| 1 | /* utf8.c |
| 2 | * |
| 3 | * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
| 4 | * by Larry Wall and others |
| 5 | * |
| 6 | * You may distribute under the terms of either the GNU General Public |
| 7 | * License or the Artistic License, as specified in the README file. |
| 8 | * |
| 9 | */ |
| 10 | |
| 11 | /* |
| 12 | * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever |
| 13 | * heard of that we don't want to see any closer; and that's the one place |
| 14 | * we're trying to get to! And that's just where we can't get, nohow.' |
| 15 | * |
| 16 | * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"] |
| 17 | * |
| 18 | * 'Well do I understand your speech,' he answered in the same language; |
| 19 | * 'yet few strangers do so. Why then do you not speak in the Common Tongue, |
| 20 | * as is the custom in the West, if you wish to be answered?' |
| 21 | * --Gandalf, addressing Théoden's door wardens |
| 22 | * |
| 23 | * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] |
| 24 | * |
| 25 | * ...the travellers perceived that the floor was paved with stones of many |
| 26 | * hues; branching runes and strange devices intertwined beneath their feet. |
| 27 | * |
| 28 | * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"] |
| 29 | */ |
| 30 | |
| 31 | #include "EXTERN.h" |
| 32 | #define PERL_IN_UTF8_C |
| 33 | #include "perl.h" |
| 34 | #include "invlist_inline.h" |
| 35 | |
| 36 | static const char malformed_text[] = "Malformed UTF-8 character"; |
| 37 | static const char unees[] = |
| 38 | "Malformed UTF-8 character (unexpected end of string)"; |
| 39 | static const char cp_above_legal_max[] = |
| 40 | "Use of code point 0x%"UVXf" is deprecated; the permissible max is 0x%"UVXf""; |
| 41 | |
| 42 | #define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX)) |
| 43 | |
| 44 | /* |
| 45 | =head1 Unicode Support |
| 46 | These are various utility functions for manipulating UTF8-encoded |
| 47 | strings. For the uninitiated, this is a method of representing arbitrary |
| 48 | Unicode characters as a variable number of bytes, in such a way that |
| 49 | characters in the ASCII range are unmodified, and a zero byte never appears |
| 50 | within non-zero characters. |
| 51 | |
| 52 | =cut |
| 53 | */ |
| 54 | |
| 55 | /* |
| 56 | =for apidoc uvoffuni_to_utf8_flags |
| 57 | |
| 58 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 59 | Instead, B<Almost all code should use L</uvchr_to_utf8> or |
| 60 | L</uvchr_to_utf8_flags>>. |
| 61 | |
| 62 | This function is like them, but the input is a strict Unicode |
| 63 | (as opposed to native) code point. Only in very rare circumstances should code |
| 64 | not be using the native code point. |
| 65 | |
| 66 | For details, see the description for L</uvchr_to_utf8_flags>. |
| 67 | |
| 68 | =cut |
| 69 | */ |
| 70 | |
| 71 | #define HANDLE_UNICODE_SURROGATE(uv, flags) \ |
| 72 | STMT_START { \ |
| 73 | if (flags & UNICODE_WARN_SURROGATE) { \ |
| 74 | Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \ |
| 75 | "UTF-16 surrogate U+%04"UVXf, uv); \ |
| 76 | } \ |
| 77 | if (flags & UNICODE_DISALLOW_SURROGATE) { \ |
| 78 | return NULL; \ |
| 79 | } \ |
| 80 | } STMT_END; |
| 81 | |
| 82 | #define HANDLE_UNICODE_NONCHAR(uv, flags) \ |
| 83 | STMT_START { \ |
| 84 | if (flags & UNICODE_WARN_NONCHAR) { \ |
| 85 | Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \ |
| 86 | "Unicode non-character U+%04"UVXf" is not " \ |
| 87 | "recommended for open interchange", uv); \ |
| 88 | } \ |
| 89 | if (flags & UNICODE_DISALLOW_NONCHAR) { \ |
| 90 | return NULL; \ |
| 91 | } \ |
| 92 | } STMT_END; |
| 93 | |
| 94 | /* Use shorter names internally in this file */ |
| 95 | #define SHIFT UTF_ACCUMULATION_SHIFT |
| 96 | #undef MARK |
| 97 | #define MARK UTF_CONTINUATION_MARK |
| 98 | #define MASK UTF_CONTINUATION_MASK |
| 99 | |
| 100 | U8 * |
| 101 | Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) |
| 102 | { |
| 103 | PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS; |
| 104 | |
| 105 | if (OFFUNI_IS_INVARIANT(uv)) { |
| 106 | *d++ = LATIN1_TO_NATIVE(uv); |
| 107 | return d; |
| 108 | } |
| 109 | |
| 110 | if (uv <= MAX_UTF8_TWO_BYTE) { |
| 111 | *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2)); |
| 112 | *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK); |
| 113 | return d; |
| 114 | } |
| 115 | |
| 116 | /* Not 2-byte; test for and handle 3-byte result. In the test immediately |
| 117 | * below, the 16 is for start bytes E0-EF (which are all the possible ones |
| 118 | * for 3 byte characters). The 2 is for 2 continuation bytes; these each |
| 119 | * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000 |
| 120 | * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC; |
| 121 | * 0x800-0xFFFF on ASCII */ |
| 122 | if (uv < (16 * (1U << (2 * SHIFT)))) { |
| 123 | *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3)); |
| 124 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); |
| 125 | *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); |
| 126 | |
| 127 | #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so |
| 128 | aren't tested here */ |
| 129 | /* The most likely code points in this range are below the surrogates. |
| 130 | * Do an extra test to quickly exclude those. */ |
| 131 | if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) { |
| 132 | if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) |
| 133 | || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) |
| 134 | { |
| 135 | HANDLE_UNICODE_NONCHAR(uv, flags); |
| 136 | } |
| 137 | else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { |
| 138 | HANDLE_UNICODE_SURROGATE(uv, flags); |
| 139 | } |
| 140 | } |
| 141 | #endif |
| 142 | return d; |
| 143 | } |
| 144 | |
| 145 | /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII |
| 146 | * platforms, and 0x4000 on EBCDIC. There are problematic cases that can |
| 147 | * happen starting with 4-byte characters on ASCII platforms. We unify the |
| 148 | * code for these with EBCDIC, even though some of them require 5-bytes on |
| 149 | * those, because khw believes the code saving is worth the very slight |
| 150 | * performance hit on these high EBCDIC code points. */ |
| 151 | |
| 152 | if (UNLIKELY(UNICODE_IS_SUPER(uv))) { |
| 153 | if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP) |
| 154 | && ckWARN_d(WARN_DEPRECATED)) |
| 155 | { |
| 156 | Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), |
| 157 | cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP); |
| 158 | } |
| 159 | if ( (flags & UNICODE_WARN_SUPER) |
| 160 | || ( UNICODE_IS_ABOVE_31_BIT(uv) |
| 161 | && (flags & UNICODE_WARN_ABOVE_31_BIT))) |
| 162 | { |
| 163 | Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), |
| 164 | |
| 165 | /* Choose the more dire applicable warning */ |
| 166 | (UNICODE_IS_ABOVE_31_BIT(uv)) |
| 167 | ? "Code point 0x%"UVXf" is not Unicode, and not portable" |
| 168 | : "Code point 0x%"UVXf" is not Unicode, may not be portable", |
| 169 | uv); |
| 170 | } |
| 171 | if (flags & UNICODE_DISALLOW_SUPER |
| 172 | || ( UNICODE_IS_ABOVE_31_BIT(uv) |
| 173 | && (flags & UNICODE_DISALLOW_ABOVE_31_BIT))) |
| 174 | { |
| 175 | return NULL; |
| 176 | } |
| 177 | } |
| 178 | else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) { |
| 179 | HANDLE_UNICODE_NONCHAR(uv, flags); |
| 180 | } |
| 181 | |
| 182 | /* Test for and handle 4-byte result. In the test immediately below, the |
| 183 | * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte |
| 184 | * characters). The 3 is for 3 continuation bytes; these each contribute |
| 185 | * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on |
| 186 | * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC; |
| 187 | * 0x1_0000-0x1F_FFFF on ASCII */ |
| 188 | if (uv < (8 * (1U << (3 * SHIFT)))) { |
| 189 | *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4)); |
| 190 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK); |
| 191 | *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK); |
| 192 | *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK); |
| 193 | |
| 194 | #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte |
| 195 | characters. The end-plane non-characters for EBCDIC were |
| 196 | handled just above */ |
| 197 | if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) { |
| 198 | HANDLE_UNICODE_NONCHAR(uv, flags); |
| 199 | } |
| 200 | else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { |
| 201 | HANDLE_UNICODE_SURROGATE(uv, flags); |
| 202 | } |
| 203 | #endif |
| 204 | |
| 205 | return d; |
| 206 | } |
| 207 | |
| 208 | /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII |
| 209 | * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop |
| 210 | * format. The unrolled version above turns out to not save all that much |
| 211 | * time, and at these high code points (well above the legal Unicode range |
| 212 | * on ASCII platforms, and well above anything in common use in EBCDIC), |
| 213 | * khw believes that less code outweighs slight performance gains. */ |
| 214 | |
| 215 | { |
| 216 | STRLEN len = OFFUNISKIP(uv); |
| 217 | U8 *p = d+len-1; |
| 218 | while (p > d) { |
| 219 | *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK); |
| 220 | uv >>= UTF_ACCUMULATION_SHIFT; |
| 221 | } |
| 222 | *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len)); |
| 223 | return d+len; |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | =for apidoc uvchr_to_utf8 |
| 229 | |
| 230 | Adds the UTF-8 representation of the native code point C<uv> to the end |
| 231 | of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to |
| 232 | C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to |
| 233 | the byte after the end of the new character. In other words, |
| 234 | |
| 235 | d = uvchr_to_utf8(d, uv); |
| 236 | |
| 237 | is the recommended wide native character-aware way of saying |
| 238 | |
| 239 | *(d++) = uv; |
| 240 | |
| 241 | This function accepts any UV as input, but very high code points (above |
| 242 | C<IV_MAX> on the platform) will raise a deprecation warning. This is |
| 243 | typically 0x7FFF_FFFF in a 32-bit word. |
| 244 | |
| 245 | It is possible to forbid or warn on non-Unicode code points, or those that may |
| 246 | be problematic by using L</uvchr_to_utf8_flags>. |
| 247 | |
| 248 | =cut |
| 249 | */ |
| 250 | |
| 251 | /* This is also a macro */ |
| 252 | PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv); |
| 253 | |
| 254 | U8 * |
| 255 | Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv) |
| 256 | { |
| 257 | return uvchr_to_utf8(d, uv); |
| 258 | } |
| 259 | |
| 260 | /* |
| 261 | =for apidoc uvchr_to_utf8_flags |
| 262 | |
| 263 | Adds the UTF-8 representation of the native code point C<uv> to the end |
| 264 | of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to |
| 265 | C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to |
| 266 | the byte after the end of the new character. In other words, |
| 267 | |
| 268 | d = uvchr_to_utf8_flags(d, uv, flags); |
| 269 | |
| 270 | or, in most cases, |
| 271 | |
| 272 | d = uvchr_to_utf8_flags(d, uv, 0); |
| 273 | |
| 274 | This is the Unicode-aware way of saying |
| 275 | |
| 276 | *(d++) = uv; |
| 277 | |
| 278 | If C<flags> is 0, this function accepts any UV as input, but very high code |
| 279 | points (above C<IV_MAX> for the platform) will raise a deprecation warning. |
| 280 | This is typically 0x7FFF_FFFF in a 32-bit word. |
| 281 | |
| 282 | Specifying C<flags> can further restrict what is allowed and not warned on, as |
| 283 | follows: |
| 284 | |
| 285 | If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set, |
| 286 | the function will raise a warning, provided UTF8 warnings are enabled. If |
| 287 | instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return |
| 288 | NULL. If both flags are set, the function will both warn and return NULL. |
| 289 | |
| 290 | Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags |
| 291 | affect how the function handles a Unicode non-character. |
| 292 | |
| 293 | And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags |
| 294 | affect the handling of code points that are above the Unicode maximum of |
| 295 | 0x10FFFF. Languages other than Perl may not be able to accept files that |
| 296 | contain these. |
| 297 | |
| 298 | The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of |
| 299 | the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all |
| 300 | three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the |
| 301 | allowed inputs to the strict UTF-8 traditionally defined by Unicode. |
| 302 | Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and |
| 303 | C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the |
| 304 | above-Unicode and surrogate flags, but not the non-character ones, as |
| 305 | defined in |
| 306 | L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. |
| 307 | See L<perlunicode/Noncharacter code points>. |
| 308 | |
| 309 | Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, |
| 310 | so using them is more problematic than other above-Unicode code points. Perl |
| 311 | invented an extension to UTF-8 to represent the ones above 2**36-1, so it is |
| 312 | likely that non-Perl languages will not be able to read files that contain |
| 313 | these that written by the perl interpreter; nor would Perl understand files |
| 314 | written by something that uses a different extension. For these reasons, there |
| 315 | is a separate set of flags that can warn and/or disallow these extremely high |
| 316 | code points, even if other above-Unicode ones are accepted. These are the |
| 317 | C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These |
| 318 | are entirely independent from the deprecation warning for code points above |
| 319 | C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any |
| 320 | code point that needs more than 31 bits to represent. When that happens, |
| 321 | effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on |
| 322 | 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all |
| 323 | above-Unicode code points, including these, as malformations; and |
| 324 | C<UNICODE_WARN_SUPER> warns on these.) |
| 325 | |
| 326 | On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing |
| 327 | extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower |
| 328 | than on ASCII. Prior to that, code points 2**31 and higher were simply |
| 329 | unrepresentable, and a different, incompatible method was used to represent |
| 330 | code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT> |
| 331 | and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII |
| 332 | platforms, warning and disallowing 2**31 and higher. |
| 333 | |
| 334 | =cut |
| 335 | */ |
| 336 | |
| 337 | /* This is also a macro */ |
| 338 | PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags); |
| 339 | |
| 340 | U8 * |
| 341 | Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) |
| 342 | { |
| 343 | return uvchr_to_utf8_flags(d, uv, flags); |
| 344 | } |
| 345 | |
| 346 | PERL_STATIC_INLINE bool |
| 347 | S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e) |
| 348 | { |
| 349 | /* Returns TRUE if the first code point represented by the Perl-extended- |
| 350 | * UTF-8-encoded string starting at 's', and looking no further than 'e - |
| 351 | * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31. |
| 352 | * |
| 353 | * The function handles the case where the input bytes do not include all |
| 354 | * the ones necessary to represent a full character. That is, they may be |
| 355 | * the intial bytes of the representation of a code point, but possibly |
| 356 | * the final ones necessary for the complete representation may be beyond |
| 357 | * 'e - 1'. |
| 358 | * |
| 359 | * The function assumes that the sequence is well-formed UTF-8 as far as it |
| 360 | * goes, and is for a UTF-8 variant code point. If the sequence is |
| 361 | * incomplete, the function returns FALSE if there is any well-formed |
| 362 | * UTF-8 byte sequence that can complete it in such a way that a code point |
| 363 | * < 2**31 is produced; otherwise it returns TRUE. |
| 364 | * |
| 365 | * Getting this exactly right is slightly tricky, and has to be done in |
| 366 | * several places in this file, so is centralized here. It is based on the |
| 367 | * following table: |
| 368 | * |
| 369 | * U+7FFFFFFF (2 ** 31 - 1) |
| 370 | * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF |
| 371 | * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73 |
| 372 | * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72 |
| 373 | * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75 |
| 374 | * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF |
| 375 | * U+80000000 (2 ** 31): |
| 376 | * ASCII: \xFE\x82\x80\x80\x80\x80\x80 |
| 377 | * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13 |
| 378 | * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 |
| 379 | * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 |
| 380 | * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41 |
| 381 | * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0 |
| 382 | */ |
| 383 | |
| 384 | #ifdef EBCDIC |
| 385 | |
| 386 | /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */ |
| 387 | const U8 * const prefix = (U8 *) "\x41\x41\x41\x41\x41\x41\x42"; |
| 388 | const STRLEN prefix_len = sizeof(prefix) - 1; |
| 389 | const STRLEN len = e - s; |
| 390 | const STRLEN cmp_len = MIN(prefix_len, len - 1); |
| 391 | |
| 392 | #else |
| 393 | |
| 394 | PERL_UNUSED_ARG(e); |
| 395 | |
| 396 | #endif |
| 397 | |
| 398 | PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS; |
| 399 | |
| 400 | assert(! UTF8_IS_INVARIANT(*s)); |
| 401 | |
| 402 | #ifndef EBCDIC |
| 403 | |
| 404 | /* Technically, a start byte of FE can be for a code point that fits into |
| 405 | * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong |
| 406 | * malformation. */ |
| 407 | return (*s >= 0xFE); |
| 408 | |
| 409 | #else |
| 410 | |
| 411 | /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or |
| 412 | * larger code point (0xFF is an invariant). For 0xFE, we need at least 2 |
| 413 | * bytes, and maybe up through 8 bytes, to be sure if the value is above 31 |
| 414 | * bits. */ |
| 415 | if (*s != 0xFE || len == 1) { |
| 416 | return FALSE; |
| 417 | } |
| 418 | |
| 419 | /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are |
| 420 | * \x41 and \x42. */ |
| 421 | return cBOOL(memGT(s + 1, prefix, cmp_len)); |
| 422 | |
| 423 | #endif |
| 424 | |
| 425 | } |
| 426 | |
| 427 | PERL_STATIC_INLINE bool |
| 428 | S_does_utf8_overflow(const U8 * const s, const U8 * e) |
| 429 | { |
| 430 | const U8 *x; |
| 431 | const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8; |
| 432 | |
| 433 | /* Returns a boolean as to if this UTF-8 string would overflow a UV on this |
| 434 | * platform, that is if it represents a code point larger than the highest |
| 435 | * representable code point. (For ASCII platforms, we could use memcmp() |
| 436 | * because we don't have to convert each byte to I8, but it's very rare |
| 437 | * input indeed that would approach overflow, so the loop below will likely |
| 438 | * only get executed once. |
| 439 | * |
| 440 | * 'e' must not be beyond a full character. If it is less than a full |
| 441 | * character, the function returns FALSE if there is any input beyond 'e' |
| 442 | * that could result in a non-overflowing code point */ |
| 443 | |
| 444 | PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW; |
| 445 | assert(s + UTF8SKIP(s) >= e); |
| 446 | |
| 447 | for (x = s; x < e; x++, y++) { |
| 448 | |
| 449 | /* If this byte is larger than the corresponding highest UTF-8 byte, it |
| 450 | * overflows */ |
| 451 | if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) { |
| 452 | return TRUE; |
| 453 | } |
| 454 | |
| 455 | /* If not the same as this byte, it must be smaller, doesn't overflow */ |
| 456 | if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) { |
| 457 | return FALSE; |
| 458 | } |
| 459 | } |
| 460 | |
| 461 | /* Got to the end and all bytes are the same. If the input is a whole |
| 462 | * character, it doesn't overflow. And if it is a partial character, |
| 463 | * there's not enough information to tell, so assume doesn't overflow */ |
| 464 | return FALSE; |
| 465 | } |
| 466 | |
| 467 | PERL_STATIC_INLINE bool |
| 468 | S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len) |
| 469 | { |
| 470 | /* Overlongs can occur whenever the number of continuation bytes |
| 471 | * changes. That means whenever the number of leading 1 bits in a start |
| 472 | * byte increases from the next lower start byte. That happens for start |
| 473 | * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following |
| 474 | * illegal start bytes have already been excluded, so don't need to be |
| 475 | * tested here; |
| 476 | * ASCII platforms: C0, C1 |
| 477 | * EBCDIC platforms C0, C1, C2, C3, C4, E0 |
| 478 | * |
| 479 | * At least a second byte is required to determine if other sequences will |
| 480 | * be an overlong. */ |
| 481 | |
| 482 | const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); |
| 483 | const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); |
| 484 | |
| 485 | PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK; |
| 486 | assert(len > 1 && UTF8_IS_START(*s)); |
| 487 | |
| 488 | /* Each platform has overlongs after the start bytes given above (expressed |
| 489 | * in I8 for EBCDIC). What constitutes an overlong varies by platform, but |
| 490 | * the logic is the same, except the E0 overlong has already been excluded |
| 491 | * on EBCDIC platforms. The values below were found by manually |
| 492 | * inspecting the UTF-8 patterns. See the tables in utf8.h and |
| 493 | * utfebcdic.h. */ |
| 494 | |
| 495 | # ifdef EBCDIC |
| 496 | # define F0_ABOVE_OVERLONG 0xB0 |
| 497 | # define F8_ABOVE_OVERLONG 0xA8 |
| 498 | # define FC_ABOVE_OVERLONG 0xA4 |
| 499 | # define FE_ABOVE_OVERLONG 0xA2 |
| 500 | # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41" |
| 501 | /* I8(0xfe) is FF */ |
| 502 | # else |
| 503 | |
| 504 | if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) { |
| 505 | return TRUE; |
| 506 | } |
| 507 | |
| 508 | # define F0_ABOVE_OVERLONG 0x90 |
| 509 | # define F8_ABOVE_OVERLONG 0x88 |
| 510 | # define FC_ABOVE_OVERLONG 0x84 |
| 511 | # define FE_ABOVE_OVERLONG 0x82 |
| 512 | # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80" |
| 513 | # endif |
