* The data structures are set up so that if 'base' is non-negative, 3605 * the case change is 1-to-1; and if 0, the change is to itself */ 3606 if (base >= 0) { 3607 IV lc; 3608 3609 if (base == 0) { 3610 goto cases_to_self; 3611 } 3612 3613 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */ 3614 lc = base + uv1 - invlist_array(invlist)[index]; 3615 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp; 3616 return lc; 3617 } 3618 3619 /* Here 'base' is negative. That means the mapping is 1-to-many, and 3620 * requires an auxiliary table look up. abs(base) gives the index into 3621 * a list of such tables which points to the proper aux table. And a 3622 * parallel list gives the length of each corresponding aux table. */ 3623 cp_list = aux_tables[-base]; 3624 3625 /* Create the string of UTF-8 from the mapped-to code points */ 3626 d = ustrp; 3627 for (i = 0; i < aux_table_lengths[-base]; i++) { 3628 d = uvchr_to_utf8(d, cp_list[i]); 3629 } 3630 *d = '\0'; 3631 *lenp = d - ustrp; 3632 3633 return cp_list[0]; 3634 } 3635 3636 /* Here, there was no mapping defined, which means that the code point maps 3637 * to itself. Return the inputs */ 3638 cases_to_self: 3639 len = UTF8SKIP(p); 3640 if (p != ustrp) { /* Don't copy onto itself */ 3641 Copy(p, ustrp, len, U8); 3642 } 3643 3644 if (lenp) 3645 *lenp = len; 3646 3647 return uv1; 3648 3649 } 3650 3651 Size_t 3652 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to, 3653 const unsigned int ** remaining_folds_to) 3654 { 3655 /* Returns the count of the number of code points that fold to the input 3656 * 'cp' (besides itself). 3657 * 3658 * If the return is 0, there is nothing else that folds to it, and 3659 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL. 3660 * 3661 * If the return is 1, '*first_folds_to' is set to the single code point, 3662 * and '*remaining_folds_to' is set to NULL. 3663 * 3664 * Otherwise, '*first_folds_to' is set to a code point, and 3665 * '*remaining_fold_to' is set to an array that contains the others. The 3666 * length of this array is the returned count minus 1. 3667 * 3668 * The reason for this convolution is to avoid having to deal with 3669 * allocating and freeing memory. The lists are already constructed, so 3670 * the return can point to them, but single code points aren't, so would 3671 * need to be constructed if we didn't employ something like this API */ 3672 3673 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp); 3674 int base = _Perl_IVCF_invmap[index]; 3675 3676 PERL_ARGS_ASSERT__INVERSE_FOLDS; 3677 3678 if (base == 0) { /* No fold */ 3679 *first_folds_to = 0; 3680 *remaining_folds_to = NULL; 3681 return 0; 3682 } 3683 3684 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */ 3685 3686 assert(base > 0); 3687 3688 #else 3689 3690 if (UNLIKELY(base < 0)) { /* Folds to more than one character */ 3691 3692 /* The data structure is set up so that the absolute value of 'base' is 3693 * an index into a table of pointers to arrays, with the array 3694 * corresponding to the index being the list of code points that fold 3695 * to 'cp', and the parallel array containing the length of the list 3696 * array */ 3697 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0]; 3698 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes 3699 *first_folds_to 3700 */ 3701 return IVCF_AUX_TABLE_lengths[-base]; 3702 } 3703 3704 #endif 3705 3706 /* Only the single code point. This works like 'fc(G) = G - A + a' */ 3707 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index]; 3708 *remaining_folds_to = NULL; 3709 return 1; 3710 } 3711 3712 STATIC UV 3713 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, 3714 U8* const ustrp, STRLEN *lenp) 3715 { 3716 /* This is called when changing the case of a UTF-8-encoded character above 3717 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the 3718 * result contains a character that crosses the 255/256 boundary, disallow 3719 * the change, and return the original code point. See L<perlfunc/lc> for 3720 * why; 3721 * 3722 * p points to the original string whose case was changed; assumed 3723 * by this routine to be well-formed 3724 * result the code point of the first character in the changed-case string 3725 * ustrp points to the changed-case string (<result> represents its 3726 * first char) 3727 * lenp points to the length of <ustrp> */ 3728 3729 UV original; /* To store the first code point of <p> */ 3730 3731 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING; 3732 3733 assert(UTF8_IS_ABOVE_LATIN1(*p)); 3734 3735 /* We know immediately if the first character in the string crosses the 3736 * boundary, so can skip testing */ 3737 if (result > 255) { 3738 3739 /* Look at every character in the result; if any cross the 3740 * boundary, the whole thing is disallowed */ 3741 U8* s = ustrp + UTF8SKIP(ustrp); 3742 U8* e = ustrp + *lenp; 3743 while (s < e) { 3744 if (! UTF8_IS_ABOVE_LATIN1(*s)) { 3745 goto bad_crossing; 3746 } 3747 s += UTF8SKIP(s); 3748 } 3749 3750 /* Here, no characters crossed, result is ok as-is, but we warn. */ 3751 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p)); 3752 return result; 3753 } 3754 3755 bad_crossing: 3756 3757 /* Failed, have to return the original */ 3758 original = valid_utf8_to_uvchr(p, lenp); 3759 3760 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 3761 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 3762 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8" 3763 " locale; resolved to \"\\x{%" UVXf "}\".", 3764 OP_DESC(PL_op), 3765 original, 3766 original); 3767 Copy(p, ustrp, *lenp, char); 3768 return original; 3769 } 3770 3771 STATIC U32 3772 S_check_and_deprecate(pTHX_ const U8 *p, 3773 const U8 **e, 3774 const unsigned int type, /* See below */ 3775 const bool use_locale, /* Is this a 'LC_' 3776 macro call? */ 3777 const char * const file, 3778 const unsigned line) 3779 { 3780 /* This is a temporary function to deprecate the unsafe calls to the case 3781 * changing macros and functions. It keeps all the special stuff in just 3782 * one place. 3783 * 3784 * It updates *e with the pointer to the end of the input string. If using 3785 * the old-style macros, *e is NULL on input, and so this function assumes 3786 * the input string is long enough to hold the entire UTF-8 sequence, and 3787 * sets *e accordingly, but it then returns a flag to pass the 3788 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid 3789 * using the full length if possible. 3790 * 3791 * It also does the assert that *e > p when *e is not NULL. This should be 3792 * migrated to the callers when this function gets deleted. 3793 * 3794 * The 'type' parameter is used for the caller to specify which case 3795 * changing function this is called from: */ 3796 3797 # define DEPRECATE_TO_UPPER 0 3798 # define DEPRECATE_TO_TITLE 1 3799 # define DEPRECATE_TO_LOWER 2 3800 # define DEPRECATE_TO_FOLD 3 3801 3802 U32 utf8n_flags = 0; 3803 const char * name; 3804 const char * alternative; 3805 3806 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE; 3807 3808 if (*e == NULL) { 3809 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN; 3810 *e = p + UTF8SKIP(p); 3811 3812 /* For mathoms.c calls, we use the function name we know is stored 3813 * there. It could be part of a larger path */ 3814 if (type == DEPRECATE_TO_UPPER) { 3815 name = instr(file, "mathoms.c") 3816 ? "to_utf8_upper" 3817 : "toUPPER_utf8"; 3818 alternative = "toUPPER_utf8_safe"; 3819 } 3820 else if (type == DEPRECATE_TO_TITLE) { 3821 name = instr(file, "mathoms.c") 3822 ? "to_utf8_title" 3823 : "toTITLE_utf8"; 3824 alternative = "toTITLE_utf8_safe"; 3825 } 3826 else if (type == DEPRECATE_TO_LOWER) { 3827 name = instr(file, "mathoms.c") 3828 ? "to_utf8_lower" 3829 : "toLOWER_utf8"; 3830 alternative = "toLOWER_utf8_safe"; 3831 } 3832 else if (type == DEPRECATE_TO_FOLD) { 3833 name = instr(file, "mathoms.c") 3834 ? "to_utf8_fold" 3835 : "toFOLD_utf8"; 3836 alternative = "toFOLD_utf8_safe"; 3837 } 3838 else Perl_croak(aTHX_ "panic: Unexpected case change type"); 3839 3840 warn_on_first_deprecated_use(name, alternative, use_locale, file, line); 3841 } 3842 else { 3843 assert (p < *e); 3844 } 3845 3846 return utf8n_flags; 3847 } 3848 3849 /* The process for changing the case is essentially the same for the four case 3850 * change types, except there are complications for folding. Otherwise the 3851 * difference is only which case to change to. To make sure that they all do 3852 * the same thing, the bodies of the functions are extracted out into the 3853 * following two macros. The functions are written with the same variable 3854 * names, and these are known and used inside these macros. It would be 3855 * better, of course, to have inline functions to do it, but since different 3856 * macros are called, depending on which case is being changed to, this is not 3857 * feasible in C (to khw's knowledge). Two macros are created so that the fold 3858 * function can start with the common start macro, then finish with its special 3859 * handling; while the other three cases can just use the common end macro. 3860 * 3861 * The algorithm is to use the proper (passed in) macro or function to change 3862 * the case for code points that are below 256. The macro is used if using 3863 * locale rules for the case change; the function if not. If the code point is 3864 * above 255, it is computed from the input UTF-8, and another macro is called 3865 * to do the conversion. If necessary, the output is converted to UTF-8. If 3866 * using a locale, we have to check that the change did not cross the 255/256 3867 * boundary, see check_locale_boundary_crossing() for further details. 3868 * 3869 * The macros are split with the correct case change for the below-256 case 3870 * stored into 'result', and in the middle of an else clause for the above-255 3871 * case. At that point in the 'else', 'result' is not the final result, but is 3872 * the input code point calculated from the UTF-8. The fold code needs to 3873 * realize all this and take it from there. 3874 * 3875 * If you read the two macros as sequential, it's easier to understand what's 3876 * going on. */ 3877 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \ 3878 L1_func_extra_param) \ 3879 \ 3880 if (flags & (locale_flags)) { \ 3881 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ 3882 /* Treat a UTF-8 locale as not being in locale at all */ \ 3883 if (IN_UTF8_CTYPE_LOCALE) { \ 3884 flags &= ~(locale_flags); \ 3885 } \ 3886 } \ 3887 \ 3888 if (UTF8_IS_INVARIANT(*p)) { \ 3889 if (flags & (locale_flags)) { \ 3890 result = LC_L1_change_macro(*p); \ 3891 } \ 3892 else { \ 3893 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \ 3894 } \ 3895 } \ 3896 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \ 3897 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \ 3898 if (flags & (locale_flags)) { \ 3899 result = LC_L1_change_macro(c); \ 3900 } \ 3901 else { \ 3902 return L1_func(c, ustrp, lenp, L1_func_extra_param); \ 3903 } \ 3904 } \ 3905 else { /* malformed UTF-8 or ord above 255 */ \ 3906 STRLEN len_result; \ 3907 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \ 3908 if (len_result == (STRLEN) -1) { \ 3909 _force_out_malformed_utf8_message(p, e, utf8n_flags, \ 3910 1 /* Die */ ); \ 3911 } 3912 3913 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \ 3914 result = change_macro(result, p, ustrp, lenp); \ 3915 \ 3916 if (flags & (locale_flags)) { \ 3917 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \ 3918 } \ 3919 return result; \ 3920 } \ 3921 \ 3922 /* Here, used locale rules. Convert back to UTF-8 */ \ 3923 if (UTF8_IS_INVARIANT(result)) { \ 3924 *ustrp = (U8) result; \ 3925 *lenp = 1; \ 3926 } \ 3927 else { \ 3928 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \ 3929 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \ 3930 *lenp = 2; \ 3931 } \ 3932 \ 3933 return result; 3934 3935 /* 3936 =for apidoc to_utf8_upper 3937 3938 Instead use L</toUPPER_utf8_safe>. 