1 /* utf8.c 2 * 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 4 * by Larry Wall and others 5 * 6 * You may distribute under the terms of either the GNU General Public 7 * License or the Artistic License, as specified in the README file. 8 * 9 */ 10 11 /* 12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever 13 * heard of that we don't want to see any closer; and that's the one place 14 * we're trying to get to! And that's just where we can't get, nohow.' 15 * 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"] 17 * 18 * 'Well do I understand your speech,' he answered in the same language; 19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue, 20 * as is the custom in the West, if you wish to be answered?' 21 * --Gandalf, addressing Théoden's door wardens 22 * 23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] 24 * 25 * ...the travellers perceived that the floor was paved with stones of many 26 * hues; branching runes and strange devices intertwined beneath their feet. 27 * 28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] 29 */ 30 31 #include "EXTERN.h" 32 #define PERL_IN_UTF8_C 33 #include "perl.h" 34 #include "invlist_inline.h" 35 36 static const char unees[] = 37 "Malformed UTF-8 character (unexpected end of string)"; 38 static const char cp_above_legal_max[] = 39 "Use of code point 0x%"UVXf" is deprecated; the permissible max is 0x%"UVXf""; 40 41 #define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX)) 42 43 /* 44 =head1 Unicode Support 45 These are various utility functions for manipulating UTF8-encoded 46 strings. For the uninitiated, this is a method of representing arbitrary 47 Unicode characters as a variable number of bytes, in such a way that 48 characters in the ASCII range are unmodified, and a zero byte never appears 49 within non-zero characters. 50 51 =cut 52 */ 53 54 /* 55 =for apidoc is_invariant_string 56 57 Returns true iff the first C<len> bytes of the string C<s> are the same 58 regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on 59 EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish 60 machines, all the ASCII characters and only the ASCII characters fit this 61 definition. On EBCDIC machines, the ASCII-range characters are invariant, but 62 so also are the C1 controls and C<\c?> (which isn't in the ASCII range on 63 EBCDIC). 64 65 If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you 66 use this option, that C<s> can't have embedded C<NUL> characters and has to 67 have a terminating C<NUL> byte). 68 69 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>(). 70 71 =cut 72 */ 73 74 bool 75 Perl_is_invariant_string(const U8 *s, STRLEN len) 76 { 77 const U8* const send = s + (len ? len : strlen((const char *)s)); 78 const U8* x = s; 79 80 PERL_ARGS_ASSERT_IS_INVARIANT_STRING; 81 82 for (; x < send; ++x) { 83 if (!UTF8_IS_INVARIANT(*x)) 84 break; 85 } 86 87 return x == send; 88 } 89 90 /* 91 =for apidoc uvoffuni_to_utf8_flags 92 93 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 94 Instead, B<Almost all code should use L</uvchr_to_utf8> or 95 L</uvchr_to_utf8_flags>>. 96 97 This function is like them, but the input is a strict Unicode 98 (as opposed to native) code point. Only in very rare circumstances should code 99 not be using the native code point. 100 101 For details, see the description for L</uvchr_to_utf8_flags>. 102 103 =cut 104 */ 105 106 #define HANDLE_UNICODE_SURROGATE(uv, flags) \ 107 STMT_START { \ 108 if (flags & UNICODE_WARN_SURROGATE) { \ 109 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \ 110 "UTF-16 surrogate U+%04"UVXf, uv); \ 111 } \ 112 if (flags & UNICODE_DISALLOW_SURROGATE) { \ 113 return NULL; \ 114 } \ 115 } STMT_END; 116 117 #define HANDLE_UNICODE_NONCHAR(uv, flags) \ 118 STMT_START { \ 119 if (flags & UNICODE_WARN_NONCHAR) { \ 120 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \ 121 "Unicode non-character U+%04"UVXf" is not " \ 122 "recommended for open interchange", uv); \ 123 } \ 124 if (flags & UNICODE_DISALLOW_NONCHAR) { \ 125 return NULL; \ 126 } \ 127 } STMT_END; 128 129 /* Use shorter names internally in this file */ 130 #define SHIFT UTF_ACCUMULATION_SHIFT 131 #undef MARK 132 #define MARK UTF_CONTINUATION_MARK 133 #define MASK UTF_CONTINUATION_MASK 134 135 U8 * 136 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) 137 { 138 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS; 139 140 if (OFFUNI_IS_INVARIANT(uv)) { 141 *d++ = LATIN1_TO_NATIVE(uv); 142 return d; 143 } 144 145 if (uv <= MAX_UTF8_TWO_BYTE) { 146 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2)); 147 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK); 148 return d; 149 } 150 151 /* Not 2-byte; test for and handle 3-byte result. In the test immediately 152 * below, the 16 is for start bytes E0-EF (which are all the possible ones 153 * for 3 byte characters). The 2 is for 2 continuation bytes; these each 154 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000 155 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC; 156 * 0x800-0xFFFF on ASCII */ 157 if (uv < (16 * (1U << (2 * SHIFT)))) { 158 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3)); 159 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); 160 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); 161 162 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so 163 aren't tested here */ 164 /* The most likely code points in this range are below the surrogates. 165 * Do an extra test to quickly exclude those. */ 166 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) { 167 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) 168 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) 169 { 170 HANDLE_UNICODE_NONCHAR(uv, flags); 171 } 172 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 173 HANDLE_UNICODE_SURROGATE(uv, flags); 174 } 175 } 176 #endif 177 return d; 178 } 179 180 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII 181 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can 182 * happen starting with 4-byte characters on ASCII platforms. We unify the 183 * code for these with EBCDIC, even though some of them require 5-bytes on 184 * those, because khw believes the code saving is worth the very slight 185 * performance hit on these high EBCDIC code points. */ 186 187 if (UNLIKELY(UNICODE_IS_SUPER(uv))) { 188 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP) 189 && ckWARN_d(WARN_DEPRECATED)) 190 { 191 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), 192 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP); 193 } 194 if ( (flags & UNICODE_WARN_SUPER) 195 || ( UNICODE_IS_ABOVE_31_BIT(uv) 196 && (flags & UNICODE_WARN_ABOVE_31_BIT))) 197 { 198 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), 199 200 /* Choose the more dire applicable warning */ 201 (UNICODE_IS_ABOVE_31_BIT(uv)) 202 ? "Code point 0x%"UVXf" is not Unicode, and not portable" 203 : "Code point 0x%"UVXf" is not Unicode, may not be portable", 204 uv); 205 } 206 if (flags & UNICODE_DISALLOW_SUPER 207 || ( UNICODE_IS_ABOVE_31_BIT(uv) 208 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT))) 209 { 210 return NULL; 211 } 212 } 213 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) { 214 HANDLE_UNICODE_NONCHAR(uv, flags); 215 } 216 217 /* Test for and handle 4-byte result. In the test immediately below, the 218 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte 219 * characters). The 3 is for 3 continuation bytes; these each contribute 220 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on 221 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC; 222 * 0x1_0000-0x1F_FFFF on ASCII */ 223 if (uv < (8 * (1U << (3 * SHIFT)))) { 224 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4)); 225 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK); 226 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); 227 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); 228 229 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte 230 characters. The end-plane non-characters for EBCDIC were 231 handled just above */ 232 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) { 233 HANDLE_UNICODE_NONCHAR(uv, flags); 234 } 235 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { 236 HANDLE_UNICODE_SURROGATE(uv, flags); 237 } 238 #endif 239 240 return d; 241 } 242 243 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII 244 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop 245 * format. The unrolled version above turns out to not save all that much 246 * time, and at these high code points (well above the legal Unicode range 247 * on ASCII platforms, and well above anything in common use in EBCDIC), 248 * khw believes that less code outweighs slight performance gains. */ 249 250 { 251 STRLEN len = OFFUNISKIP(uv); 252 U8 *p = d+len-1; 253 while (p > d) { 254 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK); 255 uv >>= UTF_ACCUMULATION_SHIFT; 256 } 257 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len)); 258 return d+len; 259 } 260 } 261 262 /* 263 =for apidoc uvchr_to_utf8 264 265 Adds the UTF-8 representation of the native code point C<uv> to the end 266 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to 267 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to 268 the byte after the end of the new character. In other words, 269 270 d = uvchr_to_utf8(d, uv); 271 272 is the recommended wide native character-aware way of saying 273 274 *(d++) = uv; 275 276 This function accepts any UV as input, but very high code points (above 277 C<IV_MAX> on the platform) will raise a deprecation warning. This is 278 typically 0x7FFF_FFFF in a 32-bit word. 279 280 It is possible to forbid or warn on non-Unicode code points, or those that may 281 be problematic by using L</uvchr_to_utf8_flags>. 282 283 =cut 284 */ 285 286 /* This is also a macro */ 287 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv); 288 289 U8 * 290 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv) 291 { 292 return uvchr_to_utf8(d, uv); 293 } 294 295 /* 296 =for apidoc uvchr_to_utf8_flags 297 298 Adds the UTF-8 representation of the native code point C<uv> to the end 299 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to 300 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to 301 the byte after the end of the new character. In other words, 302 303 d = uvchr_to_utf8_flags(d, uv, flags); 304 305 or, in most cases, 306 307 d = uvchr_to_utf8_flags(d, uv, 0); 308 309 This is the Unicode-aware way of saying 310 311 *(d++) = uv; 312 313 If C<flags> is 0, this function accepts any UV as input, but very high code 314 points (above C<IV_MAX> for the platform) will raise a deprecation warning. 315 This is typically 0x7FFF_FFFF in a 32-bit word. 316 317 Specifying C<flags> can further restrict what is allowed and not warned on, as 318 follows: 319 320 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set, 321 the function will raise a warning, provided UTF8 warnings are enabled. If 322 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return 323 NULL. If both flags are set, the function will both warn and return NULL. 324 325 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags 326 affect how the function handles a Unicode non-character. 327 328 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags 329 affect the handling of code points that are above the Unicode maximum of 330 0x10FFFF. Languages other than Perl may not be able to accept files that 331 contain these. 332 333 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of 334 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all 335 three DISALLOW flags. 336 337 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, 338 so using them is more problematic than other above-Unicode code points. Perl 339 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is 340 likely that non-Perl languages will not be able to read files that contain 341 these that written by the perl interpreter; nor would Perl understand files 342 written by something that uses a different extension. For these reasons, there 343 is a separate set of flags that can warn and/or disallow these extremely high 344 code points, even if other above-Unicode ones are accepted. These are the 345 C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These 346 are entirely independent from the deprecation warning for code points above 347 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any 348 code point that needs more than 31 bits to represent. When that happens, 349 effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on 350 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all 351 above-Unicode code points, including these, as malformations; and 352 C<UNICODE_WARN_SUPER> warns on these.) 353 354 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing 355 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower 356 than on ASCII. Prior to that, code points 2**31 and higher were simply 357 unrepresentable, and a different, incompatible method was used to represent 358 code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT> 359 and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII 360 platforms, warning and disallowing 2**31 and higher. 361 362 =cut 363 */ 364 365 /* This is also a macro */ 366 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags); 367 368 U8 * 369 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) 370 { 371 return uvchr_to_utf8_flags(d, uv, flags); 372 } 373 374 /* 375 =for apidoc is_utf8_string 376 377 Returns true if the first C<len> bytes of string C<s> form a valid 378 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated 379 using C<strlen(s)> (which means if you use this option, that C<s> can't have 380 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note 381 that all characters being ASCII constitute 'a valid UTF-8 string'. 382 383 See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>(). 384 385 =cut 386 */ 387 388 bool 389 Perl_is_utf8_string(const U8 *s, STRLEN len) 390 { 391 const U8* const send = s + (len ? len : strlen((const char *)s)); 392 const U8* x = s; 393 394 PERL_ARGS_ASSERT_IS_UTF8_STRING; 395 396 while (x < send) { 397 STRLEN len = isUTF8_CHAR(x, send); 398 if (UNLIKELY(! len)) { 399 return FALSE; 400 } 401 x += len; 402 } 403 404 return TRUE; 405 } 406 407 /* 408 Implemented as a macro in utf8.h 409 410 =for apidoc is_utf8_string_loc 411 412 Like L</is_utf8_string> but stores the location of the failure (in the 413 case of "utf8ness failure") or the location C<s>+C<len> (in the case of 414 "utf8ness success") in the C<ep>. 415 416 See also L</is_utf8_string_loclen>() and L</is_utf8_string>(). 417 418 =for apidoc is_utf8_string_loclen 419 420 Like L</is_utf8_string>() but stores the location of the failure (in the 421 case of "utf8ness failure") or the location C<s>+C<len> (in the case of 422 "utf8ness success") in the C<ep>, and the number of UTF-8 423 encoded characters in the C<el>. 424 425 See also L</is_utf8_string_loc>() and L</is_utf8_string>(). 426 427 =cut 428 */ 429 430 bool 431 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el) 432 { 433 const U8* const send = s + (len ? len : strlen((const char *)s)); 434 const U8* x = s; 435 STRLEN outlen = 0; 436 437 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN; 438 439 while (x < send) { 440 STRLEN len = isUTF8_CHAR(x, send); 441 if (UNLIKELY(! len)) { 442 goto out; 443 } 444 x += len; 445 outlen++; 446 } 447 448 out: 449 if (el) 450 *el = outlen; 451 452 if (ep) 453 *ep = x; 454 return (x == send); 455 } 456 457 /* 458 459 =for apidoc utf8n_to_uvchr 460 461 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. 462 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. 463 464 Bottom level UTF-8 decode routine. 465 Returns the native code point value of the first character in the string C<s>, 466 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than 467 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to 468 the length, in bytes, of that character. 469 470 The value of C<flags> determines the behavior when C<s> does not point to a 471 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found, 472 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the 473 next possible position in C<s> that could begin a non-malformed character. 474 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised. 475 476 Various ALLOW flags can be set in C<flags> to allow (and not warn on) 477 individual types of malformations, such as the sequence being overlong (that 478 is, when there is a shorter sequence that can express the same code point; 479 overlong sequences are expressly forbidden in the UTF-8 standard due to 480 potential security issues). Another malformation example is the first byte of 481 a character not being a legal first byte. See F<utf8.h> for the list of such 482 flags. For allowed 0 length strings, this function returns 0; for allowed 483 overlong sequences, the computed code point is returned; for all other allowed 484 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no 485 determinable reasonable value. 486 487 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other 488 flags) malformation is found. If this flag is set, the routine assumes that 489 the caller will raise a warning, and this function will silently just set 490 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero. 491 492 Note that this API requires disambiguation between successful decoding a C<NUL> 493 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as 494 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the 495 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the 496 input had an error. 497 498 Certain code points are considered problematic. These are Unicode surrogates, 499 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF. 500 By default these are considered regular code points, but certain situations 501 warrant special handling for them. If C<flags> contains 502 C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as 503 malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>, 504 C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal 505 Unicode maximum) can be set to disallow these categories individually. 506 507 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>, 508 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be 509 raised for their respective categories, but otherwise the code points are 510 considered valid (not malformations). To get a category to both be treated as 511 a malformation and raise a warning, specify both the WARN and DISALLOW flags. 512 (But note that warnings are not raised if lexically disabled nor if 513 C<UTF8_CHECK_ONLY> is also specified.) 