| 514 | |
| 515 | |
| 516 | if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG)) |
| 517 | || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG)) |
| 518 | || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG)) |
| 519 | || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG))) |
| 520 | { |
| 521 | return TRUE; |
| 522 | } |
| 523 | |
| 524 | # if defined(UV_IS_QUAD) || defined(EBCDIC) |
| 525 | |
| 526 | /* Check for the FF overlong. This happens only if all these bytes match; |
| 527 | * what comes after them doesn't matter. See tables in utf8.h, |
| 528 | * utfebcdic.h. (Can't happen on ASCII 32-bit platforms, as overflows |
| 529 | * instead.) */ |
| 530 | |
| 531 | if ( len >= sizeof(FF_OVERLONG_PREFIX) - 1 |
| 532 | && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX, |
| 533 | sizeof(FF_OVERLONG_PREFIX) - 1))) |
| 534 | { |
| 535 | return TRUE; |
| 536 | } |
| 537 | |
| 538 | #endif |
| 539 | |
| 540 | return FALSE; |
| 541 | } |
| 542 | |
| 543 | #undef F0_ABOVE_OVERLONG |
| 544 | #undef F8_ABOVE_OVERLONG |
| 545 | #undef FC_ABOVE_OVERLONG |
| 546 | #undef FE_ABOVE_OVERLONG |
| 547 | #undef FF_OVERLONG_PREFIX |
| 548 | |
| 549 | STRLEN |
| 550 | Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags) |
| 551 | { |
| 552 | STRLEN len; |
| 553 | const U8 *x; |
| 554 | |
| 555 | /* A helper function that should not be called directly. |
| 556 | * |
| 557 | * This function returns non-zero if the string beginning at 's' and |
| 558 | * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a |
| 559 | * code point; otherwise it returns 0. The examination stops after the |
| 560 | * first code point in 's' is validated, not looking at the rest of the |
| 561 | * input. If 'e' is such that there are not enough bytes to represent a |
| 562 | * complete code point, this function will return non-zero anyway, if the |
| 563 | * bytes it does have are well-formed UTF-8 as far as they go, and aren't |
| 564 | * excluded by 'flags'. |
| 565 | * |
| 566 | * A non-zero return gives the number of bytes required to represent the |
| 567 | * code point. Be aware that if the input is for a partial character, the |
| 568 | * return will be larger than 'e - s'. |
| 569 | * |
| 570 | * This function assumes that the code point represented is UTF-8 variant. |
| 571 | * The caller should have excluded this possibility before calling this |
| 572 | * function. |
| 573 | * |
| 574 | * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags |
| 575 | * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return |
| 576 | * 0 if the code point represented is well-formed Perl-extended-UTF-8, but |
| 577 | * disallowed by the flags. If the input is only for a partial character, |
| 578 | * the function will return non-zero if there is any sequence of |
| 579 | * well-formed UTF-8 that, when appended to the input sequence, could |
| 580 | * result in an allowed code point; otherwise it returns 0. Non characters |
| 581 | * cannot be determined based on partial character input. But many of the |
| 582 | * other excluded types can be determined with just the first one or two |
| 583 | * bytes. |
| 584 | * |
| 585 | */ |
| 586 | |
| 587 | PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER; |
| 588 | |
| 589 | assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE |
| 590 | |UTF8_DISALLOW_ABOVE_31_BIT))); |
| 591 | assert(! UTF8_IS_INVARIANT(*s)); |
| 592 | |
| 593 | /* A variant char must begin with a start byte */ |
| 594 | if (UNLIKELY(! UTF8_IS_START(*s))) { |
| 595 | return 0; |
| 596 | } |
| 597 | |
| 598 | /* Examine a maximum of a single whole code point */ |
| 599 | if (e - s > UTF8SKIP(s)) { |
| 600 | e = s + UTF8SKIP(s); |
| 601 | } |
| 602 | |
| 603 | len = e - s; |
| 604 | |
| 605 | if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) { |
| 606 | const U8 s0 = NATIVE_UTF8_TO_I8(s[0]); |
| 607 | |
| 608 | /* The code below is derived from this table. Keep in mind that legal |
| 609 | * continuation bytes range between \x80..\xBF for UTF-8, and |
| 610 | * \xA0..\xBF for I8. Anything above those aren't continuation bytes. |
| 611 | * Hence, we don't have to test the upper edge because if any of those |
| 612 | * are encountered, the sequence is malformed, and will fail elsewhere |
| 613 | * in this function. |
| 614 | * UTF-8 UTF-EBCDIC I8 |
| 615 | * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate |
| 616 | * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate |
| 617 | * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode |
| 618 | * |
| 619 | */ |
| 620 | |
| 621 | #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */ |
| 622 | # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA |
| 623 | # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2) |
| 624 | |
| 625 | # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \ |
| 626 | /* B6 and B7 */ \ |
| 627 | && ((s1) & 0xFE ) == 0xB6) |
| 628 | #else |
| 629 | # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5 |
| 630 | # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90) |
| 631 | # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0) |
| 632 | #endif |
| 633 | |
| 634 | if ( (flags & UTF8_DISALLOW_SUPER) |
| 635 | && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) { |
| 636 | return 0; /* Above Unicode */ |
| 637 | } |
| 638 | |
| 639 | if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT) |
| 640 | && UNLIKELY(is_utf8_cp_above_31_bits(s, e))) |
| 641 | { |
| 642 | return 0; /* Above 31 bits */ |
| 643 | } |
| 644 | |
| 645 | if (len > 1) { |
| 646 | const U8 s1 = NATIVE_UTF8_TO_I8(s[1]); |
| 647 | |
| 648 | if ( (flags & UTF8_DISALLOW_SUPER) |
| 649 | && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1))) |
| 650 | { |
| 651 | return 0; /* Above Unicode */ |
| 652 | } |
| 653 | |
| 654 | if ( (flags & UTF8_DISALLOW_SURROGATE) |
| 655 | && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1))) |
| 656 | { |
| 657 | return 0; /* Surrogate */ |
| 658 | } |
| 659 | |
| 660 | if ( (flags & UTF8_DISALLOW_NONCHAR) |
| 661 | && UNLIKELY(UTF8_IS_NONCHAR(s, e))) |
| 662 | { |
| 663 | return 0; /* Noncharacter code point */ |
| 664 | } |
| 665 | } |
| 666 | } |
| 667 | |
| 668 | /* Make sure that all that follows are continuation bytes */ |
| 669 | for (x = s + 1; x < e; x++) { |
| 670 | if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) { |
| 671 | return 0; |
| 672 | } |
| 673 | } |
| 674 | |
| 675 | /* Here is syntactically valid. Next, make sure this isn't the start of an |
| 676 | * overlong. */ |
| 677 | if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) { |
| 678 | return 0; |
| 679 | } |
| 680 | |
| 681 | /* And finally, that the code point represented fits in a word on this |
| 682 | * platform */ |
| 683 | if (does_utf8_overflow(s, e)) { |
| 684 | return 0; |
| 685 | } |
| 686 | |
| 687 | return UTF8SKIP(s); |
| 688 | } |
| 689 | |
| 690 | STATIC char * |
| 691 | S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len) |
| 692 | { |
| 693 | /* Returns a mortalized C string that is a displayable copy of the 'len' |
| 694 | * bytes starting at 's', each in a \xXY format. */ |
| 695 | |
| 696 | const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a |
| 697 | trailing NUL */ |
| 698 | const U8 * const e = s + len; |
| 699 | char * output; |
| 700 | char * d; |
| 701 | |
| 702 | PERL_ARGS_ASSERT__BYTE_DUMP_STRING; |
| 703 | |
| 704 | Newx(output, output_len, char); |
| 705 | SAVEFREEPV(output); |
| 706 | |
| 707 | d = output; |
| 708 | for (; s < e; s++) { |
| 709 | const unsigned high_nibble = (*s & 0xF0) >> 4; |
| 710 | const unsigned low_nibble = (*s & 0x0F); |
| 711 | |
| 712 | *d++ = '\\'; |
| 713 | *d++ = 'x'; |
| 714 | |
| 715 | if (high_nibble < 10) { |
| 716 | *d++ = high_nibble + '0'; |
| 717 | } |
| 718 | else { |
| 719 | *d++ = high_nibble - 10 + 'a'; |
| 720 | } |
| 721 | |
| 722 | if (low_nibble < 10) { |
| 723 | *d++ = low_nibble + '0'; |
| 724 | } |
| 725 | else { |
| 726 | *d++ = low_nibble - 10 + 'a'; |
| 727 | } |
| 728 | } |
| 729 | |
| 730 | *d = '\0'; |
| 731 | return output; |
| 732 | } |
| 733 | |
| 734 | PERL_STATIC_INLINE char * |
| 735 | S_unexpected_non_continuation_text(pTHX_ const U8 * const s, |
| 736 | |
| 737 | /* How many bytes to print */ |
| 738 | STRLEN print_len, |
| 739 | |
| 740 | /* Which one is the non-continuation */ |
| 741 | const STRLEN non_cont_byte_pos, |
| 742 | |
| 743 | /* How many bytes should there be? */ |
| 744 | const STRLEN expect_len) |
| 745 | { |
| 746 | /* Return the malformation warning text for an unexpected continuation |
| 747 | * byte. */ |
| 748 | |
| 749 | const char * const where = (non_cont_byte_pos == 1) |
| 750 | ? "immediately" |
| 751 | : Perl_form(aTHX_ "%d bytes", |
| 752 | (int) non_cont_byte_pos); |
| 753 | unsigned int i; |
| 754 | |
| 755 | PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT; |
| 756 | |
| 757 | /* We don't need to pass this parameter, but since it has already been |
| 758 | * calculated, it's likely faster to pass it; verify under DEBUGGING */ |
| 759 | assert(expect_len == UTF8SKIP(s)); |
| 760 | |
| 761 | /* It is possible that utf8n_to_uvchr() was called incorrectly, with a |
| 762 | * length that is larger than is actually available in the buffer. If we |
| 763 | * print all the bytes based on that length, we will read past the buffer |
| 764 | * end. Often, the strings are NUL terminated, so to lower the chances of |
| 765 | * this happening, print the malformed bytes only up through any NUL. */ |
| 766 | for (i = 1; i < print_len; i++) { |
| 767 | if (*(s + i) == '\0') { |
| 768 | print_len = i + 1; /* +1 gets the NUL printed */ |
| 769 | break; |
| 770 | } |
| 771 | } |
| 772 | |
| 773 | return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x," |
| 774 | " %s after start byte 0x%02x; need %d bytes, got %d)", |
| 775 | malformed_text, |
| 776 | _byte_dump_string(s, print_len), |
| 777 | *(s + non_cont_byte_pos), |
| 778 | where, |
| 779 | *s, |
| 780 | (int) expect_len, |
| 781 | (int) non_cont_byte_pos); |
| 782 | } |
| 783 | |
| 784 | /* |
| 785 | |
| 786 | =for apidoc utf8n_to_uvchr |
| 787 | |
| 788 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 789 | Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. |
| 790 | |
| 791 | Bottom level UTF-8 decode routine. |
| 792 | Returns the native code point value of the first character in the string C<s>, |
| 793 | which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than |
| 794 | C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to |
| 795 | the length, in bytes, of that character. |
| 796 | |
| 797 | The value of C<flags> determines the behavior when C<s> does not point to a |
| 798 | well-formed UTF-8 character. If C<flags> is 0, encountering a malformation |
| 799 | causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) |
| 800 | is the next possible position in C<s> that could begin a non-malformed |
| 801 | character. Also, if UTF-8 warnings haven't been lexically disabled, a warning |
| 802 | is raised. Some UTF-8 input sequences may contain multiple malformations. |
| 803 | This function tries to find every possible one in each call, so multiple |
| 804 | warnings can be raised for each sequence. |
| 805 | |
| 806 | Various ALLOW flags can be set in C<flags> to allow (and not warn on) |
| 807 | individual types of malformations, such as the sequence being overlong (that |
| 808 | is, when there is a shorter sequence that can express the same code point; |
| 809 | overlong sequences are expressly forbidden in the UTF-8 standard due to |
| 810 | potential security issues). Another malformation example is the first byte of |
| 811 | a character not being a legal first byte. See F<utf8.h> for the list of such |
| 812 | flags. For allowed 0 length strings, this function returns 0; for allowed |
| 813 | overlong sequences, the computed code point is returned; for all other allowed |
| 814 | malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no |
| 815 | determinable reasonable value. |
| 816 | |
| 817 | The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other |
| 818 | flags) malformation is found. If this flag is set, the routine assumes that |
| 819 | the caller will raise a warning, and this function will silently just set |
| 820 | C<retlen> to C<-1> (cast to C<STRLEN>) and return zero. |
| 821 | |
| 822 | Note that this API requires disambiguation between successful decoding a C<NUL> |
| 823 | character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as |
| 824 | in both cases, 0 is returned, and, depending on the malformation, C<retlen> may |
| 825 | be set to 1. To disambiguate, upon a zero return, see if the first byte of |
| 826 | C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an |
| 827 | error. Or you can use C<L</utf8n_to_uvchr_error>>. |
| 828 | |
| 829 | Certain code points are considered problematic. These are Unicode surrogates, |
| 830 | Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF. |
| 831 | By default these are considered regular code points, but certain situations |
| 832 | warrant special handling for them, which can be specified using the C<flags> |
| 833 | parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all |
| 834 | three classes are treated as malformations and handled as such. The flags |
| 835 | C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and |
| 836 | C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to |
| 837 | disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE> |
| 838 | restricts the allowed inputs to the strict UTF-8 traditionally defined by |
| 839 | Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness |
| 840 | definition given by |
| 841 | L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>. |
| 842 | The difference between traditional strictness and C9 strictness is that the |
| 843 | latter does not forbid non-character code points. (They are still discouraged, |
| 844 | however.) For more discussion see L<perlunicode/Noncharacter code points>. |
| 845 | |
| 846 | The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, |
| 847 | C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>, |
| 848 | C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be |
| 849 | raised for their respective categories, but otherwise the code points are |
| 850 | considered valid (not malformations). To get a category to both be treated as |
| 851 | a malformation and raise a warning, specify both the WARN and DISALLOW flags. |
| 852 | (But note that warnings are not raised if lexically disabled nor if |
| 853 | C<UTF8_CHECK_ONLY> is also specified.) |
| 854 | |
| 855 | It is now deprecated to have very high code points (above C<IV_MAX> on the |
| 856 | platforms) and this function will raise a deprecation warning for these (unless |
| 857 | such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1) |
| 858 | in a 32-bit word. |
| 859 | |
| 860 | Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard, |
| 861 | so using them is more problematic than other above-Unicode code points. Perl |
| 862 | invented an extension to UTF-8 to represent the ones above 2**36-1, so it is |
| 863 | likely that non-Perl languages will not be able to read files that contain |
| 864 | these; nor would Perl understand files |
| 865 | written by something that uses a different extension. For these reasons, there |
| 866 | is a separate set of flags that can warn and/or disallow these extremely high |
| 867 | code points, even if other above-Unicode ones are accepted. These are the |
| 868 | C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These |
| 869 | are entirely independent from the deprecation warning for code points above |
| 870 | C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any |
| 871 | code point that needs more than 31 bits to represent. When that happens, |
| 872 | effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on |
| 873 | 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all |
| 874 | above-Unicode code points, including these, as malformations; and |
| 875 | C<UTF8_WARN_SUPER> warns on these.) |
| 876 | |
| 877 | On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing |
| 878 | extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower |
| 879 | than on ASCII. Prior to that, code points 2**31 and higher were simply |
| 880 | unrepresentable, and a different, incompatible method was used to represent |
| 881 | code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT> |
| 882 | and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII |
| 883 | platforms, warning and disallowing 2**31 and higher. |
| 884 | |
| 885 | All other code points corresponding to Unicode characters, including private |
| 886 | use and those yet to be assigned, are never considered malformed and never |
| 887 | warn. |
| 888 | |
| 889 | =cut |
| 890 | |
| 891 | Also implemented as a macro in utf8.h |
| 892 | */ |
| 893 | |
| 894 | UV |
| 895 | Perl_utf8n_to_uvchr(pTHX_ const U8 *s, |
| 896 | STRLEN curlen, |
| 897 | STRLEN *retlen, |
| 898 | const U32 flags) |
| 899 | { |
| 900 | PERL_ARGS_ASSERT_UTF8N_TO_UVCHR; |
| 901 | |
| 902 | return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL); |
| 903 | } |
| 904 | |
| 905 | /* |
| 906 | |
| 907 | =for apidoc utf8n_to_uvchr_error |
| 908 | |
| 909 | THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES. |
| 910 | Most code should use L</utf8_to_uvchr_buf>() rather than call this directly. |
| 911 | |
| 912 | This function is for code that needs to know what the precise malformation(s) |
| 913 | are when an error is found. |
| 914 | |
| 915 | It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after |
| 916 | all the others, C<errors>. If this parameter is 0, this function behaves |
| 917 | identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer |
| 918 | to a C<U32> variable, which this function sets to indicate any errors found. |
| 919 | Upon return, if C<*errors> is 0, there were no errors found. Otherwise, |
| 920 | C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some |
| 921 | of these bits will be set if a malformation is found, even if the input |
| 922 | C<flags> parameter indicates that the given malformation is allowed; the |
| 923 | exceptions are noted: |
| 924 | |
| 925 | =over 4 |
| 926 | |
| 927 | =item C<UTF8_GOT_ABOVE_31_BIT> |
| 928 | |
| 929 | The code point represented by the input UTF-8 sequence occupies more than 31 |
| 930 | bits. |
| 931 | This bit is set only if the input C<flags> parameter contains either the |
| 932 | C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags. |
| 933 | |
| 934 | =item C<UTF8_GOT_CONTINUATION> |
| 935 | |
| 936 | The input sequence was malformed in that the first byte was a a UTF-8 |
| 937 | continuation byte. |
| 938 | |
| 939 | =item C<UTF8_GOT_EMPTY> |
| 940 | |
| 941 | The input C<curlen> parameter was 0. |
| 942 | |
| 943 | =item C<UTF8_GOT_LONG> |
| 944 | |
| 945 | The input sequence was malformed in that there is some other sequence that |
| 946 | evaluates to the same code point, but that sequence is shorter than this one. |
| 947 | |
| 948 | =item C<UTF8_GOT_NONCHAR> |
| 949 | |
| 950 | The code point represented by the input UTF-8 sequence is for a Unicode |
| 951 | non-character code point. |
| 952 | This bit is set only if the input C<flags> parameter contains either the |
| 953 | C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags. |
| 954 | |
| 955 | =item C<UTF8_GOT_NON_CONTINUATION> |
| 956 | |
| 957 | The input sequence was malformed in that a non-continuation type byte was found |
| 958 | in a position where only a continuation type one should be. |
| 959 | |
| 960 | =item C<UTF8_GOT_OVERFLOW> |
| 961 | |
| 962 | The input sequence was malformed in that it is for a code point that is not |
| 963 | representable in the number of bits available in a UV on the current platform. |
| 964 | |
| 965 | =item C<UTF8_GOT_SHORT> |
| 966 | |
| 967 | The input sequence was malformed in that C<curlen> is smaller than required for |
| 968 | a complete sequence. In other words, the input is for a partial character |
| 969 | sequence. |
| 970 | |
| 971 | =item C<UTF8_GOT_SUPER> |
| 972 | |
| 973 | The input sequence was malformed in that it is for a non-Unicode code point; |
| 974 | that is, one above the legal Unicode maximum. |
| 975 | This bit is set only if the input C<flags> parameter contains either the |
| 976 | C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags. |
| 977 | |
| 978 | =item C<UTF8_GOT_SURROGATE> |
| 979 | |
| 980 | The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate |
| 981 | code point. |
| 982 | This bit is set only if the input C<flags> parameter contains either the |
| 983 | C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags. |
| 984 | |
| 985 | =back |
| 986 | |
| 987 | =cut |
| 988 | */ |
| 989 | |
| 990 | UV |
| 991 | Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s, |
| 992 | STRLEN curlen, |
| 993 | STRLEN *retlen, |
| 994 | const U32 flags, |
| 995 | U32 * errors) |
| 996 | { |
| 997 | const U8 * const s0 = s; |
| 998 | U8 * send = NULL; /* (initialized to silence compilers' wrong |
| 999 | warning) */ |
| 1000 | U32 possible_problems = 0; /* A bit is set here for each potential problem |
| 1001 | found as we go along */ |
| 1002 | UV uv = *s; |
| 1003 | STRLEN expectlen = 0; /* How long should this sequence be? |
| 1004 | (initialized to silence compilers' wrong |
| 1005 | warning) */ |
| 1006 | U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL; |
| 1007 | this gets set and discarded */ |
| 1008 | |
| 1009 | /* The below are used only if there is both an overlong malformation and a |
| 1010 | * too short one. Otherwise the first two are set to 's0' and 'send', and |
| 1011 | * the third not used at all */ |
| 1012 | U8 * adjusted_s0 = (U8 *) s0; |
| 1013 | U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong |
| 1014 | warning) */ |
| 1015 | UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */ |
| 1016 | |
| 1017 | PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR; |
| 1018 | |
| 1019 | if (errors) { |
| 1020 | *errors = 0; |
| 1021 | } |
| 1022 | else { |
| 1023 | errors = &discard_errors; |
| 1024 | } |
| 1025 | |
| 1026 | /* The order of malformation tests here is important. We should consume as |
| 1027 | * few bytes as possible in order to not skip any valid character. This is |
| 1028 | * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also |
| 1029 | * http://unicode.org/reports/tr36 for more discussion as to why. For |