3939 3940 =cut */ 3941 3942 /* Not currently externally documented, and subject to change: 3943 * <flags> is set iff iff the rules from the current underlying locale are to 3944 * be used. */ 3945 3946 UV 3947 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, 3948 const U8 *e, 3949 U8* ustrp, 3950 STRLEN *lenp, 3951 bool flags, 3952 const char * const file, 3953 const int line) 3954 { 3955 UV result; 3956 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER, 3957 cBOOL(flags), file, line); 3958 3959 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS; 3960 3961 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */ 3962 /* 2nd char of uc(U+DF) is 'S' */ 3963 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S'); 3964 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE); 3965 } 3966 3967 /* 3968 =for apidoc to_utf8_title 3969 3970 Instead use L</toTITLE_utf8_safe>. 3971 3972 =cut */ 3973 3974 /* Not currently externally documented, and subject to change: 3975 * <flags> is set iff the rules from the current underlying locale are to be 3976 * used. Since titlecase is not defined in POSIX, for other than a 3977 * UTF-8 locale, uppercase is used instead for code points < 256. 3978 */ 3979 3980 UV 3981 Perl__to_utf8_title_flags(pTHX_ const U8 *p, 3982 const U8 *e, 3983 U8* ustrp, 3984 STRLEN *lenp, 3985 bool flags, 3986 const char * const file, 3987 const int line) 3988 { 3989 UV result; 3990 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE, 3991 cBOOL(flags), file, line); 3992 3993 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS; 3994 3995 /* 2nd char of ucfirst(U+DF) is 's' */ 3996 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's'); 3997 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE); 3998 } 3999 4000 /* 4001 =for apidoc to_utf8_lower 4002 4003 Instead use L</toLOWER_utf8_safe>. 4004 4005 =cut */ 4006 4007 /* Not currently externally documented, and subject to change: 4008 * <flags> is set iff iff the rules from the current underlying locale are to 4009 * be used. 4010 */ 4011 4012 UV 4013 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, 4014 const U8 *e, 4015 U8* ustrp, 4016 STRLEN *lenp, 4017 bool flags, 4018 const char * const file, 4019 const int line) 4020 { 4021 UV result; 4022 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER, 4023 cBOOL(flags), file, line); 4024 4025 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS; 4026 4027 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */) 4028 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE) 4029 } 4030 4031 /* 4032 =for apidoc to_utf8_fold 4033 4034 Instead use L</toFOLD_utf8_safe>. 4035 4036 =cut */ 4037 4038 /* Not currently externally documented, and subject to change, 4039 * in <flags> 4040 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying 4041 * locale are to be used. 4042 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used; 4043 * otherwise simple folds 4044 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are 4045 * prohibited 4046 */ 4047 4048 UV 4049 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, 4050 const U8 *e, 4051 U8* ustrp, 4052 STRLEN *lenp, 4053 U8 flags, 4054 const char * const file, 4055 const int line) 4056 { 4057 UV result; 4058 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD, 4059 cBOOL(flags), file, line); 4060 4061 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS; 4062 4063 /* These are mutually exclusive */ 4064 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII))); 4065 4066 assert(p != ustrp); /* Otherwise overwrites */ 4067 4068 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1, 4069 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII))); 4070 4071 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL); 4072 4073 if (flags & FOLD_FLAGS_LOCALE) { 4074 4075 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8 4076 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8 4077 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8 4078 4079 /* Special case these two characters, as what normally gets 4080 * returned under locale doesn't work */ 4081 if (memEQs((char *) p, UTF8SKIP(p), CAP_SHARP_S)) 4082 { 4083 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 4084 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 4085 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; " 4086 "resolved to \"\\x{17F}\\x{17F}\"."); 4087 goto return_long_s; 4088 } 4089 else 4090 #endif 4091 if (memEQs((char *) p, UTF8SKIP(p), LONG_S_T)) 4092 { 4093 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 4094 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 4095 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; " 4096 "resolved to \"\\x{FB06}\"."); 4097 goto return_ligature_st; 4098 } 4099 4100 #if UNICODE_MAJOR_VERSION == 3 \ 4101 && UNICODE_DOT_VERSION == 0 \ 4102 && UNICODE_DOT_DOT_VERSION == 1 4103 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8 4104 4105 /* And special case this on this Unicode version only, for the same 4106 * reaons the other two are special cased. They would cross the 4107 * 255/256 boundary which is forbidden under /l, and so the code 4108 * wouldn't catch that they are equivalent (which they are only in 4109 * this release) */ 4110 else if (memEQs((char *) p, UTF8SKIP(p), DOTTED_I)) { 4111 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 4112 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 4113 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; " 4114 "resolved to \"\\x{0131}\"."); 4115 goto return_dotless_i; 4116 } 4117 #endif 4118 4119 return check_locale_boundary_crossing(p, result, ustrp, lenp); 4120 } 4121 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) { 4122 return result; 4123 } 4124 else { 4125 /* This is called when changing the case of a UTF-8-encoded 4126 * character above the ASCII range, and the result should not 4127 * contain an ASCII character. */ 4128 4129 UV original; /* To store the first code point of <p> */ 4130 4131 /* Look at every character in the result; if any cross the 4132 * boundary, the whole thing is disallowed */ 4133 U8* s = ustrp; 4134 U8* e = ustrp + *lenp; 4135 while (s < e) { 4136 if (isASCII(*s)) { 4137 /* Crossed, have to return the original */ 4138 original = valid_utf8_to_uvchr(p, lenp); 4139 4140 /* But in these instances, there is an alternative we can 4141 * return that is valid */ 4142 if (original == LATIN_SMALL_LETTER_SHARP_S 4143 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */ 4144 || original == LATIN_CAPITAL_LETTER_SHARP_S 4145 #endif 4146 ) { 4147 goto return_long_s; 4148 } 4149 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) { 4150 goto return_ligature_st; 4151 } 4152 #if UNICODE_MAJOR_VERSION == 3 \ 4153 && UNICODE_DOT_VERSION == 0 \ 4154 && UNICODE_DOT_DOT_VERSION == 1 4155 4156 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) { 4157 goto return_dotless_i; 4158 } 4159 #endif 4160 Copy(p, ustrp, *lenp, char); 4161 return original; 4162 } 4163 s += UTF8SKIP(s); 4164 } 4165 4166 /* Here, no characters crossed, result is ok as-is */ 4167 return result; 4168 } 4169 } 4170 4171 /* Here, used locale rules. Convert back to UTF-8 */ 4172 if (UTF8_IS_INVARIANT(result)) { 4173 *ustrp = (U8) result; 4174 *lenp = 1; 4175 } 4176 else { 4177 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); 4178 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); 4179 *lenp = 2; 4180 } 4181 4182 return result; 4183 4184 return_long_s: 4185 /* Certain folds to 'ss' are prohibited by the options, but they do allow 4186 * folds to a string of two of these characters. By returning this 4187 * instead, then, e.g., 4188 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}") 4189 * works. */ 4190 4191 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; 4192 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, 4193 ustrp, *lenp, U8); 4194 return LATIN_SMALL_LETTER_LONG_S; 4195 4196 return_ligature_st: 4197 /* Two folds to 'st' are prohibited by the options; instead we pick one and 4198 * have the other one fold to it */ 4199 4200 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1; 4201 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8); 4202 return LATIN_SMALL_LIGATURE_ST; 4203 4204 #if UNICODE_MAJOR_VERSION == 3 \ 4205 && UNICODE_DOT_VERSION == 0 \ 4206 && UNICODE_DOT_DOT_VERSION == 1 4207 4208 return_dotless_i: 4209 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1; 4210 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8); 4211 return LATIN_SMALL_LETTER_DOTLESS_I; 4212 4213 #endif 4214 4215 } 4216 4217 /* Note: 4218 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch(). 4219 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8". 4220 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl. 4221 */ 4222 4223 SV* 4224 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, 4225 I32 minbits, I32 none) 4226 { 4227 PERL_ARGS_ASSERT_SWASH_INIT; 4228 4229 /* Returns a copy of a swash initiated by the called function. This is the 4230 * public interface, and returning a copy prevents others from doing 4231 * mischief on the original */ 4232 4233 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, 4234 NULL, NULL)); 4235 } 4236 4237 SV* 4238 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, 4239 I32 minbits, I32 none, SV* invlist, 4240 U8* const flags_p) 4241 { 4242 4243 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST 4244 * use the following define */ 4245 4246 #define CORE_SWASH_INIT_RETURN(x) \ 4247 PL_curpm= old_PL_curpm; \ 4248 return x 4249 4250 /* Initialize and return a swash, creating it if necessary. It does this 4251 * by calling utf8_heavy.pl in the general case. The returned value may be 4252 * the swash's inversion list instead if the input parameters allow it. 4253 * Which is returned should be immaterial to callers, as the only 4254 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(), 4255 * and swash_to_invlist() handle both these transparently. 4256 * 4257 * This interface should only be used by functions that won't destroy or 4258 * adversely change the swash, as doing so affects all other uses of the 4259 * swash in the program; the general public should use 'Perl_swash_init' 4260 * instead. 