514 515 It is now deprecated to have very high code points (above C<IV_MAX> on the 516 platforms) and this function will raise a deprecation warning for these (unless 517 such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1) 518 in a 32-bit word. 519 520 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, 521 so using them is more problematic than other above-Unicode code points. Perl 522 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is 523 likely that non-Perl languages will not be able to read files that contain 524 these that written by the perl interpreter; nor would Perl understand files 525 written by something that uses a different extension. For these reasons, there 526 is a separate set of flags that can warn and/or disallow these extremely high 527 code points, even if other above-Unicode ones are accepted. These are the 528 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These 529 are entirely independent from the deprecation warning for code points above 530 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any 531 code point that needs more than 31 bits to represent. When that happens, 532 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on 533 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all 534 above-Unicode code points, including these, as malformations; and 535 C<UTF8_WARN_SUPER> warns on these.) 536 537 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing 538 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower 539 than on ASCII. Prior to that, code points 2**31 and higher were simply 540 unrepresentable, and a different, incompatible method was used to represent 541 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT> 542 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII 543 platforms, warning and disallowing 2**31 and higher. 544 545 All other code points corresponding to Unicode characters, including private 546 use and those yet to be assigned, are never considered malformed and never 547 warn. 548 549 =cut 550 */ 551 552 UV 553 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags) 554 { 555 const U8 * const s0 = s; 556 U8 overflow_byte = '\0'; /* Save byte in case of overflow */ 557 U8 * send; 558 UV uv = *s; 559 STRLEN expectlen; 560 SV* sv = NULL; 561 UV outlier_ret = 0; /* return value when input is in error or problematic 562 */ 563 UV pack_warn = 0; /* Save result of packWARN() for later */ 564 bool unexpected_non_continuation = FALSE; 565 bool overflowed = FALSE; 566 bool do_overlong_test = TRUE; /* May have to skip this test */ 567 568 const char* const malformed_text = "Malformed UTF-8 character"; 569 570 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR; 571 572 /* The order of malformation tests here is important. We should consume as 573 * few bytes as possible in order to not skip any valid character. This is 574 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also 575 * http://unicode.org/reports/tr36 for more discussion as to why. For 576 * example, once we've done a UTF8SKIP, we can tell the expected number of 577 * bytes, and could fail right off the bat if the input parameters indicate 578 * that there are too few available. But it could be that just that first 579 * byte is garbled, and the intended character occupies fewer bytes. If we 580 * blindly assumed that the first byte is correct, and skipped based on 581 * that number, we could skip over a valid input character. So instead, we 582 * always examine the sequence byte-by-byte. 583 * 584 * We also should not consume too few bytes, otherwise someone could inject 585 * things. For example, an input could be deliberately designed to 586 * overflow, and if this code bailed out immediately upon discovering that, 587 * returning to the caller C<*retlen> pointing to the very next byte (one 588 * which is actually part of of the overflowing sequence), that could look 589 * legitimate to the caller, which could discard the initial partial 590 * sequence and process the rest, inappropriately */ 591 592 /* Zero length strings, if allowed, of necessity are zero */ 593 if (UNLIKELY(curlen == 0)) { 594 if (retlen) { 595 *retlen = 0; 596 } 597 598 if (flags & UTF8_ALLOW_EMPTY) { 599 return 0; 600 } 601 if (! (flags & UTF8_CHECK_ONLY)) { 602 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text)); 603 } 604 goto malformed; 605 } 606 607 expectlen = UTF8SKIP(s); 608 609 /* A well-formed UTF-8 character, as the vast majority of calls to this 610 * function will be for, has this expected length. For efficiency, set 611 * things up here to return it. It will be overriden only in those rare 612 * cases where a malformation is found */ 613 if (retlen) { 614 *retlen = expectlen; 615 } 616 617 /* An invariant is trivially well-formed */ 618 if (UTF8_IS_INVARIANT(uv)) { 619 return uv; 620 } 621 622 /* A continuation character can't start a valid sequence */ 623 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) { 624 if (flags & UTF8_ALLOW_CONTINUATION) { 625 if (retlen) { 626 *retlen = 1; 627 } 628 return UNICODE_REPLACEMENT; 629 } 630 631 if (! (flags & UTF8_CHECK_ONLY)) { 632 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0)); 633 } 634 curlen = 1; 635 goto malformed; 636 } 637 638 /* Here is not a continuation byte, nor an invariant. The only thing left 639 * is a start byte (possibly for an overlong) */ 640 641 #ifdef EBCDIC 642 uv = NATIVE_UTF8_TO_I8(uv); 643 #endif 644 645 /* Remove the leading bits that indicate the number of bytes in the 646 * character's whole UTF-8 sequence, leaving just the bits that are part of 647 * the value */ 648 uv &= UTF_START_MASK(expectlen); 649 650 /* Now, loop through the remaining bytes in the character's sequence, 651 * accumulating each into the working value as we go. Be sure to not look 652 * past the end of the input string */ 653 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen); 654 655 for (s = s0 + 1; s < send; s++) { 656 if (LIKELY(UTF8_IS_CONTINUATION(*s))) { 657 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) { 658 659 /* The original implementors viewed this malformation as more 660 * serious than the others (though I, khw, don't understand 661 * why, since other malformations also give very very wrong 662 * results), so there is no way to turn off checking for it. 663 * Set a flag, but keep going in the loop, so that we absorb 664 * the rest of the bytes that comprise the character. */ 665 overflowed = TRUE; 666 overflow_byte = *s; /* Save for warning message's use */ 667 } 668 uv = UTF8_ACCUMULATE(uv, *s); 669 } 670 else { 671 /* Here, found a non-continuation before processing all expected 672 * bytes. This byte begins a new character, so quit, even if 673 * allowing this malformation. */ 674 unexpected_non_continuation = TRUE; 675 break; 676 } 677 } /* End of loop through the character's bytes */ 678 679 /* Save how many bytes were actually in the character */ 680 curlen = s - s0; 681 682 /* The loop above finds two types of malformations: non-continuation and/or 683 * overflow. The non-continuation malformation is really a too-short 684 * malformation, as it means that the current character ended before it was 685 * expected to (being terminated prematurely by the beginning of the next 686 * character, whereas in the too-short malformation there just are too few 687 * bytes available to hold the character. In both cases, the check below 688 * that we have found the expected number of bytes would fail if executed.) 689 * Thus the non-continuation malformation is really unnecessary, being a 690 * subset of the too-short malformation. But there may be existing 691 * applications that are expecting the non-continuation type, so we retain 692 * it, and return it in preference to the too-short malformation. (If this 693 * code were being written from scratch, the two types might be collapsed 694 * into one.) I, khw, am also giving priority to returning the 695 * non-continuation and too-short malformations over overflow when multiple 696 * ones are present. I don't know of any real reason to prefer one over 697 * the other, except that it seems to me that multiple-byte errors trumps 698 * errors from a single byte */ 699 if (UNLIKELY(unexpected_non_continuation)) { 700 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) { 701 if (! (flags & UTF8_CHECK_ONLY)) { 702 if (curlen == 1) { 703 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0)); 704 } 705 else { 706 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, %d bytes after start byte 0x%02x, expected %d bytes)", malformed_text, *s, (int) curlen, *s0, (int)expectlen)); 707 } 708 } 709 goto malformed; 710 } 711 uv = UNICODE_REPLACEMENT; 712 713 /* Skip testing for overlongs, as the REPLACEMENT may not be the same 714 * as what the original expectations were. */ 715 do_overlong_test = FALSE; 716 if (retlen) { 717 *retlen = curlen; 718 } 719 } 720 else if (UNLIKELY(curlen < expectlen)) { 721 if (! (flags & UTF8_ALLOW_SHORT)) { 722 if (! (flags & UTF8_CHECK_ONLY)) { 723 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, *s0)); 724 } 725 goto malformed; 726 } 727 uv = UNICODE_REPLACEMENT; 728 do_overlong_test = FALSE; 729 if (retlen) { 730 *retlen = curlen; 731 } 732 } 733 734 if (UNLIKELY(overflowed)) { 735 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0)); 736 goto malformed; 737 } 738 739 if (do_overlong_test 740 && expectlen > (STRLEN) OFFUNISKIP(uv) 741 && ! (flags & UTF8_ALLOW_LONG)) 742 { 743 /* The overlong malformation has lower precedence than the others. 744 * Note that if this malformation is allowed, we return the actual 745 * value, instead of the replacement character. This is because this 746 * value is actually well-defined. */ 747 if (! (flags & UTF8_CHECK_ONLY)) { 748 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", OFFUNISKIP(uv), *s0)); 749 } 750 goto malformed; 751 } 752 753 /* Here, the input is considered to be well-formed, but it still could be a 754 * problematic code point that is not allowed by the input parameters. */ 755 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */ 756 && ((flags & ( UTF8_DISALLOW_NONCHAR 757 |UTF8_DISALLOW_SURROGATE 758 |UTF8_DISALLOW_SUPER 759 |UTF8_DISALLOW_ABOVE_31_BIT 760 |UTF8_WARN_NONCHAR 761 |UTF8_WARN_SURROGATE 762 |UTF8_WARN_SUPER 763 |UTF8_WARN_ABOVE_31_BIT)) 764 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP) 765 && ckWARN_d(WARN_DEPRECATED)))) 766 { 767 if (UNICODE_IS_SURROGATE(uv)) { 768 769 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary 770 * generation of the sv, since no warnings are raised under CHECK */ 771 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE 772 && ckWARN_d(WARN_SURROGATE)) 773 { 774 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv)); 775 pack_warn = packWARN(WARN_SURROGATE); 776 } 777 if (flags & UTF8_DISALLOW_SURROGATE) { 778 goto disallowed; 779 } 780 } 781 else if ((uv > PERL_UNICODE_MAX)) { 782 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER 783 && ckWARN_d(WARN_NON_UNICODE)) 784 { 785 sv = sv_2mortal(Perl_newSVpvf(aTHX_ 786 "Code point 0x%04"UVXf" is not Unicode, may not be portable", 787 uv)); 788 pack_warn = packWARN(WARN_NON_UNICODE); 789 } 790 791 /* The maximum code point ever specified by a standard was 792 * 2**31 - 1. Anything larger than that is a Perl extension that 793 * very well may not be understood by other applications (including 794 * earlier perl versions on EBCDIC platforms). On ASCII platforms, 795 * these code points are indicated by the first UTF-8 byte being 796 * 0xFE or 0xFF. We test for these after the regular SUPER ones, 797 * and before possibly bailing out, so that the slightly more dire 798 * warning will override the regular one. */ 799 if ( 800 #ifndef EBCDIC 801 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */ 802 #else 803 /* The I8 for 2**31 (U+80000000) is 804 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0 805 * and it turns out that on all EBCDIC pages recognized that 806 * the UTF-EBCDIC for that code point is 807 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 808 * For the next lower code point, the 1047 UTF-EBCDIC is 809 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73 810 * The other code pages differ only in the bytes following 811 * \x42. Thus the following works (the minimum continuation 812 * byte is \x41). */ 813 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41 814 || s0[2] > 0x41 815 || s0[3] > 0x41 816 || s0[4] > 0x41 817 || s0[5] > 0x41 818 || s0[6] > 0x41 819 || s0[7] > 0x42) 820 #endif 821 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER 822 |UTF8_DISALLOW_ABOVE_31_BIT))) 823 { 824 if ( ! (flags & UTF8_CHECK_ONLY) 825 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER)) 826 && ckWARN_d(WARN_UTF8)) 827 { 828 sv = sv_2mortal(Perl_newSVpvf(aTHX_ 829 "Code point 0x%"UVXf" is not Unicode, and not portable", 830 uv)); 831 pack_warn = packWARN(WARN_UTF8); 832 } 833 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) { 834 goto disallowed; 835 } 836 } 837 838 if (flags & UTF8_DISALLOW_SUPER) { 839 goto disallowed; 840 } 841 842 /* The deprecated warning overrides any non-deprecated one */ 843 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED)) 844 { 845 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max, 846 uv, MAX_NON_DEPRECATED_CP)); 847 pack_warn = packWARN(WARN_DEPRECATED); 848 } 849 } 850 else if (UNICODE_IS_NONCHAR(uv)) { 851 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR 852 && ckWARN_d(WARN_NONCHAR)) 853 { 854 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv)); 855 pack_warn = packWARN(WARN_NONCHAR); 856 } 857 if (flags & UTF8_DISALLOW_NONCHAR) { 858 goto disallowed; 859 } 860 } 861 862 if (sv) { 863 outlier_ret = uv; /* Note we don't bother to convert to native, 864 as all the outlier code points are the same 865 in both ASCII and EBCDIC */ 866 goto do_warn; 867 } 868 869 /* Here, this is not considered a malformed character, so drop through 870 * to return it */ 871 } 872 873 return UNI_TO_NATIVE(uv); 874 875 /* There are three cases which get to beyond this point. In all 3 cases: 876 * <sv> if not null points to a string to print as a warning. 877 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't 878 * set. 879 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set. 880 * This is done by initializing it to 0, and changing it only 881 * for case 1). 882 * The 3 cases are: 883 * 1) The input is valid but problematic, and to be warned about. The 884 * return value is the resultant code point; <*retlen> is set to 885 * <curlen>, the number of bytes that comprise the code point. 886 * <pack_warn> contains the result of packWARN() for the warning 887 * types. The entry point for this case is the label <do_warn>; 888 * 2) The input is a valid code point but disallowed by the parameters to 889 * this function. The return value is 0. If UTF8_CHECK_ONLY is set, 890 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that 891 * comprise the code point. <pack_warn> contains the result of 892 * packWARN() for the warning types. The entry point for this case is 893 * the label <disallowed>. 894 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY 895 * is set, <*relen> is -1; otherwise it is <curlen>, the number of 896 * bytes that comprise the malformation. All such malformations are 897 * assumed to be warning type <utf8>. The entry point for this case 898 * is the label <malformed>. 899 */ 900 901 malformed: 902 903 if (sv && ckWARN_d(WARN_UTF8)) { 904 pack_warn = packWARN(WARN_UTF8); 905 } 906 907 disallowed: 908 909 if (flags & UTF8_CHECK_ONLY) { 910 if (retlen) 911 *retlen = ((STRLEN) -1); 912 return 0; 913 } 914 915 do_warn: 916 917 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only 918 if warnings are to be raised. */ 919 const char * const string = SvPVX_const(sv); 920 921 if (PL_op) 922 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op)); 923 else 924 Perl_warner(aTHX_ pack_warn, "%s", string); 925 } 926 927 if (retlen) { 928 *retlen = curlen; 929 } 930 931 return outlier_ret; 932 } 933 934 /* 935 =for apidoc utf8_to_uvchr_buf 936 937 Returns the native code point of the first character in the string C<s> which 938 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. 939 C<*retlen> will be set to the length, in bytes, of that character. 940 941 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are 942 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 943 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined 944 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and 945 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is 946 the next possible position in C<s> that could begin a non-malformed character. 947 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is 948 returned. 949 950 Code points above the platform's C<IV_MAX> will raise a deprecation warning, 951 unless those are turned off. 952 953 =cut 954 */ 955 956 957 UV 958 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) 959 { 960 assert(s < send); 961 962 return utf8n_to_uvchr(s, send - s, retlen, 963 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY); 964 } 965 966 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that 967 * there are no malformations in the input UTF-8 string C<s>. surrogates, 968 * non-character code points, and non-Unicode code points are allowed. */ 969 970 UV 971 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen) 972 { 973 UV expectlen = UTF8SKIP(s); 974 const U8* send = s + expectlen; 975 UV uv = *s; 976 977 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR; 978 PERL_UNUSED_CONTEXT; 979 980 if (retlen) { 981 *retlen = expectlen; 982 } 983 984 /* An invariant is trivially returned */ 985 if (expectlen == 1) { 986 return uv; 987 } 988 989 #ifdef EBCDIC 990 uv = NATIVE_UTF8_TO_I8(uv); 991 #endif 992 993 /* Remove the leading bits that indicate the number of bytes, leaving just 994 * the bits that are part of the value */ 995 uv &= UTF_START_MASK(expectlen); 996 997 /* Now, loop through the remaining bytes, accumulating each into the 998 * working total as we go. (I khw tried unrolling the loop for up to 4 999 * bytes, but there was no performance improvement) */ 1000 for (++s; s < send; s++) { 1001 uv = UTF8_ACCUMULATE(uv, *s); 1002 } 1003 1004 return UNI_TO_NATIVE(uv); 1005 1006 } 1007 1008 /* 1009 =for apidoc utf8_to_uvuni_buf 1010 1011 Only in very rare circumstances should code need to be dealing in Unicode 1012 (as opposed to native) code points. In those few cases, use 1013 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. 