| 1030 | * example, once we've done a UTF8SKIP, we can tell the expected number of |
| 1031 | * bytes, and could fail right off the bat if the input parameters indicate |
| 1032 | * that there are too few available. But it could be that just that first |
| 1033 | * byte is garbled, and the intended character occupies fewer bytes. If we |
| 1034 | * blindly assumed that the first byte is correct, and skipped based on |
| 1035 | * that number, we could skip over a valid input character. So instead, we |
| 1036 | * always examine the sequence byte-by-byte. |
| 1037 | * |
| 1038 | * We also should not consume too few bytes, otherwise someone could inject |
| 1039 | * things. For example, an input could be deliberately designed to |
| 1040 | * overflow, and if this code bailed out immediately upon discovering that, |
| 1041 | * returning to the caller C<*retlen> pointing to the very next byte (one |
| 1042 | * which is actually part of of the overflowing sequence), that could look |
| 1043 | * legitimate to the caller, which could discard the initial partial |
| 1044 | * sequence and process the rest, inappropriately. |
| 1045 | * |
| 1046 | * Some possible input sequences are malformed in more than one way. This |
| 1047 | * function goes to lengths to try to find all of them. This is necessary |
| 1048 | * for correctness, as the inputs may allow one malformation but not |
| 1049 | * another, and if we abandon searching for others after finding the |
| 1050 | * allowed one, we could allow in something that shouldn't have been. |
| 1051 | */ |
| 1052 | |
| 1053 | if (UNLIKELY(curlen == 0)) { |
| 1054 | possible_problems |= UTF8_GOT_EMPTY; |
| 1055 | curlen = 0; |
| 1056 | uv = 0; /* XXX It could be argued that this should be |
| 1057 | UNICODE_REPLACEMENT? */ |
| 1058 | goto ready_to_handle_errors; |
| 1059 | } |
| 1060 | |
| 1061 | expectlen = UTF8SKIP(s); |
| 1062 | |
| 1063 | /* A well-formed UTF-8 character, as the vast majority of calls to this |
| 1064 | * function will be for, has this expected length. For efficiency, set |
| 1065 | * things up here to return it. It will be overriden only in those rare |
| 1066 | * cases where a malformation is found */ |
| 1067 | if (retlen) { |
| 1068 | *retlen = expectlen; |
| 1069 | } |
| 1070 | |
| 1071 | /* An invariant is trivially well-formed */ |
| 1072 | if (UTF8_IS_INVARIANT(uv)) { |
| 1073 | return uv; |
| 1074 | } |
| 1075 | |
| 1076 | /* A continuation character can't start a valid sequence */ |
| 1077 | if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) { |
| 1078 | possible_problems |= UTF8_GOT_CONTINUATION; |
| 1079 | curlen = 1; |
| 1080 | uv = UNICODE_REPLACEMENT; |
| 1081 | goto ready_to_handle_errors; |
| 1082 | } |
| 1083 | |
| 1084 | /* Here is not a continuation byte, nor an invariant. The only thing left |
| 1085 | * is a start byte (possibly for an overlong) */ |
| 1086 | |
| 1087 | /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits |
| 1088 | * that indicate the number of bytes in the character's whole UTF-8 |
| 1089 | * sequence, leaving just the bits that are part of the value. */ |
| 1090 | uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen); |
| 1091 | |
| 1092 | /* Now, loop through the remaining bytes in the character's sequence, |
| 1093 | * accumulating each into the working value as we go. Be sure to not look |
| 1094 | * past the end of the input string */ |
| 1095 | send = adjusted_send = (U8*) s0 + ((expectlen <= curlen) |
| 1096 | ? expectlen |
| 1097 | : curlen); |
| 1098 | for (s = s0 + 1; s < send; s++) { |
| 1099 | if (LIKELY(UTF8_IS_CONTINUATION(*s))) { |
| 1100 | uv = UTF8_ACCUMULATE(uv, *s); |
| 1101 | continue; |
| 1102 | } |
| 1103 | |
| 1104 | /* Here, found a non-continuation before processing all expected bytes. |
| 1105 | * This byte indicates the beginning of a new character, so quit, even |
| 1106 | * if allowing this malformation. */ |
| 1107 | curlen = s - s0; /* Save how many bytes we actually got */ |
| 1108 | possible_problems |= UTF8_GOT_NON_CONTINUATION; |
| 1109 | goto finish_short; |
| 1110 | } /* End of loop through the character's bytes */ |
| 1111 | |
| 1112 | /* Save how many bytes were actually in the character */ |
| 1113 | curlen = s - s0; |
| 1114 | |
| 1115 | /* Did we get all the continuation bytes that were expected? Note that we |
| 1116 | * know this result even without executing the loop above. But we had to |
| 1117 | * do the loop to see if there are unexpected non-continuations. */ |
| 1118 | if (UNLIKELY(curlen < expectlen)) { |
| 1119 | possible_problems |= UTF8_GOT_SHORT; |
| 1120 | |
| 1121 | finish_short: |
| 1122 | uv_so_far = uv; |
| 1123 | uv = UNICODE_REPLACEMENT; |
| 1124 | } |
| 1125 | |
| 1126 | /* Note that there are two types of too-short malformation. One is when |
| 1127 | * there is actual wrong data before the normal termination of the |
| 1128 | * sequence. The other is that the sequence wasn't complete before the end |
| 1129 | * of the data we are allowed to look at, based on the input 'curlen'. |
| 1130 | * This means that we were passed data for a partial character, but it is |
| 1131 | * valid as far as we saw. The other is definitely invalid. This |
| 1132 | * distinction could be important to a caller, so the two types are kept |
| 1133 | * separate. */ |
| 1134 | |
| 1135 | /* Check for overflow */ |
| 1136 | if (UNLIKELY(does_utf8_overflow(s0, send))) { |
| 1137 | possible_problems |= UTF8_GOT_OVERFLOW; |
| 1138 | uv = UNICODE_REPLACEMENT; |
| 1139 | } |
| 1140 | |
| 1141 | /* Check for overlong. If no problems so far, 'uv' is the correct code |
| 1142 | * point value. Simply see if it is expressible in fewer bytes. Otherwise |
| 1143 | * we must look at the UTF-8 byte sequence itself to see if it is for an |
| 1144 | * overlong */ |
| 1145 | if ( ( LIKELY(! possible_problems) |
| 1146 | && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv))) |
| 1147 | || ( UNLIKELY( possible_problems) |
| 1148 | && ( UNLIKELY(! UTF8_IS_START(*s0)) |
| 1149 | || ( curlen > 1 |
| 1150 | && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0, |
| 1151 | send - s0)))))) |
| 1152 | { |
| 1153 | possible_problems |= UTF8_GOT_LONG; |
| 1154 | |
| 1155 | /* A convenience macro that matches either of the too-short conditions. |
| 1156 | * */ |
| 1157 | # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION) |
| 1158 | |
| 1159 | if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) { |
| 1160 | UV min_uv = uv_so_far; |
| 1161 | STRLEN i; |
| 1162 | |
| 1163 | /* Here, the input is both overlong and is missing some trailing |
| 1164 | * bytes. There is no single code point it could be for, but there |
| 1165 | * may be enough information present to determine if what we have |
| 1166 | * so far is for an unallowed code point, such as for a surrogate. |
| 1167 | * The code below has the intelligence to determine this, but just |
| 1168 | * for non-overlong UTF-8 sequences. What we do here is calculate |
| 1169 | * the smallest code point the input could represent if there were |
| 1170 | * no too short malformation. Then we compute and save the UTF-8 |
| 1171 | * for that, which is what the code below looks at instead of the |
| 1172 | * raw input. It turns out that the smallest such code point is |
| 1173 | * all we need. */ |
| 1174 | for (i = curlen; i < expectlen; i++) { |
| 1175 | min_uv = UTF8_ACCUMULATE(min_uv, |
| 1176 | I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK)); |
| 1177 | } |
| 1178 | |
| 1179 | Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8); |
| 1180 | SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get |
| 1181 | to free it ourselves if |
| 1182 | warnings are made fatal */ |
| 1183 | adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0); |
| 1184 | } |
| 1185 | } |
| 1186 | |
| 1187 | /* Now check that the input isn't for a problematic code point not allowed |
| 1188 | * by the input parameters. */ |
| 1189 | /* isn't problematic if < this */ |
| 1190 | if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST) |
| 1191 | || ( UNLIKELY(possible_problems) |
| 1192 | && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0))) |
| 1193 | && ((flags & ( UTF8_DISALLOW_NONCHAR |
| 1194 | |UTF8_DISALLOW_SURROGATE |
| 1195 | |UTF8_DISALLOW_SUPER |
| 1196 | |UTF8_DISALLOW_ABOVE_31_BIT |
| 1197 | |UTF8_WARN_NONCHAR |
| 1198 | |UTF8_WARN_SURROGATE |
| 1199 | |UTF8_WARN_SUPER |
| 1200 | |UTF8_WARN_ABOVE_31_BIT)) |
| 1201 | /* In case of a malformation, 'uv' is not valid, and has |
| 1202 | * been changed to something in the Unicode range. |
| 1203 | * Currently we don't output a deprecation message if there |
| 1204 | * is already a malformation, so we don't have to special |
| 1205 | * case the test immediately below */ |
| 1206 | || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP) |
| 1207 | && ckWARN_d(WARN_DEPRECATED)))) |
| 1208 | { |
| 1209 | /* If there were no malformations, or the only malformation is an |
| 1210 | * overlong, 'uv' is valid */ |
| 1211 | if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) { |
| 1212 | if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) { |
| 1213 | possible_problems |= UTF8_GOT_SURROGATE; |
| 1214 | } |
| 1215 | else if (UNLIKELY(uv > PERL_UNICODE_MAX)) { |
| 1216 | possible_problems |= UTF8_GOT_SUPER; |
| 1217 | } |
| 1218 | else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) { |
| 1219 | possible_problems |= UTF8_GOT_NONCHAR; |
| 1220 | } |
| 1221 | } |
| 1222 | else { /* Otherwise, need to look at the source UTF-8, possibly |
| 1223 | adjusted to be non-overlong */ |
| 1224 | |
| 1225 | if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0) |
| 1226 | >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) |
| 1227 | { |
| 1228 | possible_problems |= UTF8_GOT_SUPER; |
| 1229 | } |
| 1230 | else if (curlen > 1) { |
| 1231 | if (UNLIKELY(IS_UTF8_2_BYTE_SUPER( |
| 1232 | NATIVE_UTF8_TO_I8(*adjusted_s0), |
| 1233 | NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) |
| 1234 | { |
| 1235 | possible_problems |= UTF8_GOT_SUPER; |
| 1236 | } |
| 1237 | else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE( |
| 1238 | NATIVE_UTF8_TO_I8(*adjusted_s0), |
| 1239 | NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1))))) |
| 1240 | { |
| 1241 | possible_problems |= UTF8_GOT_SURROGATE; |
| 1242 | } |
| 1243 | } |
| 1244 | |
| 1245 | /* We need a complete well-formed UTF-8 character to discern |
| 1246 | * non-characters, so can't look for them here */ |
| 1247 | } |
| 1248 | } |
| 1249 | |
| 1250 | ready_to_handle_errors: |
| 1251 | |
| 1252 | /* At this point: |
| 1253 | * curlen contains the number of bytes in the sequence that |
| 1254 | * this call should advance the input by. |
| 1255 | * possible_problems' is 0 if there weren't any problems; otherwise a bit |
| 1256 | * is set in it for each potential problem found. |
| 1257 | * uv contains the code point the input sequence |
| 1258 | * represents; or if there is a problem that prevents |
| 1259 | * a well-defined value from being computed, it is |
| 1260 | * some subsitute value, typically the REPLACEMENT |
| 1261 | * CHARACTER. |
| 1262 | * s0 points to the first byte of the character |
| 1263 | * send points to just after where that (potentially |
| 1264 | * partial) character ends |
| 1265 | * adjusted_s0 normally is the same as s0, but in case of an |
| 1266 | * overlong for which the UTF-8 matters below, it is |
| 1267 | * the first byte of the shortest form representation |
| 1268 | * of the input. |
| 1269 | * adjusted_send normally is the same as 'send', but if adjusted_s0 |
| 1270 | * is set to something other than s0, this points one |
| 1271 | * beyond its end |
| 1272 | */ |
| 1273 | |
| 1274 | if (UNLIKELY(possible_problems)) { |
| 1275 | bool disallowed = FALSE; |
| 1276 | const U32 orig_problems = possible_problems; |
| 1277 | |
| 1278 | while (possible_problems) { /* Handle each possible problem */ |
| 1279 | UV pack_warn = 0; |
| 1280 | char * message = NULL; |
| 1281 | |
| 1282 | /* Each 'if' clause handles one problem. They are ordered so that |
| 1283 | * the first ones' messages will be displayed before the later |
| 1284 | * ones; this is kinda in decreasing severity order */ |
| 1285 | if (possible_problems & UTF8_GOT_OVERFLOW) { |
| 1286 | |
| 1287 | /* Overflow means also got a super and above 31 bits, but we |
| 1288 | * handle all three cases here */ |
| 1289 | possible_problems |
| 1290 | &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT); |
| 1291 | *errors |= UTF8_GOT_OVERFLOW; |
| 1292 | |
| 1293 | /* But the API says we flag all errors found */ |
| 1294 | if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) { |
| 1295 | *errors |= UTF8_GOT_SUPER; |
| 1296 | } |
| 1297 | if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) { |
| 1298 | *errors |= UTF8_GOT_ABOVE_31_BIT; |
| 1299 | } |
| 1300 | |
| 1301 | disallowed = TRUE; |
| 1302 | |
| 1303 | /* The warnings code explicitly says it doesn't handle the case |
| 1304 | * of packWARN2 and two categories which have parent-child |
| 1305 | * relationship. Even if it works now to raise the warning if |
| 1306 | * either is enabled, it wouldn't necessarily do so in the |
| 1307 | * future. We output (only) the most dire warning*/ |
| 1308 | if (! (flags & UTF8_CHECK_ONLY)) { |
| 1309 | if (ckWARN_d(WARN_UTF8)) { |
| 1310 | pack_warn = packWARN(WARN_UTF8); |
| 1311 | } |
| 1312 | else if (ckWARN_d(WARN_NON_UNICODE)) { |
| 1313 | pack_warn = packWARN(WARN_NON_UNICODE); |
| 1314 | } |
| 1315 | if (pack_warn) { |
| 1316 | message = Perl_form(aTHX_ "%s: %s (overflows)", |
| 1317 | malformed_text, |
| 1318 | _byte_dump_string(s0, send - s0)); |
| 1319 | } |
| 1320 | } |
| 1321 | } |
| 1322 | else if (possible_problems & UTF8_GOT_EMPTY) { |
| 1323 | possible_problems &= ~UTF8_GOT_EMPTY; |
| 1324 | *errors |= UTF8_GOT_EMPTY; |
| 1325 | |
| 1326 | if (! (flags & UTF8_ALLOW_EMPTY)) { |
| 1327 | disallowed = TRUE; |
| 1328 | if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) { |
| 1329 | pack_warn = packWARN(WARN_UTF8); |
| 1330 | message = Perl_form(aTHX_ "%s (empty string)", |
| 1331 | malformed_text); |
| 1332 | } |
| 1333 | } |
| 1334 | } |
| 1335 | else if (possible_problems & UTF8_GOT_CONTINUATION) { |
| 1336 | possible_problems &= ~UTF8_GOT_CONTINUATION; |
| 1337 | *errors |= UTF8_GOT_CONTINUATION; |
| 1338 | |
| 1339 | if (! (flags & UTF8_ALLOW_CONTINUATION)) { |
| 1340 | disallowed = TRUE; |
| 1341 | if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) { |
| 1342 | pack_warn = packWARN(WARN_UTF8); |
| 1343 | message = Perl_form(aTHX_ |
| 1344 | "%s: %s (unexpected continuation byte 0x%02x," |
| 1345 | " with no preceding start byte)", |
| 1346 | malformed_text, |
| 1347 | _byte_dump_string(s0, 1), *s0); |
| 1348 | } |
| 1349 | } |
| 1350 | } |
| 1351 | else if (possible_problems & UTF8_GOT_NON_CONTINUATION) { |
| 1352 | possible_problems &= ~UTF8_GOT_NON_CONTINUATION; |
| 1353 | *errors |= UTF8_GOT_NON_CONTINUATION; |
| 1354 | |
| 1355 | if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) { |
| 1356 | disallowed = TRUE; |
| 1357 | if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) { |
| 1358 | pack_warn = packWARN(WARN_UTF8); |
| 1359 | message = Perl_form(aTHX_ "%s", |
| 1360 | unexpected_non_continuation_text(s0, |
| 1361 | send - s0, |
| 1362 | s - s0, |
| 1363 | (int) expectlen)); |
| 1364 | } |
| 1365 | } |
| 1366 | } |
| 1367 | else if (possible_problems & UTF8_GOT_SHORT) { |
| 1368 | possible_problems &= ~UTF8_GOT_SHORT; |
| 1369 | *errors |= UTF8_GOT_SHORT; |
| 1370 | |
| 1371 | if (! (flags & UTF8_ALLOW_SHORT)) { |
| 1372 | disallowed = TRUE; |
| 1373 | if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) { |
| 1374 | pack_warn = packWARN(WARN_UTF8); |
| 1375 | message = Perl_form(aTHX_ |
| 1376 | "%s: %s (too short; got %d byte%s, need %d)", |
| 1377 | malformed_text, |
| 1378 | _byte_dump_string(s0, send - s0), |
| 1379 | (int)curlen, |
| 1380 | curlen == 1 ? "" : "s", |
| 1381 | (int)expectlen); |
| 1382 | } |
| 1383 | } |
| 1384 | |
| 1385 | } |
| 1386 | else if (possible_problems & UTF8_GOT_LONG) { |
| 1387 | possible_problems &= ~UTF8_GOT_LONG; |
| 1388 | *errors |= UTF8_GOT_LONG; |
| 1389 | |
| 1390 | if (! (flags & UTF8_ALLOW_LONG)) { |
| 1391 | disallowed = TRUE; |
| 1392 | |
| 1393 | if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) { |
| 1394 | pack_warn = packWARN(WARN_UTF8); |
| 1395 | |
| 1396 | /* These error types cause 'uv' to be something that |
| 1397 | * isn't what was intended, so can't use it in the |
| 1398 | * message. The other error types either can't |
| 1399 | * generate an overlong, or else the 'uv' is valid */ |
| 1400 | if (orig_problems & |
| 1401 | (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW)) |
| 1402 | { |
| 1403 | message = Perl_form(aTHX_ |
| 1404 | "%s: %s (any UTF-8 sequence that starts" |
| 1405 | " with \"%s\" is overlong which can and" |
| 1406 | " should be represented with a" |
| 1407 | " different, shorter sequence)", |
| 1408 | malformed_text, |
| 1409 | _byte_dump_string(s0, send - s0), |
| 1410 | _byte_dump_string(s0, curlen)); |
| 1411 | } |
| 1412 | else { |
| 1413 | U8 tmpbuf[UTF8_MAXBYTES+1]; |
| 1414 | const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf, |
| 1415 | uv, 0); |
| 1416 | message = Perl_form(aTHX_ |
| 1417 | "%s: %s (overlong; instead use %s to represent" |
| 1418 | " U+%0*"UVXf")", |
| 1419 | malformed_text, |
| 1420 | _byte_dump_string(s0, send - s0), |
| 1421 | _byte_dump_string(tmpbuf, e - tmpbuf), |
| 1422 | ((uv < 256) ? 2 : 4), /* Field width of 2 for |
| 1423 | small code points */ |
| 1424 | uv); |
| 1425 | } |
| 1426 | } |
| 1427 | } |
| 1428 | } |
| 1429 | else if (possible_problems & UTF8_GOT_SURROGATE) { |
| 1430 | possible_problems &= ~UTF8_GOT_SURROGATE; |
| 1431 | |
| 1432 | if (flags & UTF8_WARN_SURROGATE) { |
| 1433 | *errors |= UTF8_GOT_SURROGATE; |
| 1434 | |
| 1435 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 1436 | && ckWARN_d(WARN_SURROGATE)) |
| 1437 | { |
| 1438 | pack_warn = packWARN(WARN_SURROGATE); |
| 1439 | |
| 1440 | /* These are the only errors that can occur with a |
| 1441 | * surrogate when the 'uv' isn't valid */ |
| 1442 | if (orig_problems & UTF8_GOT_TOO_SHORT) { |
| 1443 | message = Perl_form(aTHX_ |
| 1444 | "UTF-16 surrogate (any UTF-8 sequence that" |
| 1445 | " starts with \"%s\" is for a surrogate)", |
| 1446 | _byte_dump_string(s0, curlen)); |
| 1447 | } |
| 1448 | else { |
| 1449 | message = Perl_form(aTHX_ |
| 1450 | "UTF-16 surrogate U+%04"UVXf"", uv); |
| 1451 | } |
| 1452 | } |
| 1453 | } |
| 1454 | |
| 1455 | if (flags & UTF8_DISALLOW_SURROGATE) { |
| 1456 | disallowed = TRUE; |
| 1457 | *errors |= UTF8_GOT_SURROGATE; |
| 1458 | } |
| 1459 | } |
| 1460 | else if (possible_problems & UTF8_GOT_SUPER) { |
| 1461 | possible_problems &= ~UTF8_GOT_SUPER; |
| 1462 | |
| 1463 | if (flags & UTF8_WARN_SUPER) { |
| 1464 | *errors |= UTF8_GOT_SUPER; |
| 1465 | |
| 1466 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 1467 | && ckWARN_d(WARN_NON_UNICODE)) |
| 1468 | { |
| 1469 | pack_warn = packWARN(WARN_NON_UNICODE); |
| 1470 | |
| 1471 | if (orig_problems & UTF8_GOT_TOO_SHORT) { |
| 1472 | message = Perl_form(aTHX_ |
| 1473 | "Any UTF-8 sequence that starts with" |
| 1474 | " \"%s\" is for a non-Unicode code point," |
| 1475 | " may not be portable", |
| 1476 | _byte_dump_string(s0, curlen)); |
| 1477 | } |
| 1478 | else { |
| 1479 | message = Perl_form(aTHX_ |
| 1480 | "Code point 0x%04"UVXf" is not" |
| 1481 | " Unicode, may not be portable", |
| 1482 | uv); |
| 1483 | } |
| 1484 | } |
| 1485 | } |
| 1486 | |
| 1487 | /* The maximum code point ever specified by a standard was |
| 1488 | * 2**31 - 1. Anything larger than that is a Perl extension |
| 1489 | * that very well may not be understood by other applications |
| 1490 | * (including earlier perl versions on EBCDIC platforms). We |
| 1491 | * test for these after the regular SUPER ones, and before |
| 1492 | * possibly bailing out, so that the slightly more dire warning |
| 1493 | * will override the regular one. */ |
| 1494 | if ( (flags & (UTF8_WARN_ABOVE_31_BIT |
| 1495 | |UTF8_WARN_SUPER |
| 1496 | |UTF8_DISALLOW_ABOVE_31_BIT)) |
| 1497 | && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT) |
| 1498 | && UNLIKELY(is_utf8_cp_above_31_bits( |
| 1499 | adjusted_s0, |
| 1500 | adjusted_send))) |
| 1501 | || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT)) |
| 1502 | && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv))))) |
| 1503 | { |
| 1504 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 1505 | && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER)) |
| 1506 | && ckWARN_d(WARN_UTF8)) |
| 1507 | { |
| 1508 | pack_warn = packWARN(WARN_UTF8); |
| 1509 | |
| 1510 | if (orig_problems & UTF8_GOT_TOO_SHORT) { |
| 1511 | message = Perl_form(aTHX_ |
| 1512 | "Any UTF-8 sequence that starts with" |
| 1513 | " \"%s\" is for a non-Unicode code" |
| 1514 | " point, and is not portable", |
| 1515 | _byte_dump_string(s0, curlen)); |
| 1516 | } |
| 1517 | else { |
| 1518 | message = Perl_form(aTHX_ |
| 1519 | "Code point 0x%"UVXf" is not Unicode," |
| 1520 | " and not portable", |
| 1521 | uv); |
| 1522 | } |
| 1523 | } |
| 1524 | |
| 1525 | if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) { |
| 1526 | *errors |= UTF8_GOT_ABOVE_31_BIT; |
| 1527 | |
| 1528 | if (flags & UTF8_DISALLOW_ABOVE_31_BIT) { |
| 1529 | disallowed = TRUE; |
| 1530 | } |
| 1531 | } |
| 1532 | } |
| 1533 | |
| 1534 | if (flags & UTF8_DISALLOW_SUPER) { |
| 1535 | *errors |= UTF8_GOT_SUPER; |
| 1536 | disallowed = TRUE; |
| 1537 | } |
| 1538 | |
| 1539 | /* The deprecated warning overrides any non-deprecated one. If |
| 1540 | * there are other problems, a deprecation message is not |
| 1541 | * really helpful, so don't bother to raise it in that case. |
| 1542 | * This also keeps the code from having to handle the case |
| 1543 | * where 'uv' is not valid. */ |
| 1544 | if ( ! (orig_problems |
| 1545 | & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW)) |
| 1546 | && UNLIKELY(uv > MAX_NON_DEPRECATED_CP) |
| 1547 | && ckWARN_d(WARN_DEPRECATED)) |
| 1548 | { |
| 1549 | message = Perl_form(aTHX_ cp_above_legal_max, |
| 1550 | uv, MAX_NON_DEPRECATED_CP); |
| 1551 | pack_warn = packWARN(WARN_DEPRECATED); |
| 1552 | } |
| 1553 | } |
| 1554 | else if (possible_problems & UTF8_GOT_NONCHAR) { |
| 1555 | possible_problems &= ~UTF8_GOT_NONCHAR; |
| 1556 | |
| 1557 | if (flags & UTF8_WARN_NONCHAR) { |