4261 * 4262 * pkg is the name of the package that <name> should be in. 4263 * name is the name of the swash to find. Typically it is a Unicode 4264 * property name, including user-defined ones 4265 * listsv is a string to initialize the swash with. It must be of the form 4266 * documented as the subroutine return value in 4267 * L<perlunicode/User-Defined Character Properties> 4268 * minbits is the number of bits required to represent each data element. 4269 * It is '1' for binary properties. 4270 * none I (khw) do not understand this one, but it is used only in tr///. 4271 * invlist is an inversion list to initialize the swash with (or NULL) 4272 * flags_p if non-NULL is the address of various input and output flag bits 4273 * to the routine, as follows: ('I' means is input to the routine; 4274 * 'O' means output from the routine. Only flags marked O are 4275 * meaningful on return.) 4276 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash 4277 * came from a user-defined property. (I O) 4278 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking 4279 * when the swash cannot be located, to simply return NULL. (I) 4280 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a 4281 * return of an inversion list instead of a swash hash if this routine 4282 * thinks that would result in faster execution of swash_fetch() later 4283 * on. (I) 4284 * 4285 * Thus there are three possible inputs to find the swash: <name>, 4286 * <listsv>, and <invlist>. 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 * <invlist> is only valid for binary properties */ 4293 4294 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */ 4295 4296 SV* retval = &PL_sv_undef; 4297 HV* swash_hv = NULL; 4298 const bool use_invlist= (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST); 4299 4300 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist); 4301 assert(! invlist || minbits == 1); 4302 4303 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the 4304 regex that triggered the swash init and the swash init 4305 perl logic itself. See perl #122747 */ 4306 4307 /* If data was passed in to go out to utf8_heavy to find the swash of, do 4308 * so */ 4309 if (listsv != &PL_sv_undef || strNE(name, "")) { 4310 dSP; 4311 const size_t pkg_len = strlen(pkg); 4312 const size_t name_len = strlen(name); 4313 HV * const stash = gv_stashpvn(pkg, pkg_len, 0); 4314 SV* errsv_save; 4315 GV *method; 4316 4317 PERL_ARGS_ASSERT__CORE_SWASH_INIT; 4318 4319 PUSHSTACKi(PERLSI_MAGIC); 4320 ENTER; 4321 SAVEHINTS(); 4322 save_re_context(); 4323 /* We might get here via a subroutine signature which uses a utf8 4324 * parameter name, at which point PL_subname will have been set 4325 * but not yet used. */ 4326 save_item(PL_subname); 4327 if (PL_parser && PL_parser->error_count) 4328 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0; 4329 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1); 4330 if (!method) { /* demand load UTF-8 */ 4331 ENTER; 4332 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); 4333 GvSV(PL_errgv) = NULL; 4334 #ifndef NO_TAINT_SUPPORT 4335 /* It is assumed that callers of this routine are not passing in 4336 * any user derived data. */ 4337 /* Need to do this after save_re_context() as it will set 4338 * PL_tainted to 1 while saving $1 etc (see the code after getrx: 4339 * in Perl_magic_get). Even line to create errsv_save can turn on 4340 * PL_tainted. */ 4341 SAVEBOOL(TAINT_get); 4342 TAINT_NOT; 4343 #endif 4344 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len), 4345 NULL); 4346 { 4347 /* Not ERRSV, as there is no need to vivify a scalar we are 4348 about to discard. */ 4349 SV * const errsv = GvSV(PL_errgv); 4350 if (!SvTRUE(errsv)) { 4351 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); 4352 SvREFCNT_dec(errsv); 4353 } 4354 } 4355 LEAVE; 4356 } 4357 SPAGAIN; 4358 PUSHMARK(SP); 4359 EXTEND(SP,5); 4360 mPUSHp(pkg, pkg_len); 4361 mPUSHp(name, name_len); 4362 PUSHs(listsv); 4363 mPUSHi(minbits); 4364 mPUSHi(none); 4365 PUTBACK; 4366 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); 4367 GvSV(PL_errgv) = NULL; 4368 /* If we already have a pointer to the method, no need to use 4369 * call_method() to repeat the lookup. */ 4370 if (method 4371 ? call_sv(MUTABLE_SV(method), G_SCALAR) 4372 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD)) 4373 { 4374 retval = *PL_stack_sp--; 4375 SvREFCNT_inc(retval); 4376 } 4377 { 4378 /* Not ERRSV. See above. */ 4379 SV * const errsv = GvSV(PL_errgv); 4380 if (!SvTRUE(errsv)) { 4381 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); 4382 SvREFCNT_dec(errsv); 4383 } 4384 } 4385 LEAVE; 4386 POPSTACK; 4387 if (IN_PERL_COMPILETIME) { 4388 CopHINTS_set(PL_curcop, PL_hints); 4389 } 4390 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) { 4391 if (SvPOK(retval)) { 4392 4393 /* If caller wants to handle missing properties, let them */ 4394 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) { 4395 CORE_SWASH_INIT_RETURN(NULL); 4396 } 4397 Perl_croak(aTHX_ 4398 "Can't find Unicode property definition \"%" SVf "\"", 4399 SVfARG(retval)); 4400 NOT_REACHED; /* NOTREACHED */ 4401 } 4402 } 4403 } /* End of calling the module to find the swash */ 4404 4405 /* If this operation fetched a swash, and we will need it later, get it */ 4406 if (retval != &PL_sv_undef 4407 && (minbits == 1 || (flags_p 4408 && ! (*flags_p 4409 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)))) 4410 { 4411 swash_hv = MUTABLE_HV(SvRV(retval)); 4412 4413 /* If we don't already know that there is a user-defined component to 4414 * this swash, and the user has indicated they wish to know if there is 4415 * one (by passing <flags_p>), find out */ 4416 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) { 4417 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE); 4418 if (user_defined && SvUV(*user_defined)) { 4419 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY; 4420 } 4421 } 4422 } 4423 4424 /* Make sure there is an inversion list for binary properties */ 4425 if (minbits == 1) { 4426 SV** swash_invlistsvp = NULL; 4427 SV* swash_invlist = NULL; 4428 bool invlist_in_swash_is_valid = FALSE; 4429 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has 4430 an unclaimed reference count */ 4431 4432 /* If this operation fetched a swash, get its already existing 4433 * inversion list, or create one for it */ 4434 4435 if (swash_hv) { 4436 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE); 4437 if (swash_invlistsvp) { 4438 swash_invlist = *swash_invlistsvp; 4439 invlist_in_swash_is_valid = TRUE; 4440 } 4441 else { 4442 swash_invlist = _swash_to_invlist(retval); 4443 swash_invlist_unclaimed = TRUE; 4444 } 4445 } 4446 4447 /* If an inversion list was passed in, have to include it */ 4448 if (invlist) { 4449 4450 /* Any fetched swash will by now have an inversion list in it; 4451 * otherwise <swash_invlist> will be NULL, indicating that we 4452 * didn't fetch a swash */ 4453 if (swash_invlist) { 4454 4455 /* Add the passed-in inversion list, which invalidates the one 4456 * already stored in the swash */ 4457 invlist_in_swash_is_valid = FALSE; 4458 SvREADONLY_off(swash_invlist); /* Turned on again below */ 4459 _invlist_union(invlist, swash_invlist, &swash_invlist); 4460 } 4461 else { 4462 4463 /* Here, there is no swash already. Set up a minimal one, if 4464 * we are going to return a swash */ 4465 if (! use_invlist) { 4466 swash_hv = newHV(); 4467 retval = newRV_noinc(MUTABLE_SV(swash_hv)); 4468 } 4469 swash_invlist = invlist; 4470 } 4471 } 4472 4473 /* Here, we have computed the union of all the passed-in data. It may 4474 * be that there was an inversion list in the swash which didn't get 4475 * touched; otherwise save the computed one */ 4476 if (! invlist_in_swash_is_valid && ! use_invlist) { 4477 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist)) 4478 { 4479 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); 4480 } 4481 /* We just stole a reference count. */ 4482 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE; 4483 else SvREFCNT_inc_simple_void_NN(swash_invlist); 4484 } 4485 4486 /* The result is immutable. Forbid attempts to change it. */ 4487 SvREADONLY_on(swash_invlist); 4488 4489 if (use_invlist) { 4490 SvREFCNT_dec(retval); 4491 if (!swash_invlist_unclaimed) 4492 SvREFCNT_inc_simple_void_NN(swash_invlist); 4493 retval = newRV_noinc(swash_invlist); 4494 } 4495 } 4496 4497 CORE_SWASH_INIT_RETURN(retval); 4498 #undef CORE_SWASH_INIT_RETURN 4499 } 4500 4501 4502 /* This API is wrong for special case conversions since we may need to 4503 * return several Unicode characters for a single Unicode character 4504 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is 4505 * the lower-level routine, and it is similarly broken for returning 4506 * multiple values. --jhi 4507 * For those, you should use S__to_utf8_case() instead */ 4508 /* Now SWASHGET is recasted into S_swatch_get in this file. */ 4509 4510 /* Note: 4511 * Returns the value of property/mapping C<swash> for the first character 4512 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is 4513 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr> 4514 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>. 4515 * 4516 * A "swash" is a hash which contains initially the keys/values set up by 4517 * SWASHNEW. The purpose is to be able to completely represent a Unicode 4518 * property for all possible code points. Things are stored in a compact form 4519 * (see utf8_heavy.pl) so that calculation is required to find the actual 4520 * property value for a given code point. As code points are looked up, new 4521 * key/value pairs are added to the hash, so that the calculation doesn't have 4522 * to ever be re-done. Further, each calculation is done, not just for the 4523 * desired one, but for a whole block of code points adjacent to that one. 4524 * For binary properties on ASCII machines, the block is usually for 64 code 4525 * points, starting with a code point evenly divisible by 64. Thus if the 4526 * property value for code point 257 is requested, the code goes out and 4527 * calculates the property values for all 64 code points between 256 and 319, 4528 * and stores these as a single 64-bit long bit vector, called a "swatch", 4529 * under the key for code point 256. The key is the UTF-8 encoding for code 4530 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding 4531 * for a code point is 13 bytes, the key will be 12 bytes long. If the value 4532 * for code point 258 is then requested, this code realizes that it would be 4533 * stored under the key for 256, and would find that value and extract the 4534 * relevant bit, offset from 256. 