1014 1015 Returns the Unicode (not-native) code point of the first character in the 1016 string C<s> which 1017 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. 1018 C<retlen> will be set to the length, in bytes, of that character. 1019 1020 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are 1021 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't 1022 NULL) to -1. If those warnings are off, the computed value if well-defined (or 1023 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> 1024 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the 1025 next possible position in C<s> that could begin a non-malformed character. 1026 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. 1027 1028 Code points above the platform's C<IV_MAX> will raise a deprecation warning, 1029 unless those are turned off. 1030 1031 =cut 1032 */ 1033 1034 UV 1035 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) 1036 { 1037 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF; 1038 1039 assert(send > s); 1040 1041 /* Call the low level routine asking for checks */ 1042 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen, 1043 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY)); 1044 } 1045 1046 /* 1047 =for apidoc utf8_length 1048 1049 Return the length of the UTF-8 char encoded string C<s> in characters. 1050 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end 1051 up past C<e>, croaks. 1052 1053 =cut 1054 */ 1055 1056 STRLEN 1057 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e) 1058 { 1059 STRLEN len = 0; 1060 1061 PERL_ARGS_ASSERT_UTF8_LENGTH; 1062 1063 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g. 1064 * the bitops (especially ~) can create illegal UTF-8. 1065 * In other words: in Perl UTF-8 is not just for Unicode. */ 1066 1067 if (e < s) 1068 goto warn_and_return; 1069 while (s < e) { 1070 s += UTF8SKIP(s); 1071 len++; 1072 } 1073 1074 if (e != s) { 1075 len--; 1076 warn_and_return: 1077 if (PL_op) 1078 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 1079 "%s in %s", unees, OP_DESC(PL_op)); 1080 else 1081 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); 1082 } 1083 1084 return len; 1085 } 1086 1087 /* 1088 =for apidoc utf8_distance 1089 1090 Returns the number of UTF-8 characters between the UTF-8 pointers C<a> 1091 and C<b>. 1092 1093 WARNING: use only if you *know* that the pointers point inside the 1094 same UTF-8 buffer. 1095 1096 =cut 1097 */ 1098 1099 IV 1100 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b) 1101 { 1102 PERL_ARGS_ASSERT_UTF8_DISTANCE; 1103 1104 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a); 1105 } 1106 1107 /* 1108 =for apidoc utf8_hop 1109 1110 Return the UTF-8 pointer C<s> displaced by C<off> characters, either 1111 forward or backward. 1112 1113 WARNING: do not use the following unless you *know* C<off> is within 1114 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned 1115 on the first byte of character or just after the last byte of a character. 1116 1117 =cut 1118 */ 1119 1120 U8 * 1121 Perl_utf8_hop(const U8 *s, SSize_t off) 1122 { 1123 PERL_ARGS_ASSERT_UTF8_HOP; 1124 1125 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g 1126 * the bitops (especially ~) can create illegal UTF-8. 1127 * In other words: in Perl UTF-8 is not just for Unicode. */ 1128 1129 if (off >= 0) { 1130 while (off--) 1131 s += UTF8SKIP(s); 1132 } 1133 else { 1134 while (off++) { 1135 s--; 1136 while (UTF8_IS_CONTINUATION(*s)) 1137 s--; 1138 } 1139 } 1140 return (U8 *)s; 1141 } 1142 1143 /* 1144 =for apidoc bytes_cmp_utf8 1145 1146 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the 1147 sequence of characters (stored as UTF-8) 1148 in C<u>, C<ulen>. Returns 0 if they are 1149 equal, -1 or -2 if the first string is less than the second string, +1 or +2 1150 if the first string is greater than the second string. 1151 1152 -1 or +1 is returned if the shorter string was identical to the start of the 1153 longer string. -2 or +2 is returned if 1154 there was a difference between characters 1155 within the strings. 1156 1157 =cut 1158 */ 1159 1160 int 1161 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen) 1162 { 1163 const U8 *const bend = b + blen; 1164 const U8 *const uend = u + ulen; 1165 1166 PERL_ARGS_ASSERT_BYTES_CMP_UTF8; 1167 1168 while (b < bend && u < uend) { 1169 U8 c = *u++; 1170 if (!UTF8_IS_INVARIANT(c)) { 1171 if (UTF8_IS_DOWNGRADEABLE_START(c)) { 1172 if (u < uend) { 1173 U8 c1 = *u++; 1174 if (UTF8_IS_CONTINUATION(c1)) { 1175 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1); 1176 } else { 1177 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 1178 "Malformed UTF-8 character " 1179 "(unexpected non-continuation byte 0x%02x" 1180 ", immediately after start byte 0x%02x)" 1181 /* Dear diag.t, it's in the pod. */ 1182 "%s%s", c1, c, 1183 PL_op ? " in " : "", 1184 PL_op ? OP_DESC(PL_op) : ""); 1185 return -2; 1186 } 1187 } else { 1188 if (PL_op) 1189 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 1190 "%s in %s", unees, OP_DESC(PL_op)); 1191 else 1192 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); 1193 return -2; /* Really want to return undef :-) */ 1194 } 1195 } else { 1196 return -2; 1197 } 1198 } 1199 if (*b != c) { 1200 return *b < c ? -2 : +2; 1201 } 1202 ++b; 1203 } 1204 1205 if (b == bend && u == uend) 1206 return 0; 1207 1208 return b < bend ? +1 : -1; 1209 } 1210 1211 /* 1212 =for apidoc utf8_to_bytes 1213 1214 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding. 1215 Unlike L</bytes_to_utf8>, this over-writes the original string, and 1216 updates C<len> to contain the new length. 1217 Returns zero on failure, setting C<len> to -1. 1218 1219 If you need a copy of the string, see L</bytes_from_utf8>. 1220 1221 =cut 1222 */ 1223 1224 U8 * 1225 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len) 1226 { 1227 U8 * const save = s; 1228 U8 * const send = s + *len; 1229 U8 *d; 1230 1231 PERL_ARGS_ASSERT_UTF8_TO_BYTES; 1232 PERL_UNUSED_CONTEXT; 1233 1234 /* ensure valid UTF-8 and chars < 256 before updating string */ 1235 while (s < send) { 1236 if (! UTF8_IS_INVARIANT(*s)) { 1237 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { 1238 *len = ((STRLEN) -1); 1239 return 0; 1240 } 1241 s++; 1242 } 1243 s++; 1244 } 1245 1246 d = s = save; 1247 while (s < send) { 1248 U8 c = *s++; 1249 if (! UTF8_IS_INVARIANT(c)) { 1250 /* Then it is two-byte encoded */ 1251 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); 1252 s++; 1253 } 1254 *d++ = c; 1255 } 1256 *d = '\0'; 1257 *len = d - save; 1258 return save; 1259 } 1260 1261 /* 1262 =for apidoc bytes_from_utf8 1263 1264 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding. 1265 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to 1266 the newly-created string, and updates C<len> to contain the new 1267 length. Returns the original string if no conversion occurs, C<len> 1268 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to 1269 0 if C<s> is converted or consisted entirely of characters that are invariant 1270 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines). 1271 1272 =cut 1273 */ 1274 1275 U8 * 1276 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8) 1277 { 1278 U8 *d; 1279 const U8 *start = s; 1280 const U8 *send; 1281 I32 count = 0; 1282 1283 PERL_ARGS_ASSERT_BYTES_FROM_UTF8; 1284 PERL_UNUSED_CONTEXT; 1285 if (!*is_utf8) 1286 return (U8 *)start; 1287 1288 /* ensure valid UTF-8 and chars < 256 before converting string */ 1289 for (send = s + *len; s < send;) { 1290 if (! UTF8_IS_INVARIANT(*s)) { 1291 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { 1292 return (U8 *)start; 1293 } 1294 count++; 1295 s++; 1296 } 1297 s++; 1298 } 1299 1300 *is_utf8 = FALSE; 1301 1302 Newx(d, (*len) - count + 1, U8); 1303 s = start; start = d; 1304 while (s < send) { 1305 U8 c = *s++; 1306 if (! UTF8_IS_INVARIANT(c)) { 1307 /* Then it is two-byte encoded */ 1308 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); 1309 s++; 1310 } 1311 *d++ = c; 1312 } 1313 *d = '\0'; 1314 *len = d - start; 1315 return (U8 *)start; 1316 } 1317 1318 /* 1319 =for apidoc bytes_to_utf8 1320 1321 Converts a string C<s> of length C<len> bytes from the native encoding into 1322 UTF-8. 1323 Returns a pointer to the newly-created string, and sets C<len> to 1324 reflect the new length in bytes. 1325 1326 A C<NUL> character will be written after the end of the string. 1327 1328 If you want to convert to UTF-8 from encodings other than 1329 the native (Latin1 or EBCDIC), 1330 see L</sv_recode_to_utf8>(). 1331 1332 =cut 1333 */ 1334 1335 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will 1336 likewise need duplication. */ 1337 1338 U8* 1339 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len) 1340 { 1341 const U8 * const send = s + (*len); 1342 U8 *d; 1343 U8 *dst; 1344 1345 PERL_ARGS_ASSERT_BYTES_TO_UTF8; 1346 PERL_UNUSED_CONTEXT; 1347 1348 Newx(d, (*len) * 2 + 1, U8); 1349 dst = d; 1350 1351 while (s < send) { 1352 append_utf8_from_native_byte(*s, &d); 1353 s++; 1354 } 1355 *d = '\0'; 1356 *len = d-dst; 1357 return dst; 1358 } 1359 1360 /* 1361 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8. 1362 * 1363 * Destination must be pre-extended to 3/2 source. Do not use in-place. 1364 * We optimize for native, for obvious reasons. */ 1365 1366 U8* 1367 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) 1368 { 1369 U8* pend; 1370 U8* dstart = d; 1371 1372 PERL_ARGS_ASSERT_UTF16_TO_UTF8; 1373 1374 if (bytelen & 1) 1375 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen); 1376 1377 pend = p + bytelen; 1378 1379 while (p < pend) { 1380 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */ 1381 p += 2; 1382 if (OFFUNI_IS_INVARIANT(uv)) { 1383 *d++ = LATIN1_TO_NATIVE((U8) uv); 1384 continue; 1385 } 1386 if (uv <= MAX_UTF8_TWO_BYTE) { 1387 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv)); 1388 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv)); 1389 continue; 1390 } 1391 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST 1392 #define LAST_HIGH_SURROGATE 0xDBFF 1393 #define FIRST_LOW_SURROGATE 0xDC00 1394 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST 1395 1396 /* This assumes that most uses will be in the first Unicode plane, not 1397 * needing surrogates */ 1398 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST 1399 && uv <= UNICODE_SURROGATE_LAST)) 1400 { 1401 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) { 1402 Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); 1403 } 1404 else { 1405 UV low = (p[0] << 8) + p[1]; 1406 if ( UNLIKELY(low < FIRST_LOW_SURROGATE) 1407 || UNLIKELY(low > LAST_LOW_SURROGATE)) 1408 { 1409 Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); 1410 } 1411 p += 2; 1412 uv = ((uv - FIRST_HIGH_SURROGATE) << 10) 1413 + (low - FIRST_LOW_SURROGATE) + 0x10000; 1414 } 1415 } 1416 #ifdef EBCDIC 1417 d = uvoffuni_to_utf8_flags(d, uv, 0); 1418 #else 1419 if (uv < 0x10000) { 1420 *d++ = (U8)(( uv >> 12) | 0xe0); 1421 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); 1422 *d++ = (U8)(( uv & 0x3f) | 0x80); 1423 continue; 1424 } 1425 else { 1426 *d++ = (U8)(( uv >> 18) | 0xf0); 1427 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80); 1428 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); 1429 *d++ = (U8)(( uv & 0x3f) | 0x80); 1430 continue; 1431 } 1432 #endif 1433 } 1434 *newlen = d - dstart; 1435 return d; 1436 } 1437 1438 /* Note: this one is slightly destructive of the source. */ 1439 1440 U8* 1441 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) 1442 { 1443 U8* s = (U8*)p; 1444 U8* const send = s + bytelen; 1445 1446 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED; 1447 1448 if (bytelen & 1) 1449 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf, 1450 (UV)bytelen); 1451 1452 while (s < send) { 1453 const U8 tmp = s[0]; 1454 s[0] = s[1]; 1455 s[1] = tmp; 1456 s += 2; 1457 } 1458 return utf16_to_utf8(p, d, bytelen, newlen); 1459 } 1460 1461 bool 1462 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c) 1463 { 1464 U8 tmpbuf[UTF8_MAXBYTES+1]; 1465 uvchr_to_utf8(tmpbuf, c); 1466 return _is_utf8_FOO(classnum, tmpbuf); 1467 } 1468 1469 /* Internal function so we can deprecate the external one, and call 1470 this one from other deprecated functions in this file */ 1471 1472 bool 1473 Perl__is_utf8_idstart(pTHX_ const U8 *p) 1474 { 1475 PERL_ARGS_ASSERT__IS_UTF8_IDSTART; 1476 1477 if (*p == '_') 1478 return TRUE; 1479 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL); 1480 } 1481 1482 bool 1483 Perl__is_uni_perl_idcont(pTHX_ UV c) 1484 { 1485 U8 tmpbuf[UTF8_MAXBYTES+1]; 1486 uvchr_to_utf8(tmpbuf, c); 1487 return _is_utf8_perl_idcont(tmpbuf); 1488 } 1489 1490 bool 1491 Perl__is_uni_perl_idstart(pTHX_ UV c) 1492 { 1493 U8 tmpbuf[UTF8_MAXBYTES+1]; 1494 uvchr_to_utf8(tmpbuf, c); 1495 return _is_utf8_perl_idstart(tmpbuf); 1496 } 1497 1498 UV 1499 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s) 1500 { 1501 /* We have the latin1-range values compiled into the core, so just use 1502 * those, converting the result to UTF-8. The only difference between upper 1503 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is 1504 * either "SS" or "Ss". Which one to use is passed into the routine in 1505 * 'S_or_s' to avoid a test */ 1506 1507 UV converted = toUPPER_LATIN1_MOD(c); 1508 1509 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1; 1510 1511 assert(S_or_s == 'S' || S_or_s == 's'); 1512 1513 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for 1514 characters in this range */ 1515 *p = (U8) converted; 1516 *lenp = 1; 1517 return converted; 1518 } 1519 1520 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers, 1521 * which it maps to one of them, so as to only have to have one check for 1522 * it in the main case */ 1523 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { 1524 switch (c) { 1525 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: 1526 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; 1527 break; 1528 case MICRO_SIGN: 1529 converted = GREEK_CAPITAL_LETTER_MU; 1530 break; 1531 #if UNICODE_MAJOR_VERSION > 2 \ 1532 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \ 1533 && UNICODE_DOT_DOT_VERSION >= 8) 1534 case LATIN_SMALL_LETTER_SHARP_S: 1535 *(p)++ = 'S'; 1536 *p = S_or_s; 1537 *lenp = 2; 1538 return 'S'; 1539 #endif 1540 default: 1541 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS); 1542 NOT_REACHED; /* NOTREACHED */ 1543 } 1544 } 1545 1546 *(p)++ = UTF8_TWO_BYTE_HI(converted); 1547 *p = UTF8_TWO_BYTE_LO(converted); 1548 *lenp = 2; 1549 1550 return converted; 1551 } 1552 1553 /* Call the function to convert a UTF-8 encoded character to the specified case. 1554 * Note that there may be more than one character in the result. 1555 * INP is a pointer to the first byte of the input character 1556 * OUTP will be set to the first byte of the string of changed characters. It 1557 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes 1558 * LENP will be set to the length in bytes of the string of changed characters 1559 * 1560 * The functions return the ordinal of the first character in the string of OUTP */ 1561 #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "") 1562 #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "") 1563 #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "") 1564 1565 /* This additionally has the input parameter 'specials', which if non-zero will 1566 * cause this to use the specials hash for folding (meaning get full case 1567 * folding); otherwise, when zero, this implies a simple case fold */ 1568 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL) 1569 1570 UV 1571 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp) 1572 { 1573 /* Convert the Unicode character whose ordinal is <c> to its uppercase 1574 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>. 1575 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since 1576 * the changed version may be longer than the original character. 1577 * 1578 * The ordinal of the first character of the changed version is returned 1579 * (but note, as explained above, that there may be more.) */ 1580 1581 PERL_ARGS_ASSERT_TO_UNI_UPPER; 1582 1583 if (c < 256) { 1584 return _to_upper_title_latin1((U8) c, p, lenp, 'S'); 1585 } 1586 1587 uvchr_to_utf8(p, c); 1588 return CALL_UPPER_CASE(c, p, p, lenp); 1589 } 1590 1591 UV 1592 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp) 1593 { 1594 PERL_ARGS_ASSERT_TO_UNI_TITLE; 1595 1596 if (c < 256) { 1597 return _to_upper_title_latin1((U8) c, p, lenp, 's'); 1598 } 1599 1600 uvchr_to_utf8(p, c); 1601 return CALL_TITLE_CASE(c, p, p, lenp); 1602 } 1603 1604 STATIC U8 1605 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp) 1606 { 1607 /* We have the latin1-range values compiled into the core, so just use 1608 * those, converting the result to UTF-8. Since the result is always just 1609 * one character, we allow <p> to be NULL */ 1610 1611 U8 converted = toLOWER_LATIN1(c); 1612 1613 if (p != NULL) { 1614 if (NATIVE_BYTE_IS_INVARIANT(converted)) { 1615 *p = converted; 1616 *lenp = 1; 1617 } 1618 else { 1619 /* Result is known to always be < 256, so can use the EIGHT_BIT 1620 * macros */ 1621 *p = UTF8_EIGHT_BIT_HI(converted); 1622 *(p+1) = UTF8_EIGHT_BIT_LO(converted); 1623 *lenp = 2; 1624 } 1625 } 1626 return converted; 1627 } 1628 1629 UV 1630 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp) 1631 { 1632 PERL_ARGS_ASSERT_TO_UNI_LOWER; 1633 1634 if (c < 256) { 1635 return to_lower_latin1((U8) c, p, lenp); 1636 } 1637 1638 uvchr_to_utf8(p, c); 1639 return CALL_LOWER_CASE(c, p, p, lenp); 1640 } 1641 1642 UV 1643 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags) 1644 { 1645 /* Corresponds to to_lower_latin1(); <flags> bits meanings: 1646 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited 1647 * FOLD_FLAGS_FULL iff full folding is to be used; 1648 * 1649 * Not to be used for locale folds 1650 */ 1651 1652 UV converted; 1653 1654 PERL_ARGS_ASSERT__TO_FOLD_LATIN1; 1655 PERL_UNUSED_CONTEXT; 1656 1657 assert (! (flags & FOLD_FLAGS_LOCALE)); 1658 1659 if (UNLIKELY(c == MICRO_SIGN)) { 1660 converted = GREEK_SMALL_LETTER_MU; 1661 } 1662 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ 1663 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ 1664 || UNICODE_DOT_DOT_VERSION > 0) 1665 else if ( (flags & FOLD_FLAGS_FULL) 1666 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S)) 1667 { 1668 /* If can't cross 127/128 boundary, can't return "ss"; instead return 1669 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}") 1670 * under those circumstances. */ 1671 if (flags & FOLD_FLAGS_NOMIX_ASCII) { 1672 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; 1673 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, 1674 p, *lenp, U8); 1675 return LATIN_SMALL_LETTER_LONG_S; 1676 } 1677 else { 1678 *(p)++ = 's'; 1679 *p = 's'; 1680 *lenp = 2; 1681 return 's'; 1682 } 1683 } 1684 #endif 1685 else { /* In this range the fold of all other characters is their lower 1686 case */ 1687 converted = toLOWER_LATIN1(c); 1688 } 1689 1690 if (UVCHR_IS_INVARIANT(converted)) { 1691 *p = (U8) converted; 1692 *lenp = 1; 1693 } 1694 else { 1695 *(p)++ = UTF8_TWO_BYTE_HI(converted); 1696 *p = UTF8_TWO_BYTE_LO(converted); 1697 *lenp = 2; 1698 } 1699 1700 return converted; 1701 } 1702 1703 UV 1704 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags) 1705 { 1706 1707 /* Not currently externally documented, and subject to change 1708 * <flags> bits meanings: 1709 * FOLD_FLAGS_FULL iff full folding is to be used; 1710 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying 1711 * locale are to be used. 1712 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited 1713 */ 1714 1715 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS; 1716 1717 if (flags & FOLD_FLAGS_LOCALE) { 1718 /* Treat a UTF-8 locale as not being in locale at all */ 1719 if (IN_UTF8_CTYPE_LOCALE) { 1720 flags &= ~FOLD_FLAGS_LOCALE; 1721 } 1722 else { 1723 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 1724 goto needs_full_generality; 1725 } 1726 } 1727 1728 if (c < 256) { 1729 return _to_fold_latin1((U8) c, p, lenp, 1730 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); 1731 } 1732 1733 /* Here, above 255. If no special needs, just use the macro */ 1734 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) { 1735 uvchr_to_utf8(p, c); 1736 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL); 1737 } 1738 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with 1739 the special flags. */ 1740 U8 utf8_c[UTF8_MAXBYTES + 1]; 1741 1742 needs_full_generality: 1743 uvchr_to_utf8(utf8_c, c); 1744 return _to_utf8_fold_flags(utf8_c, p, lenp, flags); 1745 } 1746 } 1747 1748 PERL_STATIC_INLINE bool 1749 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash, 1750 const char *const swashname, SV* const invlist) 1751 { 1752 /* returns a boolean giving whether or not the UTF8-encoded character that 1753 * starts at <p> is in the swash indicated by <swashname>. <swash> 1754 * contains a pointer to where the swash indicated by <swashname> 1755 * is to be stored; which this routine will do, so that future calls will 1756 * look at <*swash> and only generate a swash if it is not null. <invlist> 1757 * is NULL or an inversion list that defines the swash. If not null, it 1758 * saves time during initialization of the swash. 