| 1558 | *errors |= UTF8_GOT_NONCHAR; |
| 1559 | |
| 1560 | if ( ! (flags & UTF8_CHECK_ONLY) |
| 1561 | && ckWARN_d(WARN_NONCHAR)) |
| 1562 | { |
| 1563 | /* The code above should have guaranteed that we don't |
| 1564 | * get here with errors other than overlong */ |
| 1565 | assert (! (orig_problems |
| 1566 | & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR))); |
| 1567 | |
| 1568 | pack_warn = packWARN(WARN_NONCHAR); |
| 1569 | message = Perl_form(aTHX_ "Unicode non-character" |
| 1570 | " U+%04"UVXf" is not recommended" |
| 1571 | " for open interchange", uv); |
| 1572 | } |
| 1573 | } |
| 1574 | |
| 1575 | if (flags & UTF8_DISALLOW_NONCHAR) { |
| 1576 | disallowed = TRUE; |
| 1577 | *errors |= UTF8_GOT_NONCHAR; |
| 1578 | } |
| 1579 | } /* End of looking through the possible flags */ |
| 1580 | |
| 1581 | /* Display the message (if any) for the problem being handled in |
| 1582 | * this iteration of the loop */ |
| 1583 | if (message) { |
| 1584 | if (PL_op) |
| 1585 | Perl_warner(aTHX_ pack_warn, "%s in %s", message, |
| 1586 | OP_DESC(PL_op)); |
| 1587 | else |
| 1588 | Perl_warner(aTHX_ pack_warn, "%s", message); |
| 1589 | } |
| 1590 | } /* End of 'while (possible_problems) {' */ |
| 1591 | |
| 1592 | /* Since there was a possible problem, the returned length may need to |
| 1593 | * be changed from the one stored at the beginning of this function. |
| 1594 | * Instead of trying to figure out if that's needed, just do it. */ |
| 1595 | if (retlen) { |
| 1596 | *retlen = curlen; |
| 1597 | } |
| 1598 | |
| 1599 | if (disallowed) { |
| 1600 | if (flags & UTF8_CHECK_ONLY && retlen) { |
| 1601 | *retlen = ((STRLEN) -1); |
| 1602 | } |
| 1603 | return 0; |
| 1604 | } |
| 1605 | } |
| 1606 | |
| 1607 | return UNI_TO_NATIVE(uv); |
| 1608 | } |
| 1609 | |
| 1610 | /* |
| 1611 | =for apidoc utf8_to_uvchr_buf |
| 1612 | |
| 1613 | Returns the native code point of the first character in the string C<s> which |
| 1614 | is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. |
| 1615 | C<*retlen> will be set to the length, in bytes, of that character. |
| 1616 | |
| 1617 | If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are |
| 1618 | enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't |
| 1619 | C<NULL>) to -1. If those warnings are off, the computed value, if well-defined |
| 1620 | (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and |
| 1621 | C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is |
| 1622 | the next possible position in C<s> that could begin a non-malformed character. |
| 1623 | See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is |
| 1624 | returned. |
| 1625 | |
| 1626 | Code points above the platform's C<IV_MAX> will raise a deprecation warning, |
| 1627 | unless those are turned off. |
| 1628 | |
| 1629 | =cut |
| 1630 | |
| 1631 | Also implemented as a macro in utf8.h |
| 1632 | |
| 1633 | */ |
| 1634 | |
| 1635 | |
| 1636 | UV |
| 1637 | Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) |
| 1638 | { |
| 1639 | assert(s < send); |
| 1640 | |
| 1641 | return utf8n_to_uvchr(s, send - s, retlen, |
| 1642 | ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY); |
| 1643 | } |
| 1644 | |
| 1645 | /* This is marked as deprecated |
| 1646 | * |
| 1647 | =for apidoc utf8_to_uvuni_buf |
| 1648 | |
| 1649 | Only in very rare circumstances should code need to be dealing in Unicode |
| 1650 | (as opposed to native) code points. In those few cases, use |
| 1651 | C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. |
| 1652 | |
| 1653 | Returns the Unicode (not-native) code point of the first character in the |
| 1654 | string C<s> which |
| 1655 | is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>. |
| 1656 | C<retlen> will be set to the length, in bytes, of that character. |
| 1657 | |
| 1658 | If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are |
| 1659 | enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't |
| 1660 | NULL) to -1. If those warnings are off, the computed value if well-defined (or |
| 1661 | the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen> |
| 1662 | is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the |
| 1663 | next possible position in C<s> that could begin a non-malformed character. |
| 1664 | See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned. |
| 1665 | |
| 1666 | Code points above the platform's C<IV_MAX> will raise a deprecation warning, |
| 1667 | unless those are turned off. |
| 1668 | |
| 1669 | =cut |
| 1670 | */ |
| 1671 | |
| 1672 | UV |
| 1673 | Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen) |
| 1674 | { |
| 1675 | PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF; |
| 1676 | |
| 1677 | assert(send > s); |
| 1678 | |
| 1679 | /* Call the low level routine, asking for checks */ |
| 1680 | return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen)); |
| 1681 | } |
| 1682 | |
| 1683 | /* |
| 1684 | =for apidoc utf8_length |
| 1685 | |
| 1686 | Return the length of the UTF-8 char encoded string C<s> in characters. |
| 1687 | Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end |
| 1688 | up past C<e>, croaks. |
| 1689 | |
| 1690 | =cut |
| 1691 | */ |
| 1692 | |
| 1693 | STRLEN |
| 1694 | Perl_utf8_length(pTHX_ const U8 *s, const U8 *e) |
| 1695 | { |
| 1696 | STRLEN len = 0; |
| 1697 | |
| 1698 | PERL_ARGS_ASSERT_UTF8_LENGTH; |
| 1699 | |
| 1700 | /* Note: cannot use UTF8_IS_...() too eagerly here since e.g. |
| 1701 | * the bitops (especially ~) can create illegal UTF-8. |
| 1702 | * In other words: in Perl UTF-8 is not just for Unicode. */ |
| 1703 | |
| 1704 | if (e < s) |
| 1705 | goto warn_and_return; |
| 1706 | while (s < e) { |
| 1707 | s += UTF8SKIP(s); |
| 1708 | len++; |
| 1709 | } |
| 1710 | |
| 1711 | if (e != s) { |
| 1712 | len--; |
| 1713 | warn_and_return: |
| 1714 | if (PL_op) |
| 1715 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 1716 | "%s in %s", unees, OP_DESC(PL_op)); |
| 1717 | else |
| 1718 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); |
| 1719 | } |
| 1720 | |
| 1721 | return len; |
| 1722 | } |
| 1723 | |
| 1724 | /* |
| 1725 | =for apidoc bytes_cmp_utf8 |
| 1726 | |
| 1727 | Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the |
| 1728 | sequence of characters (stored as UTF-8) |
| 1729 | in C<u>, C<ulen>. Returns 0 if they are |
| 1730 | equal, -1 or -2 if the first string is less than the second string, +1 or +2 |
| 1731 | if the first string is greater than the second string. |
| 1732 | |
| 1733 | -1 or +1 is returned if the shorter string was identical to the start of the |
| 1734 | longer string. -2 or +2 is returned if |
| 1735 | there was a difference between characters |
| 1736 | within the strings. |
| 1737 | |
| 1738 | =cut |
| 1739 | */ |
| 1740 | |
| 1741 | int |
| 1742 | Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen) |
| 1743 | { |
| 1744 | const U8 *const bend = b + blen; |
| 1745 | const U8 *const uend = u + ulen; |
| 1746 | |
| 1747 | PERL_ARGS_ASSERT_BYTES_CMP_UTF8; |
| 1748 | |
| 1749 | while (b < bend && u < uend) { |
| 1750 | U8 c = *u++; |
| 1751 | if (!UTF8_IS_INVARIANT(c)) { |
| 1752 | if (UTF8_IS_DOWNGRADEABLE_START(c)) { |
| 1753 | if (u < uend) { |
| 1754 | U8 c1 = *u++; |
| 1755 | if (UTF8_IS_CONTINUATION(c1)) { |
| 1756 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1); |
| 1757 | } else { |
| 1758 | /* diag_listed_as: Malformed UTF-8 character%s */ |
| 1759 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 1760 | "%s %s%s", |
| 1761 | unexpected_non_continuation_text(u - 1, 2, 1, 2), |
| 1762 | PL_op ? " in " : "", |
| 1763 | PL_op ? OP_DESC(PL_op) : ""); |
| 1764 | return -2; |
| 1765 | } |
| 1766 | } else { |
| 1767 | if (PL_op) |
| 1768 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 1769 | "%s in %s", unees, OP_DESC(PL_op)); |
| 1770 | else |
| 1771 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees); |
| 1772 | return -2; /* Really want to return undef :-) */ |
| 1773 | } |
| 1774 | } else { |
| 1775 | return -2; |
| 1776 | } |
| 1777 | } |
| 1778 | if (*b != c) { |
| 1779 | return *b < c ? -2 : +2; |
| 1780 | } |
| 1781 | ++b; |
| 1782 | } |
| 1783 | |
| 1784 | if (b == bend && u == uend) |
| 1785 | return 0; |
| 1786 | |
| 1787 | return b < bend ? +1 : -1; |
| 1788 | } |
| 1789 | |
| 1790 | /* |
| 1791 | =for apidoc utf8_to_bytes |
| 1792 | |
| 1793 | Converts a string C<s> of length C<len> from UTF-8 into native byte encoding. |
| 1794 | Unlike L</bytes_to_utf8>, this over-writes the original string, and |
| 1795 | updates C<len> to contain the new length. |
| 1796 | Returns zero on failure, setting C<len> to -1. |
| 1797 | |
| 1798 | If you need a copy of the string, see L</bytes_from_utf8>. |
| 1799 | |
| 1800 | =cut |
| 1801 | */ |
| 1802 | |
| 1803 | U8 * |
| 1804 | Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len) |
| 1805 | { |
| 1806 | U8 * const save = s; |
| 1807 | U8 * const send = s + *len; |
| 1808 | U8 *d; |
| 1809 | |
| 1810 | PERL_ARGS_ASSERT_UTF8_TO_BYTES; |
| 1811 | PERL_UNUSED_CONTEXT; |
| 1812 | |
| 1813 | /* ensure valid UTF-8 and chars < 256 before updating string */ |
| 1814 | while (s < send) { |
| 1815 | if (! UTF8_IS_INVARIANT(*s)) { |
| 1816 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { |
| 1817 | *len = ((STRLEN) -1); |
| 1818 | return 0; |
| 1819 | } |
| 1820 | s++; |
| 1821 | } |
| 1822 | s++; |
| 1823 | } |
| 1824 | |
| 1825 | d = s = save; |
| 1826 | while (s < send) { |
| 1827 | U8 c = *s++; |
| 1828 | if (! UTF8_IS_INVARIANT(c)) { |
| 1829 | /* Then it is two-byte encoded */ |
| 1830 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); |
| 1831 | s++; |
| 1832 | } |
| 1833 | *d++ = c; |
| 1834 | } |
| 1835 | *d = '\0'; |
| 1836 | *len = d - save; |
| 1837 | return save; |
| 1838 | } |
| 1839 | |
| 1840 | /* |
| 1841 | =for apidoc bytes_from_utf8 |
| 1842 | |
| 1843 | Converts a string C<s> of length C<len> from UTF-8 into native byte encoding. |
| 1844 | Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to |
| 1845 | the newly-created string, and updates C<len> to contain the new |
| 1846 | length. Returns the original string if no conversion occurs, C<len> |
| 1847 | is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to |
| 1848 | 0 if C<s> is converted or consisted entirely of characters that are invariant |
| 1849 | in UTF-8 (i.e., US-ASCII on non-EBCDIC machines). |
| 1850 | |
| 1851 | =cut |
| 1852 | */ |
| 1853 | |
| 1854 | U8 * |
| 1855 | Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8) |
| 1856 | { |
| 1857 | U8 *d; |
| 1858 | const U8 *start = s; |
| 1859 | const U8 *send; |
| 1860 | I32 count = 0; |
| 1861 | |
| 1862 | PERL_ARGS_ASSERT_BYTES_FROM_UTF8; |
| 1863 | PERL_UNUSED_CONTEXT; |
| 1864 | if (!*is_utf8) |
| 1865 | return (U8 *)start; |
| 1866 | |
| 1867 | /* ensure valid UTF-8 and chars < 256 before converting string */ |
| 1868 | for (send = s + *len; s < send;) { |
| 1869 | if (! UTF8_IS_INVARIANT(*s)) { |
| 1870 | if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) { |
| 1871 | return (U8 *)start; |
| 1872 | } |
| 1873 | count++; |
| 1874 | s++; |
| 1875 | } |
| 1876 | s++; |
| 1877 | } |
| 1878 | |
| 1879 | *is_utf8 = FALSE; |
| 1880 | |
| 1881 | Newx(d, (*len) - count + 1, U8); |
| 1882 | s = start; start = d; |
| 1883 | while (s < send) { |
| 1884 | U8 c = *s++; |
| 1885 | if (! UTF8_IS_INVARIANT(c)) { |
| 1886 | /* Then it is two-byte encoded */ |
| 1887 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s); |
| 1888 | s++; |
| 1889 | } |
| 1890 | *d++ = c; |
| 1891 | } |
| 1892 | *d = '\0'; |
| 1893 | *len = d - start; |
| 1894 | return (U8 *)start; |
| 1895 | } |
| 1896 | |
| 1897 | /* |
| 1898 | =for apidoc bytes_to_utf8 |
| 1899 | |
| 1900 | Converts a string C<s> of length C<len> bytes from the native encoding into |
| 1901 | UTF-8. |
| 1902 | Returns a pointer to the newly-created string, and sets C<len> to |
| 1903 | reflect the new length in bytes. |
| 1904 | |
| 1905 | A C<NUL> character will be written after the end of the string. |
| 1906 | |
| 1907 | If you want to convert to UTF-8 from encodings other than |
| 1908 | the native (Latin1 or EBCDIC), |
| 1909 | see L</sv_recode_to_utf8>(). |
| 1910 | |
| 1911 | =cut |
| 1912 | */ |
| 1913 | |
| 1914 | /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will |
| 1915 | likewise need duplication. */ |
| 1916 | |
| 1917 | U8* |
| 1918 | Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len) |
| 1919 | { |
| 1920 | const U8 * const send = s + (*len); |
| 1921 | U8 *d; |
| 1922 | U8 *dst; |
| 1923 | |
| 1924 | PERL_ARGS_ASSERT_BYTES_TO_UTF8; |
| 1925 | PERL_UNUSED_CONTEXT; |
| 1926 | |
| 1927 | Newx(d, (*len) * 2 + 1, U8); |
| 1928 | dst = d; |
| 1929 | |
| 1930 | while (s < send) { |
| 1931 | append_utf8_from_native_byte(*s, &d); |
| 1932 | s++; |
| 1933 | } |
| 1934 | *d = '\0'; |
| 1935 | *len = d-dst; |
| 1936 | return dst; |
| 1937 | } |
| 1938 | |
| 1939 | /* |
| 1940 | * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8. |
| 1941 | * |
| 1942 | * Destination must be pre-extended to 3/2 source. Do not use in-place. |
| 1943 | * We optimize for native, for obvious reasons. */ |
| 1944 | |
| 1945 | U8* |
| 1946 | Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) |
| 1947 | { |
| 1948 | U8* pend; |
| 1949 | U8* dstart = d; |
| 1950 | |
| 1951 | PERL_ARGS_ASSERT_UTF16_TO_UTF8; |
| 1952 | |
| 1953 | if (bytelen & 1) |
| 1954 | Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen); |
| 1955 | |
| 1956 | pend = p + bytelen; |
| 1957 | |
| 1958 | while (p < pend) { |
| 1959 | UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */ |
| 1960 | p += 2; |
| 1961 | if (OFFUNI_IS_INVARIANT(uv)) { |
| 1962 | *d++ = LATIN1_TO_NATIVE((U8) uv); |
| 1963 | continue; |
| 1964 | } |
| 1965 | if (uv <= MAX_UTF8_TWO_BYTE) { |
| 1966 | *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv)); |
| 1967 | *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv)); |
| 1968 | continue; |
| 1969 | } |
| 1970 | #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST |
| 1971 | #define LAST_HIGH_SURROGATE 0xDBFF |
| 1972 | #define FIRST_LOW_SURROGATE 0xDC00 |
| 1973 | #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST |
| 1974 | |
| 1975 | /* This assumes that most uses will be in the first Unicode plane, not |
| 1976 | * needing surrogates */ |
| 1977 | if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST |
| 1978 | && uv <= UNICODE_SURROGATE_LAST)) |
| 1979 | { |
| 1980 | if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) { |
| 1981 | Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); |
| 1982 | } |
| 1983 | else { |
| 1984 | UV low = (p[0] << 8) + p[1]; |
| 1985 | if ( UNLIKELY(low < FIRST_LOW_SURROGATE) |
| 1986 | || UNLIKELY(low > LAST_LOW_SURROGATE)) |
| 1987 | { |
| 1988 | Perl_croak(aTHX_ "Malformed UTF-16 surrogate"); |
| 1989 | } |
| 1990 | p += 2; |
| 1991 | uv = ((uv - FIRST_HIGH_SURROGATE) << 10) |
| 1992 | + (low - FIRST_LOW_SURROGATE) + 0x10000; |
| 1993 | } |
| 1994 | } |
| 1995 | #ifdef EBCDIC |
| 1996 | d = uvoffuni_to_utf8_flags(d, uv, 0); |
| 1997 | #else |
| 1998 | if (uv < 0x10000) { |
| 1999 | *d++ = (U8)(( uv >> 12) | 0xe0); |
| 2000 | *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); |
| 2001 | *d++ = (U8)(( uv & 0x3f) | 0x80); |
| 2002 | continue; |
| 2003 | } |
| 2004 | else { |
| 2005 | *d++ = (U8)(( uv >> 18) | 0xf0); |
| 2006 | *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80); |
| 2007 | *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80); |
| 2008 | *d++ = (U8)(( uv & 0x3f) | 0x80); |
| 2009 | continue; |
| 2010 | } |
| 2011 | #endif |
| 2012 | } |
| 2013 | *newlen = d - dstart; |
| 2014 | return d; |
| 2015 | } |
| 2016 | |
| 2017 | /* Note: this one is slightly destructive of the source. */ |
| 2018 | |
| 2019 | U8* |
| 2020 | Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen) |
| 2021 | { |
| 2022 | U8* s = (U8*)p; |
| 2023 | U8* const send = s + bytelen; |
| 2024 | |
| 2025 | PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED; |
| 2026 | |
| 2027 | if (bytelen & 1) |
| 2028 | Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf, |
| 2029 | (UV)bytelen); |
| 2030 | |
| 2031 | while (s < send) { |
| 2032 | const U8 tmp = s[0]; |
| 2033 | s[0] = s[1]; |
| 2034 | s[1] = tmp; |
| 2035 | s += 2; |
| 2036 | } |
| 2037 | return utf16_to_utf8(p, d, bytelen, newlen); |
| 2038 | } |
| 2039 | |
| 2040 | bool |
| 2041 | Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c) |
| 2042 | { |
| 2043 | U8 tmpbuf[UTF8_MAXBYTES+1]; |
| 2044 | uvchr_to_utf8(tmpbuf, c); |
| 2045 | return _is_utf8_FOO(classnum, tmpbuf); |
| 2046 | } |
| 2047 | |
| 2048 | /* Internal function so we can deprecate the external one, and call |
| 2049 | this one from other deprecated functions in this file */ |
| 2050 | |
| 2051 | bool |
| 2052 | Perl__is_utf8_idstart(pTHX_ const U8 *p) |
| 2053 | { |
| 2054 | PERL_ARGS_ASSERT__IS_UTF8_IDSTART; |
| 2055 | |
| 2056 | if (*p == '_') |
| 2057 | return TRUE; |
| 2058 | return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL); |
| 2059 | } |
| 2060 | |
| 2061 | bool |
| 2062 | Perl__is_uni_perl_idcont(pTHX_ UV c) |
| 2063 | { |
| 2064 | U8 tmpbuf[UTF8_MAXBYTES+1]; |
| 2065 | uvchr_to_utf8(tmpbuf, c); |
| 2066 | return _is_utf8_perl_idcont(tmpbuf); |
| 2067 | } |
| 2068 | |
| 2069 | bool |
| 2070 | Perl__is_uni_perl_idstart(pTHX_ UV c) |
| 2071 | { |
| 2072 | U8 tmpbuf[UTF8_MAXBYTES+1]; |
| 2073 | uvchr_to_utf8(tmpbuf, c); |
| 2074 | return _is_utf8_perl_idstart(tmpbuf); |
| 2075 | } |
| 2076 | |
| 2077 | UV |
| 2078 | Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s) |
| 2079 | { |
| 2080 | /* We have the latin1-range values compiled into the core, so just use |
| 2081 | * those, converting the result to UTF-8. The only difference between upper |
| 2082 | * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is |
| 2083 | * either "SS" or "Ss". Which one to use is passed into the routine in |
| 2084 | * 'S_or_s' to avoid a test */ |
| 2085 | |
| 2086 | UV converted = toUPPER_LATIN1_MOD(c); |
| 2087 | |
| 2088 | PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1; |
| 2089 | |
| 2090 | assert(S_or_s == 'S' || S_or_s == 's'); |
| 2091 | |
| 2092 | if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for |
| 2093 | characters in this range */ |
| 2094 | *p = (U8) converted; |
| 2095 | *lenp = 1; |
| 2096 | return converted; |
| 2097 | } |
| 2098 | |
| 2099 | /* toUPPER_LATIN1_MOD gives the correct results except for three outliers, |
| 2100 | * which it maps to one of them, so as to only have to have one check for |
| 2101 | * it in the main case */ |
| 2102 | if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { |
| 2103 | switch (c) { |
| 2104 | case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: |
| 2105 | converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; |
| 2106 | break; |
| 2107 | case MICRO_SIGN: |
| 2108 | converted = GREEK_CAPITAL_LETTER_MU; |
| 2109 | break; |
| 2110 | #if UNICODE_MAJOR_VERSION > 2 \ |
| 2111 | || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \ |
| 2112 | && UNICODE_DOT_DOT_VERSION >= 8) |
| 2113 | case LATIN_SMALL_LETTER_SHARP_S: |
| 2114 | *(p)++ = 'S'; |
| 2115 | *p = S_or_s; |
| 2116 | *lenp = 2; |
| 2117 | return 'S'; |
| 2118 | #endif |
| 2119 | default: |
| 2120 | Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS); |
| 2121 | NOT_REACHED; /* NOTREACHED */ |
| 2122 | } |
| 2123 | } |
| 2124 | |
| 2125 | *(p)++ = UTF8_TWO_BYTE_HI(converted); |
| 2126 | *p = UTF8_TWO_BYTE_LO(converted); |
| 2127 | *lenp = 2; |
| 2128 | |
| 2129 | return converted; |
| 2130 | } |
| 2131 | |
| 2132 | /* Call the function to convert a UTF-8 encoded character to the specified case. |
| 2133 | * Note that there may be more than one character in the result. |
| 2134 | * INP is a pointer to the first byte of the input character |
| 2135 | * OUTP will be set to the first byte of the string of changed characters. It |
| 2136 | * needs to have space for UTF8_MAXBYTES_CASE+1 bytes |
| 2137 | * LENP will be set to the length in bytes of the string of changed characters |
| 2138 | * |
| 2139 | * The functions return the ordinal of the first character in the string of OUTP */ |
| 2140 | #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "") |
| 2141 | #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "") |
| 2142 | #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "") |
| 2143 | |
| 2144 | /* This additionally has the input parameter 'specials', which if non-zero will |
| 2145 | * cause this to use the specials hash for folding (meaning get full case |
| 2146 | * folding); otherwise, when zero, this implies a simple case fold */ |
| 2147 | #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL) |
| 2148 | |
| 2149 | UV |
| 2150 | Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp) |
| 2151 | { |
| 2152 | /* Convert the Unicode character whose ordinal is <c> to its uppercase |
| 2153 | * version and store that in UTF-8 in <p> and its length in bytes in <lenp>. |
| 2154 | * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since |
| 2155 | * the changed version may be longer than the original character. |
| 2156 | * |
| 2157 | * The ordinal of the first character of the changed version is returned |
| 2158 | * (but note, as explained above, that there may be more.) */ |
| 2159 | |
| 2160 | PERL_ARGS_ASSERT_TO_UNI_UPPER; |
| 2161 | |
| 2162 | if (c < 256) { |
| 2163 | return _to_upper_title_latin1((U8) c, p, lenp, 'S'); |
| 2164 | } |
| 2165 | |
| 2166 | uvchr_to_utf8(p, c); |
| 2167 | return CALL_UPPER_CASE(c, p, p, lenp); |
| 2168 | } |
| 2169 | |
| 2170 | UV |
| 2171 | Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp) |
| 2172 | { |
| 2173 | PERL_ARGS_ASSERT_TO_UNI_TITLE; |
| 2174 | |
| 2175 | if (c < 256) { |
| 2176 | return _to_upper_title_latin1((U8) c, p, lenp, 's'); |
| 2177 | } |
| 2178 | |
| 2179 | uvchr_to_utf8(p, c); |
| 2180 | return CALL_TITLE_CASE(c, p, p, lenp); |
| 2181 | } |
| 2182 | |
| 2183 | STATIC U8 |
| 2184 | S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp) |
| 2185 | { |
| 2186 | /* We have the latin1-range values compiled into the core, so just use |
| 2187 | * those, converting the result to UTF-8. Since the result is always just |
| 2188 | * one character, we allow <p> to be NULL */ |
| 2189 | |
| 2190 | U8 converted = toLOWER_LATIN1(c); |
| 2191 | |
| 2192 | if (p != NULL) { |
| 2193 | if (NATIVE_BYTE_IS_INVARIANT(converted)) { |
| 2194 | *p = converted; |
| 2195 | *lenp = 1; |
| 2196 | } |
| 2197 | else { |