4535 * 4536 * Non-binary properties are stored in as many bits as necessary to represent 4537 * their values (32 currently, though the code is more general than that), not 4538 * as single bits, but the principle is the same: the value for each key is a 4539 * vector that encompasses the property values for all code points whose UTF-8 4540 * representations are represented by the key. That is, for all code points 4541 * whose UTF-8 representations are length N bytes, and the key is the first N-1 4542 * bytes of that. 4543 */ 4544 UV 4545 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8) 4546 { 4547 HV *const hv = MUTABLE_HV(SvRV(swash)); 4548 U32 klen; 4549 U32 off; 4550 STRLEN slen = 0; 4551 STRLEN needents; 4552 const U8 *tmps = NULL; 4553 SV *swatch; 4554 const U8 c = *ptr; 4555 4556 PERL_ARGS_ASSERT_SWASH_FETCH; 4557 4558 /* If it really isn't a hash, it isn't really swash; must be an inversion 4559 * list */ 4560 if (SvTYPE(hv) != SVt_PVHV) { 4561 return _invlist_contains_cp((SV*)hv, 4562 (do_utf8) 4563 ? valid_utf8_to_uvchr(ptr, NULL) 4564 : c); 4565 } 4566 4567 /* We store the values in a "swatch" which is a vec() value in a swash 4568 * hash. Code points 0-255 are a single vec() stored with key length 4569 * (klen) 0. All other code points have a UTF-8 representation 4570 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which 4571 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key 4572 * length for them is the length of the encoded char - 1. ptr[klen] is the 4573 * final byte in the sequence representing the character */ 4574 if (!do_utf8 || UTF8_IS_INVARIANT(c)) { 4575 klen = 0; 4576 needents = 256; 4577 off = c; 4578 } 4579 else if (UTF8_IS_DOWNGRADEABLE_START(c)) { 4580 klen = 0; 4581 needents = 256; 4582 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1)); 4583 } 4584 else { 4585 klen = UTF8SKIP(ptr) - 1; 4586 4587 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into 4588 * the vec is the final byte in the sequence. (In EBCDIC this is 4589 * converted to I8 to get consecutive values.) To help you visualize 4590 * all this: 4591 * Straight 1047 After final byte 4592 * UTF-8 UTF-EBCDIC I8 transform 4593 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0 4594 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1 4595 * ... 4596 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9 4597 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA 4598 * ... 4599 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2 4600 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3 4601 * ... 4602 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB 4603 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC 4604 * ... 4605 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF 4606 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41 4607 * 4608 * (There are no discontinuities in the elided (...) entries.) 4609 * The UTF-8 key for these 33 code points is '\xD0' (which also is the 4610 * key for the next 31, up through U+043F, whose UTF-8 final byte is 4611 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points. 4612 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an 4613 * index into the vec() swatch (after subtracting 0x80, which we 4614 * actually do with an '&'). 4615 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32 4616 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has 4617 * dicontinuities which go away by transforming it into I8, and we 4618 * effectively subtract 0xA0 to get the index. */ 4619 needents = (1 << UTF_ACCUMULATION_SHIFT); 4620 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK; 4621 } 4622 4623 /* 4624 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test 4625 * suite. (That is, only 7-8% overall over just a hash cache. Still, 4626 * it's nothing to sniff at.) Pity we usually come through at least 4627 * two function calls to get here... 4628 * 4629 * NB: this code assumes that swatches are never modified, once generated! 4630 */ 4631 4632 if (hv == PL_last_swash_hv && 4633 klen == PL_last_swash_klen && 4634 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) ) 4635 { 4636 tmps = PL_last_swash_tmps; 4637 slen = PL_last_swash_slen; 4638 } 4639 else { 4640 /* Try our second-level swatch cache, kept in a hash. */ 4641 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE); 4642 4643 /* If not cached, generate it via swatch_get */ 4644 if (!svp || !SvPOK(*svp) 4645 || !(tmps = (const U8*)SvPV_const(*svp, slen))) 4646 { 4647 if (klen) { 4648 const UV code_point = valid_utf8_to_uvchr(ptr, NULL); 4649 swatch = swatch_get(swash, 4650 code_point & ~((UV)needents - 1), 4651 needents); 4652 } 4653 else { /* For the first 256 code points, the swatch has a key of 4654 length 0 */ 4655 swatch = swatch_get(swash, 0, needents); 4656 } 4657 4658 if (IN_PERL_COMPILETIME) 4659 CopHINTS_set(PL_curcop, PL_hints); 4660 4661 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0); 4662 4663 if (!svp || !(tmps = (U8*)SvPV(*svp, slen)) 4664 || (slen << 3) < needents) 4665 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, " 4666 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf, 4667 svp, tmps, (UV)slen, (UV)needents); 4668 } 4669 4670 PL_last_swash_hv = hv; 4671 assert(klen <= sizeof(PL_last_swash_key)); 4672 PL_last_swash_klen = (U8)klen; 4673 /* FIXME change interpvar.h? */ 4674 PL_last_swash_tmps = (U8 *) tmps; 4675 PL_last_swash_slen = slen; 4676 if (klen) 4677 Copy(ptr, PL_last_swash_key, klen, U8); 4678 } 4679 4680 switch ((int)((slen << 3) / needents)) { 4681 case 1: 4682 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0; 4683 case 8: 4684 return ((UV) tmps[off]); 4685 case 16: 4686 off <<= 1; 4687 return 4688 ((UV) tmps[off ] << 8) + 4689 ((UV) tmps[off + 1]); 4690 case 32: 4691 off <<= 2; 4692 return 4693 ((UV) tmps[off ] << 24) + 4694 ((UV) tmps[off + 1] << 16) + 4695 ((UV) tmps[off + 2] << 8) + 4696 ((UV) tmps[off + 3]); 4697 } 4698 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, " 4699 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents); 4700 NORETURN_FUNCTION_END; 4701 } 4702 4703 /* Read a single line of the main body of the swash input text. These are of 4704 * the form: 4705 * 0053 0056 0073 4706 * where each number is hex. The first two numbers form the minimum and 4707 * maximum of a range, and the third is the value associated with the range. 4708 * Not all swashes should have a third number 4709 * 4710 * On input: l points to the beginning of the line to be examined; it points 4711 * to somewhere in the string of the whole input text, and is 4712 * terminated by a \n or the null string terminator. 4713 * lend points to the null terminator of that string 4714 * wants_value is non-zero if the swash expects a third number 4715 * typestr is the name of the swash's mapping, like 'ToLower' 4716 * On output: *min, *max, and *val are set to the values read from the line. 4717 * returns a pointer just beyond the line examined. If there was no 4718 * valid min number on the line, returns lend+1 4719 */ 4720 4721 STATIC U8* 4722 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val, 4723 const bool wants_value, const U8* const typestr) 4724 { 4725 const int typeto = typestr[0] == 'T' && typestr[1] == 'o'; 4726 STRLEN numlen; /* Length of the number */ 4727 I32 flags = PERL_SCAN_SILENT_ILLDIGIT 4728 | PERL_SCAN_DISALLOW_PREFIX 4729 | PERL_SCAN_SILENT_NON_PORTABLE; 4730 4731 /* nl points to the next \n in the scan */ 4732 U8* const nl = (U8*)memchr(l, '\n', lend - l); 4733 4734 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE; 4735 4736 /* Get the first number on the line: the range minimum */ 4737 numlen = lend - l; 4738 *min = grok_hex((char *)l, &numlen, &flags, NULL); 4739 *max = *min; /* So can never return without setting max */ 4740 if (numlen) /* If found a hex number, position past it */ 4741 l += numlen; 4742 else if (nl) { /* Else, go handle next line, if any */ 4743 return nl + 1; /* 1 is length of "\n" */ 4744 } 4745 else { /* Else, no next line */ 4746 return lend + 1; /* to LIST's end at which \n is not found */ 4747 } 4748 4749 /* The max range value follows, separated by a BLANK */ 4750 if (isBLANK(*l)) { 4751 ++l; 4752 flags = PERL_SCAN_SILENT_ILLDIGIT 4753 | PERL_SCAN_DISALLOW_PREFIX 4754 | PERL_SCAN_SILENT_NON_PORTABLE; 4755 numlen = lend - l; 4756 *max = grok_hex((char *)l, &numlen, &flags, NULL); 4757 if (numlen) 4758 l += numlen; 4759 else /* If no value here, it is a single element range */ 4760 *max = *min; 4761 4762 /* Non-binary tables have a third entry: what the first element of the 4763 * range maps to. The map for those currently read here is in hex */ 4764 if (wants_value) { 4765 if (isBLANK(*l)) { 4766 ++l; 4767 flags = PERL_SCAN_SILENT_ILLDIGIT 4768 | PERL_SCAN_DISALLOW_PREFIX 4769 | PERL_SCAN_SILENT_NON_PORTABLE; 4770 numlen = lend - l; 4771 *val = grok_hex((char *)l, &numlen, &flags, NULL); 4772 if (numlen) 4773 l += numlen; 4774 else 4775 *val = 0; 4776 } 4777 else { 4778 *val = 0; 4779 if (typeto) { 4780 /* diag_listed_as: To%s: illegal mapping '%s' */ 4781 Perl_croak(aTHX_ "%s: illegal mapping '%s'", 4782 typestr, l); 4783 } 4784 } 4785 } 4786 else 4787 *val = 0; /* bits == 1, then any val should be ignored */ 4788 } 4789 else { /* Nothing following range min, should be single element with no 4790 mapping expected */ 4791 if (wants_value) { 4792 *val = 0; 4793 if (typeto) { 4794 /* diag_listed_as: To%s: illegal mapping '%s' */ 4795 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l); 4796 } 4797 } 4798 else 4799 *val = 0; /* bits == 1, then val should be ignored */ 4800 } 4801 4802 /* Position to next line if any, or EOF */ 4803 if (nl) 4804 l = nl + 1; 4805 else 4806 l = lend; 4807 4808 return l; 4809 } 4810 4811 /* Note: 4812 * Returns a swatch (a bit vector string) for a code point sequence 4813 * that starts from the value C<start> and comprises the number C<span>. 4814 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl). 4815 * Should be used via swash_fetch, which will cache the swatch in C<swash>. 