1759 * 1760 * Note that it is assumed that the buffer length of <p> is enough to 1761 * contain all the bytes that comprise the character. Thus, <*p> should 1762 * have been checked before this call for mal-formedness enough to assure 1763 * that. */ 1764 1765 PERL_ARGS_ASSERT_IS_UTF8_COMMON; 1766 1767 /* The API should have included a length for the UTF-8 character in <p>, 1768 * but it doesn't. We therefore assume that p has been validated at least 1769 * as far as there being enough bytes available in it to accommodate the 1770 * character without reading beyond the end, and pass that number on to the 1771 * validating routine */ 1772 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) { 1773 if (ckWARN_d(WARN_UTF8)) { 1774 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8), 1775 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname); 1776 if (ckWARN(WARN_UTF8)) { /* This will output details as to the 1777 what the malformation is */ 1778 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL); 1779 } 1780 } 1781 return FALSE; 1782 } 1783 if (!*swash) { 1784 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 1785 *swash = _core_swash_init("utf8", 1786 1787 /* Only use the name if there is no inversion 1788 * list; otherwise will go out to disk */ 1789 (invlist) ? "" : swashname, 1790 1791 &PL_sv_undef, 1, 0, invlist, &flags); 1792 } 1793 1794 return swash_fetch(*swash, p, TRUE) != 0; 1795 } 1796 1797 bool 1798 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p) 1799 { 1800 PERL_ARGS_ASSERT__IS_UTF8_FOO; 1801 1802 assert(classnum < _FIRST_NON_SWASH_CC); 1803 1804 return is_utf8_common(p, 1805 &PL_utf8_swash_ptrs[classnum], 1806 swash_property_names[classnum], 1807 PL_XPosix_ptrs[classnum]); 1808 } 1809 1810 bool 1811 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p) 1812 { 1813 SV* invlist = NULL; 1814 1815 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART; 1816 1817 if (! PL_utf8_perl_idstart) { 1818 invlist = _new_invlist_C_array(_Perl_IDStart_invlist); 1819 } 1820 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist); 1821 } 1822 1823 bool 1824 Perl__is_utf8_xidstart(pTHX_ const U8 *p) 1825 { 1826 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART; 1827 1828 if (*p == '_') 1829 return TRUE; 1830 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL); 1831 } 1832 1833 bool 1834 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p) 1835 { 1836 SV* invlist = NULL; 1837 1838 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT; 1839 1840 if (! PL_utf8_perl_idcont) { 1841 invlist = _new_invlist_C_array(_Perl_IDCont_invlist); 1842 } 1843 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist); 1844 } 1845 1846 bool 1847 Perl__is_utf8_idcont(pTHX_ const U8 *p) 1848 { 1849 PERL_ARGS_ASSERT__IS_UTF8_IDCONT; 1850 1851 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL); 1852 } 1853 1854 bool 1855 Perl__is_utf8_xidcont(pTHX_ const U8 *p) 1856 { 1857 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT; 1858 1859 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL); 1860 } 1861 1862 bool 1863 Perl__is_utf8_mark(pTHX_ const U8 *p) 1864 { 1865 PERL_ARGS_ASSERT__IS_UTF8_MARK; 1866 1867 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL); 1868 } 1869 1870 /* 1871 =for apidoc to_utf8_case 1872 1873 Instead use the appropriate one of L</toUPPER_utf8>, 1874 L</toTITLE_utf8>, 1875 L</toLOWER_utf8>, 1876 or L</toFOLD_utf8>. 1877 1878 C<p> contains the pointer to the UTF-8 string encoding 1879 the character that is being converted. This routine assumes that the character 1880 at C<p> is well-formed. 1881 1882 C<ustrp> is a pointer to the character buffer to put the 1883 conversion result to. C<lenp> is a pointer to the length 1884 of the result. 1885 1886 C<swashp> is a pointer to the swash to use. 1887 1888 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>, 1889 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually, 1890 but not always, a multicharacter mapping), is tried first. 1891 1892 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use 1893 any special mappings; C<""> means to use the special mappings. Values other 1894 than these two are treated as the name of the hash containing the special 1895 mappings, like C<"utf8::ToSpecLower">. 1896 1897 C<normal> is a string like C<"ToLower"> which means the swash 1898 C<%utf8::ToLower>. 1899 1900 Code points above the platform's C<IV_MAX> will raise a deprecation warning, 1901 unless those are turned off. 1902 1903 =cut */ 1904 1905 UV 1906 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, 1907 SV **swashp, const char *normal, const char *special) 1908 { 1909 PERL_ARGS_ASSERT_TO_UTF8_CASE; 1910 1911 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special); 1912 } 1913 1914 /* change namve uv1 to 'from' */ 1915 STATIC UV 1916 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp, 1917 SV **swashp, const char *normal, const char *special) 1918 { 1919 STRLEN len = 0; 1920 1921 PERL_ARGS_ASSERT__TO_UTF8_CASE; 1922 1923 /* For code points that don't change case, we already know that the output 1924 * of this function is the unchanged input, so we can skip doing look-ups 1925 * for them. Unfortunately the case-changing code points are scattered 1926 * around. But there are some long consecutive ranges where there are no 1927 * case changing code points. By adding tests, we can eliminate the lookup 1928 * for all the ones in such ranges. This is currently done here only for 1929 * just a few cases where the scripts are in common use in modern commerce 1930 * (and scripts adjacent to those which can be included without additional 1931 * tests). */ 1932 1933 if (uv1 >= 0x0590) { 1934 /* This keeps from needing further processing the code points most 1935 * likely to be used in the following non-cased scripts: Hebrew, 1936 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari, 1937 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada, 1938 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */ 1939 if (uv1 < 0x10A0) { 1940 goto cases_to_self; 1941 } 1942 1943 /* The following largish code point ranges also don't have case 1944 * changes, but khw didn't think they warranted extra tests to speed 1945 * them up (which would slightly slow down everything else above them): 1946 * 1100..139F Hangul Jamo, Ethiopic 1947 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic, 1948 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian, 1949 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham, 1950 * Combining Diacritical Marks Extended, Balinese, 1951 * Sundanese, Batak, Lepcha, Ol Chiki 1952 * 2000..206F General Punctuation 1953 */ 1954 1955 if (uv1 >= 0x2D30) { 1956 1957 /* This keeps the from needing further processing the code points 1958 * most likely to be used in the following non-cased major scripts: 1959 * CJK, Katakana, Hiragana, plus some less-likely scripts. 1960 * 1961 * (0x2D30 above might have to be changed to 2F00 in the unlikely 1962 * event that Unicode eventually allocates the unused block as of 1963 * v8.0 2FE0..2FEF to code points that are cased. khw has verified 1964 * that the test suite will start having failures to alert you 1965 * should that happen) */ 1966 if (uv1 < 0xA640) { 1967 goto cases_to_self; 1968 } 1969 1970 if (uv1 >= 0xAC00) { 1971 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) { 1972 if (ckWARN_d(WARN_SURROGATE)) { 1973 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; 1974 Perl_warner(aTHX_ packWARN(WARN_SURROGATE), 1975 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1); 1976 } 1977 goto cases_to_self; 1978 } 1979 1980 /* AC00..FAFF Catches Hangul syllables and private use, plus 1981 * some others */ 1982 if (uv1 < 0xFB00) { 1983 goto cases_to_self; 1984 1985 } 1986 1987 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) { 1988 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP) 1989 && ckWARN_d(WARN_DEPRECATED)) 1990 { 1991 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), 1992 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP); 1993 } 1994 if (ckWARN_d(WARN_NON_UNICODE)) { 1995 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; 1996 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), 1997 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1); 1998 } 1999 goto cases_to_self; 2000 } 2001 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C 2002 if (UNLIKELY(uv1 2003 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C)) 2004 { 2005 2006 /* As of this writing, this means we avoid swash creation 2007 * for anything beyond low Plane 1 */ 2008 goto cases_to_self; 2009 } 2010 #endif 2011 } 2012 } 2013 2014 /* Note that non-characters are perfectly legal, so no warning should 2015 * be given. There are so few of them, that it isn't worth the extra 2016 * tests to avoid swash creation */ 2017 } 2018 2019 if (!*swashp) /* load on-demand */ 2020 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL); 2021 2022 if (special) { 2023 /* It might be "special" (sometimes, but not always, 2024 * a multicharacter mapping) */ 2025 HV *hv = NULL; 2026 SV **svp; 2027 2028 /* If passed in the specials name, use that; otherwise use any 2029 * given in the swash */ 2030 if (*special != '\0') { 2031 hv = get_hv(special, 0); 2032 } 2033 else { 2034 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0); 2035 if (svp) { 2036 hv = MUTABLE_HV(SvRV(*svp)); 2037 } 2038 } 2039 2040 if (hv 2041 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE)) 2042 && (*svp)) 2043 { 2044 const char *s; 2045 2046 s = SvPV_const(*svp, len); 2047 if (len == 1) 2048 /* EIGHTBIT */ 2049 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp; 2050 else { 2051 Copy(s, ustrp, len, U8); 2052 } 2053 } 2054 } 2055 2056 if (!len && *swashp) { 2057 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */); 2058 2059 if (uv2) { 2060 /* It was "normal" (a single character mapping). */ 2061 len = uvchr_to_utf8(ustrp, uv2) - ustrp; 2062 } 2063 } 2064 2065 if (len) { 2066 if (lenp) { 2067 *lenp = len; 2068 } 2069 return valid_utf8_to_uvchr(ustrp, 0); 2070 } 2071 2072 /* Here, there was no mapping defined, which means that the code point maps 2073 * to itself. Return the inputs */ 2074 cases_to_self: 2075 len = UTF8SKIP(p); 2076 if (p != ustrp) { /* Don't copy onto itself */ 2077 Copy(p, ustrp, len, U8); 2078 } 2079 2080 if (lenp) 2081 *lenp = len; 2082 2083 return uv1; 2084 2085 } 2086 2087 STATIC UV 2088 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp) 2089 { 2090 /* This is called when changing the case of a UTF-8-encoded character above 2091 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the 2092 * result contains a character that crosses the 255/256 boundary, disallow 2093 * the change, and return the original code point. See L<perlfunc/lc> for 2094 * why; 2095 * 2096 * p points to the original string whose case was changed; assumed 2097 * by this routine to be well-formed 2098 * result the code point of the first character in the changed-case string 2099 * ustrp points to the changed-case string (<result> represents its first char) 2100 * lenp points to the length of <ustrp> */ 2101 2102 UV original; /* To store the first code point of <p> */ 2103 2104 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING; 2105 2106 assert(UTF8_IS_ABOVE_LATIN1(*p)); 2107 2108 /* We know immediately if the first character in the string crosses the 2109 * boundary, so can skip */ 2110 if (result > 255) { 2111 2112 /* Look at every character in the result; if any cross the 2113 * boundary, the whole thing is disallowed */ 2114 U8* s = ustrp + UTF8SKIP(ustrp); 2115 U8* e = ustrp + *lenp; 2116 while (s < e) { 2117 if (! UTF8_IS_ABOVE_LATIN1(*s)) { 2118 goto bad_crossing; 2119 } 2120 s += UTF8SKIP(s); 2121 } 2122 2123 /* Here, no characters crossed, result is ok as-is, but we warn. */ 2124 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p)); 2125 return result; 2126 } 2127 2128 bad_crossing: 2129 2130 /* Failed, have to return the original */ 2131 original = valid_utf8_to_uvchr(p, lenp); 2132 2133 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 2134 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 2135 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; " 2136 "resolved to \"\\x{%"UVXf"}\".", 2137 OP_DESC(PL_op), 2138 original, 2139 original); 2140 Copy(p, ustrp, *lenp, char); 2141 return original; 2142 } 2143 2144 /* 2145 =for apidoc to_utf8_upper 2146 2147 Instead use L</toUPPER_utf8>. 2148 2149 =cut */ 2150 2151 /* Not currently externally documented, and subject to change: 2152 * <flags> is set iff iff the rules from the current underlying locale are to 2153 * be used. */ 2154 2155 UV 2156 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags) 2157 { 2158 UV result; 2159 2160 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS; 2161 2162 if (flags) { 2163 /* Treat a UTF-8 locale as not being in locale at all */ 2164 if (IN_UTF8_CTYPE_LOCALE) { 2165 flags = FALSE; 2166 } 2167 else { 2168 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 2169 } 2170 } 2171 2172 if (UTF8_IS_INVARIANT(*p)) { 2173 if (flags) { 2174 result = toUPPER_LC(*p); 2175 } 2176 else { 2177 return _to_upper_title_latin1(*p, ustrp, lenp, 'S'); 2178 } 2179 } 2180 else if UTF8_IS_DOWNGRADEABLE_START(*p) { 2181 if (flags) { 2182 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); 2183 result = toUPPER_LC(c); 2184 } 2185 else { 2186 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), 2187 ustrp, lenp, 'S'); 2188 } 2189 } 2190 else { /* UTF-8, ord above 255 */ 2191 result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp); 2192 2193 if (flags) { 2194 result = check_locale_boundary_crossing(p, result, ustrp, lenp); 2195 } 2196 return result; 2197 } 2198 2199 /* Here, used locale rules. Convert back to UTF-8 */ 2200 if (UTF8_IS_INVARIANT(result)) { 2201 *ustrp = (U8) result; 2202 *lenp = 1; 2203 } 2204 else { 2205 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); 2206 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); 2207 *lenp = 2; 2208 } 2209 2210 return result; 2211 } 2212 2213 /* 2214 =for apidoc to_utf8_title 2215 2216 Instead use L</toTITLE_utf8>. 2217 2218 =cut */ 2219 2220 /* Not currently externally documented, and subject to change: 2221 * <flags> is set iff the rules from the current underlying locale are to be 2222 * used. Since titlecase is not defined in POSIX, for other than a 2223 * UTF-8 locale, uppercase is used instead for code points < 256. 2224 */ 2225 2226 UV 2227 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags) 2228 { 2229 UV result; 2230 2231 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS; 2232 2233 if (flags) { 2234 /* Treat a UTF-8 locale as not being in locale at all */ 2235 if (IN_UTF8_CTYPE_LOCALE) { 2236 flags = FALSE; 2237 } 2238 else { 2239 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 2240 } 2241 } 2242 2243 if (UTF8_IS_INVARIANT(*p)) { 2244 if (flags) { 2245 result = toUPPER_LC(*p); 2246 } 2247 else { 2248 return _to_upper_title_latin1(*p, ustrp, lenp, 's'); 2249 } 2250 } 2251 else if UTF8_IS_DOWNGRADEABLE_START(*p) { 2252 if (flags) { 2253 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); 2254 result = toUPPER_LC(c); 2255 } 2256 else { 2257 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), 2258 ustrp, lenp, 's'); 2259 } 2260 } 2261 else { /* UTF-8, ord above 255 */ 2262 result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp); 2263 2264 if (flags) { 2265 result = check_locale_boundary_crossing(p, result, ustrp, lenp); 2266 } 2267 return result; 2268 } 2269 2270 /* Here, used locale rules. Convert back to UTF-8 */ 2271 if (UTF8_IS_INVARIANT(result)) { 2272 *ustrp = (U8) result; 2273 *lenp = 1; 2274 } 2275 else { 2276 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); 2277 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); 2278 *lenp = 2; 2279 } 2280 2281 return result; 2282 } 2283 2284 /* 2285 =for apidoc to_utf8_lower 2286 2287 Instead use L</toLOWER_utf8>. 