| 2198 | /* Result is known to always be < 256, so can use the EIGHT_BIT |
| 2199 | * macros */ |
| 2200 | *p = UTF8_EIGHT_BIT_HI(converted); |
| 2201 | *(p+1) = UTF8_EIGHT_BIT_LO(converted); |
| 2202 | *lenp = 2; |
| 2203 | } |
| 2204 | } |
| 2205 | return converted; |
| 2206 | } |
| 2207 | |
| 2208 | UV |
| 2209 | Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp) |
| 2210 | { |
| 2211 | PERL_ARGS_ASSERT_TO_UNI_LOWER; |
| 2212 | |
| 2213 | if (c < 256) { |
| 2214 | return to_lower_latin1((U8) c, p, lenp); |
| 2215 | } |
| 2216 | |
| 2217 | uvchr_to_utf8(p, c); |
| 2218 | return CALL_LOWER_CASE(c, p, p, lenp); |
| 2219 | } |
| 2220 | |
| 2221 | UV |
| 2222 | Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags) |
| 2223 | { |
| 2224 | /* Corresponds to to_lower_latin1(); <flags> bits meanings: |
| 2225 | * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited |
| 2226 | * FOLD_FLAGS_FULL iff full folding is to be used; |
| 2227 | * |
| 2228 | * Not to be used for locale folds |
| 2229 | */ |
| 2230 | |
| 2231 | UV converted; |
| 2232 | |
| 2233 | PERL_ARGS_ASSERT__TO_FOLD_LATIN1; |
| 2234 | PERL_UNUSED_CONTEXT; |
| 2235 | |
| 2236 | assert (! (flags & FOLD_FLAGS_LOCALE)); |
| 2237 | |
| 2238 | if (UNLIKELY(c == MICRO_SIGN)) { |
| 2239 | converted = GREEK_SMALL_LETTER_MU; |
| 2240 | } |
| 2241 | #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ |
| 2242 | || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ |
| 2243 | || UNICODE_DOT_DOT_VERSION > 0) |
| 2244 | else if ( (flags & FOLD_FLAGS_FULL) |
| 2245 | && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S)) |
| 2246 | { |
| 2247 | /* If can't cross 127/128 boundary, can't return "ss"; instead return |
| 2248 | * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}") |
| 2249 | * under those circumstances. */ |
| 2250 | if (flags & FOLD_FLAGS_NOMIX_ASCII) { |
| 2251 | *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; |
| 2252 | Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, |
| 2253 | p, *lenp, U8); |
| 2254 | return LATIN_SMALL_LETTER_LONG_S; |
| 2255 | } |
| 2256 | else { |
| 2257 | *(p)++ = 's'; |
| 2258 | *p = 's'; |
| 2259 | *lenp = 2; |
| 2260 | return 's'; |
| 2261 | } |
| 2262 | } |
| 2263 | #endif |
| 2264 | else { /* In this range the fold of all other characters is their lower |
| 2265 | case */ |
| 2266 | converted = toLOWER_LATIN1(c); |
| 2267 | } |
| 2268 | |
| 2269 | if (UVCHR_IS_INVARIANT(converted)) { |
| 2270 | *p = (U8) converted; |
| 2271 | *lenp = 1; |
| 2272 | } |
| 2273 | else { |
| 2274 | *(p)++ = UTF8_TWO_BYTE_HI(converted); |
| 2275 | *p = UTF8_TWO_BYTE_LO(converted); |
| 2276 | *lenp = 2; |
| 2277 | } |
| 2278 | |
| 2279 | return converted; |
| 2280 | } |
| 2281 | |
| 2282 | UV |
| 2283 | Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags) |
| 2284 | { |
| 2285 | |
| 2286 | /* Not currently externally documented, and subject to change |
| 2287 | * <flags> bits meanings: |
| 2288 | * FOLD_FLAGS_FULL iff full folding is to be used; |
| 2289 | * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying |
| 2290 | * locale are to be used. |
| 2291 | * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited |
| 2292 | */ |
| 2293 | |
| 2294 | PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS; |
| 2295 | |
| 2296 | if (flags & FOLD_FLAGS_LOCALE) { |
| 2297 | /* Treat a UTF-8 locale as not being in locale at all */ |
| 2298 | if (IN_UTF8_CTYPE_LOCALE) { |
| 2299 | flags &= ~FOLD_FLAGS_LOCALE; |
| 2300 | } |
| 2301 | else { |
| 2302 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2303 | goto needs_full_generality; |
| 2304 | } |
| 2305 | } |
| 2306 | |
| 2307 | if (c < 256) { |
| 2308 | return _to_fold_latin1((U8) c, p, lenp, |
| 2309 | flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); |
| 2310 | } |
| 2311 | |
| 2312 | /* Here, above 255. If no special needs, just use the macro */ |
| 2313 | if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) { |
| 2314 | uvchr_to_utf8(p, c); |
| 2315 | return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL); |
| 2316 | } |
| 2317 | else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with |
| 2318 | the special flags. */ |
| 2319 | U8 utf8_c[UTF8_MAXBYTES + 1]; |
| 2320 | |
| 2321 | needs_full_generality: |
| 2322 | uvchr_to_utf8(utf8_c, c); |
| 2323 | return _to_utf8_fold_flags(utf8_c, p, lenp, flags); |
| 2324 | } |
| 2325 | } |
| 2326 | |
| 2327 | PERL_STATIC_INLINE bool |
| 2328 | S_is_utf8_common(pTHX_ const U8 *const p, SV **swash, |
| 2329 | const char *const swashname, SV* const invlist) |
| 2330 | { |
| 2331 | /* returns a boolean giving whether or not the UTF8-encoded character that |
| 2332 | * starts at <p> is in the swash indicated by <swashname>. <swash> |
| 2333 | * contains a pointer to where the swash indicated by <swashname> |
| 2334 | * is to be stored; which this routine will do, so that future calls will |
| 2335 | * look at <*swash> and only generate a swash if it is not null. <invlist> |
| 2336 | * is NULL or an inversion list that defines the swash. If not null, it |
| 2337 | * saves time during initialization of the swash. |
| 2338 | * |
| 2339 | * Note that it is assumed that the buffer length of <p> is enough to |
| 2340 | * contain all the bytes that comprise the character. Thus, <*p> should |
| 2341 | * have been checked before this call for mal-formedness enough to assure |
| 2342 | * that. */ |
| 2343 | |
| 2344 | PERL_ARGS_ASSERT_IS_UTF8_COMMON; |
| 2345 | |
| 2346 | /* The API should have included a length for the UTF-8 character in <p>, |
| 2347 | * but it doesn't. We therefore assume that p has been validated at least |
| 2348 | * as far as there being enough bytes available in it to accommodate the |
| 2349 | * character without reading beyond the end, and pass that number on to the |
| 2350 | * validating routine */ |
| 2351 | if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) { |
| 2352 | if (ckWARN_d(WARN_UTF8)) { |
| 2353 | Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8), |
| 2354 | "Passing malformed UTF-8 to \"%s\" is deprecated", swashname); |
| 2355 | if (ckWARN(WARN_UTF8)) { /* This will output details as to the |
| 2356 | what the malformation is */ |
| 2357 | utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL); |
| 2358 | } |
| 2359 | } |
| 2360 | return FALSE; |
| 2361 | } |
| 2362 | if (!*swash) { |
| 2363 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 2364 | *swash = _core_swash_init("utf8", |
| 2365 | |
| 2366 | /* Only use the name if there is no inversion |
| 2367 | * list; otherwise will go out to disk */ |
| 2368 | (invlist) ? "" : swashname, |
| 2369 | |
| 2370 | &PL_sv_undef, 1, 0, invlist, &flags); |
| 2371 | } |
| 2372 | |
| 2373 | return swash_fetch(*swash, p, TRUE) != 0; |
| 2374 | } |
| 2375 | |
| 2376 | bool |
| 2377 | Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p) |
| 2378 | { |
| 2379 | PERL_ARGS_ASSERT__IS_UTF8_FOO; |
| 2380 | |
| 2381 | assert(classnum < _FIRST_NON_SWASH_CC); |
| 2382 | |
| 2383 | return is_utf8_common(p, |
| 2384 | &PL_utf8_swash_ptrs[classnum], |
| 2385 | swash_property_names[classnum], |
| 2386 | PL_XPosix_ptrs[classnum]); |
| 2387 | } |
| 2388 | |
| 2389 | bool |
| 2390 | Perl__is_utf8_perl_idstart(pTHX_ const U8 *p) |
| 2391 | { |
| 2392 | SV* invlist = NULL; |
| 2393 | |
| 2394 | PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART; |
| 2395 | |
| 2396 | if (! PL_utf8_perl_idstart) { |
| 2397 | invlist = _new_invlist_C_array(_Perl_IDStart_invlist); |
| 2398 | } |
| 2399 | return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist); |
| 2400 | } |
| 2401 | |
| 2402 | bool |
| 2403 | Perl__is_utf8_xidstart(pTHX_ const U8 *p) |
| 2404 | { |
| 2405 | PERL_ARGS_ASSERT__IS_UTF8_XIDSTART; |
| 2406 | |
| 2407 | if (*p == '_') |
| 2408 | return TRUE; |
| 2409 | return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL); |
| 2410 | } |
| 2411 | |
| 2412 | bool |
| 2413 | Perl__is_utf8_perl_idcont(pTHX_ const U8 *p) |
| 2414 | { |
| 2415 | SV* invlist = NULL; |
| 2416 | |
| 2417 | PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT; |
| 2418 | |
| 2419 | if (! PL_utf8_perl_idcont) { |
| 2420 | invlist = _new_invlist_C_array(_Perl_IDCont_invlist); |
| 2421 | } |
| 2422 | return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist); |
| 2423 | } |
| 2424 | |
| 2425 | bool |
| 2426 | Perl__is_utf8_idcont(pTHX_ const U8 *p) |
| 2427 | { |
| 2428 | PERL_ARGS_ASSERT__IS_UTF8_IDCONT; |
| 2429 | |
| 2430 | return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL); |
| 2431 | } |
| 2432 | |
| 2433 | bool |
| 2434 | Perl__is_utf8_xidcont(pTHX_ const U8 *p) |
| 2435 | { |
| 2436 | PERL_ARGS_ASSERT__IS_UTF8_XIDCONT; |
| 2437 | |
| 2438 | return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL); |
| 2439 | } |
| 2440 | |
| 2441 | bool |
| 2442 | Perl__is_utf8_mark(pTHX_ const U8 *p) |
| 2443 | { |
| 2444 | PERL_ARGS_ASSERT__IS_UTF8_MARK; |
| 2445 | |
| 2446 | return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL); |
| 2447 | } |
| 2448 | |
| 2449 | /* |
| 2450 | =for apidoc to_utf8_case |
| 2451 | |
| 2452 | Instead use the appropriate one of L</toUPPER_utf8>, |
| 2453 | L</toTITLE_utf8>, |
| 2454 | L</toLOWER_utf8>, |
| 2455 | or L</toFOLD_utf8>. |
| 2456 | |
| 2457 | C<p> contains the pointer to the UTF-8 string encoding |
| 2458 | the character that is being converted. This routine assumes that the character |
| 2459 | at C<p> is well-formed. |
| 2460 | |
| 2461 | C<ustrp> is a pointer to the character buffer to put the |
| 2462 | conversion result to. C<lenp> is a pointer to the length |
| 2463 | of the result. |
| 2464 | |
| 2465 | C<swashp> is a pointer to the swash to use. |
| 2466 | |
| 2467 | Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>, |
| 2468 | and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually, |
| 2469 | but not always, a multicharacter mapping), is tried first. |
| 2470 | |
| 2471 | C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use |
| 2472 | any special mappings; C<""> means to use the special mappings. Values other |
| 2473 | than these two are treated as the name of the hash containing the special |
| 2474 | mappings, like C<"utf8::ToSpecLower">. |
| 2475 | |
| 2476 | C<normal> is a string like C<"ToLower"> which means the swash |
| 2477 | C<%utf8::ToLower>. |
| 2478 | |
| 2479 | Code points above the platform's C<IV_MAX> will raise a deprecation warning, |
| 2480 | unless those are turned off. |
| 2481 | |
| 2482 | =cut */ |
| 2483 | |
| 2484 | UV |
| 2485 | Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, |
| 2486 | SV **swashp, const char *normal, const char *special) |
| 2487 | { |
| 2488 | PERL_ARGS_ASSERT_TO_UTF8_CASE; |
| 2489 | |
| 2490 | return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special); |
| 2491 | } |
| 2492 | |
| 2493 | /* change namve uv1 to 'from' */ |
| 2494 | STATIC UV |
| 2495 | S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp, |
| 2496 | SV **swashp, const char *normal, const char *special) |
| 2497 | { |
| 2498 | STRLEN len = 0; |
| 2499 | |
| 2500 | PERL_ARGS_ASSERT__TO_UTF8_CASE; |
| 2501 | |
| 2502 | /* For code points that don't change case, we already know that the output |
| 2503 | * of this function is the unchanged input, so we can skip doing look-ups |
| 2504 | * for them. Unfortunately the case-changing code points are scattered |
| 2505 | * around. But there are some long consecutive ranges where there are no |
| 2506 | * case changing code points. By adding tests, we can eliminate the lookup |
| 2507 | * for all the ones in such ranges. This is currently done here only for |
| 2508 | * just a few cases where the scripts are in common use in modern commerce |
| 2509 | * (and scripts adjacent to those which can be included without additional |
| 2510 | * tests). */ |
| 2511 | |
| 2512 | if (uv1 >= 0x0590) { |
| 2513 | /* This keeps from needing further processing the code points most |
| 2514 | * likely to be used in the following non-cased scripts: Hebrew, |
| 2515 | * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari, |
| 2516 | * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada, |
| 2517 | * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */ |
| 2518 | if (uv1 < 0x10A0) { |
| 2519 | goto cases_to_self; |
| 2520 | } |
| 2521 | |
| 2522 | /* The following largish code point ranges also don't have case |
| 2523 | * changes, but khw didn't think they warranted extra tests to speed |
| 2524 | * them up (which would slightly slow down everything else above them): |
| 2525 | * 1100..139F Hangul Jamo, Ethiopic |
| 2526 | * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic, |
| 2527 | * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian, |
| 2528 | * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham, |
| 2529 | * Combining Diacritical Marks Extended, Balinese, |
| 2530 | * Sundanese, Batak, Lepcha, Ol Chiki |
| 2531 | * 2000..206F General Punctuation |
| 2532 | */ |
| 2533 | |
| 2534 | if (uv1 >= 0x2D30) { |
| 2535 | |
| 2536 | /* This keeps the from needing further processing the code points |
| 2537 | * most likely to be used in the following non-cased major scripts: |
| 2538 | * CJK, Katakana, Hiragana, plus some less-likely scripts. |
| 2539 | * |
| 2540 | * (0x2D30 above might have to be changed to 2F00 in the unlikely |
| 2541 | * event that Unicode eventually allocates the unused block as of |
| 2542 | * v8.0 2FE0..2FEF to code points that are cased. khw has verified |
| 2543 | * that the test suite will start having failures to alert you |
| 2544 | * should that happen) */ |
| 2545 | if (uv1 < 0xA640) { |
| 2546 | goto cases_to_self; |
| 2547 | } |
| 2548 | |
| 2549 | if (uv1 >= 0xAC00) { |
| 2550 | if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) { |
| 2551 | if (ckWARN_d(WARN_SURROGATE)) { |
| 2552 | const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; |
| 2553 | Perl_warner(aTHX_ packWARN(WARN_SURROGATE), |
| 2554 | "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1); |
| 2555 | } |
| 2556 | goto cases_to_self; |
| 2557 | } |
| 2558 | |
| 2559 | /* AC00..FAFF Catches Hangul syllables and private use, plus |
| 2560 | * some others */ |
| 2561 | if (uv1 < 0xFB00) { |
| 2562 | goto cases_to_self; |
| 2563 | |
| 2564 | } |
| 2565 | |
| 2566 | if (UNLIKELY(UNICODE_IS_SUPER(uv1))) { |
| 2567 | if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP) |
| 2568 | && ckWARN_d(WARN_DEPRECATED)) |
| 2569 | { |
| 2570 | Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), |
| 2571 | cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP); |
| 2572 | } |
| 2573 | if (ckWARN_d(WARN_NON_UNICODE)) { |
| 2574 | const char* desc = (PL_op) ? OP_DESC(PL_op) : normal; |
| 2575 | Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), |
| 2576 | "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1); |
| 2577 | } |
| 2578 | goto cases_to_self; |
| 2579 | } |
| 2580 | #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C |
| 2581 | if (UNLIKELY(uv1 |
| 2582 | > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C)) |
| 2583 | { |
| 2584 | |
| 2585 | /* As of this writing, this means we avoid swash creation |
| 2586 | * for anything beyond low Plane 1 */ |
| 2587 | goto cases_to_self; |
| 2588 | } |
| 2589 | #endif |
| 2590 | } |
| 2591 | } |
| 2592 | |
| 2593 | /* Note that non-characters are perfectly legal, so no warning should |
| 2594 | * be given. There are so few of them, that it isn't worth the extra |
| 2595 | * tests to avoid swash creation */ |
| 2596 | } |
| 2597 | |
| 2598 | if (!*swashp) /* load on-demand */ |
| 2599 | *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL); |
| 2600 | |
| 2601 | if (special) { |
| 2602 | /* It might be "special" (sometimes, but not always, |
| 2603 | * a multicharacter mapping) */ |
| 2604 | HV *hv = NULL; |
| 2605 | SV **svp; |
| 2606 | |
| 2607 | /* If passed in the specials name, use that; otherwise use any |
| 2608 | * given in the swash */ |
| 2609 | if (*special != '\0') { |
| 2610 | hv = get_hv(special, 0); |
| 2611 | } |
| 2612 | else { |
| 2613 | svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0); |
| 2614 | if (svp) { |
| 2615 | hv = MUTABLE_HV(SvRV(*svp)); |
| 2616 | } |
| 2617 | } |
| 2618 | |
| 2619 | if (hv |
| 2620 | && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE)) |
| 2621 | && (*svp)) |
| 2622 | { |
| 2623 | const char *s; |
| 2624 | |
| 2625 | s = SvPV_const(*svp, len); |
| 2626 | if (len == 1) |
| 2627 | /* EIGHTBIT */ |
| 2628 | len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp; |
| 2629 | else { |
| 2630 | Copy(s, ustrp, len, U8); |
| 2631 | } |
| 2632 | } |
| 2633 | } |
| 2634 | |
| 2635 | if (!len && *swashp) { |
| 2636 | const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */); |
| 2637 | |
| 2638 | if (uv2) { |
| 2639 | /* It was "normal" (a single character mapping). */ |
| 2640 | len = uvchr_to_utf8(ustrp, uv2) - ustrp; |
| 2641 | } |
| 2642 | } |
| 2643 | |
| 2644 | if (len) { |
| 2645 | if (lenp) { |
| 2646 | *lenp = len; |
| 2647 | } |
| 2648 | return valid_utf8_to_uvchr(ustrp, 0); |
| 2649 | } |
| 2650 | |
| 2651 | /* Here, there was no mapping defined, which means that the code point maps |
| 2652 | * to itself. Return the inputs */ |
| 2653 | cases_to_self: |
| 2654 | len = UTF8SKIP(p); |
| 2655 | if (p != ustrp) { /* Don't copy onto itself */ |
| 2656 | Copy(p, ustrp, len, U8); |
| 2657 | } |
| 2658 | |
| 2659 | if (lenp) |
| 2660 | *lenp = len; |
| 2661 | |
| 2662 | return uv1; |
| 2663 | |
| 2664 | } |
| 2665 | |
| 2666 | STATIC UV |
| 2667 | S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp) |
| 2668 | { |
| 2669 | /* This is called when changing the case of a UTF-8-encoded character above |
| 2670 | * the Latin1 range, and the operation is in a non-UTF-8 locale. If the |
| 2671 | * result contains a character that crosses the 255/256 boundary, disallow |
| 2672 | * the change, and return the original code point. See L<perlfunc/lc> for |
| 2673 | * why; |
| 2674 | * |
| 2675 | * p points to the original string whose case was changed; assumed |
| 2676 | * by this routine to be well-formed |
| 2677 | * result the code point of the first character in the changed-case string |
| 2678 | * ustrp points to the changed-case string (<result> represents its first char) |
| 2679 | * lenp points to the length of <ustrp> */ |
| 2680 | |
| 2681 | UV original; /* To store the first code point of <p> */ |
| 2682 | |
| 2683 | PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING; |
| 2684 | |
| 2685 | assert(UTF8_IS_ABOVE_LATIN1(*p)); |
| 2686 | |
| 2687 | /* We know immediately if the first character in the string crosses the |
| 2688 | * boundary, so can skip */ |
| 2689 | if (result > 255) { |
| 2690 | |
| 2691 | /* Look at every character in the result; if any cross the |
| 2692 | * boundary, the whole thing is disallowed */ |
| 2693 | U8* s = ustrp + UTF8SKIP(ustrp); |
| 2694 | U8* e = ustrp + *lenp; |
| 2695 | while (s < e) { |
| 2696 | if (! UTF8_IS_ABOVE_LATIN1(*s)) { |
| 2697 | goto bad_crossing; |
| 2698 | } |
| 2699 | s += UTF8SKIP(s); |
| 2700 | } |
| 2701 | |
| 2702 | /* Here, no characters crossed, result is ok as-is, but we warn. */ |
| 2703 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p)); |
| 2704 | return result; |
| 2705 | } |
| 2706 | |
| 2707 | bad_crossing: |
| 2708 | |
| 2709 | /* Failed, have to return the original */ |
| 2710 | original = valid_utf8_to_uvchr(p, lenp); |
| 2711 | |
| 2712 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 2713 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 2714 | "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; " |
| 2715 | "resolved to \"\\x{%"UVXf"}\".", |
| 2716 | OP_DESC(PL_op), |
| 2717 | original, |
| 2718 | original); |
| 2719 | Copy(p, ustrp, *lenp, char); |
| 2720 | return original; |
| 2721 | } |
| 2722 | |
| 2723 | /* |
| 2724 | =for apidoc to_utf8_upper |
| 2725 | |
| 2726 | Instead use L</toUPPER_utf8>. |
| 2727 | |
| 2728 | =cut */ |
| 2729 | |
| 2730 | /* Not currently externally documented, and subject to change: |
| 2731 | * <flags> is set iff iff the rules from the current underlying locale are to |
| 2732 | * be used. */ |
| 2733 | |
| 2734 | UV |
| 2735 | Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags) |
| 2736 | { |
| 2737 | UV result; |
| 2738 | |
| 2739 | PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS; |
| 2740 | |
| 2741 | if (flags) { |
| 2742 | /* Treat a UTF-8 locale as not being in locale at all */ |
| 2743 | if (IN_UTF8_CTYPE_LOCALE) { |
| 2744 | flags = FALSE; |
| 2745 | } |
| 2746 | else { |
| 2747 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2748 | } |
| 2749 | } |
| 2750 | |
| 2751 | if (UTF8_IS_INVARIANT(*p)) { |
| 2752 | if (flags) { |
| 2753 | result = toUPPER_LC(*p); |
| 2754 | } |
| 2755 | else { |
| 2756 | return _to_upper_title_latin1(*p, ustrp, lenp, 'S'); |
| 2757 | } |
| 2758 | } |
| 2759 | else if UTF8_IS_DOWNGRADEABLE_START(*p) { |
| 2760 | if (flags) { |
| 2761 | U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); |
| 2762 | result = toUPPER_LC(c); |
| 2763 | } |
| 2764 | else { |
| 2765 | return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), |
| 2766 | ustrp, lenp, 'S'); |
| 2767 | } |
| 2768 | } |
| 2769 | else { /* UTF-8, ord above 255 */ |
| 2770 | result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp); |
| 2771 | |
| 2772 | if (flags) { |
| 2773 | result = check_locale_boundary_crossing(p, result, ustrp, lenp); |
| 2774 | } |
| 2775 | return result; |
| 2776 | } |
| 2777 | |
| 2778 | /* Here, used locale rules. Convert back to UTF-8 */ |
| 2779 | if (UTF8_IS_INVARIANT(result)) { |
| 2780 | *ustrp = (U8) result; |
| 2781 | *lenp = 1; |
| 2782 | } |
| 2783 | else { |
| 2784 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); |
| 2785 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); |
| 2786 | *lenp = 2; |
| 2787 | } |
| 2788 | |
| 2789 | return result; |
| 2790 | } |
| 2791 | |
| 2792 | /* |
| 2793 | =for apidoc to_utf8_title |
| 2794 | |
| 2795 | Instead use L</toTITLE_utf8>. |
| 2796 | |
| 2797 | =cut */ |
| 2798 | |
| 2799 | /* Not currently externally documented, and subject to change: |
| 2800 | * <flags> is set iff the rules from the current underlying locale are to be |
| 2801 | * used. Since titlecase is not defined in POSIX, for other than a |
| 2802 | * UTF-8 locale, uppercase is used instead for code points < 256. |
| 2803 | */ |
| 2804 | |
| 2805 | UV |
| 2806 | Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags) |
| 2807 | { |
| 2808 | UV result; |