4816 */ 4817 STATIC SV* 4818 S_swatch_get(pTHX_ SV* swash, UV start, UV span) 4819 { 4820 SV *swatch; 4821 U8 *l, *lend, *x, *xend, *s, *send; 4822 STRLEN lcur, xcur, scur; 4823 HV *const hv = MUTABLE_HV(SvRV(swash)); 4824 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE); 4825 4826 SV** listsvp = NULL; /* The string containing the main body of the table */ 4827 SV** extssvp = NULL; 4828 SV** invert_it_svp = NULL; 4829 U8* typestr = NULL; 4830 STRLEN bits; 4831 STRLEN octets; /* if bits == 1, then octets == 0 */ 4832 UV none; 4833 UV end = start + span; 4834 4835 if (invlistsvp == NULL) { 4836 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE); 4837 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE); 4838 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE); 4839 extssvp = hv_fetchs(hv, "EXTRAS", FALSE); 4840 listsvp = hv_fetchs(hv, "LIST", FALSE); 4841 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); 4842 4843 bits = SvUV(*bitssvp); 4844 none = SvUV(*nonesvp); 4845 typestr = (U8*)SvPV_nolen(*typesvp); 4846 } 4847 else { 4848 bits = 1; 4849 none = 0; 4850 } 4851 octets = bits >> 3; /* if bits == 1, then octets == 0 */ 4852 4853 PERL_ARGS_ASSERT_SWATCH_GET; 4854 4855 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) { 4856 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf, 4857 (UV)bits); 4858 } 4859 4860 /* If overflowed, use the max possible */ 4861 if (end < start) { 4862 end = UV_MAX; 4863 span = end - start; 4864 } 4865 4866 /* create and initialize $swatch */ 4867 scur = octets ? (span * octets) : (span + 7) / 8; 4868 swatch = newSV(scur); 4869 SvPOK_on(swatch); 4870 s = (U8*)SvPVX(swatch); 4871 if (octets && none) { 4872 const U8* const e = s + scur; 4873 while (s < e) { 4874 if (bits == 8) 4875 *s++ = (U8)(none & 0xff); 4876 else if (bits == 16) { 4877 *s++ = (U8)((none >> 8) & 0xff); 4878 *s++ = (U8)( none & 0xff); 4879 } 4880 else if (bits == 32) { 4881 *s++ = (U8)((none >> 24) & 0xff); 4882 *s++ = (U8)((none >> 16) & 0xff); 4883 *s++ = (U8)((none >> 8) & 0xff); 4884 *s++ = (U8)( none & 0xff); 4885 } 4886 } 4887 *s = '\0'; 4888 } 4889 else { 4890 (void)memzero((U8*)s, scur + 1); 4891 } 4892 SvCUR_set(swatch, scur); 4893 s = (U8*)SvPVX(swatch); 4894 4895 if (invlistsvp) { /* If has an inversion list set up use that */ 4896 _invlist_populate_swatch(*invlistsvp, start, end, s); 4897 return swatch; 4898 } 4899 4900 /* read $swash->{LIST} */ 4901 l = (U8*)SvPV(*listsvp, lcur); 4902 lend = l + lcur; 4903 while (l < lend) { 4904 UV min, max, val, upper; 4905 l = swash_scan_list_line(l, lend, &min, &max, &val, 4906 cBOOL(octets), typestr); 4907 if (l > lend) { 4908 break; 4909 } 4910 4911 /* If looking for something beyond this range, go try the next one */ 4912 if (max < start) 4913 continue; 4914 4915 /* <end> is generally 1 beyond where we want to set things, but at the 4916 * platform's infinity, where we can't go any higher, we want to 4917 * include the code point at <end> */ 4918 upper = (max < end) 4919 ? max 4920 : (max != UV_MAX || end != UV_MAX) 4921 ? end - 1 4922 : end; 4923 4924 if (octets) { 4925 UV key; 4926 if (min < start) { 4927 if (!none || val < none) { 4928 val += start - min; 4929 } 4930 min = start; 4931 } 4932 for (key = min; key <= upper; key++) { 4933 STRLEN offset; 4934 /* offset must be non-negative (start <= min <= key < end) */ 4935 offset = octets * (key - start); 4936 if (bits == 8) 4937 s[offset] = (U8)(val & 0xff); 4938 else if (bits == 16) { 4939 s[offset ] = (U8)((val >> 8) & 0xff); 4940 s[offset + 1] = (U8)( val & 0xff); 4941 } 4942 else if (bits == 32) { 4943 s[offset ] = (U8)((val >> 24) & 0xff); 4944 s[offset + 1] = (U8)((val >> 16) & 0xff); 4945 s[offset + 2] = (U8)((val >> 8) & 0xff); 4946 s[offset + 3] = (U8)( val & 0xff); 4947 } 4948 4949 if (!none || val < none) 4950 ++val; 4951 } 4952 } 4953 else { /* bits == 1, then val should be ignored */ 4954 UV key; 4955 if (min < start) 4956 min = start; 4957 4958 for (key = min; key <= upper; key++) { 4959 const STRLEN offset = (STRLEN)(key - start); 4960 s[offset >> 3] |= 1 << (offset & 7); 4961 } 4962 } 4963 } /* while */ 4964 4965 /* Invert if the data says it should be. Assumes that bits == 1 */ 4966 if (invert_it_svp && SvUV(*invert_it_svp)) { 4967 4968 /* Unicode properties should come with all bits above PERL_UNICODE_MAX 4969 * be 0, and their inversion should also be 0, as we don't succeed any 4970 * Unicode property matches for non-Unicode code points */ 4971 if (start <= PERL_UNICODE_MAX) { 4972 4973 /* The code below assumes that we never cross the 4974 * Unicode/above-Unicode boundary in a range, as otherwise we would 4975 * have to figure out where to stop flipping the bits. Since this 4976 * boundary is divisible by a large power of 2, and swatches comes 4977 * in small powers of 2, this should be a valid assumption */ 4978 assert(start + span - 1 <= PERL_UNICODE_MAX); 4979 4980 send = s + scur; 4981 while (s < send) { 4982 *s = ~(*s); 4983 s++; 4984 } 4985 } 4986 } 4987 4988 /* read $swash->{EXTRAS} 4989 * This code also copied to swash_to_invlist() below */ 4990 x = (U8*)SvPV(*extssvp, xcur); 4991 xend = x + xcur; 4992 while (x < xend) { 4993 STRLEN namelen; 4994 U8 *namestr; 4995 SV** othersvp; 4996 HV* otherhv; 4997 STRLEN otherbits; 4998 SV **otherbitssvp, *other; 4999 U8 *s, *o, *nl; 5000 STRLEN slen, olen; 5001 5002 const U8 opc = *x++; 5003 if (opc == '\n') 5004 continue; 5005 5006 nl = (U8*)memchr(x, '\n', xend - x); 5007 5008 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { 5009 if (nl) { 5010 x = nl + 1; /* 1 is length of "\n" */ 5011 continue; 5012 } 5013 else { 5014 x = xend; /* to EXTRAS' end at which \n is not found */ 5015 break; 5016 } 5017 } 5018 5019 namestr = x; 5020 if (nl) { 5021 namelen = nl - namestr; 5022 x = nl + 1; 5023 } 5024 else { 5025 namelen = xend - namestr; 5026 x = xend; 5027 } 5028 5029 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); 5030 otherhv = MUTABLE_HV(SvRV(*othersvp)); 5031 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); 5032 otherbits = (STRLEN)SvUV(*otherbitssvp); 5033 if (bits < otherbits) 5034 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, " 5035 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits); 5036 5037 /* The "other" swatch must be destroyed after. */ 5038 other = swatch_get(*othersvp, start, span); 5039 o = (U8*)SvPV(other, olen); 5040 5041 if (!olen) 5042 Perl_croak(aTHX_ "panic: swatch_get got improper swatch"); 5043 5044 s = (U8*)SvPV(swatch, slen); 5045 if (bits == 1 && otherbits == 1) { 5046 if (slen != olen) 5047 Perl_croak(aTHX_ "panic: swatch_get found swatch length " 5048 "mismatch, slen=%" UVuf ", olen=%" UVuf, 5049 (UV)slen, (UV)olen); 5050 5051 switch (opc) { 5052 case '+': 5053 while (slen--) 5054 *s++ |= *o++; 5055 break; 5056 case '!': 5057 while (slen--) 5058 *s++ |= ~*o++; 5059 break; 5060 case '-': 5061 while (slen--) 5062 *s++ &= ~*o++; 5063 break; 5064 case '&': 5065 while (slen--) 5066 *s++ &= *o++; 5067 break; 5068 default: 5069 break; 5070 } 5071 } 5072 else { 5073 STRLEN otheroctets = otherbits >> 3; 5074 STRLEN offset = 0; 5075 U8* const send = s + slen; 5076 5077 while (s < send) { 5078 UV otherval = 0; 5079 5080 if (otherbits == 1) { 5081 otherval = (o[offset >> 3] >> (offset & 7)) & 1; 5082 ++offset; 5083 } 5084 else { 5085 STRLEN vlen = otheroctets; 5086 otherval = *o++; 5087 while (--vlen) { 5088 otherval <<= 8; 5089 otherval |= *o++; 5090 } 5091 } 5092 5093 if (opc == '+' && otherval) 5094 NOOP; /* replace with otherval */ 5095 else if (opc == '!' && !otherval) 5096 otherval = 1; 5097 else if (opc == '-' && otherval) 5098 otherval = 0; 5099 else if (opc == '&' && !otherval) 5100 otherval = 0; 5101 else { 5102 s += octets; /* no replacement */ 5103 continue; 5104 } 5105 5106 if (bits == 8) 5107 *s++ = (U8)( otherval & 0xff); 5108 else if (bits == 16) { 5109 *s++ = (U8)((otherval >> 8) & 0xff); 5110 *s++ = (U8)( otherval & 0xff); 5111 } 5112 else if (bits == 32) { 5113 *s++ = (U8)((otherval >> 24) & 0xff); 5114 *s++ = (U8)((otherval >> 16) & 0xff); 5115 *s++ = (U8)((otherval >> 8) & 0xff); 5116 *s++ = (U8)( otherval & 0xff); 5117 } 5118 } 5119 } 5120 sv_free(other); /* through with it! */ 5121 } /* while */ 5122 return swatch; 5123 } 5124 5125 SV* 5126 Perl__swash_to_invlist(pTHX_ SV* const swash) 5127 { 5128 5129 /* Subject to change or removal. For use only in one place in regcomp.c. 5130 * Ownership is given to one reference count in the returned SV* */ 5131 5132 U8 *l, *lend; 5133 char *loc; 5134 STRLEN lcur; 5135 HV *const hv = MUTABLE_HV(SvRV(swash)); 5136 UV elements = 0; /* Number of elements in the inversion list */ 5137 U8 empty[] = ""; 5138 SV** listsvp; 5139 SV** typesvp; 5140 SV** bitssvp; 5141 SV** extssvp; 5142 SV** invert_it_svp; 5143 5144 U8* typestr; 5145 STRLEN bits; 5146 STRLEN octets; /* if bits == 1, then octets == 0 */ 5147 U8 *x, *xend; 5148 STRLEN xcur; 5149 5150 SV* invlist; 5151 5152 PERL_ARGS_ASSERT__SWASH_TO_INVLIST; 5153 5154 /* If not a hash, it must be the swash's inversion list instead */ 5155 if (SvTYPE(hv) != SVt_PVHV) { 5156 return SvREFCNT_inc_simple_NN((SV*) hv); 5157 } 5158 5159 /* The string containing the main body of the table */ 5160 listsvp = hv_fetchs(hv, "LIST", FALSE); 5161 typesvp = hv_fetchs(hv, "TYPE", FALSE); 5162 bitssvp = hv_fetchs(hv, "BITS", FALSE); 5163 extssvp = hv_fetchs(hv, "EXTRAS", FALSE); 5164 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); 5165 5166 typestr = (U8*)SvPV_nolen(*typesvp); 5167 bits = SvUV(*bitssvp); 5168 octets = bits >> 3; /* if bits == 1, then octets == 0 */ 5169 5170 /* read $swash->{LIST} */ 5171 if (SvPOK(*listsvp)) { 5172 l = (U8*)SvPV(*listsvp, lcur); 5173 } 5174 else { 5175 /* LIST legitimately doesn't contain a string during compilation phases 5176 * of Perl itself, before the Unicode tables are generated. In this 5177 * case, just fake things up by creating an empty list */ 5178 l = empty; 5179 lcur = 0; 5180 } 5181 loc = (char *) l; 5182 lend = l + lcur; 5183 5184 if (*l == 'V') { /* Inversion list format */ 5185 const char *after_atou = (char *) lend; 5186 UV element0; 5187 UV* other_elements_ptr; 5188 5189 /* The first number is a count of the rest */ 5190 l++; 5191 if (!grok_atoUV((const char *)l, &elements, &after_atou)) { 5192 Perl_croak(aTHX_ "panic: Expecting a valid count of elements" 5193 " at start of inversion list"); 5194 } 5195 if (elements == 0) { 5196 invlist = _new_invlist(0); 5197 } 5198 else { 5199 l = (U8 *) after_atou; 5200 5201 /* Get the 0th element, which is needed to setup the inversion list 5202 * */ 5203 while (isSPACE(*l)) l++; 5204 if (!grok_atoUV((const char *)l, &element0, &after_atou)) { 5205 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for" 5206 " inversion list"); 5207 } 5208 l = (U8 *) after_atou; 5209 invlist = _setup_canned_invlist(elements, element0, 5210 &other_elements_ptr); 5211 elements--; 5212 5213 /* Then just populate the rest of the input */ 5214 while (elements-- > 0) { 5215 if (l > lend) { 5216 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more" 5217 " elements than available", elements); 5218 } 5219 while (isSPACE(*l)) l++; 5220 if (!grok_atoUV((const char *)l, other_elements_ptr++, 5221 &after_atou)) 5222 { 5223 Perl_croak(aTHX_ "panic: Expecting a valid element" 5224 " in inversion list"); 5225 } 5226 l = (U8 *) after_atou; 5227 } 5228 } 5229 } 5230 else { 5231 5232 /* Scan the input to count the number of lines to preallocate array 5233 * size based on worst possible case, which is each line in the input 5234 * creates 2 elements in the inversion list: 1) the beginning of a 5235 * range in the list; 2) the beginning of a range not in the list. */ 5236 while ((loc = (char *) memchr(loc, '\n', lend - (U8 *) loc)) != NULL) { 5237 elements += 2; 5238 loc++; 5239 } 5240 5241 /* If the ending is somehow corrupt and isn't a new line, add another 5242 * element for the final range that isn't in the inversion list */ 5243 if (! (*lend == '\n' 5244 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n')))) 5245 { 5246 elements++; 5247 } 5248 5249 invlist = _new_invlist(elements); 5250 5251 /* Now go through the input again, adding each range to the list */ 5252 while (l < lend) { 5253 UV start, end; 5254 UV val; /* Not used by this function */ 5255 5256 l = swash_scan_list_line(l, lend, &start, &end, &val, 5257 cBOOL(octets), typestr); 5258 5259 if (l > lend) { 5260 break; 5261 } 5262 5263 invlist = _add_range_to_invlist(invlist, start, end); 5264 } 5265 } 5266 5267 /* Invert if the data says it should be */ 5268 if (invert_it_svp && SvUV(*invert_it_svp)) { 5269 _invlist_invert(invlist); 5270 } 5271 5272 /* This code is copied from swatch_get() 5273 * read $swash->{EXTRAS} */ 5274 x = (U8*)SvPV(*extssvp, xcur); 5275 xend = x + xcur; 5276 