2288 2289 =cut */ 2290 2291 /* Not currently externally documented, and subject to change: 2292 * <flags> is set iff iff the rules from the current underlying locale are to 2293 * be used. 2294 */ 2295 2296 UV 2297 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags) 2298 { 2299 UV result; 2300 2301 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS; 2302 2303 if (flags) { 2304 /* Treat a UTF-8 locale as not being in locale at all */ 2305 if (IN_UTF8_CTYPE_LOCALE) { 2306 flags = FALSE; 2307 } 2308 else { 2309 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 2310 } 2311 } 2312 2313 if (UTF8_IS_INVARIANT(*p)) { 2314 if (flags) { 2315 result = toLOWER_LC(*p); 2316 } 2317 else { 2318 return to_lower_latin1(*p, ustrp, lenp); 2319 } 2320 } 2321 else if UTF8_IS_DOWNGRADEABLE_START(*p) { 2322 if (flags) { 2323 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); 2324 result = toLOWER_LC(c); 2325 } 2326 else { 2327 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), 2328 ustrp, lenp); 2329 } 2330 } 2331 else { /* UTF-8, ord above 255 */ 2332 result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp); 2333 2334 if (flags) { 2335 result = check_locale_boundary_crossing(p, result, ustrp, lenp); 2336 } 2337 2338 return result; 2339 } 2340 2341 /* Here, used locale rules. Convert back to UTF-8 */ 2342 if (UTF8_IS_INVARIANT(result)) { 2343 *ustrp = (U8) result; 2344 *lenp = 1; 2345 } 2346 else { 2347 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); 2348 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); 2349 *lenp = 2; 2350 } 2351 2352 return result; 2353 } 2354 2355 /* 2356 =for apidoc to_utf8_fold 2357 2358 Instead use L</toFOLD_utf8>. 2359 2360 =cut */ 2361 2362 /* Not currently externally documented, and subject to change, 2363 * in <flags> 2364 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying 2365 * locale are to be used. 2366 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used; 2367 * otherwise simple folds 2368 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are 2369 * prohibited 2370 */ 2371 2372 UV 2373 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags) 2374 { 2375 UV result; 2376 2377 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS; 2378 2379 /* These are mutually exclusive */ 2380 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII))); 2381 2382 assert(p != ustrp); /* Otherwise overwrites */ 2383 2384 if (flags & FOLD_FLAGS_LOCALE) { 2385 /* Treat a UTF-8 locale as not being in locale at all */ 2386 if (IN_UTF8_CTYPE_LOCALE) { 2387 flags &= ~FOLD_FLAGS_LOCALE; 2388 } 2389 else { 2390 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 2391 } 2392 } 2393 2394 if (UTF8_IS_INVARIANT(*p)) { 2395 if (flags & FOLD_FLAGS_LOCALE) { 2396 result = toFOLD_LC(*p); 2397 } 2398 else { 2399 return _to_fold_latin1(*p, ustrp, lenp, 2400 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); 2401 } 2402 } 2403 else if UTF8_IS_DOWNGRADEABLE_START(*p) { 2404 if (flags & FOLD_FLAGS_LOCALE) { 2405 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); 2406 result = toFOLD_LC(c); 2407 } 2408 else { 2409 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), 2410 ustrp, lenp, 2411 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); 2412 } 2413 } 2414 else { /* UTF-8, ord above 255 */ 2415 result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL); 2416 2417 if (flags & FOLD_FLAGS_LOCALE) { 2418 2419 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8 2420 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1; 2421 2422 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8 2423 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8 2424 2425 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1; 2426 2427 /* Special case these two characters, as what normally gets 2428 * returned under locale doesn't work */ 2429 if (UTF8SKIP(p) == cap_sharp_s_len 2430 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len)) 2431 { 2432 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 2433 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 2434 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; " 2435 "resolved to \"\\x{17F}\\x{17F}\"."); 2436 goto return_long_s; 2437 } 2438 else 2439 #endif 2440 if (UTF8SKIP(p) == long_s_t_len 2441 && memEQ((char *) p, LONG_S_T, long_s_t_len)) 2442 { 2443 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 2444 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 2445 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; " 2446 "resolved to \"\\x{FB06}\"."); 2447 goto return_ligature_st; 2448 } 2449 2450 #if UNICODE_MAJOR_VERSION == 3 \ 2451 && UNICODE_DOT_VERSION == 0 \ 2452 && UNICODE_DOT_DOT_VERSION == 1 2453 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8 2454 2455 /* And special case this on this Unicode version only, for the same 2456 * reaons the other two are special cased. They would cross the 2457 * 255/256 boundary which is forbidden under /l, and so the code 2458 * wouldn't catch that they are equivalent (which they are only in 2459 * this release) */ 2460 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1 2461 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1)) 2462 { 2463 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ 2464 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 2465 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; " 2466 "resolved to \"\\x{0131}\"."); 2467 goto return_dotless_i; 2468 } 2469 #endif 2470 2471 return check_locale_boundary_crossing(p, result, ustrp, lenp); 2472 } 2473 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) { 2474 return result; 2475 } 2476 else { 2477 /* This is called when changing the case of a UTF-8-encoded 2478 * character above the ASCII range, and the result should not 2479 * contain an ASCII character. */ 2480 2481 UV original; /* To store the first code point of <p> */ 2482 2483 /* Look at every character in the result; if any cross the 2484 * boundary, the whole thing is disallowed */ 2485 U8* s = ustrp; 2486 U8* e = ustrp + *lenp; 2487 while (s < e) { 2488 if (isASCII(*s)) { 2489 /* Crossed, have to return the original */ 2490 original = valid_utf8_to_uvchr(p, lenp); 2491 2492 /* But in these instances, there is an alternative we can 2493 * return that is valid */ 2494 if (original == LATIN_SMALL_LETTER_SHARP_S 2495 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */ 2496 || original == LATIN_CAPITAL_LETTER_SHARP_S 2497 #endif 2498 ) { 2499 goto return_long_s; 2500 } 2501 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) { 2502 goto return_ligature_st; 2503 } 2504 #if UNICODE_MAJOR_VERSION == 3 \ 2505 && UNICODE_DOT_VERSION == 0 \ 2506 && UNICODE_DOT_DOT_VERSION == 1 2507 2508 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) { 2509 goto return_dotless_i; 2510 } 2511 #endif 2512 Copy(p, ustrp, *lenp, char); 2513 return original; 2514 } 2515 s += UTF8SKIP(s); 2516 } 2517 2518 /* Here, no characters crossed, result is ok as-is */ 2519 return result; 2520 } 2521 } 2522 2523 /* Here, used locale rules. Convert back to UTF-8 */ 2524 if (UTF8_IS_INVARIANT(result)) { 2525 *ustrp = (U8) result; 2526 *lenp = 1; 2527 } 2528 else { 2529 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); 2530 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); 2531 *lenp = 2; 2532 } 2533 2534 return result; 2535 2536 return_long_s: 2537 /* Certain folds to 'ss' are prohibited by the options, but they do allow 2538 * folds to a string of two of these characters. By returning this 2539 * instead, then, e.g., 2540 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}") 2541 * works. */ 2542 2543 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; 2544 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, 2545 ustrp, *lenp, U8); 2546 return LATIN_SMALL_LETTER_LONG_S; 2547 2548 return_ligature_st: 2549 /* Two folds to 'st' are prohibited by the options; instead we pick one and 2550 * have the other one fold to it */ 2551 2552 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1; 2553 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8); 2554 return LATIN_SMALL_LIGATURE_ST; 2555 2556 #if UNICODE_MAJOR_VERSION == 3 \ 2557 && UNICODE_DOT_VERSION == 0 \ 2558 && UNICODE_DOT_DOT_VERSION == 1 2559 2560 return_dotless_i: 2561 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1; 2562 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8); 2563 return LATIN_SMALL_LETTER_DOTLESS_I; 2564 2565 #endif 2566 2567 } 2568 2569 /* Note: 2570 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch(). 2571 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8". 2572 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl. 2573 */ 2574 2575 SV* 2576 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none) 2577 { 2578 PERL_ARGS_ASSERT_SWASH_INIT; 2579 2580 /* Returns a copy of a swash initiated by the called function. This is the 2581 * public interface, and returning a copy prevents others from doing 2582 * mischief on the original */ 2583 2584 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL)); 2585 } 2586 2587 SV* 2588 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p) 2589 { 2590 2591 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST 2592 * use the following define */ 2593 2594 #define CORE_SWASH_INIT_RETURN(x) \ 2595 PL_curpm= old_PL_curpm; \ 2596 return x 2597 2598 /* Initialize and return a swash, creating it if necessary. It does this 2599 * by calling utf8_heavy.pl in the general case. The returned value may be 2600 * the swash's inversion list instead if the input parameters allow it. 2601 * Which is returned should be immaterial to callers, as the only 2602 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(), 2603 * and swash_to_invlist() handle both these transparently. 2604 * 2605 * This interface should only be used by functions that won't destroy or 2606 * adversely change the swash, as doing so affects all other uses of the 2607 * swash in the program; the general public should use 'Perl_swash_init' 2608 * instead. 2609 * 2610 * pkg is the name of the package that <name> should be in. 2611 * name is the name of the swash to find. Typically it is a Unicode 2612 * property name, including user-defined ones 2613 * listsv is a string to initialize the swash with. It must be of the form 2614 * documented as the subroutine return value in 2615 * L<perlunicode/User-Defined Character Properties> 2616 * minbits is the number of bits required to represent each data element. 2617 * It is '1' for binary properties. 2618 * none I (khw) do not understand this one, but it is used only in tr///. 2619 * invlist is an inversion list to initialize the swash with (or NULL) 2620 * flags_p if non-NULL is the address of various input and output flag bits 2621 * to the routine, as follows: ('I' means is input to the routine; 2622 * 'O' means output from the routine. Only flags marked O are 2623 * meaningful on return.) 2624 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash 2625 * came from a user-defined property. (I O) 2626 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking 2627 * when the swash cannot be located, to simply return NULL. (I) 2628 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a 2629 * return of an inversion list instead of a swash hash if this routine 2630 * thinks that would result in faster execution of swash_fetch() later 2631 * on. (I) 2632 * 2633 * Thus there are three possible inputs to find the swash: <name>, 2634 * <listsv>, and <invlist>. At least one must be specified. The result 2635 * will be the union of the specified ones, although <listsv>'s various 2636 * actions can intersect, etc. what <name> gives. To avoid going out to 2637 * disk at all, <invlist> should specify completely what the swash should 2638 * have, and <listsv> should be &PL_sv_undef and <name> should be "". 2639 * 2640 * <invlist> is only valid for binary properties */ 2641 2642 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */ 2643 2644 SV* retval = &PL_sv_undef; 2645 HV* swash_hv = NULL; 2646 const int invlist_swash_boundary = 2647 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST) 2648 ? 512 /* Based on some benchmarking, but not extensive, see commit 2649 message */ 2650 : -1; /* Never return just an inversion list */ 2651 2652 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist); 2653 assert(! invlist || minbits == 1); 2654 2655 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex 2656 that triggered the swash init and the swash init perl logic itself. 2657 See perl #122747 */ 2658 2659 /* If data was passed in to go out to utf8_heavy to find the swash of, do 2660 * so */ 2661 if (listsv != &PL_sv_undef || strNE(name, "")) { 2662 dSP; 2663 const size_t pkg_len = strlen(pkg); 2664 const size_t name_len = strlen(name); 2665 HV * const stash = gv_stashpvn(pkg, pkg_len, 0); 2666 SV* errsv_save; 2667 GV *method; 2668 2669 PERL_ARGS_ASSERT__CORE_SWASH_INIT; 2670 2671 PUSHSTACKi(PERLSI_MAGIC); 2672 ENTER; 2673 SAVEHINTS(); 2674 save_re_context(); 2675 /* We might get here via a subroutine signature which uses a utf8 2676 * parameter name, at which point PL_subname will have been set 2677 * but not yet used. */ 2678 save_item(PL_subname); 2679 if (PL_parser && PL_parser->error_count) 2680 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0; 2681 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1); 2682 if (!method) { /* demand load UTF-8 */ 2683 ENTER; 2684 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); 2685 GvSV(PL_errgv) = NULL; 2686 #ifndef NO_TAINT_SUPPORT 2687 /* It is assumed that callers of this routine are not passing in 2688 * any user derived data. */ 2689 /* Need to do this after save_re_context() as it will set 2690 * PL_tainted to 1 while saving $1 etc (see the code after getrx: 2691 * in Perl_magic_get). Even line to create errsv_save can turn on 2692 * PL_tainted. */ 2693 SAVEBOOL(TAINT_get); 2694 TAINT_NOT; 2695 #endif 2696 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len), 2697 NULL); 2698 { 2699 /* Not ERRSV, as there is no need to vivify a scalar we are 2700 about to discard. */ 2701 SV * const errsv = GvSV(PL_errgv); 2702 if (!SvTRUE(errsv)) { 2703 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); 2704 SvREFCNT_dec(errsv); 2705 } 2706 } 2707 LEAVE; 2708 } 2709 SPAGAIN; 2710 PUSHMARK(SP); 2711 EXTEND(SP,5); 2712 mPUSHp(pkg, pkg_len); 2713 mPUSHp(name, name_len); 2714 PUSHs(listsv); 2715 mPUSHi(minbits); 2716 mPUSHi(none); 2717 PUTBACK; 2718 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); 2719 GvSV(PL_errgv) = NULL; 2720 /* If we already have a pointer to the method, no need to use 2721 * call_method() to repeat the lookup. */ 2722 if (method 2723 ? call_sv(MUTABLE_SV(method), G_SCALAR) 2724 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD)) 2725 { 2726 retval = *PL_stack_sp--; 2727 SvREFCNT_inc(retval); 2728 } 2729 { 2730 /* Not ERRSV. See above. */ 2731 SV * const errsv = GvSV(PL_errgv); 2732 if (!SvTRUE(errsv)) { 2733 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); 2734 SvREFCNT_dec(errsv); 2735 } 2736 } 2737 LEAVE; 2738 POPSTACK; 2739 if (IN_PERL_COMPILETIME) { 2740 CopHINTS_set(PL_curcop, PL_hints); 2741 } 2742 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) { 2743 if (SvPOK(retval)) { 2744 2745 /* If caller wants to handle missing properties, let them */ 2746 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) { 2747 CORE_SWASH_INIT_RETURN(NULL); 2748 } 2749 Perl_croak(aTHX_ 2750 "Can't find Unicode property definition \"%"SVf"\"", 2751 SVfARG(retval)); 2752 NOT_REACHED; /* NOTREACHED */ 2753 } 2754 } 2755 } /* End of calling the module to find the swash */ 2756 2757 /* If this operation fetched a swash, and we will need it later, get it */ 2758 if (retval != &PL_sv_undef 2759 && (minbits == 1 || (flags_p 2760 && ! (*flags_p 2761 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)))) 2762 { 2763 swash_hv = MUTABLE_HV(SvRV(retval)); 2764 2765 /* If we don't already know that there is a user-defined component to 2766 * this swash, and the user has indicated they wish to know if there is 2767 * one (by passing <flags_p>), find out */ 2768 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) { 2769 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE); 2770 if (user_defined && SvUV(*user_defined)) { 2771 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY; 2772 } 2773 } 2774 } 2775 2776 /* Make sure there is an inversion list for binary properties */ 2777 if (minbits == 1) { 2778 SV** swash_invlistsvp = NULL; 2779 SV* swash_invlist = NULL; 2780 bool invlist_in_swash_is_valid = FALSE; 2781 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has 2782 an unclaimed reference count */ 2783 2784 /* If this operation fetched a swash, get its already existing 2785 * inversion list, or create one for it */ 2786 2787 if (swash_hv) { 2788 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE); 2789 if (swash_invlistsvp) { 2790 swash_invlist = *swash_invlistsvp; 2791 invlist_in_swash_is_valid = TRUE; 2792 } 2793 else { 2794 swash_invlist = _swash_to_invlist(retval); 2795 swash_invlist_unclaimed = TRUE; 2796 } 2797 } 2798 2799 /* If an inversion list was passed in, have to include it */ 2800 if (invlist) { 2801 2802 /* Any fetched swash will by now have an inversion list in it; 2803 * otherwise <swash_invlist> will be NULL, indicating that we 2804 * didn't fetch a swash */ 2805 if (swash_invlist) { 2806 2807 /* Add the passed-in inversion list, which invalidates the one 2808 * already stored in the swash */ 2809 invlist_in_swash_is_valid = FALSE; 2810 _invlist_union(invlist, swash_invlist, &swash_invlist); 2811 } 2812 else { 2813 2814 /* Here, there is no swash already. Set up a minimal one, if 2815 * we are going to return a swash */ 2816 if ((int) _invlist_len(invlist) > invlist_swash_boundary) { 2817 swash_hv = newHV(); 2818 retval = newRV_noinc(MUTABLE_SV(swash_hv)); 2819 } 2820 swash_invlist = invlist; 2821 } 2822 } 2823 2824 /* Here, we have computed the union of all the passed-in data. It may 2825 * be that there was an inversion list in the swash which didn't get 2826 * touched; otherwise save the computed one */ 2827 if (! invlist_in_swash_is_valid 2828 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary) 2829 { 2830 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist)) 2831 { 2832 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); 2833 } 2834 /* We just stole a reference count. */ 2835 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE; 2836 else SvREFCNT_inc_simple_void_NN(swash_invlist); 2837 } 2838 2839 SvREADONLY_on(swash_invlist); 2840 2841 /* Use the inversion list stand-alone if small enough */ 2842 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) { 2843 SvREFCNT_dec(retval); 2844 if (!swash_invlist_unclaimed) 2845 SvREFCNT_inc_simple_void_NN(swash_invlist); 2846 retval = newRV_noinc(swash_invlist); 2847 } 2848 } 2849 2850 CORE_SWASH_INIT_RETURN(retval); 2851 #undef CORE_SWASH_INIT_RETURN 2852 } 2853 2854 2855 /* This API is wrong for special case conversions since we may need to 2856 * return several Unicode characters for a single Unicode character 2857 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is 2858 * the lower-level routine, and it is similarly broken for returning 2859 * multiple values. --jhi 2860 * For those, you should use S__to_utf8_case() instead */ 2861 /* Now SWASHGET is recasted into S_swatch_get in this file. */ 2862 2863 /* Note: 2864 * Returns the value of property/mapping C<swash> for the first character 2865 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is 2866 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr> 2867 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>. 