| 2809 | |
| 2810 | PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS; |
| 2811 | |
| 2812 | if (flags) { |
| 2813 | /* Treat a UTF-8 locale as not being in locale at all */ |
| 2814 | if (IN_UTF8_CTYPE_LOCALE) { |
| 2815 | flags = FALSE; |
| 2816 | } |
| 2817 | else { |
| 2818 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2819 | } |
| 2820 | } |
| 2821 | |
| 2822 | if (UTF8_IS_INVARIANT(*p)) { |
| 2823 | if (flags) { |
| 2824 | result = toUPPER_LC(*p); |
| 2825 | } |
| 2826 | else { |
| 2827 | return _to_upper_title_latin1(*p, ustrp, lenp, 's'); |
| 2828 | } |
| 2829 | } |
| 2830 | else if UTF8_IS_DOWNGRADEABLE_START(*p) { |
| 2831 | if (flags) { |
| 2832 | U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); |
| 2833 | result = toUPPER_LC(c); |
| 2834 | } |
| 2835 | else { |
| 2836 | return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), |
| 2837 | ustrp, lenp, 's'); |
| 2838 | } |
| 2839 | } |
| 2840 | else { /* UTF-8, ord above 255 */ |
| 2841 | result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp); |
| 2842 | |
| 2843 | if (flags) { |
| 2844 | result = check_locale_boundary_crossing(p, result, ustrp, lenp); |
| 2845 | } |
| 2846 | return result; |
| 2847 | } |
| 2848 | |
| 2849 | /* Here, used locale rules. Convert back to UTF-8 */ |
| 2850 | if (UTF8_IS_INVARIANT(result)) { |
| 2851 | *ustrp = (U8) result; |
| 2852 | *lenp = 1; |
| 2853 | } |
| 2854 | else { |
| 2855 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); |
| 2856 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); |
| 2857 | *lenp = 2; |
| 2858 | } |
| 2859 | |
| 2860 | return result; |
| 2861 | } |
| 2862 | |
| 2863 | /* |
| 2864 | =for apidoc to_utf8_lower |
| 2865 | |
| 2866 | Instead use L</toLOWER_utf8>. |
| 2867 | |
| 2868 | =cut */ |
| 2869 | |
| 2870 | /* Not currently externally documented, and subject to change: |
| 2871 | * <flags> is set iff iff the rules from the current underlying locale are to |
| 2872 | * be used. |
| 2873 | */ |
| 2874 | |
| 2875 | UV |
| 2876 | Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags) |
| 2877 | { |
| 2878 | UV result; |
| 2879 | |
| 2880 | PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS; |
| 2881 | |
| 2882 | if (flags) { |
| 2883 | /* Treat a UTF-8 locale as not being in locale at all */ |
| 2884 | if (IN_UTF8_CTYPE_LOCALE) { |
| 2885 | flags = FALSE; |
| 2886 | } |
| 2887 | else { |
| 2888 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2889 | } |
| 2890 | } |
| 2891 | |
| 2892 | if (UTF8_IS_INVARIANT(*p)) { |
| 2893 | if (flags) { |
| 2894 | result = toLOWER_LC(*p); |
| 2895 | } |
| 2896 | else { |
| 2897 | return to_lower_latin1(*p, ustrp, lenp); |
| 2898 | } |
| 2899 | } |
| 2900 | else if UTF8_IS_DOWNGRADEABLE_START(*p) { |
| 2901 | if (flags) { |
| 2902 | U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); |
| 2903 | result = toLOWER_LC(c); |
| 2904 | } |
| 2905 | else { |
| 2906 | return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), |
| 2907 | ustrp, lenp); |
| 2908 | } |
| 2909 | } |
| 2910 | else { /* UTF-8, ord above 255 */ |
| 2911 | result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp); |
| 2912 | |
| 2913 | if (flags) { |
| 2914 | result = check_locale_boundary_crossing(p, result, ustrp, lenp); |
| 2915 | } |
| 2916 | |
| 2917 | return result; |
| 2918 | } |
| 2919 | |
| 2920 | /* Here, used locale rules. Convert back to UTF-8 */ |
| 2921 | if (UTF8_IS_INVARIANT(result)) { |
| 2922 | *ustrp = (U8) result; |
| 2923 | *lenp = 1; |
| 2924 | } |
| 2925 | else { |
| 2926 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); |
| 2927 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); |
| 2928 | *lenp = 2; |
| 2929 | } |
| 2930 | |
| 2931 | return result; |
| 2932 | } |
| 2933 | |
| 2934 | /* |
| 2935 | =for apidoc to_utf8_fold |
| 2936 | |
| 2937 | Instead use L</toFOLD_utf8>. |
| 2938 | |
| 2939 | =cut */ |
| 2940 | |
| 2941 | /* Not currently externally documented, and subject to change, |
| 2942 | * in <flags> |
| 2943 | * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying |
| 2944 | * locale are to be used. |
| 2945 | * bit FOLD_FLAGS_FULL is set iff full case folds are to be used; |
| 2946 | * otherwise simple folds |
| 2947 | * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are |
| 2948 | * prohibited |
| 2949 | */ |
| 2950 | |
| 2951 | UV |
| 2952 | Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags) |
| 2953 | { |
| 2954 | UV result; |
| 2955 | |
| 2956 | PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS; |
| 2957 | |
| 2958 | /* These are mutually exclusive */ |
| 2959 | assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII))); |
| 2960 | |
| 2961 | assert(p != ustrp); /* Otherwise overwrites */ |
| 2962 | |
| 2963 | if (flags & FOLD_FLAGS_LOCALE) { |
| 2964 | /* Treat a UTF-8 locale as not being in locale at all */ |
| 2965 | if (IN_UTF8_CTYPE_LOCALE) { |
| 2966 | flags &= ~FOLD_FLAGS_LOCALE; |
| 2967 | } |
| 2968 | else { |
| 2969 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2970 | } |
| 2971 | } |
| 2972 | |
| 2973 | if (UTF8_IS_INVARIANT(*p)) { |
| 2974 | if (flags & FOLD_FLAGS_LOCALE) { |
| 2975 | result = toFOLD_LC(*p); |
| 2976 | } |
| 2977 | else { |
| 2978 | return _to_fold_latin1(*p, ustrp, lenp, |
| 2979 | flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); |
| 2980 | } |
| 2981 | } |
| 2982 | else if UTF8_IS_DOWNGRADEABLE_START(*p) { |
| 2983 | if (flags & FOLD_FLAGS_LOCALE) { |
| 2984 | U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); |
| 2985 | result = toFOLD_LC(c); |
| 2986 | } |
| 2987 | else { |
| 2988 | return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), |
| 2989 | ustrp, lenp, |
| 2990 | flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)); |
| 2991 | } |
| 2992 | } |
| 2993 | else { /* UTF-8, ord above 255 */ |
| 2994 | result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL); |
| 2995 | |
| 2996 | if (flags & FOLD_FLAGS_LOCALE) { |
| 2997 | |
| 2998 | # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8 |
| 2999 | const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1; |
| 3000 | |
| 3001 | # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8 |
| 3002 | # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8 |
| 3003 | |
| 3004 | const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1; |
| 3005 | |
| 3006 | /* Special case these two characters, as what normally gets |
| 3007 | * returned under locale doesn't work */ |
| 3008 | if (UTF8SKIP(p) == cap_sharp_s_len |
| 3009 | && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len)) |
| 3010 | { |
| 3011 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 3012 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 3013 | "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; " |
| 3014 | "resolved to \"\\x{17F}\\x{17F}\"."); |
| 3015 | goto return_long_s; |
| 3016 | } |
| 3017 | else |
| 3018 | #endif |
| 3019 | if (UTF8SKIP(p) == long_s_t_len |
| 3020 | && memEQ((char *) p, LONG_S_T, long_s_t_len)) |
| 3021 | { |
| 3022 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 3023 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 3024 | "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; " |
| 3025 | "resolved to \"\\x{FB06}\"."); |
| 3026 | goto return_ligature_st; |
| 3027 | } |
| 3028 | |
| 3029 | #if UNICODE_MAJOR_VERSION == 3 \ |
| 3030 | && UNICODE_DOT_VERSION == 0 \ |
| 3031 | && UNICODE_DOT_DOT_VERSION == 1 |
| 3032 | # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8 |
| 3033 | |
| 3034 | /* And special case this on this Unicode version only, for the same |
| 3035 | * reaons the other two are special cased. They would cross the |
| 3036 | * 255/256 boundary which is forbidden under /l, and so the code |
| 3037 | * wouldn't catch that they are equivalent (which they are only in |
| 3038 | * this release) */ |
| 3039 | else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1 |
| 3040 | && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1)) |
| 3041 | { |
| 3042 | /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */ |
| 3043 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 3044 | "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; " |
| 3045 | "resolved to \"\\x{0131}\"."); |
| 3046 | goto return_dotless_i; |
| 3047 | } |
| 3048 | #endif |
| 3049 | |
| 3050 | return check_locale_boundary_crossing(p, result, ustrp, lenp); |
| 3051 | } |
| 3052 | else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) { |
| 3053 | return result; |
| 3054 | } |
| 3055 | else { |
| 3056 | /* This is called when changing the case of a UTF-8-encoded |
| 3057 | * character above the ASCII range, and the result should not |
| 3058 | * contain an ASCII character. */ |
| 3059 | |
| 3060 | UV original; /* To store the first code point of <p> */ |
| 3061 | |
| 3062 | /* Look at every character in the result; if any cross the |
| 3063 | * boundary, the whole thing is disallowed */ |
| 3064 | U8* s = ustrp; |
| 3065 | U8* e = ustrp + *lenp; |
| 3066 | while (s < e) { |
| 3067 | if (isASCII(*s)) { |
| 3068 | /* Crossed, have to return the original */ |
| 3069 | original = valid_utf8_to_uvchr(p, lenp); |
| 3070 | |
| 3071 | /* But in these instances, there is an alternative we can |
| 3072 | * return that is valid */ |
| 3073 | if (original == LATIN_SMALL_LETTER_SHARP_S |
| 3074 | #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */ |
| 3075 | || original == LATIN_CAPITAL_LETTER_SHARP_S |
| 3076 | #endif |
| 3077 | ) { |
| 3078 | goto return_long_s; |
| 3079 | } |
| 3080 | else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) { |
| 3081 | goto return_ligature_st; |
| 3082 | } |
| 3083 | #if UNICODE_MAJOR_VERSION == 3 \ |
| 3084 | && UNICODE_DOT_VERSION == 0 \ |
| 3085 | && UNICODE_DOT_DOT_VERSION == 1 |
| 3086 | |
| 3087 | else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) { |
| 3088 | goto return_dotless_i; |
| 3089 | } |
| 3090 | #endif |
| 3091 | Copy(p, ustrp, *lenp, char); |
| 3092 | return original; |
| 3093 | } |
| 3094 | s += UTF8SKIP(s); |
| 3095 | } |
| 3096 | |
| 3097 | /* Here, no characters crossed, result is ok as-is */ |
| 3098 | return result; |
| 3099 | } |
| 3100 | } |
| 3101 | |
| 3102 | /* Here, used locale rules. Convert back to UTF-8 */ |
| 3103 | if (UTF8_IS_INVARIANT(result)) { |
| 3104 | *ustrp = (U8) result; |
| 3105 | *lenp = 1; |
| 3106 | } |
| 3107 | else { |
| 3108 | *ustrp = UTF8_EIGHT_BIT_HI((U8) result); |
| 3109 | *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); |
| 3110 | *lenp = 2; |
| 3111 | } |
| 3112 | |
| 3113 | return result; |
| 3114 | |
| 3115 | return_long_s: |
| 3116 | /* Certain folds to 'ss' are prohibited by the options, but they do allow |
| 3117 | * folds to a string of two of these characters. By returning this |
| 3118 | * instead, then, e.g., |
| 3119 | * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}") |
| 3120 | * works. */ |
| 3121 | |
| 3122 | *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2; |
| 3123 | Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8, |
| 3124 | ustrp, *lenp, U8); |
| 3125 | return LATIN_SMALL_LETTER_LONG_S; |
| 3126 | |
| 3127 | return_ligature_st: |
| 3128 | /* Two folds to 'st' are prohibited by the options; instead we pick one and |
| 3129 | * have the other one fold to it */ |
| 3130 | |
| 3131 | *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1; |
| 3132 | Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8); |
| 3133 | return LATIN_SMALL_LIGATURE_ST; |
| 3134 | |
| 3135 | #if UNICODE_MAJOR_VERSION == 3 \ |
| 3136 | && UNICODE_DOT_VERSION == 0 \ |
| 3137 | && UNICODE_DOT_DOT_VERSION == 1 |
| 3138 | |
| 3139 | return_dotless_i: |
| 3140 | *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1; |
| 3141 | Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8); |
| 3142 | return LATIN_SMALL_LETTER_DOTLESS_I; |
| 3143 | |
| 3144 | #endif |
| 3145 | |
| 3146 | } |
| 3147 | |
| 3148 | /* Note: |
| 3149 | * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch(). |
| 3150 | * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8". |
| 3151 | * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl. |
| 3152 | */ |
| 3153 | |
| 3154 | SV* |
| 3155 | Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none) |
| 3156 | { |
| 3157 | PERL_ARGS_ASSERT_SWASH_INIT; |
| 3158 | |
| 3159 | /* Returns a copy of a swash initiated by the called function. This is the |
| 3160 | * public interface, and returning a copy prevents others from doing |
| 3161 | * mischief on the original */ |
| 3162 | |
| 3163 | return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL)); |
| 3164 | } |
| 3165 | |
| 3166 | SV* |
| 3167 | Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p) |
| 3168 | { |
| 3169 | |
| 3170 | /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST |
| 3171 | * use the following define */ |
| 3172 | |
| 3173 | #define CORE_SWASH_INIT_RETURN(x) \ |
| 3174 | PL_curpm= old_PL_curpm; \ |
| 3175 | return x |
| 3176 | |
| 3177 | /* Initialize and return a swash, creating it if necessary. It does this |
| 3178 | * by calling utf8_heavy.pl in the general case. The returned value may be |
| 3179 | * the swash's inversion list instead if the input parameters allow it. |
| 3180 | * Which is returned should be immaterial to callers, as the only |
| 3181 | * operations permitted on a swash, swash_fetch(), _get_swash_invlist(), |
| 3182 | * and swash_to_invlist() handle both these transparently. |
| 3183 | * |
| 3184 | * This interface should only be used by functions that won't destroy or |
| 3185 | * adversely change the swash, as doing so affects all other uses of the |
| 3186 | * swash in the program; the general public should use 'Perl_swash_init' |
| 3187 | * instead. |
| 3188 | * |
| 3189 | * pkg is the name of the package that <name> should be in. |
| 3190 | * name is the name of the swash to find. Typically it is a Unicode |
| 3191 | * property name, including user-defined ones |
| 3192 | * listsv is a string to initialize the swash with. It must be of the form |
| 3193 | * documented as the subroutine return value in |
| 3194 | * L<perlunicode/User-Defined Character Properties> |
| 3195 | * minbits is the number of bits required to represent each data element. |
| 3196 | * It is '1' for binary properties. |
| 3197 | * none I (khw) do not understand this one, but it is used only in tr///. |
| 3198 | * invlist is an inversion list to initialize the swash with (or NULL) |
| 3199 | * flags_p if non-NULL is the address of various input and output flag bits |
| 3200 | * to the routine, as follows: ('I' means is input to the routine; |
| 3201 | * 'O' means output from the routine. Only flags marked O are |
| 3202 | * meaningful on return.) |
| 3203 | * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash |
| 3204 | * came from a user-defined property. (I O) |
| 3205 | * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking |
| 3206 | * when the swash cannot be located, to simply return NULL. (I) |
| 3207 | * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a |
| 3208 | * return of an inversion list instead of a swash hash if this routine |
| 3209 | * thinks that would result in faster execution of swash_fetch() later |
| 3210 | * on. (I) |
| 3211 | * |
| 3212 | * Thus there are three possible inputs to find the swash: <name>, |
| 3213 | * <listsv>, and <invlist>. At least one must be specified. The result |
| 3214 | * will be the union of the specified ones, although <listsv>'s various |
| 3215 | * actions can intersect, etc. what <name> gives. To avoid going out to |
| 3216 | * disk at all, <invlist> should specify completely what the swash should |
| 3217 | * have, and <listsv> should be &PL_sv_undef and <name> should be "". |
| 3218 | * |
| 3219 | * <invlist> is only valid for binary properties */ |
| 3220 | |
| 3221 | PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */ |
| 3222 | |
| 3223 | SV* retval = &PL_sv_undef; |
| 3224 | HV* swash_hv = NULL; |
| 3225 | const int invlist_swash_boundary = |
| 3226 | (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST) |
| 3227 | ? 512 /* Based on some benchmarking, but not extensive, see commit |
| 3228 | message */ |
| 3229 | : -1; /* Never return just an inversion list */ |
| 3230 | |
| 3231 | assert(listsv != &PL_sv_undef || strNE(name, "") || invlist); |
| 3232 | assert(! invlist || minbits == 1); |
| 3233 | |
| 3234 | PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex |
| 3235 | that triggered the swash init and the swash init perl logic itself. |
| 3236 | See perl #122747 */ |
| 3237 | |
| 3238 | /* If data was passed in to go out to utf8_heavy to find the swash of, do |
| 3239 | * so */ |
| 3240 | if (listsv != &PL_sv_undef || strNE(name, "")) { |
| 3241 | dSP; |
| 3242 | const size_t pkg_len = strlen(pkg); |
| 3243 | const size_t name_len = strlen(name); |
| 3244 | HV * const stash = gv_stashpvn(pkg, pkg_len, 0); |
| 3245 | SV* errsv_save; |
| 3246 | GV *method; |
| 3247 | |
| 3248 | PERL_ARGS_ASSERT__CORE_SWASH_INIT; |
| 3249 | |
| 3250 | PUSHSTACKi(PERLSI_MAGIC); |
| 3251 | ENTER; |
| 3252 | SAVEHINTS(); |
| 3253 | save_re_context(); |
| 3254 | /* We might get here via a subroutine signature which uses a utf8 |
| 3255 | * parameter name, at which point PL_subname will have been set |
| 3256 | * but not yet used. */ |
| 3257 | save_item(PL_subname); |
| 3258 | if (PL_parser && PL_parser->error_count) |
| 3259 | SAVEI8(PL_parser->error_count), PL_parser->error_count = 0; |
| 3260 | method = gv_fetchmeth(stash, "SWASHNEW", 8, -1); |
| 3261 | if (!method) { /* demand load UTF-8 */ |
| 3262 | ENTER; |
| 3263 | if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); |
| 3264 | GvSV(PL_errgv) = NULL; |
| 3265 | #ifndef NO_TAINT_SUPPORT |
| 3266 | /* It is assumed that callers of this routine are not passing in |
| 3267 | * any user derived data. */ |
| 3268 | /* Need to do this after save_re_context() as it will set |
| 3269 | * PL_tainted to 1 while saving $1 etc (see the code after getrx: |
| 3270 | * in Perl_magic_get). Even line to create errsv_save can turn on |
| 3271 | * PL_tainted. */ |
| 3272 | SAVEBOOL(TAINT_get); |
| 3273 | TAINT_NOT; |
| 3274 | #endif |
| 3275 | Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len), |
| 3276 | NULL); |
| 3277 | { |
| 3278 | /* Not ERRSV, as there is no need to vivify a scalar we are |
| 3279 | about to discard. */ |
| 3280 | SV * const errsv = GvSV(PL_errgv); |
| 3281 | if (!SvTRUE(errsv)) { |
| 3282 | GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); |
| 3283 | SvREFCNT_dec(errsv); |
| 3284 | } |
| 3285 | } |
| 3286 | LEAVE; |
| 3287 | } |
| 3288 | SPAGAIN; |
| 3289 | PUSHMARK(SP); |
| 3290 | EXTEND(SP,5); |
| 3291 | mPUSHp(pkg, pkg_len); |
| 3292 | mPUSHp(name, name_len); |
| 3293 | PUSHs(listsv); |
| 3294 | mPUSHi(minbits); |
| 3295 | mPUSHi(none); |
| 3296 | PUTBACK; |
| 3297 | if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save); |
| 3298 | GvSV(PL_errgv) = NULL; |
| 3299 | /* If we already have a pointer to the method, no need to use |
| 3300 | * call_method() to repeat the lookup. */ |
| 3301 | if (method |
| 3302 | ? call_sv(MUTABLE_SV(method), G_SCALAR) |
| 3303 | : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD)) |
| 3304 | { |
| 3305 | retval = *PL_stack_sp--; |
| 3306 | SvREFCNT_inc(retval); |
| 3307 | } |
| 3308 | { |
| 3309 | /* Not ERRSV. See above. */ |
| 3310 | SV * const errsv = GvSV(PL_errgv); |
| 3311 | if (!SvTRUE(errsv)) { |
| 3312 | GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save); |
| 3313 | SvREFCNT_dec(errsv); |
| 3314 | } |
| 3315 | } |
| 3316 | LEAVE; |
| 3317 | POPSTACK; |
| 3318 | if (IN_PERL_COMPILETIME) { |
| 3319 | CopHINTS_set(PL_curcop, PL_hints); |
| 3320 | } |
| 3321 | if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) { |
| 3322 | if (SvPOK(retval)) { |
| 3323 | |
| 3324 | /* If caller wants to handle missing properties, let them */ |
| 3325 | if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) { |
| 3326 | CORE_SWASH_INIT_RETURN(NULL); |
| 3327 | } |
| 3328 | Perl_croak(aTHX_ |
| 3329 | "Can't find Unicode property definition \"%"SVf"\"", |
| 3330 | SVfARG(retval)); |
| 3331 | NOT_REACHED; /* NOTREACHED */ |
| 3332 | } |
| 3333 | } |
| 3334 | } /* End of calling the module to find the swash */ |
| 3335 | |
| 3336 | /* If this operation fetched a swash, and we will need it later, get it */ |
| 3337 | if (retval != &PL_sv_undef |
| 3338 | && (minbits == 1 || (flags_p |
| 3339 | && ! (*flags_p |
| 3340 | & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)))) |
| 3341 | { |
| 3342 | swash_hv = MUTABLE_HV(SvRV(retval)); |
| 3343 | |
| 3344 | /* If we don't already know that there is a user-defined component to |
| 3345 | * this swash, and the user has indicated they wish to know if there is |
| 3346 | * one (by passing <flags_p>), find out */ |
| 3347 | if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) { |
| 3348 | SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE); |
| 3349 | if (user_defined && SvUV(*user_defined)) { |
| 3350 | *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY; |
| 3351 | } |
| 3352 | } |
| 3353 | } |
| 3354 | |
| 3355 | /* Make sure there is an inversion list for binary properties */ |
| 3356 | if (minbits == 1) { |
| 3357 | SV** swash_invlistsvp = NULL; |
| 3358 | SV* swash_invlist = NULL; |
| 3359 | bool invlist_in_swash_is_valid = FALSE; |
| 3360 | bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has |
| 3361 | an unclaimed reference count */ |
| 3362 | |
| 3363 | /* If this operation fetched a swash, get its already existing |
| 3364 | * inversion list, or create one for it */ |
| 3365 | |
| 3366 | if (swash_hv) { |
| 3367 | swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE); |
| 3368 | if (swash_invlistsvp) { |
| 3369 | swash_invlist = *swash_invlistsvp; |
| 3370 | invlist_in_swash_is_valid = TRUE; |
| 3371 | } |
| 3372 | else { |
| 3373 | swash_invlist = _swash_to_invlist(retval); |
| 3374 | swash_invlist_unclaimed = TRUE; |
| 3375 | } |
| 3376 | } |
| 3377 | |
| 3378 | /* If an inversion list was passed in, have to include it */ |
| 3379 | if (invlist) { |
| 3380 | |
| 3381 | /* Any fetched swash will by now have an inversion list in it; |
| 3382 | * otherwise <swash_invlist> will be NULL, indicating that we |
| 3383 | * didn't fetch a swash */ |
| 3384 | if (swash_invlist) { |
| 3385 | |
| 3386 | /* Add the passed-in inversion list, which invalidates the one |
| 3387 | * already stored in the swash */ |
| 3388 | invlist_in_swash_is_valid = FALSE; |