while (x < xend) { 5277 STRLEN namelen; 5278 U8 *namestr; 5279 SV** othersvp; 5280 HV* otherhv; 5281 STRLEN otherbits; 5282 SV **otherbitssvp, *other; 5283 U8 *nl; 5284 5285 const U8 opc = *x++; 5286 if (opc == '\n') 5287 continue; 5288 5289 nl = (U8*)memchr(x, '\n', xend - x); 5290 5291 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { 5292 if (nl) { 5293 x = nl + 1; /* 1 is length of "\n" */ 5294 continue; 5295 } 5296 else { 5297 x = xend; /* to EXTRAS' end at which \n is not found */ 5298 break; 5299 } 5300 } 5301 5302 namestr = x; 5303 if (nl) { 5304 namelen = nl - namestr; 5305 x = nl + 1; 5306 } 5307 else { 5308 namelen = xend - namestr; 5309 x = xend; 5310 } 5311 5312 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); 5313 otherhv = MUTABLE_HV(SvRV(*othersvp)); 5314 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); 5315 otherbits = (STRLEN)SvUV(*otherbitssvp); 5316 5317 if (bits != otherbits || bits != 1) { 5318 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean " 5319 "properties, bits=%" UVuf ", otherbits=%" UVuf, 5320 (UV)bits, (UV)otherbits); 5321 } 5322 5323 /* The "other" swatch must be destroyed after. */ 5324 other = _swash_to_invlist((SV *)*othersvp); 5325 5326 /* End of code copied from swatch_get() */ 5327 switch (opc) { 5328 case '+': 5329 _invlist_union(invlist, other, &invlist); 5330 break; 5331 case '!': 5332 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist); 5333 break; 5334 case '-': 5335 _invlist_subtract(invlist, other, &invlist); 5336 break; 5337 case '&': 5338 _invlist_intersection(invlist, other, &invlist); 5339 break; 5340 default: 5341 break; 5342 } 5343 sv_free(other); /* through with it! */ 5344 } 5345 5346 SvREADONLY_on(invlist); 5347 return invlist; 5348 } 5349 5350 SV* 5351 Perl__get_swash_invlist(pTHX_ SV* const swash) 5352 { 5353 SV** ptr; 5354 5355 PERL_ARGS_ASSERT__GET_SWASH_INVLIST; 5356 5357 if (! SvROK(swash)) { 5358 return NULL; 5359 } 5360 5361 /* If it really isn't a hash, it isn't really swash; must be an inversion 5362 * list */ 5363 if (SvTYPE(SvRV(swash)) != SVt_PVHV) { 5364 return SvRV(swash); 5365 } 5366 5367 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE); 5368 if (! ptr) { 5369 return NULL; 5370 } 5371 5372 return *ptr; 5373 } 5374 5375 bool 5376 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len) 5377 { 5378 /* May change: warns if surrogates, non-character code points, or 5379 * non-Unicode code points are in 's' which has length 'len' bytes. 5380 * Returns TRUE if none found; FALSE otherwise. The only other validity 5381 * check is to make sure that this won't exceed the string's length nor 5382 * overflow */ 5383 5384 const U8* const e = s + len; 5385 bool ok = TRUE; 5386 5387 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT; 5388 5389 while (s < e) { 5390 if (UTF8SKIP(s) > len) { 5391 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 5392 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print"); 5393 return FALSE; 5394 } 5395 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) { 5396 if (UNLIKELY(UTF8_IS_SUPER(s, e))) { 5397 if ( ckWARN_d(WARN_NON_UNICODE) 5398 || UNLIKELY(0 < does_utf8_overflow(s, s + len, 5399 0 /* Don't consider overlongs */ 5400 ))) 5401 { 5402 /* A side effect of this function will be to warn */ 5403 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER); 5404 ok = FALSE; 5405 } 5406 } 5407 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) { 5408 if (ckWARN_d(WARN_SURROGATE)) { 5409 /* This has a different warning than the one the called 5410 * function would output, so can't just call it, unlike we 5411 * do for the non-chars and above-unicodes */ 5412 UV uv = utf8_to_uvchr_buf(s, e, NULL); 5413 Perl_warner(aTHX_ packWARN(WARN_SURROGATE), 5414 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", 5415 uv); 5416 ok = FALSE; 5417 } 5418 } 5419 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e)) 5420 && (ckWARN_d(WARN_NONCHAR))) 5421 { 5422 /* A side effect of this function will be to warn */ 5423 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR); 5424 ok = FALSE; 5425 } 5426 } 5427 s += UTF8SKIP(s); 5428 } 5429 5430 return ok; 5431 } 5432 5433 /* 5434 =for apidoc pv_uni_display 5435 5436 Build to the scalar C<dsv> a displayable version of the string C<spv>, 5437 length C<len>, the displayable version being at most C<pvlim> bytes long 5438 (if longer, the rest is truncated and C<"..."> will be appended). 5439 5440 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display 5441 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH> 5442 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">) 5443 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">). 5444 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both 5445 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on. 5446 5447 The pointer to the PV of the C<dsv> is returned. 5448 5449 See also L</sv_uni_display>. 5450 5451 =cut */ 5452 char * 5453 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, 5454 UV flags) 5455 { 5456 int truncated = 0; 5457 const char *s, *e; 5458 5459 PERL_ARGS_ASSERT_PV_UNI_DISPLAY; 5460 5461 SvPVCLEAR(dsv); 5462 SvUTF8_off(dsv); 5463 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) { 5464 UV u; 5465 /* This serves double duty as a flag and a character to print after 5466 a \ when flags & UNI_DISPLAY_BACKSLASH is true. 5467 */ 5468 char ok = 0; 5469 5470 if (pvlim && SvCUR(dsv) >= pvlim) { 5471 truncated++; 5472 break; 5473 } 5474 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0); 5475 if (u < 256) { 5476 const unsigned char c = (unsigned char)u & 0xFF; 5477 if (flags & UNI_DISPLAY_BACKSLASH) { 5478 switch (c) { 5479 case '\n': 5480 ok = 'n'; break; 5481 case '\r': 5482 ok = 'r'; break; 5483 case '\t': 5484 ok = 't'; break; 5485 case '\f': 5486 ok = 'f'; break; 5487 case '\a': 5488 ok = 'a'; break; 5489 case '\\': 5490 ok = '\\'; break; 5491 default: break; 5492 } 5493 if (ok) { 5494 const char string = ok; 5495 sv_catpvs(dsv, "\\"); 5496 sv_catpvn(dsv, &string, 1); 5497 } 5498 } 5499 /* isPRINT() is the locale-blind version. */ 5500 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) { 5501 const char string = c; 5502 sv_catpvn(dsv, &string, 1); 5503 ok = 1; 5504 } 5505 } 5506 if (!ok) 5507 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u); 5508 } 5509 if (truncated) 5510 sv_catpvs(dsv, "..."); 5511 5512 return SvPVX(dsv); 5513 } 5514 5515 /* 5516 =for apidoc sv_uni_display 5517 5518 Build to the scalar C<dsv> a displayable version of the scalar C<sv>, 5519 the displayable version being at most C<pvlim> bytes long 5520 (if longer, the rest is truncated and "..." will be appended). 5521 5522 The C<flags> argument is as in L</pv_uni_display>(). 5523 5524 The pointer to the PV of the C<dsv> is returned. 5525 5526 =cut 5527 */ 5528 char * 5529 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags) 5530 { 5531 const char * const ptr = 5532 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv); 5533 5534 PERL_ARGS_ASSERT_SV_UNI_DISPLAY; 5535 5536 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr, 5537 SvCUR(ssv), pvlim, flags); 5538 } 5539 5540 /* 5541 =for apidoc foldEQ_utf8 5542 5543 Returns true if the leading portions of the strings C<s1> and C<s2> (either or 5544 both of which may be in UTF-8) are the same case-insensitively; false 5545 otherwise. How far into the strings to compare is determined by other input 5546 parameters. 5547 5548 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode; 5549 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for 5550 C<u2> with respect to C<s2>. 5551 5552 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for 5553 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach. 5554 The scan will not be considered to be a match unless the goal is reached, and 5555 scanning won't continue past that goal. Correspondingly for C<l2> with respect 5556 to C<s2>. 5557 5558 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that 5559 pointer is considered an end pointer to the position 1 byte past the maximum 5560 point in C<s1> beyond which scanning will not continue under any circumstances. 5561 (This routine assumes that UTF-8 encoded input strings are not malformed; 5562 malformed input can cause it to read past C<pe1>). This means that if both 5563 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match 5564 will never be successful because it can never 5565 get as far as its goal (and in fact is asserted against). Correspondingly for 5566 C<pe2> with respect to C<s2>. 5567 5568 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and 5569 C<l2> must be non-zero), and if both do, both have to be 5570 reached for a successful match. Also, if the fold of a character is multiple 5571 characters, all of them must be matched (see tr21 reference below for 5572 'folding'). 5573 5574 Upon a successful match, if C<pe1> is non-C<NULL>, 5575 it will be set to point to the beginning of the I<next> character of C<s1> 5576 beyond what was matched. Correspondingly for C<pe2> and C<s2>. 5577 5578 For case-insensitiveness, the "casefolding" of Unicode is used 5579 instead of upper/lowercasing both the characters, see 5580 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings). 5581 5582 =cut */ 5583 5584 /* A flags parameter has been added which may change, and hence isn't 5585 * externally documented. Currently it is: 5586 * 0 for as-documented above 5587 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an 5588 ASCII one, to not match 5589 * FOLDEQ_LOCALE is set iff the rules from the current underlying 5590 * locale are to be used. 5591 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this 5592 * routine. This allows that step to be skipped. 5593 * Currently, this requires s1 to be encoded as UTF-8 5594 * (u1 must be true), which is asserted for. 5595 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may 5596 * cross certain boundaries. Hence, the caller should 5597 * let this function do the folding instead of 5598 * pre-folding. This code contains an assertion to 5599 * that effect. However, if the caller knows what 5600 * it's doing, it can pass this flag to indicate that, 5601 * and the assertion is skipped. 5602 * FOLDEQ_S2_ALREADY_FOLDED Similarly. 5603 * FOLDEQ_S2_FOLDS_SANE 5604 */ 5605 I32 5606 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, 5607 const char *s2, char **pe2, UV l2, bool u2, 5608 U32 flags) 5609 { 5610 const U8 *p1 = (const U8*)s1; /* Point to current char */ 5611 const U8 *p2 = (const U8*)s2; 5612 const U8 *g1 = NULL; /* goal for s1 */ 5613 const U8 *g2 = NULL; 5614 const U8 *e1 = NULL; /* Don't scan s1 past this */ 5615 U8 *f1 = NULL; /* Point to current folded */ 5616 const U8 *e2 = NULL; 5617 U8 *f2 = NULL; 5618 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */ 5619 U8 foldbuf1[UTF8_MAXBYTES_CASE+1]; 5620 U8 foldbuf2[UTF8_MAXBYTES_CASE+1]; 5621 U8 flags_for_folder = FOLD_FLAGS_FULL; 5622 5623 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS; 5624 5625 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE)) 5626 && (((flags & FOLDEQ_S1_ALREADY_FOLDED) 5627 && !