2868 * 2869 * A "swash" is a hash which contains initially the keys/values set up by 2870 * SWASHNEW. The purpose is to be able to completely represent a Unicode 2871 * property for all possible code points. Things are stored in a compact form 2872 * (see utf8_heavy.pl) so that calculation is required to find the actual 2873 * property value for a given code point. As code points are looked up, new 2874 * key/value pairs are added to the hash, so that the calculation doesn't have 2875 * to ever be re-done. Further, each calculation is done, not just for the 2876 * desired one, but for a whole block of code points adjacent to that one. 2877 * For binary properties on ASCII machines, the block is usually for 64 code 2878 * points, starting with a code point evenly divisible by 64. Thus if the 2879 * property value for code point 257 is requested, the code goes out and 2880 * calculates the property values for all 64 code points between 256 and 319, 2881 * and stores these as a single 64-bit long bit vector, called a "swatch", 2882 * under the key for code point 256. The key is the UTF-8 encoding for code 2883 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding 2884 * for a code point is 13 bytes, the key will be 12 bytes long. If the value 2885 * for code point 258 is then requested, this code realizes that it would be 2886 * stored under the key for 256, and would find that value and extract the 2887 * relevant bit, offset from 256. 2888 * 2889 * Non-binary properties are stored in as many bits as necessary to represent 2890 * their values (32 currently, though the code is more general than that), not 2891 * as single bits, but the principal is the same: the value for each key is a 2892 * vector that encompasses the property values for all code points whose UTF-8 2893 * representations are represented by the key. That is, for all code points 2894 * whose UTF-8 representations are length N bytes, and the key is the first N-1 2895 * bytes of that. 2896 */ 2897 UV 2898 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8) 2899 { 2900 HV *const hv = MUTABLE_HV(SvRV(swash)); 2901 U32 klen; 2902 U32 off; 2903 STRLEN slen = 0; 2904 STRLEN needents; 2905 const U8 *tmps = NULL; 2906 SV *swatch; 2907 const U8 c = *ptr; 2908 2909 PERL_ARGS_ASSERT_SWASH_FETCH; 2910 2911 /* If it really isn't a hash, it isn't really swash; must be an inversion 2912 * list */ 2913 if (SvTYPE(hv) != SVt_PVHV) { 2914 return _invlist_contains_cp((SV*)hv, 2915 (do_utf8) 2916 ? valid_utf8_to_uvchr(ptr, NULL) 2917 : c); 2918 } 2919 2920 /* We store the values in a "swatch" which is a vec() value in a swash 2921 * hash. Code points 0-255 are a single vec() stored with key length 2922 * (klen) 0. All other code points have a UTF-8 representation 2923 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which 2924 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key 2925 * length for them is the length of the encoded char - 1. ptr[klen] is the 2926 * final byte in the sequence representing the character */ 2927 if (!do_utf8 || UTF8_IS_INVARIANT(c)) { 2928 klen = 0; 2929 needents = 256; 2930 off = c; 2931 } 2932 else if (UTF8_IS_DOWNGRADEABLE_START(c)) { 2933 klen = 0; 2934 needents = 256; 2935 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1)); 2936 } 2937 else { 2938 klen = UTF8SKIP(ptr) - 1; 2939 2940 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into 2941 * the vec is the final byte in the sequence. (In EBCDIC this is 2942 * converted to I8 to get consecutive values.) To help you visualize 2943 * all this: 2944 * Straight 1047 After final byte 2945 * UTF-8 UTF-EBCDIC I8 transform 2946 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0 2947 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1 2948 * ... 2949 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9 2950 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA 2951 * ... 2952 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2 2953 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3 2954 * ... 2955 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB 2956 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC 2957 * ... 2958 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF 2959 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41 2960 * 2961 * (There are no discontinuities in the elided (...) entries.) 2962 * The UTF-8 key for these 33 code points is '\xD0' (which also is the 2963 * key for the next 31, up through U+043F, whose UTF-8 final byte is 2964 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points. 2965 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an 2966 * index into the vec() swatch (after subtracting 0x80, which we 2967 * actually do with an '&'). 2968 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32 2969 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has 2970 * dicontinuities which go away by transforming it into I8, and we 2971 * effectively subtract 0xA0 to get the index. */ 2972 needents = (1 << UTF_ACCUMULATION_SHIFT); 2973 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK; 2974 } 2975 2976 /* 2977 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test 2978 * suite. (That is, only 7-8% overall over just a hash cache. Still, 2979 * it's nothing to sniff at.) Pity we usually come through at least 2980 * two function calls to get here... 2981 * 2982 * NB: this code assumes that swatches are never modified, once generated! 2983 */ 2984 2985 if (hv == PL_last_swash_hv && 2986 klen == PL_last_swash_klen && 2987 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) ) 2988 { 2989 tmps = PL_last_swash_tmps; 2990 slen = PL_last_swash_slen; 2991 } 2992 else { 2993 /* Try our second-level swatch cache, kept in a hash. */ 2994 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE); 2995 2996 /* If not cached, generate it via swatch_get */ 2997 if (!svp || !SvPOK(*svp) 2998 || !(tmps = (const U8*)SvPV_const(*svp, slen))) 2999 { 3000 if (klen) { 3001 const UV code_point = valid_utf8_to_uvchr(ptr, NULL); 3002 swatch = swatch_get(swash, 3003 code_point & ~((UV)needents - 1), 3004 needents); 3005 } 3006 else { /* For the first 256 code points, the swatch has a key of 3007 length 0 */ 3008 swatch = swatch_get(swash, 0, needents); 3009 } 3010 3011 if (IN_PERL_COMPILETIME) 3012 CopHINTS_set(PL_curcop, PL_hints); 3013 3014 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0); 3015 3016 if (!svp || !(tmps = (U8*)SvPV(*svp, slen)) 3017 || (slen << 3) < needents) 3018 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, " 3019 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf, 3020 svp, tmps, (UV)slen, (UV)needents); 3021 } 3022 3023 PL_last_swash_hv = hv; 3024 assert(klen <= sizeof(PL_last_swash_key)); 3025 PL_last_swash_klen = (U8)klen; 3026 /* FIXME change interpvar.h? */ 3027 PL_last_swash_tmps = (U8 *) tmps; 3028 PL_last_swash_slen = slen; 3029 if (klen) 3030 Copy(ptr, PL_last_swash_key, klen, U8); 3031 } 3032 3033 switch ((int)((slen << 3) / needents)) { 3034 case 1: 3035 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0; 3036 case 8: 3037 return ((UV) tmps[off]); 3038 case 16: 3039 off <<= 1; 3040 return 3041 ((UV) tmps[off ] << 8) + 3042 ((UV) tmps[off + 1]); 3043 case 32: 3044 off <<= 2; 3045 return 3046 ((UV) tmps[off ] << 24) + 3047 ((UV) tmps[off + 1] << 16) + 3048 ((UV) tmps[off + 2] << 8) + 3049 ((UV) tmps[off + 3]); 3050 } 3051 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, " 3052 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents); 3053 NORETURN_FUNCTION_END; 3054 } 3055 3056 /* Read a single line of the main body of the swash input text. These are of 3057 * the form: 3058 * 0053 0056 0073 3059 * where each number is hex. The first two numbers form the minimum and 3060 * maximum of a range, and the third is the value associated with the range. 3061 * Not all swashes should have a third number 3062 * 3063 * On input: l points to the beginning of the line to be examined; it points 3064 * to somewhere in the string of the whole input text, and is 3065 * terminated by a \n or the null string terminator. 3066 * lend points to the null terminator of that string 3067 * wants_value is non-zero if the swash expects a third number 3068 * typestr is the name of the swash's mapping, like 'ToLower' 3069 * On output: *min, *max, and *val are set to the values read from the line. 3070 * returns a pointer just beyond the line examined. If there was no 3071 * valid min number on the line, returns lend+1 3072 */ 3073 3074 STATIC U8* 3075 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val, 3076 const bool wants_value, const U8* const typestr) 3077 { 3078 const int typeto = typestr[0] == 'T' && typestr[1] == 'o'; 3079 STRLEN numlen; /* Length of the number */ 3080 I32 flags = PERL_SCAN_SILENT_ILLDIGIT 3081 | PERL_SCAN_DISALLOW_PREFIX 3082 | PERL_SCAN_SILENT_NON_PORTABLE; 3083 3084 /* nl points to the next \n in the scan */ 3085 U8* const nl = (U8*)memchr(l, '\n', lend - l); 3086 3087 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE; 3088 3089 /* Get the first number on the line: the range minimum */ 3090 numlen = lend - l; 3091 *min = grok_hex((char *)l, &numlen, &flags, NULL); 3092 *max = *min; /* So can never return without setting max */ 3093 if (numlen) /* If found a hex number, position past it */ 3094 l += numlen; 3095 else if (nl) { /* Else, go handle next line, if any */ 3096 return nl + 1; /* 1 is length of "\n" */ 3097 } 3098 else { /* Else, no next line */ 3099 return lend + 1; /* to LIST's end at which \n is not found */ 3100 } 3101 3102 /* The max range value follows, separated by a BLANK */ 3103 if (isBLANK(*l)) { 3104 ++l; 3105 flags = PERL_SCAN_SILENT_ILLDIGIT 3106 | PERL_SCAN_DISALLOW_PREFIX 3107 | PERL_SCAN_SILENT_NON_PORTABLE; 3108 numlen = lend - l; 3109 *max = grok_hex((char *)l, &numlen, &flags, NULL); 3110 if (numlen) 3111 l += numlen; 3112 else /* If no value here, it is a single element range */ 3113 *max = *min; 3114 3115 /* Non-binary tables have a third entry: what the first element of the 3116 * range maps to. The map for those currently read here is in hex */ 3117 if (wants_value) { 3118 if (isBLANK(*l)) { 3119 ++l; 3120 flags = PERL_SCAN_SILENT_ILLDIGIT 3121 | PERL_SCAN_DISALLOW_PREFIX 3122 | PERL_SCAN_SILENT_NON_PORTABLE; 3123 numlen = lend - l; 3124 *val = grok_hex((char *)l, &numlen, &flags, NULL); 3125 if (numlen) 3126 l += numlen; 3127 else 3128 *val = 0; 3129 } 3130 else { 3131 *val = 0; 3132 if (typeto) { 3133 /* diag_listed_as: To%s: illegal mapping '%s' */ 3134 Perl_croak(aTHX_ "%s: illegal mapping '%s'", 3135 typestr, l); 3136 } 3137 } 3138 } 3139 else 3140 *val = 0; /* bits == 1, then any val should be ignored */ 3141 } 3142 else { /* Nothing following range min, should be single element with no 3143 mapping expected */ 3144 if (wants_value) { 3145 *val = 0; 3146 if (typeto) { 3147 /* diag_listed_as: To%s: illegal mapping '%s' */ 3148 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l); 3149 } 3150 } 3151 else 3152 *val = 0; /* bits == 1, then val should be ignored */ 3153 } 3154 3155 /* Position to next line if any, or EOF */ 3156 if (nl) 3157 l = nl + 1; 3158 else 3159 l = lend; 3160 3161 return l; 3162 } 3163 3164 /* Note: 3165 * Returns a swatch (a bit vector string) for a code point sequence 3166 * that starts from the value C<start> and comprises the number C<span>. 3167 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl). 3168 * Should be used via swash_fetch, which will cache the swatch in C<swash>. 3169 */ 3170 STATIC SV* 3171 S_swatch_get(pTHX_ SV* swash, UV start, UV span) 3172 { 3173 SV *swatch; 3174 U8 *l, *lend, *x, *xend, *s, *send; 3175 STRLEN lcur, xcur, scur; 3176 HV *const hv = MUTABLE_HV(SvRV(swash)); 3177 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE); 3178 3179 SV** listsvp = NULL; /* The string containing the main body of the table */ 3180 SV** extssvp = NULL; 3181 SV** invert_it_svp = NULL; 3182 U8* typestr = NULL; 3183 STRLEN bits; 3184 STRLEN octets; /* if bits == 1, then octets == 0 */ 3185 UV none; 3186 UV end = start + span; 3187 3188 if (invlistsvp == NULL) { 3189 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE); 3190 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE); 3191 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE); 3192 extssvp = hv_fetchs(hv, "EXTRAS", FALSE); 3193 listsvp = hv_fetchs(hv, "LIST", FALSE); 3194 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); 3195 3196 bits = SvUV(*bitssvp); 3197 none = SvUV(*nonesvp); 3198 typestr = (U8*)SvPV_nolen(*typesvp); 3199 } 3200 else { 3201 bits = 1; 3202 none = 0; 3203 } 3204 octets = bits >> 3; /* if bits == 1, then octets == 0 */ 3205 3206 PERL_ARGS_ASSERT_SWATCH_GET; 3207 3208 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) { 3209 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf, 3210 (UV)bits); 3211 } 3212 3213 /* If overflowed, use the max possible */ 3214 if (end < start) { 3215 end = UV_MAX; 3216 span = end - start; 3217 } 3218 3219 /* create and initialize $swatch */ 3220 scur = octets ? (span * octets) : (span + 7) / 8; 3221 swatch = newSV(scur); 3222 SvPOK_on(swatch); 3223 s = (U8*)SvPVX(swatch); 3224 if (octets && none) { 3225 const U8* const e = s + scur; 3226 while (s < e) { 3227 if (bits == 8) 3228 *s++ = (U8)(none & 0xff); 3229 else if (bits == 16) { 3230 *s++ = (U8)((none >> 8) & 0xff); 3231 *s++ = (U8)( none & 0xff); 3232 } 3233 else if (bits == 32) { 3234 *s++ = (U8)((none >> 24) & 0xff); 3235 *s++ = (U8)((none >> 16) & 0xff); 3236 *s++ = (U8)((none >> 8) & 0xff); 3237 *s++ = (U8)( none & 0xff); 3238 } 3239 } 3240 *s = '\0'; 3241 } 3242 else { 3243 (void)memzero((U8*)s, scur + 1); 3244 } 3245 SvCUR_set(swatch, scur); 3246 s = (U8*)SvPVX(swatch); 3247 3248 if (invlistsvp) { /* If has an inversion list set up use that */ 3249 _invlist_populate_swatch(*invlistsvp, start, end, s); 3250 return swatch; 3251 } 3252 3253 /* read $swash->{LIST} */ 3254 l = (U8*)SvPV(*listsvp, lcur); 3255 lend = l + lcur; 3256 while (l < lend) { 3257 UV min, max, val, upper; 3258 l = swash_scan_list_line(l, lend, &min, &max, &val, 3259 cBOOL(octets), typestr); 3260 if (l > lend) { 3261 break; 3262 } 3263 3264 /* If looking for something beyond this range, go try the next one */ 3265 if (max < start) 3266 continue; 3267 3268 /* <end> is generally 1 beyond where we want to set things, but at the 3269 * platform's infinity, where we can't go any higher, we want to 3270 * include the code point at <end> */ 3271 upper = (max < end) 3272 ? max 3273 : (max != UV_MAX || end != UV_MAX) 3274 ? end - 1 3275 : end; 3276 3277 if (octets) { 3278 UV key; 3279 if (min < start) { 3280 if (!none || val < none) { 3281 val += start - min; 3282 } 3283 min = start; 3284 } 3285 for (key = min; key <= upper; key++) { 3286 STRLEN offset; 3287 /* offset must be non-negative (start <= min <= key < end) */ 3288 offset = octets * (key - start); 3289 if (bits == 8) 3290 s[offset] = (U8)(val & 0xff); 3291 else if (bits == 16) { 3292 s[offset ] = (U8)((val >> 8) & 0xff); 3293 s[offset + 1] = (U8)( val & 0xff); 3294 } 3295 else if (bits == 32) { 3296 s[offset ] = (U8)((val >> 24) & 0xff); 3297 s[offset + 1] = (U8)((val >> 16) & 0xff); 3298 s[offset + 2] = (U8)((val >> 8) & 0xff); 3299 s[offset + 3] = (U8)( val & 0xff); 3300 } 3301 3302 if (!none || val < none) 3303 ++val; 3304 } 3305 } 3306 else { /* bits == 1, then val should be ignored */ 3307 UV key; 3308 if (min < start) 3309 min = start; 3310 3311 for (key = min; key <= upper; key++) { 3312 const STRLEN offset = (STRLEN)(key - start); 3313 s[offset >> 3] |= 1 << (offset & 7); 3314 } 3315 } 3316 } /* while */ 3317 3318 /* Invert if the data says it should be. Assumes that bits == 1 */ 3319 if (invert_it_svp && SvUV(*invert_it_svp)) { 3320 3321 /* Unicode properties should come with all bits above PERL_UNICODE_MAX 3322 * be 0, and their inversion should also be 0, as we don't succeed any 3323 * Unicode property matches for non-Unicode code points */ 3324 if (start <= PERL_UNICODE_MAX) { 3325 3326 /* The code below assumes that we never cross the 3327 * Unicode/above-Unicode boundary in a range, as otherwise we would 3328 * have to figure out where to stop flipping the bits. Since this 3329 * boundary is divisible by a large power of 2, and swatches comes 3330 * in small powers of 2, this should be a valid assumption */ 3331 assert(start + span - 1 <= PERL_UNICODE_MAX); 3332 3333 send = s + scur; 3334 while (s < send) { 3335 *s = ~(*s); 3336 s++; 3337 } 3338 } 3339 } 3340 3341 /* read $swash->{EXTRAS} 3342 * This code also copied to swash_to_invlist() below */ 3343 x = (U8*)SvPV(*extssvp, xcur); 3344 xend = x + xcur; 3345 while (x < xend) { 3346 STRLEN namelen; 3347 U8 *namestr; 3348 SV** othersvp; 3349 HV* otherhv; 3350 STRLEN otherbits; 3351 SV **otherbitssvp, *other; 3352 U8 *s, *o, *nl; 3353 STRLEN slen, olen; 3354 3355 const U8 opc = *x++; 3356 if (opc == '\n') 3357 continue; 3358 3359 nl = (U8*)memchr(x, '\n', xend - x); 3360 3361 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { 3362 if (nl) { 3363 x = nl + 1; /* 1 is length of "\n" */ 3364 continue; 3365 } 3366 else { 3367 x = xend; /* to EXTRAS' end at which \n is not found */ 3368 break; 3369 } 3370 } 3371 3372 namestr = x; 3373 if (nl) { 3374 namelen = nl - namestr; 3375 x = nl + 1; 3376 } 3377 else { 3378 namelen = xend - namestr; 3379 x = xend; 3380 } 3381 3382 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); 3383 otherhv = MUTABLE_HV(SvRV(*othersvp)); 3384 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); 3385 otherbits = (STRLEN)SvUV(*otherbitssvp); 3386 if (bits < otherbits) 3387 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, " 3388 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits); 3389 3390 /* The "other" swatch must be destroyed after. */ 3391 other = swatch_get(*othersvp, start, span); 3392 o = (U8*)SvPV(other, olen); 3393 3394 if (!olen) 3395 Perl_croak(aTHX_ "panic: swatch_get got improper swatch"); 3396 3397 s = (U8*)SvPV(swatch, slen); 3398 if (bits == 1 && otherbits == 1) { 3399 if (slen != olen) 3400 Perl_croak(aTHX_ "panic: swatch_get found swatch length " 3401 "mismatch, slen=%"UVuf", olen=%"UVuf, 3402 (UV)slen, (UV)olen); 3403 3404 switch (opc) { 3405 case '+': 3406 while (slen--) 3407 *s++ |= *o++; 3408 break; 3409 case '!': 3410 while (slen--) 3411 *s++ |= ~*o++; 3412 break; 3413 case '-': 3414 while (slen--) 3415 *s++ &= ~*o++; 3416 break; 3417 case '&': 3418 while (slen--) 3419 *s++ &= *o++; 3420 break; 3421 default: 3422 break; 3423 } 3424 } 3425 else { 3426 STRLEN otheroctets = otherbits >> 3; 3427 STRLEN offset = 0; 3428 U8* const send = s + slen; 3429 3430 while (s < send) { 3431 UV otherval = 0; 3432 3433 if (otherbits == 1) { 3434 otherval = (o[offset >> 3] >> (offset & 7)) & 1; 3435 ++offset; 3436 } 3437 else { 3438 STRLEN vlen = otheroctets; 3439 otherval = *o++; 3440 while (--vlen) { 3441 otherval <<= 8; 3442 otherval |= *o++; 3443 } 3444 } 3445 3446 if (opc == '+' && otherval) 3447 NOOP; /* replace with otherval */ 3448 else if (opc == '!' && !otherval) 3449 otherval = 1; 3450 else if (opc == '-' && otherval) 3451 otherval = 0; 3452 else if (opc == '&' && !otherval) 3453 otherval = 0; 3454 else { 3455 s += octets; /* no replacement */ 3456 continue; 3457 } 3458 3459 if (bits == 8) 3460 *s++ = (U8)( otherval & 0xff); 3461 else if (bits == 16) { 3462 *s++ = (U8)((otherval >> 8) & 0xff); 3463 *s++ = (U8)( otherval & 0xff); 3464 } 3465 else if (bits == 32) { 3466 *s++ = (U8)((otherval >> 24) & 0xff); 3467 *s++ = (U8)((otherval >> 16) & 0xff); 3468 *s++ = (U8)((otherval >> 8) & 0xff); 3469 *s++ = (U8)( otherval & 0xff); 3470 } 3471 } 3472 } 3473 sv_free(other); /* through with it! */ 3474 } /* while */ 3475 return swatch; 3476 } 3477 3478 HV* 3479 Perl__swash_inversion_hash(pTHX_ SV* const swash) 3480 { 3481 3482 /* Subject to change or removal. For use only in regcomp.c and regexec.c 3483 * Can't be used on a property that is subject to user override, as it 3484 * relies on the value of SPECIALS in the swash which would be set by 3485 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set 3486 * for overridden properties 3487 * 3488 * Returns a hash which is the inversion and closure of a swash mapping. 