(flags & FOLDEQ_S1_FOLDS_SANE)) 5628 || ((flags & FOLDEQ_S2_ALREADY_FOLDED) 5629 && !(flags & FOLDEQ_S2_FOLDS_SANE))))); 5630 /* The algorithm is to trial the folds without regard to the flags on 5631 * the first line of the above assert(), and then see if the result 5632 * violates them. This means that the inputs can't be pre-folded to a 5633 * violating result, hence the assert. This could be changed, with the 5634 * addition of extra tests here for the already-folded case, which would 5635 * slow it down. That cost is more than any possible gain for when these 5636 * flags are specified, as the flags indicate /il or /iaa matching which 5637 * is less common than /iu, and I (khw) also believe that real-world /il 5638 * and /iaa matches are most likely to involve code points 0-255, and this 5639 * function only under rare conditions gets called for 0-255. */ 5640 5641 if (flags & FOLDEQ_LOCALE) { 5642 if (IN_UTF8_CTYPE_LOCALE) { 5643 flags &= ~FOLDEQ_LOCALE; 5644 } 5645 else { 5646 flags_for_folder |= FOLD_FLAGS_LOCALE; 5647 } 5648 } 5649 5650 if (pe1) { 5651 e1 = *(U8**)pe1; 5652 } 5653 5654 if (l1) { 5655 g1 = (const U8*)s1 + l1; 5656 } 5657 5658 if (pe2) { 5659 e2 = *(U8**)pe2; 5660 } 5661 5662 if (l2) { 5663 g2 = (const U8*)s2 + l2; 5664 } 5665 5666 /* Must have at least one goal */ 5667 assert(g1 || g2); 5668 5669 if (g1) { 5670 5671 /* Will never match if goal is out-of-bounds */ 5672 assert(! e1 || e1 >= g1); 5673 5674 /* Here, there isn't an end pointer, or it is beyond the goal. We 5675 * only go as far as the goal */ 5676 e1 = g1; 5677 } 5678 else { 5679 assert(e1); /* Must have an end for looking at s1 */ 5680 } 5681 5682 /* Same for goal for s2 */ 5683 if (g2) { 5684 assert(! e2 || e2 >= g2); 5685 e2 = g2; 5686 } 5687 else { 5688 assert(e2); 5689 } 5690 5691 /* If both operands are already folded, we could just do a memEQ on the 5692 * whole strings at once, but it would be better if the caller realized 5693 * this and didn't even call us */ 5694 5695 /* Look through both strings, a character at a time */ 5696 while (p1 < e1 && p2 < e2) { 5697 5698 /* If at the beginning of a new character in s1, get its fold to use 5699 * and the length of the fold. */ 5700 if (n1 == 0) { 5701 if (flags & FOLDEQ_S1_ALREADY_FOLDED) { 5702 f1 = (U8 *) p1; 5703 assert(u1); 5704 n1 = UTF8SKIP(f1); 5705 } 5706 else { 5707 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) { 5708 5709 /* We have to forbid mixing ASCII with non-ASCII if the 5710 * flags so indicate. And, we can short circuit having to 5711 * call the general functions for this common ASCII case, 5712 * all of whose non-locale folds are also ASCII, and hence 5713 * UTF-8 invariants, so the UTF8ness of the strings is not 5714 * relevant. */ 5715 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) { 5716 return 0; 5717 } 5718 n1 = 1; 5719 *foldbuf1 = toFOLD(*p1); 5720 } 5721 else if (u1) { 5722 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder); 5723 } 5724 else { /* Not UTF-8, get UTF-8 fold */ 5725 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder); 5726 } 5727 f1 = foldbuf1; 5728 } 5729 } 5730 5731 if (n2 == 0) { /* Same for s2 */ 5732 if (flags & FOLDEQ_S2_ALREADY_FOLDED) { 5733 f2 = (U8 *) p2; 5734 assert(u2); 5735 n2 = UTF8SKIP(f2); 5736 } 5737 else { 5738 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) { 5739 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) { 5740 return 0; 5741 } 5742 n2 = 1; 5743 *foldbuf2 = toFOLD(*p2); 5744 } 5745 else if (u2) { 5746 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder); 5747 } 5748 else { 5749 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder); 5750 } 5751 f2 = foldbuf2; 5752 } 5753 } 5754 5755 /* Here f1 and f2 point to the beginning of the strings to compare. 5756 * These strings are the folds of the next character from each input 5757 * string, stored in UTF-8. */ 5758 5759 /* While there is more to look for in both folds, see if they 5760 * continue to match */ 5761 while (n1 && n2) { 5762 U8 fold_length = UTF8SKIP(f1); 5763 if (fold_length != UTF8SKIP(f2) 5764 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE 5765 function call for single 5766 byte */ 5767 || memNE((char*)f1, (char*)f2, fold_length)) 5768 { 5769 return 0; /* mismatch */ 5770 } 5771 5772 /* Here, they matched, advance past them */ 5773 n1 -= fold_length; 5774 f1 += fold_length; 5775 n2 -= fold_length; 5776 f2 += fold_length; 5777 } 5778 5779 /* When reach the end of any fold, advance the input past it */ 5780 if (n1 == 0) { 5781 p1 += u1 ? UTF8SKIP(p1) : 1; 5782 } 5783 if (n2 == 0) { 5784 p2 += u2 ? UTF8SKIP(p2) : 1; 5785 } 5786 } /* End of loop through both strings */ 5787 5788 /* A match is defined by each scan that specified an explicit length 5789 * reaching its final goal, and the other not having matched a partial 5790 * character (which can happen when the fold of a character is more than one 5791 * character). */ 5792 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) { 5793 return 0; 5794 } 5795 5796 /* Successful match. Set output pointers */ 5797 if (pe1) { 5798 *pe1 = (char*)p1; 5799 } 5800 if (pe2) { 5801 *pe2 = (char*)p2; 5802 } 5803 return 1; 5804 } 5805 5806 /* XXX The next two functions should likely be moved to mathoms.c once all 5807 * occurrences of them are removed from the core; some cpan-upstream modules 5808 * still use them */ 5809 5810 U8 * 5811 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv) 5812 { 5813 PERL_ARGS_ASSERT_UVUNI_TO_UTF8; 5814 5815 return uvoffuni_to_utf8_flags(d, uv, 0); 5816 } 5817 5818 /* 5819 =for apidoc utf8n_to_uvuni 5820 5821 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>. 5822 5823 This function was useful for code that wanted to handle both EBCDIC and 5824 ASCII platforms with Unicode properties, but starting in Perl v5.20, the 5825 distinctions between the platforms have mostly been made invisible to most 5826 code, so this function is quite unlikely to be what you want. If you do need 5827 this precise functionality, use instead 5828 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> 5829 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>. 5830 5831 =cut 5832 */ 5833 5834 UV 5835 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags) 5836 { 5837 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI; 5838 5839 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags)); 5840 } 5841 5842 /* 5843 =for apidoc uvuni_to_utf8_flags 5844 5845 Instead you almost certainly want to use L</uvchr_to_utf8> or 5846 L</uvchr_to_utf8_flags>. 5847 5848 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>, 5849 which itself, while not deprecated, should be used only in isolated 5850 circumstances. These functions were useful for code that wanted to handle 5851 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl 5852 v5.20, the distinctions between the platforms have mostly been made invisible 5853 to most code, so this function is quite unlikely to be what you want. 5854 5855 =cut 5856 */ 5857 5858 U8 * 5859 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) 5860 { 5861 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS; 5862 5863 return uvoffuni_to_utf8_flags(d, uv, flags); 5864 } 5865 5866 void 5867 Perl_init_uniprops(pTHX) 5868 { 5869 /* Set up the inversion list global variables */ 5870 5871 PL_XPosix_ptrs[_CC_ASCII] = _new_invlist_C_array(PL_ASCII_invlist); 5872 PL_XPosix_ptrs[_CC_ALPHANUMERIC] = _new_invlist_C_array(PL_XPOSIXALNUM_invlist); 5873 PL_XPosix_ptrs[_CC_ALPHA] = _new_invlist_C_array(PL_XPOSIXALPHA_invlist); 5874 PL_XPosix_ptrs[_CC_BLANK] = _new_invlist_C_array(PL_XPOSIXBLANK_invlist); 5875 PL_XPosix_ptrs[_CC_CASED] = _new_invlist_C_array(PL_CASED_invlist); 5876 PL_XPosix_ptrs[_CC_CNTRL] = _new_invlist_C_array(PL_XPOSIXCNTRL_invlist); 5877 PL_XPosix_ptrs[_CC_DIGIT] = _new_invlist_C_array(PL_XPOSIXDIGIT_invlist); 5878 PL_XPosix_ptrs[_CC_GRAPH] = _new_invlist_C_array(PL_XPOSIXGRAPH_invlist); 5879 PL_XPosix_ptrs[_CC_LOWER] = _new_invlist_C_array(PL_XPOSIXLOWER_invlist); 5880 PL_XPosix_ptrs[_CC_PRINT] = _new_invlist_C_array(PL_XPOSIXPRINT_invlist); 5881 PL_XPosix_ptrs[_CC_PUNCT] = _new_invlist_C_array(PL_XPOSIXPUNCT_invlist); 5882 PL_XPosix_ptrs[_CC_SPACE] = _new_invlist_C_array(PL_XPOSIXSPACE_invlist); 5883 PL_XPosix_ptrs[_CC_UPPER] = _new_invlist_C_array(PL_XPOSIXUPPER_invlist); 5884 PL_XPosix_ptrs[_CC_VERTSPACE] = _new_invlist_C_array(PL_VERTSPACE_invlist); 5885 PL_XPosix_ptrs[_CC_WORDCHAR] = _new_invlist_C_array(PL_XPOSIXWORD_invlist); 5886 PL_XPosix_ptrs[_CC_XDIGIT] = _new_invlist_C_array(PL_XPOSIXXDIGIT_invlist); 5887 PL_GCB_invlist = _new_invlist_C_array(_Perl_GCB_invlist); 5888 PL_SB_invlist = _new_invlist_C_array(_Perl_SB_invlist); 5889 PL_WB_invlist = _new_invlist_C_array(_Perl_WB_invlist); 5890 PL_LB_invlist = _new_invlist_C_array(_Perl_LB_invlist); 5891 PL_Assigned_invlist = _new_invlist_C_array(PL_ASSIGNED_invlist); 5892 PL_SCX_invlist = _new_invlist_C_array(_Perl_SCX_invlist); 5893 PL_utf8_toupper = _new_invlist_C_array(Uppercase_Mapping_invlist); 5894 PL_utf8_tolower = _new_invlist_C_array(Lowercase_Mapping_invlist); 5895 PL_utf8_totitle = _new_invlist_C_array(Titlecase_Mapping_invlist); 5896 PL_utf8_tofold = _new_invlist_C_array(Case_Folding_invlist); 5897 PL_utf8_tosimplefold = _new_invlist_C_array(Simple_Case_Folding_invlist); 5898 PL_utf8_perl_idstart = _new_invlist_C_array(PL__PERL_IDSTART_invlist); 5899 PL_utf8_perl_idcont = _new_invlist_C_array(PL__PERL_IDCONT_invlist); 5900 PL_AboveLatin1 = _new_invlist_C_array(AboveLatin1_invlist); 5901 PL_Latin1 = _new_invlist_C_array(Latin1_invlist); 5902 PL_UpperLatin1 = _new_invlist_C_array(UpperLatin1_invlist); 5903 PL_utf8_foldable = _new_invlist_C_array(PL__PERL_ANY_FOLDS_invlist); 5904 PL_HasMultiCharFold = _new_invlist_C_array( 5905 PL__PERL_FOLDS_TO_MULTI_CHAR_invlist); 5906 PL_NonL1NonFinalFold = _new_invlist_C_array( 5907 NonL1_Perl_Non_Final_Folds_invlist); 5908 PL_utf8_charname_begin = _new_invlist_C_array(PL__PERL_CHARNAME_BEGIN_invlist); 5909 PL_utf8_charname_continue = _new_invlist_C_array(PL__PERL_CHARNAME_CONTINUE_invlist); 5910 PL_utf8_foldclosures = _new_invlist_C_array(_Perl_IVCF_invlist); 5911 } 5912 5913 SV * 5914 Perl_parse_uniprop_string(pTHX_ const char * const name, const Size_t len, const bool to_fold, bool * invert) 5915 { 5916 /* Parse the interior meat of \p{} passed to this in 'name' with length 'len', 5917 * and return an inversion list if a property with 'name' is found, or NULL 5918 * if not. 'name' point to the input with leading and trailing space trimmed. 5919 * 'to_fold' indicates if /i is in effect. 5920 * 5921 * When the return is an inversion list, '*invert' will be set to a boolean 5922 * indicating if it should be inverted or not 5923 * 5924 * This currently doesn't handle all cases. A NULL return indicates the 5925 * caller should try a different approach 5926 */ 5927 5928 char* lookup_name; 5929 bool stricter = FALSE; 5930 unsigned int i; 5931 unsigned int j = 0; 5932 int equals_pos = -1; /* Where the '=' is found, or negative if none */ 5933 int table_index = 0; 5934 bool starts_with_In_or_Is = FALSE; 5935 Size_t lookup_offset = 0; 5936 5937 PERL_ARGS_ASSERT_PARSE_UNIPROP_STRING; 5938 5939 /* The input will be modified into 'lookup_name' */ 5940 Newx(lookup_name, len, char); 5941 SAVEFREEPV(lookup_name); 5942 5943 /* Parse the input. */ 5944 for (i = 0; i < len; i++) { 5945 char cur = name[i]; 5946 5947 /* These characters can be freely ignored in most situations. Later it 5948 * may turn out we shouldn't have ignored them, and we have to reparse, 5949 * but we don't have enough information yet to make that decision */ 5950 if (cur == '-' || cur == '_' || isSPACE(cur)) { 5951 continue; 5952 } 5953 5954 /* Case differences are also ignored. Our lookup routine assumes 5955 * everything is lowercase */ 5956 if (isUPPER(cur)) { 5957 lookup_name[j++] = toLOWER(cur); 5958 continue; 5959 } 5960 5961 /* A double colon is either an error, or a package qualifier to a 5962 * subroutine user-defined property; neither of which do we currently 5963 * handle 5964 * 5965 * But a single colon is a synonym for '=' */ 5966 if (cur == ':') { 5967 if (i < len - 1 && name[i+1] == ':') { 5968 return NULL; 5969 } 5970 cur = '='; 5971 } 5972 5973 /* Otherwise, this character is part of the name. */ 5974 lookup_name[j++] = cur; 5975 5976 /* Only the equals sign needs further processing */ 5977 if (cur == '=') { 5978 equals_pos = j; /* Note where it occurred in the input */ 5979 break; 5980 } 5981 } 5982 5983 /* Here, we are either done with the whole property name, if it was simple; 5984 * or are positioned just after the '=' if it is compound. */ 5985 5986 if (equals_pos >= 0) { 5987 assert(! stricter); /* We shouldn't have set this yet */ 5988 5989 /* Space immediately after the '=' is ignored */ 5990 i++; 5991 for (; i < len; i++) { 5992 if (! isSPACE(name[i])) { 5993 break; 5994 } 5995 } 5996 5997 /* Certain properties need special handling. They may optionally be 5998 * prefixed by 'is'. Ignore that prefix for the purposes of checking 5999 * if this is one of those properties */ 6000 if (memBEGINPs(lookup_name, len, "is")) { 6001 lookup_offset = 2; 6002 } 6003 6004 /* Then check if it is one of these properties. This is hard-coded 6005 * because easier this way, and the list is unlikely to change */ 6006 if ( memEQs(lookup_name + lookup_offset, 6007 j - 1 - lookup_offset, "canonicalcombiningclass") 6008 || memEQs(lookup_name + lookup_offset, 6009 j - 1 - lookup_offset, "ccc") 6010 || memEQs(lookup_name + lookup_offset, 6011 j - 1 - lookup_offset, "numericvalue") 6012 || memEQs(lookup_name + lookup_offset, 6013 j - 1 - lookup_offset, "nv") 6014 || memEQs(lookup_name + lookup_offset, 6015 j - 1 - lookup_offset, "age") 6016 || memEQs(lookup_name + lookup_offset, 6017 j - 1 - lookup_offset, "in") 6018 || memEQs(lookup_name + lookup_offset, 6019 j - 1 - lookup_offset, "presentin")) 6020 { 6021 unsigned int k; 6022 6023 /* What makes these properties special is that the stuff after the 6024 * '=' is a number. Therefore, we can't throw away '-' 6025 * willy-nilly, as those could be a minus sign. Other stricter 6026 * rules also apply. However, these properties all can have the 6027 * rhs not be a number, in which case they contain at least one 6028 * alphabetic. In those cases, the stricter rules don't apply. We 6029 * first parse to look for alphas */ 6030 stricter = TRUE; 6031 for (k = i; k < len; k++) { 6032 if (isALPHA(name[k])) { 6033 stricter = FALSE; 6034 break; 6035 } 6036 } 6037 } 6038 6039 if (stricter) { 6040 6041 /* A number may have a leading '+' or '-'. The latter is retained 6042 * */ 6043 if (name[i] == '+') { 6044 i++; 6045 } 6046 else if (name[i] == '-') { 6047 lookup_name[j++] = '-'; 6048 i++; 6049 } 6050 6051 /* Skip leading zeros including single underscores separating the 6052 * zeros, or between the final leading zero and the first other 6053 * digit */ 6054 for (; i < len - 1; i++) { 6055 if ( name[i] != '0' 6056 && (name[i] != '_' || ! isDIGIT(name[i+1]))) 6057 { 6058 break; 6059 } 6060 } 6061 } 6062 } 6063 else { /* No '=' */ 6064 6065 /* We are now in a position to determine if this property should have 6066 * been parsed using stricter rules. Only a few are like that, and 6067 * unlikely to change. */ 6068 if ( ( memBEGINPs(lookup_name, j, "perl") 6069 && memNEs(lookup_name + 4, j - 4, "space") 6070 && memNEs(lookup_name + 4, j - 4, "word")) 6071 || memEQs(lookup_name, j, "canondcij") 6072 || memEQs(lookup_name, j, "combabove")) 6073 { 6074 stricter = TRUE; 6075 6076 /* We set the inputs back to 0 and the code below will reparse, 6077 * using strict */ 6078 i = j = 0; 6079 } 6080 } 6081 6082 /* Here, we have either finished the property, or are positioned to parse 6083 * the remainder, and we know if stricter rules apply. Finish out, if not 6084 * already done */ 6085 for (; i < len; i++) { 6086 char cur = name[i]; 6087 6088 /* In all instances, case differences are ignored, and we normalize to 6089 * lowercase */ 6090 if (isUPPER(cur)) { 6091 lookup_name[j++] = toLOWER(cur); 6092 continue; 6093 } 6094 6095 /* An underscore is skipped, but not under strict rules unless it 6096 * separates two digits */ 6097 if (cur == '_') { 6098 if ( stricter 6099 && ( i == 0 || (int) i == equals_pos || i == len- 1 6100 || ! isDIGIT(name[i-1]) || ! isDIGIT(name[i+1]))) 6101 { 6102 lookup_name[j++] = '_'; 6103 } 6104 continue; 6105 } 6106 6107 /* Hyphens are skipped except under strict */ 6108 if (cur == '-' && ! stricter) { 6109 continue; 6110 } 6111 6112 /* XXX Bug in documentation. It says white space skipped adjacent to 6113 * non-word char. Maybe we should, but shouldn't skip it next to a dot 6114 * in a number */ 6115 if (isSPACE(cur) && ! stricter) { 6116 continue; 6117 } 6118 6119 lookup_name[j++] = cur; 6120 6121 /* Unless this is a non-trailing slash, we are done with it */ 6122 if (i >= len - 1 || cur != '/') { 6123 continue; 6124 } 6125 6126 /* A slash in the 'numeric value' property indicates that what follows 6127 * is a denominator. It can have a leading '+' and '0's that should be 6128 * skipped. But we have never allowed a negative denominator, so treat 6129 * a minus like every other character. (No need to rule out a second 6130 * '/', as that won't match anything anyway */ 6131 if ( memEQs(lookup_name + lookup_offset, equals_pos - lookup_offset, 6132 "nv=") 6133 || memEQs(lookup_name + lookup_offset, equals_pos - lookup_offset, 6134 "numericvalue=")) 6135 { 6136 i++; 6137 if (i < len && name[i] == '+') { 6138 i++; 6139 } 6140 6141 /* Skip leading zeros including underscores separating digits */ 6142 for (; i < len - 1; i++) { 6143 if ( name[i] != '0' 6144 && (name[i] != '_' || ! isDIGIT(name[i+1]))) 6145 { 6146 break; 6147 } 6148 } 6149 6150 /* Store the first real character in the denominator */ 6151 lookup_name[j++] = name[i]; 6152 } 6153 } 6154 6155 /* Here are completely done parsing the input 'name', and 'lookup_name' 6156 * contains a copy, normalized. 6157 * 6158 * This special case is grandfathered in: 'L_' and 'GC=L_' are accepted and 6159 * different from without the underscores. */ 6160 if ( ( UNLIKELY(memEQs(lookup_name, j, "l")) 6161 || UNLIKELY(memEQs(lookup_name, j, "gc=l"))) 6162 && UNLIKELY(name[len-1] == '_')) 6163 { 6164 lookup_name[j++] = '&'; 6165 } 6166 else if (len > 2 && name[0] == 'I' && ( name[1] == 'n' || name[1] == 's')) 6167 { 6168 6169 /* Also, if the original input began with 'In' or 'Is', it could be a 6170 * subroutine call instead of a property names, which currently isn't 6171 * handled by this function. Subroutine calls can't happen if there is 6172 * an '=' in the name */ 6173 if (equals_pos < 0 && get_cvn_flags(name, len, GV_NOTQUAL) != NULL) { 6174 return NULL; 6175 } 6176 6177 starts_with_In_or_Is = TRUE; 6178 } 6179 6180 /* Get the index into our pointer table of the inversion list corresponding 6181 * to the property */ 6182 table_index = match_uniprop((U8 *) lookup_name, j); 6183 6184 /* If it didn't find the property */ 6185 if (table_index == 0) { 6186 6187 /* If didn't find the property, we try again stripping off any initial 6188 * 'In' or 'Is' */ 6189 if (! starts_with_In_or_Is) { 6190 return NULL; 6191 } 6192 6193 lookup_name += 2; 6194 j -= 2; 6195 6196 /* If still didn't find it, give up */ 6197 table_index = match_uniprop((U8 *) lookup_name, j); 6198 if (table_index == 0) { 6199 return NULL; 6200 } 6201 } 6202 6203 /* The return is an index into a table of ptrs. A negative return 6204 * signifies that the real index is the absolute value, but the result 6205 * needs to be inverted */ 6206 if (table_index < 0) { 6207 *invert = TRUE; 6208 table_index = -table_index; 6209 } 6210 else { 6211 *invert = FALSE; 6212 } 6213 6214 /* Out-of band indices indicate a deprecated property. The proper index is 6215 * modulo it with the table size. And dividing by the table size yields 6216 * an offset into a table constructed to contain the corresponding warning 6217 * message */ 6218 if (table_index > MAX_UNI_KEYWORD_INDEX) { 6219 Size_t warning_offset = table_index / MAX_UNI_KEYWORD_INDEX; 6220 table_index %= MAX_UNI_KEYWORD_INDEX; 6221 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), 6222 "Use of '%.*s' in \\p{} or \\P{} is deprecated because: %s", 6223 (int) len, name, deprecated_property_msgs[warning_offset]); 6224 } 6225 6226 /* In a few properties, a different property is used under /i. These are 6227 * unlikely to change, so are hard-coded here. */ 6228 if (to_fold) { 6229 if ( table_index == PL_XPOSIXUPPER 6230 || table_index == PL_XPOSIXLOWER 6231 || table_index == PL_LT) 6232 { 6233 table_index = PL_CASED; 6234 } 6235 else if ( table_index == PL_LU 6236 || table_index == PL_LL 6237 || table_index == PL_LT) 6238 { 6239 table_index = PL_L_AMP_; 6240 } 6241 else if ( table_index == PL_POSIXUPPER 6242 || table_index == PL_POSIXLOWER) 6243 { 6244 table_index = PL_POSIXALPHA; 6245 } 6246 } 6247 6248 /* Create and return the inversion list */ 6249 return _new_invlist_C_array(PL_uni_prop_ptrs[table_index]); 6250 } 6251 6252 /* 6253 =for apidoc utf8_to_uvchr 6254 6255 Returns the native code point of the first character in the string C<s> 6256 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the 6257 length, in bytes, of that character. 6258 6259 Some, but not all, UTF-8 malformations are detected, and in fact, some 6260 malformed input could cause reading beyond the end of the input buffer, which 6261 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead. 6262 6263 If C<s> points to one of the detected malformations, and UTF8 warnings are 6264 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 6265 C<NULL>) to -1. If those warnings are off, the computed value if well-defined (or 6266 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> 6267 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the 6268 next possible position in C<s> that could begin a non-malformed character. 6269 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. 6270 6271 =cut 6272 */ 6273 6274 UV 6275 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen) 6276 { 6277 PERL_ARGS_ASSERT_UTF8_TO_UVCHR; 6278 6279 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen); 6280 } 6281 6282 /* 6283 * ex: set ts=8 sts=4 sw=4 et: 6284 */ 6285