3489 * For example, consider the input lines: 3490 * 004B 006B 3491 * 004C 006C 3492 * 212A 006B 3493 * 3494 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for 3495 * 006C. The value for each key is an array. For 006C, the array would 3496 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there 3497 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A. 3498 * 3499 * Note that there are no elements in the hash for 004B, 004C, 212A. The 3500 * keys are only code points that are folded-to, so it isn't a full closure. 3501 * 3502 * Essentially, for any code point, it gives all the code points that map to 3503 * it, or the list of 'froms' for that point. 3504 * 3505 * Currently it ignores any additions or deletions from other swashes, 3506 * looking at just the main body of the swash, and if there are SPECIALS 3507 * in the swash, at that hash 3508 * 3509 * The specials hash can be extra code points, and most likely consists of 3510 * maps from single code points to multiple ones (each expressed as a string 3511 * of UTF-8 characters). This function currently returns only 1-1 mappings. 3512 * However consider this possible input in the specials hash: 3513 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074 3514 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074 3515 * 3516 * Both FB05 and FB06 map to the same multi-char sequence, which we don't 3517 * currently handle. But it also means that FB05 and FB06 are equivalent in 3518 * a 1-1 mapping which we should handle, and this relationship may not be in 3519 * the main table. Therefore this function examines all the multi-char 3520 * sequences and adds the 1-1 mappings that come out of that. 3521 * 3522 * XXX This function was originally intended to be multipurpose, but its 3523 * only use is quite likely to remain for constructing the inversion of 3524 * the CaseFolding (//i) property. If it were more general purpose for 3525 * regex patterns, it would have to do the FB05/FB06 game for simple folds, 3526 * because certain folds are prohibited under /iaa and /il. As an example, 3527 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both 3528 * equivalent under /i. But under /iaa and /il, the folds to 'i' are 3529 * prohibited, so we would not figure out that they fold to each other. 3530 * Code could be written to automatically figure this out, similar to the 3531 * code that does this for multi-character folds, but this is the only case 3532 * where something like this is ever likely to happen, as all the single 3533 * char folds to the 0-255 range are now quite settled. Instead there is a 3534 * little special code that is compiled only for this Unicode version. This 3535 * is smaller and didn't require much coding time to do. But this makes 3536 * this routine strongly tied to being used just for CaseFolding. If ever 3537 * it should be generalized, this would have to be fixed */ 3538 3539 U8 *l, *lend; 3540 STRLEN lcur; 3541 HV *const hv = MUTABLE_HV(SvRV(swash)); 3542 3543 /* The string containing the main body of the table. This will have its 3544 * assertion fail if the swash has been converted to its inversion list */ 3545 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE); 3546 3547 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE); 3548 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE); 3549 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE); 3550 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/ 3551 const U8* const typestr = (U8*)SvPV_nolen(*typesvp); 3552 const STRLEN bits = SvUV(*bitssvp); 3553 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */ 3554 const UV none = SvUV(*nonesvp); 3555 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0); 3556 3557 HV* ret = newHV(); 3558 3559 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH; 3560 3561 /* Must have at least 8 bits to get the mappings */ 3562 if (bits != 8 && bits != 16 && bits != 32) { 3563 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf, 3564 (UV)bits); 3565 } 3566 3567 if (specials_p) { /* It might be "special" (sometimes, but not always, a 3568 mapping to more than one character */ 3569 3570 /* Construct an inverse mapping hash for the specials */ 3571 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p)); 3572 HV * specials_inverse = newHV(); 3573 char *char_from; /* the lhs of the map */ 3574 I32 from_len; /* its byte length */ 3575 char *char_to; /* the rhs of the map */ 3576 I32 to_len; /* its byte length */ 3577 SV *sv_to; /* and in a sv */ 3578 AV* from_list; /* list of things that map to each 'to' */ 3579 3580 hv_iterinit(specials_hv); 3581 3582 /* The keys are the characters (in UTF-8) that map to the corresponding 3583 * UTF-8 string value. Iterate through the list creating the inverse 3584 * list. */ 3585 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) { 3586 SV** listp; 3587 if (! SvPOK(sv_to)) { 3588 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() " 3589 "unexpectedly is not a string, flags=%lu", 3590 (unsigned long)SvFLAGS(sv_to)); 3591 } 3592 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %"UVXf", First char of to is %"UVXf"\n", valid_utf8_to_uvchr((U8*) char_from, 0), valid_utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/ 3593 3594 /* Each key in the inverse list is a mapped-to value, and the key's 3595 * hash value is a list of the strings (each in UTF-8) that map to 3596 * it. Those strings are all one character long */ 3597 if ((listp = hv_fetch(specials_inverse, 3598 SvPVX(sv_to), 3599 SvCUR(sv_to), 0))) 3600 { 3601 from_list = (AV*) *listp; 3602 } 3603 else { /* No entry yet for it: create one */ 3604 from_list = newAV(); 3605 if (! hv_store(specials_inverse, 3606 SvPVX(sv_to), 3607 SvCUR(sv_to), 3608 (SV*) from_list, 0)) 3609 { 3610 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); 3611 } 3612 } 3613 3614 /* Here have the list associated with this 'to' (perhaps newly 3615 * created and empty). Just add to it. Note that we ASSUME that 3616 * the input is guaranteed to not have duplications, so we don't 3617 * check for that. Duplications just slow down execution time. */ 3618 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE)); 3619 } 3620 3621 /* Here, 'specials_inverse' contains the inverse mapping. Go through 3622 * it looking for cases like the FB05/FB06 examples above. There would 3623 * be an entry in the hash like 3624 * 'st' => [ FB05, FB06 ] 3625 * In this example we will create two lists that get stored in the 3626 * returned hash, 'ret': 3627 * FB05 => [ FB05, FB06 ] 3628 * FB06 => [ FB05, FB06 ] 3629 * 3630 * Note that there is nothing to do if the array only has one element. 3631 * (In the normal 1-1 case handled below, we don't have to worry about 3632 * two lists, as everything gets tied to the single list that is 3633 * generated for the single character 'to'. But here, we are omitting 3634 * that list, ('st' in the example), so must have multiple lists.) */ 3635 while ((from_list = (AV *) hv_iternextsv(specials_inverse, 3636 &char_to, &to_len))) 3637 { 3638 if (av_tindex_nomg(from_list) > 0) { 3639 SSize_t i; 3640 3641 /* We iterate over all combinations of i,j to place each code 3642 * point on each list */ 3643 for (i = 0; i <= av_tindex_nomg(from_list); i++) { 3644 SSize_t j; 3645 AV* i_list = newAV(); 3646 SV** entryp = av_fetch(from_list, i, FALSE); 3647 if (entryp == NULL) { 3648 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed"); 3649 } 3650 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) { 3651 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp)); 3652 } 3653 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp), 3654 (SV*) i_list, FALSE)) 3655 { 3656 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); 3657 } 3658 3659 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/ 3660 for (j = 0; j <= av_tindex_nomg(from_list); j++) { 3661 entryp = av_fetch(from_list, j, FALSE); 3662 if (entryp == NULL) { 3663 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed"); 3664 } 3665 3666 /* When i==j this adds itself to the list */ 3667 av_push(i_list, newSVuv(utf8_to_uvchr_buf( 3668 (U8*) SvPVX(*entryp), 3669 (U8*) SvPVX(*entryp) + SvCUR(*entryp), 3670 0))); 3671 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/ 3672 } 3673 } 3674 } 3675 } 3676 SvREFCNT_dec(specials_inverse); /* done with it */ 3677 } /* End of specials */ 3678 3679 /* read $swash->{LIST} */ 3680 3681 #if UNICODE_MAJOR_VERSION == 3 \ 3682 && UNICODE_DOT_VERSION == 0 \ 3683 && UNICODE_DOT_DOT_VERSION == 1 3684 3685 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a 3686 * rule so that things work under /iaa and /il */ 3687 3688 SV * mod_listsv = sv_mortalcopy(*listsvp); 3689 sv_catpv(mod_listsv, "130\t130\t131\n"); 3690 l = (U8*)SvPV(mod_listsv, lcur); 3691 3692 #else 3693 3694 l = (U8*)SvPV(*listsvp, lcur); 3695 3696 #endif 3697 3698 lend = l + lcur; 3699 3700 /* Go through each input line */ 3701 while (l < lend) { 3702 UV min, max, val; 3703 UV inverse; 3704 l = swash_scan_list_line(l, lend, &min, &max, &val, 3705 cBOOL(octets), typestr); 3706 if (l > lend) { 3707 break; 3708 } 3709 3710 /* Each element in the range is to be inverted */ 3711 for (inverse = min; inverse <= max; inverse++) { 3712 AV* list; 3713 SV** listp; 3714 IV i; 3715 bool found_key = FALSE; 3716 bool found_inverse = FALSE; 3717 3718 /* The key is the inverse mapping */ 3719 char key[UTF8_MAXBYTES+1]; 3720 char* key_end = (char *) uvchr_to_utf8((U8*) key, val); 3721 STRLEN key_len = key_end - key; 3722 3723 /* Get the list for the map */ 3724 if ((listp = hv_fetch(ret, key, key_len, FALSE))) { 3725 list = (AV*) *listp; 3726 } 3727 else { /* No entry yet for it: create one */ 3728 list = newAV(); 3729 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) { 3730 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); 3731 } 3732 } 3733 3734 /* Look through list to see if this inverse mapping already is 3735 * listed, or if there is a mapping to itself already */ 3736 for (i = 0; i <= av_tindex_nomg(list); i++) { 3737 SV** entryp = av_fetch(list, i, FALSE); 3738 SV* entry; 3739 UV uv; 3740 if (entryp == NULL) { 3741 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed"); 3742 } 3743 entry = *entryp; 3744 uv = SvUV(entry); 3745 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/ 3746 if (uv == val) { 3747 found_key = TRUE; 3748 } 3749 if (uv == inverse) { 3750 found_inverse = TRUE; 3751 } 3752 3753 /* No need to continue searching if found everything we are 3754 * looking for */ 3755 if (found_key && found_inverse) { 3756 break; 3757 } 3758 } 3759 3760 /* Make sure there is a mapping to itself on the list */ 3761 if (! found_key) { 3762 av_push(list, newSVuv(val)); 3763 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/ 3764 } 3765 3766 3767 /* Simply add the value to the list */ 3768 if (! found_inverse) { 3769 av_push(list, newSVuv(inverse)); 3770 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/ 3771 } 3772 3773 /* swatch_get() increments the value of val for each element in the 3774 * range. That makes more compact tables possible. You can 3775 * express the capitalization, for example, of all consecutive 3776 * letters with a single line: 0061\t007A\t0041 This maps 0061 to 3777 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none', 3778 * and it's not documented; it appears to be used only in 3779 * implementing tr//; I copied the semantics from swatch_get(), just 3780 * in case */ 3781 if (!none || val < none) { 3782 ++val; 3783 } 3784 } 3785 } 3786 3787 return ret; 3788 } 3789 3790 SV* 3791 Perl__swash_to_invlist(pTHX_ SV* const swash) 3792 { 3793 3794 /* Subject to change or removal. For use only in one place in regcomp.c. 3795 * Ownership is given to one reference count in the returned SV* */ 3796 3797 U8 *l, *lend; 3798 char *loc; 3799 STRLEN lcur; 3800 HV *const hv = MUTABLE_HV(SvRV(swash)); 3801 UV elements = 0; /* Number of elements in the inversion list */ 3802 U8 empty[] = ""; 3803 SV** listsvp; 3804 SV** typesvp; 3805 SV** bitssvp; 3806 SV** extssvp; 3807 SV** invert_it_svp; 3808 3809 U8* typestr; 3810 STRLEN bits; 3811 STRLEN octets; /* if bits == 1, then octets == 0 */ 3812 U8 *x, *xend; 3813 STRLEN xcur; 3814 3815 SV* invlist; 3816 3817 PERL_ARGS_ASSERT__SWASH_TO_INVLIST; 3818 3819 /* If not a hash, it must be the swash's inversion list instead */ 3820 if (SvTYPE(hv) != SVt_PVHV) { 3821 return SvREFCNT_inc_simple_NN((SV*) hv); 3822 } 3823 3824 /* The string containing the main body of the table */ 3825 listsvp = hv_fetchs(hv, "LIST", FALSE); 3826 typesvp = hv_fetchs(hv, "TYPE", FALSE); 3827 bitssvp = hv_fetchs(hv, "BITS", FALSE); 3828 extssvp = hv_fetchs(hv, "EXTRAS", FALSE); 3829 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); 3830 3831 typestr = (U8*)SvPV_nolen(*typesvp); 3832 bits = SvUV(*bitssvp); 3833 octets = bits >> 3; /* if bits == 1, then octets == 0 */ 3834 3835 /* read $swash->{LIST} */ 3836 if (SvPOK(*listsvp)) { 3837 l = (U8*)SvPV(*listsvp, lcur); 3838 } 3839 else { 3840 /* LIST legitimately doesn't contain a string during compilation phases 3841 * of Perl itself, before the Unicode tables are generated. In this 3842 * case, just fake things up by creating an empty list */ 3843 l = empty; 3844 lcur = 0; 3845 } 3846 loc = (char *) l; 3847 lend = l + lcur; 3848 3849 if (*l == 'V') { /* Inversion list format */ 3850 const char *after_atou = (char *) lend; 3851 UV element0; 3852 UV* other_elements_ptr; 3853 3854 /* The first number is a count of the rest */ 3855 l++; 3856 if (!grok_atoUV((const char *)l, &elements, &after_atou)) { 3857 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list"); 3858 } 3859 if (elements == 0) { 3860 invlist = _new_invlist(0); 3861 } 3862 else { 3863 while (isSPACE(*l)) l++; 3864 l = (U8 *) after_atou; 3865 3866 /* Get the 0th element, which is needed to setup the inversion list */ 3867 while (isSPACE(*l)) l++; 3868 if (!grok_atoUV((const char *)l, &element0, &after_atou)) { 3869 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list"); 3870 } 3871 l = (U8 *) after_atou; 3872 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr); 3873 elements--; 3874 3875 /* Then just populate the rest of the input */ 3876 while (elements-- > 0) { 3877 if (l > lend) { 3878 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements); 3879 } 3880 while (isSPACE(*l)) l++; 3881 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) { 3882 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list"); 3883 } 3884 l = (U8 *) after_atou; 3885 } 3886 } 3887 } 3888 else { 3889 3890 /* Scan the input to count the number of lines to preallocate array 3891 * size based on worst possible case, which is each line in the input 3892 * creates 2 elements in the inversion list: 1) the beginning of a 3893 * range in the list; 2) the beginning of a range not in the list. */ 3894 while ((loc = (strchr(loc, '\n'))) != NULL) { 3895 elements += 2; 3896 loc++; 3897 } 3898 3899 /* If the ending is somehow corrupt and isn't a new line, add another 3900 * element for the final range that isn't in the inversion list */ 3901 if (! (*lend == '\n' 3902 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n')))) 3903 { 3904 elements++; 3905 } 3906 3907 invlist = _new_invlist(elements); 3908 3909 /* Now go through the input again, adding each range to the list */ 3910 while (l < lend) { 3911 UV start, end; 3912 UV val; /* Not used by this function */ 3913 3914 l = swash_scan_list_line(l, lend, &start, &end, &val, 3915 cBOOL(octets), typestr); 3916 3917 if (l > lend) { 3918 break; 3919 } 3920 3921 invlist = _add_range_to_invlist(invlist, start, end); 3922 } 3923 } 3924 3925 /* Invert if the data says it should be */ 3926 if (invert_it_svp && SvUV(*invert_it_svp)) { 3927 _invlist_invert(invlist); 3928 } 3929 3930 /* This code is copied from swatch_get() 3931 * read $swash->{EXTRAS} */ 3932 x = (U8*)SvPV(*extssvp, xcur); 3933 xend = x + xcur; 3934 while (x < xend) { 3935 STRLEN namelen; 3936 U8 *namestr; 3937 SV** othersvp; 3938 HV* otherhv; 3939 STRLEN otherbits; 3940 SV **otherbitssvp, *other; 3941 U8 *nl; 3942 3943 const U8 opc = *x++; 3944 if (opc == '\n') 3945 continue; 3946 3947 nl = (U8*)memchr(x, '\n', xend - x); 3948 3949 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { 3950 if (nl) { 3951 x = nl + 1; /* 1 is length of "\n" */ 3952 continue; 3953 } 3954 else { 3955 x = xend; /* to EXTRAS' end at which \n is not found */ 3956 break; 3957 } 3958 } 3959 3960 namestr = x; 3961 if (nl) { 3962 namelen = nl - namestr; 3963 x = nl + 1; 3964 } 3965 else { 3966 namelen = xend - namestr; 3967 x = xend; 3968 } 3969 3970 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); 3971 otherhv = MUTABLE_HV(SvRV(*othersvp)); 3972 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); 3973 otherbits = (STRLEN)SvUV(*otherbitssvp); 3974 3975 if (bits != otherbits || bits != 1) { 3976 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean " 3977 "properties, bits=%"UVuf", otherbits=%"UVuf, 3978 (UV)bits, (UV)otherbits); 3979 } 3980 3981 /* The "other" swatch must be destroyed after. */ 3982 other = _swash_to_invlist((SV *)*othersvp); 3983 3984 /* End of code copied from swatch_get() */ 3985 switch (opc) { 3986 case '+': 3987 _invlist_union(invlist, other, &invlist); 3988 break; 3989 case '!': 3990 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist); 3991 break; 3992 case '-': 3993 _invlist_subtract(invlist, other, &invlist); 3994 break; 3995 case '&': 3996 _invlist_intersection(invlist, other, &invlist); 3997 break; 3998 default: 3999 break; 4000 } 4001 sv_free(other); /* through with it! */ 4002 } 4003 4004 SvREADONLY_on(invlist); 4005 return invlist; 4006 } 4007 4008 SV* 4009 Perl__get_swash_invlist(pTHX_ SV* const swash) 4010 { 4011 SV** ptr; 4012 4013 PERL_ARGS_ASSERT__GET_SWASH_INVLIST; 4014 4015 if (! SvROK(swash)) { 4016 return NULL; 4017 } 4018 4019 /* If it really isn't a hash, it isn't really swash; must be an inversion 4020 * list */ 4021 if (SvTYPE(SvRV(swash)) != SVt_PVHV) { 4022 return SvRV(swash); 4023 } 4024 4025 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE); 4026 if (! ptr) { 4027 return NULL; 4028 } 4029 4030 return *ptr; 4031 } 4032 4033 bool 4034 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len) 4035 { 4036 /* May change: warns if surrogates, non-character code points, or 4037 * non-Unicode code points are in s which has length len bytes. Returns 4038 * TRUE if none found; FALSE otherwise. The only other validity check is 4039 * to make sure that this won't exceed the string's length. 4040 * 4041 * Code points above the platform's C<IV_MAX> will raise a deprecation 4042 * warning, unless those are turned off. */ 4043 4044 const U8* const e = s + len; 4045 bool ok = TRUE; 4046 4047 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT; 4048 4049 while (s < e) { 4050 if (UTF8SKIP(s) > len) { 4051 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), 4052 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print"); 4053 return FALSE; 4054 } 4055 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) { 4056 STRLEN char_len; 4057 if (UTF8_IS_SUPER(s, e)) { 4058 if ( ckWARN_d(WARN_NON_UNICODE) 4059 || ( ckWARN_d(WARN_DEPRECATED) 4060 #if defined(UV_IS_QUAD) 4061 /* 2**63 and up meet these conditions provided we have 4062 * a 64-bit word. */ 4063 # ifdef EBCDIC 4064 && *s == 0xFE && e - s >= UTF8_MAXBYTES 4065 && s[1] >= 0x49 4066 # else 4067 && *s == 0xFF && e -s >= UTF8_MAXBYTES 4068 && s[2] >= 0x88 4069 # endif 4070 #else /* Below is 32-bit words */ 4071 /* 2**31 and above meet these conditions on all EBCDIC 4072 * pages recognized for 32-bit platforms */ 4073 # ifdef EBCDIC 4074 && *s == 0xFE && e - s >= UTF8_MAXBYTES 4075 && s[6] >= 0x43 4076 # else 4077 && *s >= 0xFE 4078 # endif 4079 #endif 4080 )) { 4081 /* A side effect of this function will be to warn */ 4082 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_SUPER); 4083 ok = FALSE; 4084 } 4085 } 4086 else if (UTF8_IS_SURROGATE(s, e)) { 4087 if (ckWARN_d(WARN_SURROGATE)) { 4088 /* This has a different warning than the one the called 4089 * function would output, so can't just call it, unlike we 4090 * do for the non-chars and above-unicodes */ 4091 UV uv = utf8_to_uvchr_buf(s, e, &char_len); 4092 Perl_warner(aTHX_ packWARN(WARN_SURROGATE), 4093 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv); 4094 ok = FALSE; 4095 } 4096 } 4097 else if ((UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) { 4098 /* A side effect of this function will be to warn */ 4099 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_NONCHAR); 4100 ok = FALSE; 4101 } 4102 } 4103 s += UTF8SKIP(s); 4104 } 4105 4106 return ok; 4107 } 4108 4109 /* 4110 =for apidoc pv_uni_display 4111 4112 Build to the scalar C<dsv> a displayable version of the string C<spv>, 4113 length C<len>, the displayable version being at most C<pvlim> bytes long 4114 (if longer, the rest is truncated and C<"..."> will be appended). 4115 4116 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display 4117 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH> 4118 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">) 4119 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">). 4120 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both 4121 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on. 4122 4123 The pointer to the PV of the C<dsv> is returned. 4124 4125 See also L</sv_uni_display>. 4126 4127 =cut */ 4128 char * 4129 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags) 4130 { 4131 int truncated = 0; 4132 const char *s, *e; 4133 4134 PERL_ARGS_ASSERT_PV_UNI_DISPLAY; 4135 4136 sv_setpvs(dsv, ""); 4137 SvUTF8_off(dsv); 4138 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) { 4139 UV u; 4140 /* This serves double duty as a flag and a character to print after 4141 a \ when flags & UNI_DISPLAY_BACKSLASH is true. 4142 */ 4143 char ok = 0; 4144 4145 if (pvlim && SvCUR(dsv) >= pvlim) { 4146 truncated++; 4147 break; 4148 } 4149 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0); 4150 if (u < 256) { 4151 const unsigned char c = (unsigned char)u & 0xFF; 4152 if (flags & UNI_DISPLAY_BACKSLASH) { 4153 switch (c) { 4154 case '\n': 4155 ok = 'n'; break; 4156 case '\r': 4157 ok = 'r'; break; 4158 case '\t': 4159 ok = 't'; break; 4160 case '\f': 4161 ok = 'f'; break; 4162 case '\a': 4163 ok = 'a'; break; 4164 case '\\': 4165 ok = '\\'; break; 4166 default: break; 4167 } 4168 if (ok) { 4169 const char string = ok; 4170 sv_catpvs(dsv, "\\"); 4171 sv_catpvn(dsv, &string, 1); 4172 } 4173 } 4174 /* isPRINT() is the locale-blind version. */ 4175 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) { 4176 const char string = c; 4177 sv_catpvn(dsv, &string, 1); 4178 ok = 1; 4179 } 4180 } 4181 if (!ok) 4182 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u); 4183 } 4184 if (truncated) 4185 sv_catpvs(dsv, "..."); 4186 4187 return SvPVX(dsv); 4188 } 4189 4190 /* 4191 =for apidoc sv_uni_display 4192 4193 Build to the scalar C<dsv> a displayable version of the scalar C<sv>, 4194 the displayable version being at most C<pvlim> bytes long 4195 (if longer, the rest is truncated and "..." will be appended). 4196 4197 The C<flags> argument is as in L</pv_uni_display>(). 4198 4199 The pointer to the PV of the C<dsv> is returned. 4200 4201 =cut 4202 */ 4203 char * 4204 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags) 4205 { 4206 const char * const ptr = 4207 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv); 4208 4209 PERL_ARGS_ASSERT_SV_UNI_DISPLAY; 4210 4211 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr, 4212 SvCUR(ssv), pvlim, flags); 4213 } 4214 4215 /* 4216 =for apidoc foldEQ_utf8 4217 4218 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both 4219 of which may be in UTF-8) are the same case-insensitively; false otherwise. 4220 How far into the strings to compare is determined by other input parameters. 4221 4222 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode; 4223 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2> 4224 with respect to C<s2>. 4225 4226 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold 4227 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The 4228 scan will not be considered to be a match unless the goal is reached, and 4229 scanning won't continue past that goal. Correspondingly for C<l2> with respect to 4230 C<s2>. 4231 4232 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is 4233 considered an end pointer to the position 1 byte past the maximum point 4234 in C<s1> beyond which scanning will not continue under any circumstances. 4235 (This routine assumes that UTF-8 encoded input strings are not malformed; 4236 malformed input can cause it to read past C<pe1>). 4237 This means that if both C<l1> and C<pe1> are specified, and C<pe1> 4238 is less than C<s1>+C<l1>, the match will never be successful because it can 4239 never 4240 get as far as its goal (and in fact is asserted against). Correspondingly for 4241 C<pe2> with respect to C<s2>. 4242 4243 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and 4244 C<l2> must be non-zero), and if both do, both have to be 4245 reached for a successful match. Also, if the fold of a character is multiple 4246 characters, all of them must be matched (see tr21 reference below for 4247 'folding'). 4248 4249 Upon a successful match, if C<pe1> is non-C<NULL>, 4250 it will be set to point to the beginning of the I<next> character of C<s1> 4251 beyond what was matched. Correspondingly for C<pe2> and C<s2>. 4252 4253 For case-insensitiveness, the "casefolding" of Unicode is used 4254 instead of upper/lowercasing both the characters, see 4255 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings). 4256 4257 =cut */ 4258 4259 /* A flags parameter has been added which may change, and hence isn't 4260 * externally documented. Currently it is: 4261 * 0 for as-documented above 4262 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an 4263 ASCII one, to not match 4264 * FOLDEQ_LOCALE is set iff the rules from the current underlying 4265 * locale are to be used. 4266 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this 4267 * routine. This allows that step to be skipped. 4268 * Currently, this requires s1 to be encoded as UTF-8 4269 * (u1 must be true), which is asserted for. 4270 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may 4271 * cross certain boundaries. Hence, the caller should 4272 * let this function do the folding instead of 4273 * pre-folding. This code contains an assertion to 4274 * that effect. However, if the caller knows what 4275 * it's doing, it can pass this flag to indicate that, 4276 * and the assertion is skipped. 4277 * FOLDEQ_S2_ALREADY_FOLDED Similarly. 4278 * FOLDEQ_S2_FOLDS_SANE 4279 */ 4280 I32 4281 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, const char *s2, char **pe2, UV l2, bool u2, U32 flags) 4282 { 4283 const U8 *p1 = (const U8*)s1; /* Point to current char */ 4284 const U8 *p2 = (const U8*)s2; 4285 const U8 *g1 = NULL; /* goal for s1 */ 4286 const U8 *g2 = NULL; 4287 const U8 *e1 = NULL; /* Don't scan s1 past this */ 4288 U8 *f1 = NULL; /* Point to current folded */ 4289 const U8 *e2 = NULL; 4290 U8 *f2 = NULL; 4291 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */ 4292 U8 foldbuf1[UTF8_MAXBYTES_CASE+1]; 4293 U8 foldbuf2[UTF8_MAXBYTES_CASE+1]; 4294 U8 flags_for_folder = FOLD_FLAGS_FULL; 4295 4296 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS; 4297 4298 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE)) 4299 && (((flags & FOLDEQ_S1_ALREADY_FOLDED) 4300 && !(flags & FOLDEQ_S1_FOLDS_SANE)) 4301 || ((flags & FOLDEQ_S2_ALREADY_FOLDED) 4302 && !(flags & FOLDEQ_S2_FOLDS_SANE))))); 4303 /* The algorithm is to trial the folds without regard to the flags on 4304 * the first line of the above assert(), and then see if the result 4305 * violates them. This means that the inputs can't be pre-folded to a 4306 * violating result, hence the assert. This could be changed, with the 4307 * addition of extra tests here for the already-folded case, which would 4308 * slow it down. That cost is more than any possible gain for when these 4309 * flags are specified, as the flags indicate /il or /iaa matching which 4310 * is less common than /iu, and I (khw) also believe that real-world /il 4311 * and /iaa matches are most likely to involve code points 0-255, and this 4312 * function only under rare conditions gets called for 0-255. */ 4313 4314 if (flags & FOLDEQ_LOCALE) { 4315 if (IN_UTF8_CTYPE_LOCALE) { 4316 flags &= ~FOLDEQ_LOCALE; 4317 } 4318 else { 4319 flags_for_folder |= FOLD_FLAGS_LOCALE; 4320 } 4321 } 4322 4323 if (pe1) { 4324 e1 = *(U8**)pe1; 4325 } 4326 4327 if (l1) { 4328 g1 = (const U8*)s1 + l1; 4329 } 4330 4331 if (pe2) { 4332 e2 = *(U8**)pe2; 4333 } 4334 4335 if (l2) { 4336 g2 = (const U8*)s2 + l2; 4337 } 4338 4339 /* Must have at least one goal */ 4340 assert(g1 || g2); 4341 4342 if (g1) { 4343 4344 /* Will never match if goal is out-of-bounds */ 4345 assert(! e1 || e1 >= g1); 4346 4347 /* Here, there isn't an end pointer, or it is beyond the goal. We 4348 * only go as far as the goal */ 4349 e1 = g1; 4350 } 4351 else { 4352 assert(e1); /* Must have an end for looking at s1 */ 4353 } 4354 4355 /* Same for goal for s2 */ 4356 if (g2) { 4357 assert(! e2 || e2 >= g2); 4358 e2 = g2; 4359 } 4360 else { 4361 assert(e2); 4362 } 4363 4364 /* If both operands are already folded, we could just do a memEQ on the 4365 * whole strings at once, but it would be better if the caller realized 4366 * this and didn't even call us */ 4367 4368 /* Look through both strings, a character at a time */ 4369 while (p1 < e1 && p2 < e2) { 4370 4371 /* If at the beginning of a new character in s1, get its fold to use 4372 * and the length of the fold. */ 4373 if (n1 == 0) { 4374 if (flags & FOLDEQ_S1_ALREADY_FOLDED) { 4375 f1 = (U8 *) p1; 4376 assert(u1); 4377 n1 = UTF8SKIP(f1); 4378 } 4379 else { 4380 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) { 4381 4382 /* We have to forbid mixing ASCII with non-ASCII if the 4383 * flags so indicate. And, we can short circuit having to 4384 * call the general functions for this common ASCII case, 4385 * all of whose non-locale folds are also ASCII, and hence 4386 * UTF-8 invariants, so the UTF8ness of the strings is not 4387 * relevant. */ 4388 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) { 4389 return 0; 4390 } 4391 n1 = 1; 4392 *foldbuf1 = toFOLD(*p1); 4393 } 4394 else if (u1) { 4395 _to_utf8_fold_flags(p1, foldbuf1, &n1, flags_for_folder); 4396 } 4397 else { /* Not UTF-8, get UTF-8 fold */ 4398 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder); 4399 } 4400 f1 = foldbuf1; 4401 } 4402 } 4403 4404 if (n2 == 0) { /* Same for s2 */ 4405 if (flags & FOLDEQ_S2_ALREADY_FOLDED) { 4406 f2 = (U8 *) p2; 4407 assert(u2); 4408 n2 = UTF8SKIP(f2); 4409 } 4410 else { 4411 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) { 4412 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) { 4413 return 0; 4414 } 4415 n2 = 1; 4416 *foldbuf2 = toFOLD(*p2); 4417 } 4418 else if (u2) { 4419 _to_utf8_fold_flags(p2, foldbuf2, &n2, flags_for_folder); 4420 } 4421 else { 4422 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder); 4423 } 4424 f2 = foldbuf2; 4425 } 4426 } 4427 4428 /* Here f1 and f2 point to the beginning of the strings to compare. 4429 * These strings are the folds of the next character from each input 4430 * string, stored in UTF-8. */ 4431 4432 /* While there is more to look for in both folds, see if they 4433 * continue to match */ 4434 while (n1 && n2) { 4435 U8 fold_length = UTF8SKIP(f1); 4436 if (fold_length != UTF8SKIP(f2) 4437 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE 4438 function call for single 4439 byte */ 4440 || memNE((char*)f1, (char*)f2, fold_length)) 4441 { 4442 return 0; /* mismatch */ 4443 } 4444 4445 /* Here, they matched, advance past them */ 4446 n1 -= fold_length; 4447 f1 += fold_length; 4448 n2 -= fold_length; 4449 f2 += fold_length; 4450 } 4451 4452 /* When reach the end of any fold, advance the input past it */ 4453 if (n1 == 0) { 4454 p1 += u1 ? UTF8SKIP(p1) : 1; 4455 } 4456 if (n2 == 0) { 4457 p2 += u2 ? UTF8SKIP(p2) : 1; 4458 } 4459 } /* End of loop through both strings */ 4460 4461 /* A match is defined by each scan that specified an explicit length 4462 * reaching its final goal, and the other not having matched a partial 4463 * character (which can happen when the fold of a character is more than one 4464 * character). */ 4465 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) { 4466 return 0; 4467 } 4468 4469 /* Successful match. Set output pointers */ 4470 if (pe1) { 4471 *pe1 = (char*)p1; 4472 } 4473 if (pe2) { 4474 *pe2 = (char*)p2; 4475 } 4476 return 1; 4477 } 4478 4479 /* XXX The next two functions should likely be moved to mathoms.c once all 4480 * occurrences of them are removed from the core; some cpan-upstream modules 4481 * still use them */ 4482 4483 U8 * 4484 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv) 4485 { 4486 PERL_ARGS_ASSERT_UVUNI_TO_UTF8; 4487 4488 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0); 4489 } 4490 4491 /* 4492 =for apidoc utf8n_to_uvuni 4493 4494 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>. 4495 4496 This function was useful for code that wanted to handle both EBCDIC and 4497 ASCII platforms with Unicode properties, but starting in Perl v5.20, the 4498 distinctions between the platforms have mostly been made invisible to most 4499 code, so this function is quite unlikely to be what you want. If you do need 4500 this precise functionality, use instead 4501 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> 4502 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>. 4503 4504 =cut 4505 */ 4506 4507 UV 4508 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags) 4509 { 4510 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI; 4511 4512 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags)); 4513 } 4514 4515 /* 4516 =for apidoc uvuni_to_utf8_flags 4517 4518 Instead you almost certainly want to use L</uvchr_to_utf8> or 4519 L</uvchr_to_utf8_flags>. 4520 4521 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>, 4522 which itself, while not deprecated, should be used only in isolated 4523 circumstances. These functions were useful for code that wanted to handle 4524 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl 4525 v5.20, the distinctions between the platforms have mostly been made invisible 4526 to most code, so this function is quite unlikely to be what you want. 4527 4528 =cut 4529 */ 4530 4531 U8 * 4532 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) 4533 { 4534 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS; 4535 4536 return uvoffuni_to_utf8_flags(d, uv, flags); 4537 } 4538 4539 /* 4540 * ex: set ts=8 sts=4 sw=4 et: 4541 */ 4542