| 3389 | _invlist_union(invlist, swash_invlist, &swash_invlist); |
| 3390 | } |
| 3391 | else { |
| 3392 | |
| 3393 | /* Here, there is no swash already. Set up a minimal one, if |
| 3394 | * we are going to return a swash */ |
| 3395 | if ((int) _invlist_len(invlist) > invlist_swash_boundary) { |
| 3396 | swash_hv = newHV(); |
| 3397 | retval = newRV_noinc(MUTABLE_SV(swash_hv)); |
| 3398 | } |
| 3399 | swash_invlist = invlist; |
| 3400 | } |
| 3401 | } |
| 3402 | |
| 3403 | /* Here, we have computed the union of all the passed-in data. It may |
| 3404 | * be that there was an inversion list in the swash which didn't get |
| 3405 | * touched; otherwise save the computed one */ |
| 3406 | if (! invlist_in_swash_is_valid |
| 3407 | && (int) _invlist_len(swash_invlist) > invlist_swash_boundary) |
| 3408 | { |
| 3409 | if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist)) |
| 3410 | { |
| 3411 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); |
| 3412 | } |
| 3413 | /* We just stole a reference count. */ |
| 3414 | if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE; |
| 3415 | else SvREFCNT_inc_simple_void_NN(swash_invlist); |
| 3416 | } |
| 3417 | |
| 3418 | SvREADONLY_on(swash_invlist); |
| 3419 | |
| 3420 | /* Use the inversion list stand-alone if small enough */ |
| 3421 | if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) { |
| 3422 | SvREFCNT_dec(retval); |
| 3423 | if (!swash_invlist_unclaimed) |
| 3424 | SvREFCNT_inc_simple_void_NN(swash_invlist); |
| 3425 | retval = newRV_noinc(swash_invlist); |
| 3426 | } |
| 3427 | } |
| 3428 | |
| 3429 | CORE_SWASH_INIT_RETURN(retval); |
| 3430 | #undef CORE_SWASH_INIT_RETURN |
| 3431 | } |
| 3432 | |
| 3433 | |
| 3434 | /* This API is wrong for special case conversions since we may need to |
| 3435 | * return several Unicode characters for a single Unicode character |
| 3436 | * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is |
| 3437 | * the lower-level routine, and it is similarly broken for returning |
| 3438 | * multiple values. --jhi |
| 3439 | * For those, you should use S__to_utf8_case() instead */ |
| 3440 | /* Now SWASHGET is recasted into S_swatch_get in this file. */ |
| 3441 | |
| 3442 | /* Note: |
| 3443 | * Returns the value of property/mapping C<swash> for the first character |
| 3444 | * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is |
| 3445 | * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr> |
| 3446 | * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>. |
| 3447 | * |
| 3448 | * A "swash" is a hash which contains initially the keys/values set up by |
| 3449 | * SWASHNEW. The purpose is to be able to completely represent a Unicode |
| 3450 | * property for all possible code points. Things are stored in a compact form |
| 3451 | * (see utf8_heavy.pl) so that calculation is required to find the actual |
| 3452 | * property value for a given code point. As code points are looked up, new |
| 3453 | * key/value pairs are added to the hash, so that the calculation doesn't have |
| 3454 | * to ever be re-done. Further, each calculation is done, not just for the |
| 3455 | * desired one, but for a whole block of code points adjacent to that one. |
| 3456 | * For binary properties on ASCII machines, the block is usually for 64 code |
| 3457 | * points, starting with a code point evenly divisible by 64. Thus if the |
| 3458 | * property value for code point 257 is requested, the code goes out and |
| 3459 | * calculates the property values for all 64 code points between 256 and 319, |
| 3460 | * and stores these as a single 64-bit long bit vector, called a "swatch", |
| 3461 | * under the key for code point 256. The key is the UTF-8 encoding for code |
| 3462 | * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding |
| 3463 | * for a code point is 13 bytes, the key will be 12 bytes long. If the value |
| 3464 | * for code point 258 is then requested, this code realizes that it would be |
| 3465 | * stored under the key for 256, and would find that value and extract the |
| 3466 | * relevant bit, offset from 256. |
| 3467 | * |
| 3468 | * Non-binary properties are stored in as many bits as necessary to represent |
| 3469 | * their values (32 currently, though the code is more general than that), not |
| 3470 | * as single bits, but the principle is the same: the value for each key is a |
| 3471 | * vector that encompasses the property values for all code points whose UTF-8 |
| 3472 | * representations are represented by the key. That is, for all code points |
| 3473 | * whose UTF-8 representations are length N bytes, and the key is the first N-1 |
| 3474 | * bytes of that. |
| 3475 | */ |
| 3476 | UV |
| 3477 | Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8) |
| 3478 | { |
| 3479 | HV *const hv = MUTABLE_HV(SvRV(swash)); |
| 3480 | U32 klen; |
| 3481 | U32 off; |
| 3482 | STRLEN slen = 0; |
| 3483 | STRLEN needents; |
| 3484 | const U8 *tmps = NULL; |
| 3485 | SV *swatch; |
| 3486 | const U8 c = *ptr; |
| 3487 | |
| 3488 | PERL_ARGS_ASSERT_SWASH_FETCH; |
| 3489 | |
| 3490 | /* If it really isn't a hash, it isn't really swash; must be an inversion |
| 3491 | * list */ |
| 3492 | if (SvTYPE(hv) != SVt_PVHV) { |
| 3493 | return _invlist_contains_cp((SV*)hv, |
| 3494 | (do_utf8) |
| 3495 | ? valid_utf8_to_uvchr(ptr, NULL) |
| 3496 | : c); |
| 3497 | } |
| 3498 | |
| 3499 | /* We store the values in a "swatch" which is a vec() value in a swash |
| 3500 | * hash. Code points 0-255 are a single vec() stored with key length |
| 3501 | * (klen) 0. All other code points have a UTF-8 representation |
| 3502 | * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which |
| 3503 | * share 0xAA..0xYY, which is the key in the hash to that vec. So the key |
| 3504 | * length for them is the length of the encoded char - 1. ptr[klen] is the |
| 3505 | * final byte in the sequence representing the character */ |
| 3506 | if (!do_utf8 || UTF8_IS_INVARIANT(c)) { |
| 3507 | klen = 0; |
| 3508 | needents = 256; |
| 3509 | off = c; |
| 3510 | } |
| 3511 | else if (UTF8_IS_DOWNGRADEABLE_START(c)) { |
| 3512 | klen = 0; |
| 3513 | needents = 256; |
| 3514 | off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1)); |
| 3515 | } |
| 3516 | else { |
| 3517 | klen = UTF8SKIP(ptr) - 1; |
| 3518 | |
| 3519 | /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into |
| 3520 | * the vec is the final byte in the sequence. (In EBCDIC this is |
| 3521 | * converted to I8 to get consecutive values.) To help you visualize |
| 3522 | * all this: |
| 3523 | * Straight 1047 After final byte |
| 3524 | * UTF-8 UTF-EBCDIC I8 transform |
| 3525 | * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0 |
| 3526 | * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1 |
| 3527 | * ... |
| 3528 | * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9 |
| 3529 | * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA |
| 3530 | * ... |
| 3531 | * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2 |
| 3532 | * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3 |
| 3533 | * ... |
| 3534 | * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB |
| 3535 | * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC |
| 3536 | * ... |
| 3537 | * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF |
| 3538 | * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41 |
| 3539 | * |
| 3540 | * (There are no discontinuities in the elided (...) entries.) |
| 3541 | * The UTF-8 key for these 33 code points is '\xD0' (which also is the |
| 3542 | * key for the next 31, up through U+043F, whose UTF-8 final byte is |
| 3543 | * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points. |
| 3544 | * The final UTF-8 byte, which ranges between \x80 and \xBF, is an |
| 3545 | * index into the vec() swatch (after subtracting 0x80, which we |
| 3546 | * actually do with an '&'). |
| 3547 | * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32 |
| 3548 | * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has |
| 3549 | * dicontinuities which go away by transforming it into I8, and we |
| 3550 | * effectively subtract 0xA0 to get the index. */ |
| 3551 | needents = (1 << UTF_ACCUMULATION_SHIFT); |
| 3552 | off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK; |
| 3553 | } |
| 3554 | |
| 3555 | /* |
| 3556 | * This single-entry cache saves about 1/3 of the UTF-8 overhead in test |
| 3557 | * suite. (That is, only 7-8% overall over just a hash cache. Still, |
| 3558 | * it's nothing to sniff at.) Pity we usually come through at least |
| 3559 | * two function calls to get here... |
| 3560 | * |
| 3561 | * NB: this code assumes that swatches are never modified, once generated! |
| 3562 | */ |
| 3563 | |
| 3564 | if (hv == PL_last_swash_hv && |
| 3565 | klen == PL_last_swash_klen && |
| 3566 | (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) ) |
| 3567 | { |
| 3568 | tmps = PL_last_swash_tmps; |
| 3569 | slen = PL_last_swash_slen; |
| 3570 | } |
| 3571 | else { |
| 3572 | /* Try our second-level swatch cache, kept in a hash. */ |
| 3573 | SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE); |
| 3574 | |
| 3575 | /* If not cached, generate it via swatch_get */ |
| 3576 | if (!svp || !SvPOK(*svp) |
| 3577 | || !(tmps = (const U8*)SvPV_const(*svp, slen))) |
| 3578 | { |
| 3579 | if (klen) { |
| 3580 | const UV code_point = valid_utf8_to_uvchr(ptr, NULL); |
| 3581 | swatch = swatch_get(swash, |
| 3582 | code_point & ~((UV)needents - 1), |
| 3583 | needents); |
| 3584 | } |
| 3585 | else { /* For the first 256 code points, the swatch has a key of |
| 3586 | length 0 */ |
| 3587 | swatch = swatch_get(swash, 0, needents); |
| 3588 | } |
| 3589 | |
| 3590 | if (IN_PERL_COMPILETIME) |
| 3591 | CopHINTS_set(PL_curcop, PL_hints); |
| 3592 | |
| 3593 | svp = hv_store(hv, (const char *)ptr, klen, swatch, 0); |
| 3594 | |
| 3595 | if (!svp || !(tmps = (U8*)SvPV(*svp, slen)) |
| 3596 | || (slen << 3) < needents) |
| 3597 | Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, " |
| 3598 | "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf, |
| 3599 | svp, tmps, (UV)slen, (UV)needents); |
| 3600 | } |
| 3601 | |
| 3602 | PL_last_swash_hv = hv; |
| 3603 | assert(klen <= sizeof(PL_last_swash_key)); |
| 3604 | PL_last_swash_klen = (U8)klen; |
| 3605 | /* FIXME change interpvar.h? */ |
| 3606 | PL_last_swash_tmps = (U8 *) tmps; |
| 3607 | PL_last_swash_slen = slen; |
| 3608 | if (klen) |
| 3609 | Copy(ptr, PL_last_swash_key, klen, U8); |
| 3610 | } |
| 3611 | |
| 3612 | switch ((int)((slen << 3) / needents)) { |
| 3613 | case 1: |
| 3614 | return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0; |
| 3615 | case 8: |
| 3616 | return ((UV) tmps[off]); |
| 3617 | case 16: |
| 3618 | off <<= 1; |
| 3619 | return |
| 3620 | ((UV) tmps[off ] << 8) + |
| 3621 | ((UV) tmps[off + 1]); |
| 3622 | case 32: |
| 3623 | off <<= 2; |
| 3624 | return |
| 3625 | ((UV) tmps[off ] << 24) + |
| 3626 | ((UV) tmps[off + 1] << 16) + |
| 3627 | ((UV) tmps[off + 2] << 8) + |
| 3628 | ((UV) tmps[off + 3]); |
| 3629 | } |
| 3630 | Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, " |
| 3631 | "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents); |
| 3632 | NORETURN_FUNCTION_END; |
| 3633 | } |
| 3634 | |
| 3635 | /* Read a single line of the main body of the swash input text. These are of |
| 3636 | * the form: |
| 3637 | * 0053 0056 0073 |
| 3638 | * where each number is hex. The first two numbers form the minimum and |
| 3639 | * maximum of a range, and the third is the value associated with the range. |
| 3640 | * Not all swashes should have a third number |
| 3641 | * |
| 3642 | * On input: l points to the beginning of the line to be examined; it points |
| 3643 | * to somewhere in the string of the whole input text, and is |
| 3644 | * terminated by a \n or the null string terminator. |
| 3645 | * lend points to the null terminator of that string |
| 3646 | * wants_value is non-zero if the swash expects a third number |
| 3647 | * typestr is the name of the swash's mapping, like 'ToLower' |
| 3648 | * On output: *min, *max, and *val are set to the values read from the line. |
| 3649 | * returns a pointer just beyond the line examined. If there was no |
| 3650 | * valid min number on the line, returns lend+1 |
| 3651 | */ |
| 3652 | |
| 3653 | STATIC U8* |
| 3654 | S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val, |
| 3655 | const bool wants_value, const U8* const typestr) |
| 3656 | { |
| 3657 | const int typeto = typestr[0] == 'T' && typestr[1] == 'o'; |
| 3658 | STRLEN numlen; /* Length of the number */ |
| 3659 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT |
| 3660 | | PERL_SCAN_DISALLOW_PREFIX |
| 3661 | | PERL_SCAN_SILENT_NON_PORTABLE; |
| 3662 | |
| 3663 | /* nl points to the next \n in the scan */ |
| 3664 | U8* const nl = (U8*)memchr(l, '\n', lend - l); |
| 3665 | |
| 3666 | PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE; |
| 3667 | |
| 3668 | /* Get the first number on the line: the range minimum */ |
| 3669 | numlen = lend - l; |
| 3670 | *min = grok_hex((char *)l, &numlen, &flags, NULL); |
| 3671 | *max = *min; /* So can never return without setting max */ |
| 3672 | if (numlen) /* If found a hex number, position past it */ |
| 3673 | l += numlen; |
| 3674 | else if (nl) { /* Else, go handle next line, if any */ |
| 3675 | return nl + 1; /* 1 is length of "\n" */ |
| 3676 | } |
| 3677 | else { /* Else, no next line */ |
| 3678 | return lend + 1; /* to LIST's end at which \n is not found */ |
| 3679 | } |
| 3680 | |
| 3681 | /* The max range value follows, separated by a BLANK */ |
| 3682 | if (isBLANK(*l)) { |
| 3683 | ++l; |
| 3684 | flags = PERL_SCAN_SILENT_ILLDIGIT |
| 3685 | | PERL_SCAN_DISALLOW_PREFIX |
| 3686 | | PERL_SCAN_SILENT_NON_PORTABLE; |
| 3687 | numlen = lend - l; |
| 3688 | *max = grok_hex((char *)l, &numlen, &flags, NULL); |
| 3689 | if (numlen) |
| 3690 | l += numlen; |
| 3691 | else /* If no value here, it is a single element range */ |
| 3692 | *max = *min; |
| 3693 | |
| 3694 | /* Non-binary tables have a third entry: what the first element of the |
| 3695 | * range maps to. The map for those currently read here is in hex */ |
| 3696 | if (wants_value) { |
| 3697 | if (isBLANK(*l)) { |
| 3698 | ++l; |
| 3699 | flags = PERL_SCAN_SILENT_ILLDIGIT |
| 3700 | | PERL_SCAN_DISALLOW_PREFIX |
| 3701 | | PERL_SCAN_SILENT_NON_PORTABLE; |
| 3702 | numlen = lend - l; |
| 3703 | *val = grok_hex((char *)l, &numlen, &flags, NULL); |
| 3704 | if (numlen) |
| 3705 | l += numlen; |
| 3706 | else |
| 3707 | *val = 0; |
| 3708 | } |
| 3709 | else { |
| 3710 | *val = 0; |
| 3711 | if (typeto) { |
| 3712 | /* diag_listed_as: To%s: illegal mapping '%s' */ |
| 3713 | Perl_croak(aTHX_ "%s: illegal mapping '%s'", |
| 3714 | typestr, l); |
| 3715 | } |
| 3716 | } |
| 3717 | } |
| 3718 | else |
| 3719 | *val = 0; /* bits == 1, then any val should be ignored */ |
| 3720 | } |
| 3721 | else { /* Nothing following range min, should be single element with no |
| 3722 | mapping expected */ |
| 3723 | if (wants_value) { |
| 3724 | *val = 0; |
| 3725 | if (typeto) { |
| 3726 | /* diag_listed_as: To%s: illegal mapping '%s' */ |
| 3727 | Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l); |
| 3728 | } |
| 3729 | } |
| 3730 | else |
| 3731 | *val = 0; /* bits == 1, then val should be ignored */ |
| 3732 | } |
| 3733 | |
| 3734 | /* Position to next line if any, or EOF */ |
| 3735 | if (nl) |
| 3736 | l = nl + 1; |
| 3737 | else |
| 3738 | l = lend; |
| 3739 | |
| 3740 | return l; |
| 3741 | } |
| 3742 | |
| 3743 | /* Note: |
| 3744 | * Returns a swatch (a bit vector string) for a code point sequence |
| 3745 | * that starts from the value C<start> and comprises the number C<span>. |
| 3746 | * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl). |
| 3747 | * Should be used via swash_fetch, which will cache the swatch in C<swash>. |
| 3748 | */ |
| 3749 | STATIC SV* |
| 3750 | S_swatch_get(pTHX_ SV* swash, UV start, UV span) |
| 3751 | { |
| 3752 | SV *swatch; |
| 3753 | U8 *l, *lend, *x, *xend, *s, *send; |
| 3754 | STRLEN lcur, xcur, scur; |
| 3755 | HV *const hv = MUTABLE_HV(SvRV(swash)); |
| 3756 | SV** const invlistsvp = hv_fetchs(hv, "V", FALSE); |
| 3757 | |
| 3758 | SV** listsvp = NULL; /* The string containing the main body of the table */ |
| 3759 | SV** extssvp = NULL; |
| 3760 | SV** invert_it_svp = NULL; |
| 3761 | U8* typestr = NULL; |
| 3762 | STRLEN bits; |
| 3763 | STRLEN octets; /* if bits == 1, then octets == 0 */ |
| 3764 | UV none; |
| 3765 | UV end = start + span; |
| 3766 | |
| 3767 | if (invlistsvp == NULL) { |
| 3768 | SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE); |
| 3769 | SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE); |
| 3770 | SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE); |
| 3771 | extssvp = hv_fetchs(hv, "EXTRAS", FALSE); |
| 3772 | listsvp = hv_fetchs(hv, "LIST", FALSE); |
| 3773 | invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); |
| 3774 | |
| 3775 | bits = SvUV(*bitssvp); |
| 3776 | none = SvUV(*nonesvp); |
| 3777 | typestr = (U8*)SvPV_nolen(*typesvp); |
| 3778 | } |
| 3779 | else { |
| 3780 | bits = 1; |
| 3781 | none = 0; |
| 3782 | } |
| 3783 | octets = bits >> 3; /* if bits == 1, then octets == 0 */ |
| 3784 | |
| 3785 | PERL_ARGS_ASSERT_SWATCH_GET; |
| 3786 | |
| 3787 | if (bits != 1 && bits != 8 && bits != 16 && bits != 32) { |
| 3788 | Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf, |
| 3789 | (UV)bits); |
| 3790 | } |
| 3791 | |
| 3792 | /* If overflowed, use the max possible */ |
| 3793 | if (end < start) { |
| 3794 | end = UV_MAX; |
| 3795 | span = end - start; |
| 3796 | } |
| 3797 | |
| 3798 | /* create and initialize $swatch */ |
| 3799 | scur = octets ? (span * octets) : (span + 7) / 8; |
| 3800 | swatch = newSV(scur); |
| 3801 | SvPOK_on(swatch); |
| 3802 | s = (U8*)SvPVX(swatch); |
| 3803 | if (octets && none) { |
| 3804 | const U8* const e = s + scur; |
| 3805 | while (s < e) { |
| 3806 | if (bits == 8) |
| 3807 | *s++ = (U8)(none & 0xff); |
| 3808 | else if (bits == 16) { |
| 3809 | *s++ = (U8)((none >> 8) & 0xff); |
| 3810 | *s++ = (U8)( none & 0xff); |
| 3811 | } |
| 3812 | else if (bits == 32) { |
| 3813 | *s++ = (U8)((none >> 24) & 0xff); |
| 3814 | *s++ = (U8)((none >> 16) & 0xff); |
| 3815 | *s++ = (U8)((none >> 8) & 0xff); |
| 3816 | *s++ = (U8)( none & 0xff); |
| 3817 | } |
| 3818 | } |
| 3819 | *s = '\0'; |
| 3820 | } |
| 3821 | else { |
| 3822 | (void)memzero((U8*)s, scur + 1); |
| 3823 | } |
| 3824 | SvCUR_set(swatch, scur); |
| 3825 | s = (U8*)SvPVX(swatch); |
| 3826 | |
| 3827 | if (invlistsvp) { /* If has an inversion list set up use that */ |
| 3828 | _invlist_populate_swatch(*invlistsvp, start, end, s); |
| 3829 | return swatch; |
| 3830 | } |
| 3831 | |
| 3832 | /* read $swash->{LIST} */ |
| 3833 | l = (U8*)SvPV(*listsvp, lcur); |
| 3834 | lend = l + lcur; |
| 3835 | while (l < lend) { |
| 3836 | UV min, max, val, upper; |
| 3837 | l = swash_scan_list_line(l, lend, &min, &max, &val, |
| 3838 | cBOOL(octets), typestr); |
| 3839 | if (l > lend) { |
| 3840 | break; |
| 3841 | } |
| 3842 | |
| 3843 | /* If looking for something beyond this range, go try the next one */ |
| 3844 | if (max < start) |
| 3845 | continue; |
| 3846 | |
| 3847 | /* <end> is generally 1 beyond where we want to set things, but at the |
| 3848 | * platform's infinity, where we can't go any higher, we want to |
| 3849 | * include the code point at <end> */ |
| 3850 | upper = (max < end) |
| 3851 | ? max |
| 3852 | : (max != UV_MAX || end != UV_MAX) |
| 3853 | ? end - 1 |
| 3854 | : end; |
| 3855 | |
| 3856 | if (octets) { |
| 3857 | UV key; |
| 3858 | if (min < start) { |
| 3859 | if (!none || val < none) { |
| 3860 | val += start - min; |
| 3861 | } |
| 3862 | min = start; |
| 3863 | } |
| 3864 | for (key = min; key <= upper; key++) { |
| 3865 | STRLEN offset; |
| 3866 | /* offset must be non-negative (start <= min <= key < end) */ |
| 3867 | offset = octets * (key - start); |
| 3868 | if (bits == 8) |
| 3869 | s[offset] = (U8)(val & 0xff); |
| 3870 | else if (bits == 16) { |
| 3871 | s[offset ] = (U8)((val >> 8) & 0xff); |
| 3872 | s[offset + 1] = (U8)( val & 0xff); |
| 3873 | } |
| 3874 | else if (bits == 32) { |
| 3875 | s[offset ] = (U8)((val >> 24) & 0xff); |
| 3876 | s[offset + 1] = (U8)((val >> 16) & 0xff); |
| 3877 | s[offset + 2] = (U8)((val >> 8) & 0xff); |
| 3878 | s[offset + 3] = (U8)( val & 0xff); |
| 3879 | } |
| 3880 | |
| 3881 | if (!none || val < none) |
| 3882 | ++val; |
| 3883 | } |
| 3884 | } |
| 3885 | else { /* bits == 1, then val should be ignored */ |
| 3886 | UV key; |
| 3887 | if (min < start) |
| 3888 | min = start; |
| 3889 | |
| 3890 | for (key = min; key <= upper; key++) { |
| 3891 | const STRLEN offset = (STRLEN)(key - start); |
| 3892 | s[offset >> 3] |= 1 << (offset & 7); |
| 3893 | } |
| 3894 | } |
| 3895 | } /* while */ |
| 3896 | |
| 3897 | /* Invert if the data says it should be. Assumes that bits == 1 */ |
| 3898 | if (invert_it_svp && SvUV(*invert_it_svp)) { |
| 3899 | |
| 3900 | /* Unicode properties should come with all bits above PERL_UNICODE_MAX |
| 3901 | * be 0, and their inversion should also be 0, as we don't succeed any |
| 3902 | * Unicode property matches for non-Unicode code points */ |
| 3903 | if (start <= PERL_UNICODE_MAX) { |
| 3904 | |
| 3905 | /* The code below assumes that we never cross the |
| 3906 | * Unicode/above-Unicode boundary in a range, as otherwise we would |
| 3907 | * have to figure out where to stop flipping the bits. Since this |
| 3908 | * boundary is divisible by a large power of 2, and swatches comes |
| 3909 | * in small powers of 2, this should be a valid assumption */ |
| 3910 | assert(start + span - 1 <= PERL_UNICODE_MAX); |
| 3911 | |
| 3912 | send = s + scur; |
| 3913 | while (s < send) { |
| 3914 | *s = ~(*s); |
| 3915 | s++; |
| 3916 | } |
| 3917 | } |
| 3918 | } |
| 3919 | |
| 3920 | /* read $swash->{EXTRAS} |
| 3921 | * This code also copied to swash_to_invlist() below */ |
| 3922 | x = (U8*)SvPV(*extssvp, xcur); |
| 3923 | xend = x + xcur; |
| 3924 | while (x < xend) { |
| 3925 | STRLEN namelen; |
| 3926 | U8 *namestr; |
| 3927 | SV** othersvp; |
| 3928 | HV* otherhv; |
| 3929 | STRLEN otherbits; |
| 3930 | SV **otherbitssvp, *other; |
| 3931 | U8 *s, *o, *nl; |
| 3932 | STRLEN slen, olen; |
| 3933 | |
| 3934 | const U8 opc = *x++; |
| 3935 | if (opc == '\n') |
| 3936 | continue; |
| 3937 | |
| 3938 | nl = (U8*)memchr(x, '\n', xend - x); |
| 3939 | |
| 3940 | if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { |
| 3941 | if (nl) { |
| 3942 | x = nl + 1; /* 1 is length of "\n" */ |
| 3943 | continue; |
| 3944 | } |
| 3945 | else { |
| 3946 | x = xend; /* to EXTRAS' end at which \n is not found */ |
| 3947 | break; |
| 3948 | } |
| 3949 | } |
| 3950 | |
| 3951 | namestr = x; |
| 3952 | if (nl) { |
| 3953 | namelen = nl - namestr; |
| 3954 | x = nl + 1; |
| 3955 | } |
| 3956 | else { |
| 3957 | namelen = xend - namestr; |
| 3958 | x = xend; |
| 3959 | } |
| 3960 | |
| 3961 | othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); |
| 3962 | otherhv = MUTABLE_HV(SvRV(*othersvp)); |
| 3963 | otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); |
| 3964 | otherbits = (STRLEN)SvUV(*otherbitssvp); |
| 3965 | if (bits < otherbits) |
| 3966 | Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, " |
| 3967 | "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits); |
| 3968 | |
| 3969 | /* The "other" swatch must be destroyed after. */ |
| 3970 | other = swatch_get(*othersvp, start, span); |
| 3971 | o = (U8*)SvPV(other, olen); |
| 3972 | |
| 3973 | if (!olen) |
| 3974 | Perl_croak(aTHX_ "panic: swatch_get got improper swatch"); |
| 3975 | |
| 3976 | s = (U8*)SvPV(swatch, slen); |
| 3977 | if (bits == 1 && otherbits == 1) { |
| 3978 | if (slen != olen) |
| 3979 | Perl_croak(aTHX_ "panic: swatch_get found swatch length " |
| 3980 | "mismatch, slen=%"UVuf", olen=%"UVuf, |
| 3981 | (UV)slen, (UV)olen); |
| 3982 | |
| 3983 | switch (opc) { |
| 3984 | case '+': |
| 3985 | while (slen--) |
| 3986 | *s++ |= *o++; |
| 3987 | break; |
| 3988 | case '!': |
| 3989 | while (slen--) |
| 3990 | *s++ |= ~*o++; |
| 3991 | break; |
| 3992 | case '-': |
| 3993 | while (slen--) |
| 3994 | *s++ &= ~*o++; |
| 3995 | break; |
| 3996 | case '&': |
| 3997 | while (slen--) |
| 3998 | *s++ &= *o++; |
| 3999 | break; |
| 4000 | default: |
| 4001 | break; |
| 4002 | } |
| 4003 | } |
| 4004 | else { |
| 4005 | STRLEN otheroctets = otherbits >> 3; |
| 4006 | STRLEN offset = 0; |
| 4007 | U8* const send = s + slen; |
| 4008 | |
| 4009 | while (s < send) { |
| 4010 | UV otherval = 0; |
| 4011 | |
| 4012 | if (otherbits == 1) { |
| 4013 | otherval = (o[offset >> 3] >> (offset & 7)) & 1; |
| 4014 | ++offset; |
| 4015 | } |
| 4016 | else { |
| 4017 | STRLEN vlen = otheroctets; |
| 4018 | otherval = *o++; |
| 4019 | while (--vlen) { |
| 4020 | otherval <<= 8; |
| 4021 | otherval |= *o++; |
| 4022 | } |
| 4023 | } |
| 4024 | |
| 4025 | if (opc == '+' && otherval) |
| 4026 | NOOP; /* replace with otherval */ |
| 4027 | else if (opc == '!' && !otherval) |
| 4028 | otherval = 1; |
| 4029 | else if (opc == '-' && otherval) |
| 4030 | otherval = 0; |
| 4031 | else if (opc == '&' && !otherval) |
| 4032 | otherval = 0; |
| 4033 | else { |
| 4034 | s += octets; /* no replacement */ |
| 4035 | continue; |
| 4036 | } |
| 4037 | |
| 4038 | if (bits == 8) |
| 4039 | *s++ = (U8)( otherval & 0xff); |
| 4040 | else if (bits == 16) { |
| 4041 | *s++ = (U8)((otherval >> 8) & 0xff); |
| 4042 | *s++ = (U8)( otherval & 0xff); |
| 4043 | } |
| 4044 | else if (bits == 32) { |
| 4045 | *s++ = (U8)((otherval >> 24) & 0xff); |
| 4046 | *s++ = (U8)((otherval >> 16) & 0xff); |
| 4047 | *s++ = (U8)((otherval >> 8) & 0xff); |
| 4048 | *s++ = (U8)( otherval & 0xff); |
| 4049 | } |
| 4050 | } |
| 4051 | } |
| 4052 | sv_free(other); /* through with it! */ |
| 4053 | } /* while */ |
| 4054 | return swatch; |
| 4055 | } |
| 4056 | |
| 4057 | HV* |
| 4058 | Perl__swash_inversion_hash(pTHX_ SV* const swash) |
| 4059 | { |
| 4060 | |
| 4061 | /* Subject to change or removal. For use only in regcomp.c and regexec.c |
| 4062 | * Can't be used on a property that is subject to user override, as it |
| 4063 | * relies on the value of SPECIALS in the swash which would be set by |
| 4064 | * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set |
| 4065 | * for overridden properties |
| 4066 | * |
| 4067 | * Returns a hash which is the inversion and closure of a swash mapping. |
| 4068 | * For example, consider the input lines: |
| 4069 | * 004B 006B |
| 4070 | * 004C 006C |
| 4071 | * 212A 006B |
| 4072 | * |
| 4073 | * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for |
| 4074 | * 006C. The value for each key is an array. For 006C, the array would |
| 4075 | * have two elements, the UTF-8 for itself, and for 004C. For 006B, there |
| 4076 | * would be three elements in its array, the UTF-8 for 006B, 004B and 212A. |
| 4077 | * |
| 4078 | * Note that there are no elements in the hash for 004B, 004C, 212A. The |
| 4079 | * keys are only code points that are folded-to, so it isn't a full closure. |
| 4080 | * |
| 4081 | * Essentially, for any code point, it gives all the code points that map to |
| 4082 | * it, or the list of 'froms' for that point. |
| 4083 | * |
| 4084 | * Currently it ignores any additions or deletions from other swashes, |
| 4085 | * looking at just the main body of the swash, and if there are SPECIALS |
| 4086 | * in the swash, at that hash |
| 4087 | * |
| 4088 | * The specials hash can be extra code points, and most likely consists of |
| 4089 | * maps from single code points to multiple ones (each expressed as a string |
| 4090 | * of UTF-8 characters). This function currently returns only 1-1 mappings. |
| 4091 | * However consider this possible input in the specials hash: |
| 4092 | * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074 |
| 4093 | * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074 |
| 4094 | * |
| 4095 | * Both FB05 and FB06 map to the same multi-char sequence, which we don't |
| 4096 | * currently handle. But it also means that FB05 and FB06 are equivalent in |
| 4097 | * a 1-1 mapping which we should handle, and this relationship may not be in |
| 4098 | * the main table. Therefore this function examines all the multi-char |
| 4099 | * sequences and adds the 1-1 mappings that come out of that. |
| 4100 | * |
| 4101 | * XXX This function was originally intended to be multipurpose, but its |
| 4102 | * only use is quite likely to remain for constructing the inversion of |
| 4103 | * the CaseFolding (//i) property. If it were more general purpose for |
| 4104 | * regex patterns, it would have to do the FB05/FB06 game for simple folds, |
| 4105 | * because certain folds are prohibited under /iaa and /il. As an example, |
| 4106 | * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both |
| 4107 | * equivalent under /i. But under /iaa and /il, the folds to 'i' are |
| 4108 | * prohibited, so we would not figure out that they fold to each other. |
| 4109 | * Code could be written to automatically figure this out, similar to the |
| 4110 | * code that does this for multi-character folds, but this is the only case |
| 4111 | * where something like this is ever likely to happen, as all the single |
| 4112 | * char folds to the 0-255 range are now quite settled. Instead there is a |
| 4113 | * little special code that is compiled only for this Unicode version. This |
| 4114 | * is smaller and didn't require much coding time to do. But this makes |
| 4115 | * this routine strongly tied to being used just for CaseFolding. If ever |
| 4116 | * it should be generalized, this would have to be fixed */ |
| 4117 | |
| 4118 | U8 *l, *lend; |
| 4119 | STRLEN lcur; |
| 4120 | HV *const hv = MUTABLE_HV(SvRV(swash)); |
| 4121 | |
| 4122 | /* The string containing the main body of the table. This will have its |
| 4123 | * assertion fail if the swash has been converted to its inversion list */ |
| 4124 | SV** const listsvp = hv_fetchs(hv, "LIST", FALSE); |
| 4125 | |
| 4126 | SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE); |
| 4127 | SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE); |
| 4128 | SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE); |
| 4129 | /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/ |
| 4130 | const U8* const typestr = (U8*)SvPV_nolen(*typesvp); |
| 4131 | const STRLEN bits = SvUV(*bitssvp); |
| 4132 | const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */ |
| 4133 | const UV none = SvUV(*nonesvp); |
| 4134 | SV **specials_p = hv_fetchs(hv, "SPECIALS", 0); |
| 4135 | |
| 4136 | HV* ret = newHV(); |
| 4137 | |
| 4138 | PERL_ARGS_ASSERT__SWASH_INVERSION_HASH; |
| 4139 | |
| 4140 | /* Must have at least 8 bits to get the mappings */ |
| 4141 | if (bits != 8 && bits != 16 && bits != 32) { |
| 4142 | Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf, |
| 4143 | (UV)bits); |
| 4144 | } |
| 4145 | |
| 4146 | if (specials_p) { /* It might be "special" (sometimes, but not always, a |
| 4147 | mapping to more than one character */ |
| 4148 | |
| 4149 | /* Construct an inverse mapping hash for the specials */ |
| 4150 | HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p)); |
| 4151 | HV * specials_inverse = newHV(); |
| 4152 | char *char_from; /* the lhs of the map */ |
| 4153 | I32 from_len; /* its byte length */ |
| 4154 | char *char_to; /* the rhs of the map */ |
| 4155 | I32 to_len; /* its byte length */ |
| 4156 | SV *sv_to; /* and in a sv */ |
| 4157 | AV* from_list; /* list of things that map to each 'to' */ |
| 4158 | |
| 4159 | hv_iterinit(specials_hv); |
| 4160 | |
| 4161 | /* The keys are the characters (in UTF-8) that map to the corresponding |
| 4162 | * UTF-8 string value. Iterate through the list creating the inverse |
| 4163 | * list. */ |
| 4164 | while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) { |
| 4165 | SV** listp; |
| 4166 | if (! SvPOK(sv_to)) { |
| 4167 | Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() " |
| 4168 | "unexpectedly is not a string, flags=%lu", |
| 4169 | (unsigned long)SvFLAGS(sv_to)); |
| 4170 | } |
| 4171 | /*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)));*/ |
| 4172 | |
| 4173 | /* Each key in the inverse list is a mapped-to value, and the key's |
| 4174 | * hash value is a list of the strings (each in UTF-8) that map to |
| 4175 | * it. Those strings are all one character long */ |
| 4176 | if ((listp = hv_fetch(specials_inverse, |
| 4177 | SvPVX(sv_to), |
| 4178 | SvCUR(sv_to), 0))) |
| 4179 | { |
| 4180 | from_list = (AV*) *listp; |
| 4181 | } |
| 4182 | else { /* No entry yet for it: create one */ |
| 4183 | from_list = newAV(); |
| 4184 | if (! hv_store(specials_inverse, |
| 4185 | SvPVX(sv_to), |
| 4186 | SvCUR(sv_to), |
| 4187 | (SV*) from_list, 0)) |
| 4188 | { |
| 4189 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); |
| 4190 | } |
| 4191 | } |
| 4192 | |
| 4193 | /* Here have the list associated with this 'to' (perhaps newly |
| 4194 | * created and empty). Just add to it. Note that we ASSUME that |
| 4195 | * the input is guaranteed to not have duplications, so we don't |
| 4196 | * check for that. Duplications just slow down execution time. */ |
| 4197 | av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE)); |
| 4198 | } |
| 4199 | |
| 4200 | /* Here, 'specials_inverse' contains the inverse mapping. Go through |
| 4201 | * it looking for cases like the FB05/FB06 examples above. There would |
| 4202 | * be an entry in the hash like |
| 4203 | * 'st' => [ FB05, FB06 ] |
| 4204 | * In this example we will create two lists that get stored in the |
| 4205 | * returned hash, 'ret': |
| 4206 | * FB05 => [ FB05, FB06 ] |
| 4207 | * FB06 => [ FB05, FB06 ] |
| 4208 | * |
| 4209 | * Note that there is nothing to do if the array only has one element. |
| 4210 | * (In the normal 1-1 case handled below, we don't have to worry about |
| 4211 | * two lists, as everything gets tied to the single list that is |
| 4212 | * generated for the single character 'to'. But here, we are omitting |
| 4213 | * that list, ('st' in the example), so must have multiple lists.) */ |
| 4214 | while ((from_list = (AV *) hv_iternextsv(specials_inverse, |
| 4215 | &char_to, &to_len))) |
| 4216 | { |
| 4217 | if (av_tindex_nomg(from_list) > 0) { |
| 4218 | SSize_t i; |
| 4219 | |
| 4220 | /* We iterate over all combinations of i,j to place each code |
| 4221 | * point on each list */ |
| 4222 | for (i = 0; i <= av_tindex_nomg(from_list); i++) { |
| 4223 | SSize_t j; |
| 4224 | AV* i_list = newAV(); |
| 4225 | SV** entryp = av_fetch(from_list, i, FALSE); |
| 4226 | if (entryp == NULL) { |
| 4227 | Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed"); |
| 4228 | } |
| 4229 | if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) { |
| 4230 | Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp)); |
| 4231 | } |
| 4232 | if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp), |
| 4233 | (SV*) i_list, FALSE)) |
| 4234 | { |
| 4235 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); |
| 4236 | } |
| 4237 | |
| 4238 | /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/ |
| 4239 | for (j = 0; j <= av_tindex_nomg(from_list); j++) { |
| 4240 | entryp = av_fetch(from_list, j, FALSE); |
| 4241 | if (entryp == NULL) { |
| 4242 | Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed"); |
| 4243 | } |
| 4244 | |
| 4245 | /* When i==j this adds itself to the list */ |
| 4246 | av_push(i_list, newSVuv(utf8_to_uvchr_buf( |
| 4247 | (U8*) SvPVX(*entryp), |
| 4248 | (U8*) SvPVX(*entryp) + SvCUR(*entryp), |
| 4249 | 0))); |
| 4250 | /*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));*/ |
| 4251 | } |
| 4252 | } |
| 4253 | } |
| 4254 | } |
| 4255 | SvREFCNT_dec(specials_inverse); /* done with it */ |
| 4256 | } /* End of specials */ |
| 4257 | |
| 4258 | /* read $swash->{LIST} */ |
| 4259 | |
| 4260 | #if UNICODE_MAJOR_VERSION == 3 \ |
| 4261 | && UNICODE_DOT_VERSION == 0 \ |
| 4262 | && UNICODE_DOT_DOT_VERSION == 1 |
| 4263 | |
| 4264 | /* For this version only U+130 and U+131 are equivalent under qr//i. Add a |
| 4265 | * rule so that things work under /iaa and /il */ |
| 4266 | |
| 4267 | SV * mod_listsv = sv_mortalcopy(*listsvp); |
| 4268 | sv_catpv(mod_listsv, "130\t130\t131\n"); |
| 4269 | l = (U8*)SvPV(mod_listsv, lcur); |
| 4270 | |
| 4271 | #else |
| 4272 | |
| 4273 | l = (U8*)SvPV(*listsvp, lcur); |
| 4274 | |
| 4275 | #endif |
| 4276 | |
| 4277 | lend = l + lcur; |
| 4278 | |
| 4279 | /* Go through each input line */ |
| 4280 | while (l < lend) { |
| 4281 | UV min, max, val; |
| 4282 | UV inverse; |
| 4283 | l = swash_scan_list_line(l, lend, &min, &max, &val, |
| 4284 | cBOOL(octets), typestr); |
| 4285 | if (l > lend) { |
| 4286 | break; |
| 4287 | } |
| 4288 | |
| 4289 | /* Each element in the range is to be inverted */ |
| 4290 | for (inverse = min; inverse <= max; inverse++) { |
| 4291 | AV* list; |
| 4292 | SV** listp; |
| 4293 | IV i; |
| 4294 | bool found_key = FALSE; |
| 4295 | bool found_inverse = FALSE; |
| 4296 | |
| 4297 | /* The key is the inverse mapping */ |
| 4298 | char key[UTF8_MAXBYTES+1]; |
| 4299 | char* key_end = (char *) uvchr_to_utf8((U8*) key, val); |
| 4300 | STRLEN key_len = key_end - key; |
| 4301 | |
| 4302 | /* Get the list for the map */ |
| 4303 | if ((listp = hv_fetch(ret, key, key_len, FALSE))) { |
| 4304 | list = (AV*) *listp; |
| 4305 | } |
| 4306 | else { /* No entry yet for it: create one */ |
| 4307 | list = newAV(); |
| 4308 | if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) { |
| 4309 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); |
| 4310 | } |
| 4311 | } |
| 4312 | |
| 4313 | /* Look through list to see if this inverse mapping already is |
| 4314 | * listed, or if there is a mapping to itself already */ |
| 4315 | for (i = 0; i <= av_tindex_nomg(list); i++) { |
| 4316 | SV** entryp = av_fetch(list, i, FALSE); |
| 4317 | SV* entry; |
| 4318 | UV uv; |
| 4319 | if (entryp == NULL) { |
| 4320 | Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed"); |
| 4321 | } |
| 4322 | entry = *entryp; |
| 4323 | uv = SvUV(entry); |
| 4324 | /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/ |
| 4325 | if (uv == val) { |
| 4326 | found_key = TRUE; |
| 4327 | } |
| 4328 | if (uv == inverse) { |
| 4329 | found_inverse = TRUE; |
| 4330 | } |
| 4331 | |
| 4332 | /* No need to continue searching if found everything we are |
| 4333 | * looking for */ |
| 4334 | if (found_key && found_inverse) { |
| 4335 | break; |
| 4336 | } |
| 4337 | } |
| 4338 | |
| 4339 | /* Make sure there is a mapping to itself on the list */ |
| 4340 | if (! found_key) { |
| 4341 | av_push(list, newSVuv(val)); |
| 4342 | /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/ |
| 4343 | } |
| 4344 | |
| 4345 | |
| 4346 | /* Simply add the value to the list */ |
| 4347 | if (! found_inverse) { |
| 4348 | av_push(list, newSVuv(inverse)); |
| 4349 | /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/ |
| 4350 | } |
| 4351 | |
| 4352 | /* swatch_get() increments the value of val for each element in the |
| 4353 | * range. That makes more compact tables possible. You can |
| 4354 | * express the capitalization, for example, of all consecutive |
| 4355 | * letters with a single line: 0061\t007A\t0041 This maps 0061 to |
| 4356 | * 0041, 0062 to 0042, etc. I (khw) have never understood 'none', |
| 4357 | * and it's not documented; it appears to be used only in |
| 4358 | * implementing tr//; I copied the semantics from swatch_get(), just |
| 4359 | * in case */ |
| 4360 | if (!none || val < none) { |
| 4361 | ++val; |
| 4362 | } |
| 4363 | } |
| 4364 | } |
| 4365 | |
| 4366 | return ret; |
| 4367 | } |
| 4368 | |
| 4369 | SV* |
| 4370 | Perl__swash_to_invlist(pTHX_ SV* const swash) |
| 4371 | { |
| 4372 | |
| 4373 | /* Subject to change or removal. For use only in one place in regcomp.c. |
| 4374 | * Ownership is given to one reference count in the returned SV* */ |
| 4375 | |
| 4376 | U8 *l, *lend; |
| 4377 | char *loc; |
| 4378 | STRLEN lcur; |
| 4379 | HV *const hv = MUTABLE_HV(SvRV(swash)); |
| 4380 | UV elements = 0; /* Number of elements in the inversion list */ |
| 4381 | U8 empty[] = ""; |
| 4382 | SV** listsvp; |
| 4383 | SV** typesvp; |
| 4384 | SV** bitssvp; |
| 4385 | SV** extssvp; |
| 4386 | SV** invert_it_svp; |
| 4387 | |
| 4388 | U8* typestr; |
| 4389 | STRLEN bits; |
| 4390 | STRLEN octets; /* if bits == 1, then octets == 0 */ |
| 4391 | U8 *x, *xend; |
| 4392 | STRLEN xcur; |
| 4393 | |
| 4394 | SV* invlist; |
| 4395 | |
| 4396 | PERL_ARGS_ASSERT__SWASH_TO_INVLIST; |
| 4397 | |
| 4398 | /* If not a hash, it must be the swash's inversion list instead */ |
| 4399 | if (SvTYPE(hv) != SVt_PVHV) { |
| 4400 | return SvREFCNT_inc_simple_NN((SV*) hv); |
| 4401 | } |
| 4402 | |
| 4403 | /* The string containing the main body of the table */ |
| 4404 | listsvp = hv_fetchs(hv, "LIST", FALSE); |
| 4405 | typesvp = hv_fetchs(hv, "TYPE", FALSE); |
| 4406 | bitssvp = hv_fetchs(hv, "BITS", FALSE); |
| 4407 | extssvp = hv_fetchs(hv, "EXTRAS", FALSE); |
| 4408 | invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE); |
| 4409 | |
| 4410 | typestr = (U8*)SvPV_nolen(*typesvp); |
| 4411 | bits = SvUV(*bitssvp); |
| 4412 | octets = bits >> 3; /* if bits == 1, then octets == 0 */ |
| 4413 | |
| 4414 | /* read $swash->{LIST} */ |
| 4415 | if (SvPOK(*listsvp)) { |
| 4416 | l = (U8*)SvPV(*listsvp, lcur); |
| 4417 | } |
| 4418 | else { |
| 4419 | /* LIST legitimately doesn't contain a string during compilation phases |
| 4420 | * of Perl itself, before the Unicode tables are generated. In this |
| 4421 | * case, just fake things up by creating an empty list */ |
| 4422 | l = empty; |
| 4423 | lcur = 0; |
| 4424 | } |
| 4425 | loc = (char *) l; |
| 4426 | lend = l + lcur; |
| 4427 | |
| 4428 | if (*l == 'V') { /* Inversion list format */ |
| 4429 | const char *after_atou = (char *) lend; |
| 4430 | UV element0; |
| 4431 | UV* other_elements_ptr; |
| 4432 | |
| 4433 | /* The first number is a count of the rest */ |
| 4434 | l++; |
| 4435 | if (!grok_atoUV((const char *)l, &elements, &after_atou)) { |
| 4436 | Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list"); |
| 4437 | } |
| 4438 | if (elements == 0) { |
| 4439 | invlist = _new_invlist(0); |
| 4440 | } |
| 4441 | else { |
| 4442 | while (isSPACE(*l)) l++; |
| 4443 | l = (U8 *) after_atou; |
| 4444 | |
| 4445 | /* Get the 0th element, which is needed to setup the inversion list */ |
| 4446 | while (isSPACE(*l)) l++; |
| 4447 | if (!grok_atoUV((const char *)l, &element0, &after_atou)) { |
| 4448 | Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list"); |
| 4449 | } |
| 4450 | l = (U8 *) after_atou; |
| 4451 | invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr); |
| 4452 | elements--; |
| 4453 | |
| 4454 | /* Then just populate the rest of the input */ |
| 4455 | while (elements-- > 0) { |
| 4456 | if (l > lend) { |
| 4457 | Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements); |
| 4458 | } |
| 4459 | while (isSPACE(*l)) l++; |
| 4460 | if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) { |
| 4461 | Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list"); |
| 4462 | } |
| 4463 | l = (U8 *) after_atou; |
| 4464 | } |
| 4465 | } |
| 4466 | } |
| 4467 | else { |
| 4468 | |
| 4469 | /* Scan the input to count the number of lines to preallocate array |
| 4470 | * size based on worst possible case, which is each line in the input |
| 4471 | * creates 2 elements in the inversion list: 1) the beginning of a |
| 4472 | * range in the list; 2) the beginning of a range not in the list. */ |
| 4473 | while ((loc = (strchr(loc, '\n'))) != NULL) { |
| 4474 | elements += 2; |
| 4475 | loc++; |
| 4476 | } |
| 4477 | |
| 4478 | /* If the ending is somehow corrupt and isn't a new line, add another |
| 4479 | * element for the final range that isn't in the inversion list */ |
| 4480 | if (! (*lend == '\n' |
| 4481 | || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n')))) |
| 4482 | { |
| 4483 | elements++; |
| 4484 | } |
| 4485 | |
| 4486 | invlist = _new_invlist(elements); |
| 4487 | |
| 4488 | /* Now go through the input again, adding each range to the list */ |
| 4489 | while (l < lend) { |
| 4490 | UV start, end; |
| 4491 | UV val; /* Not used by this function */ |
| 4492 | |
| 4493 | l = swash_scan_list_line(l, lend, &start, &end, &val, |
| 4494 | cBOOL(octets), typestr); |
| 4495 | |
| 4496 | if (l > lend) { |
| 4497 | break; |
| 4498 | } |
| 4499 | |
| 4500 | invlist = _add_range_to_invlist(invlist, start, end); |
| 4501 | } |
| 4502 | } |
| 4503 | |
| 4504 | /* Invert if the data says it should be */ |
| 4505 | if (invert_it_svp && SvUV(*invert_it_svp)) { |
| 4506 | _invlist_invert(invlist); |
| 4507 | } |
| 4508 | |
| 4509 | /* This code is copied from swatch_get() |
| 4510 | * read $swash->{EXTRAS} */ |
| 4511 | x = (U8*)SvPV(*extssvp, xcur); |
| 4512 | xend = x + xcur; |
| 4513 | while (x < xend) { |
| 4514 | STRLEN namelen; |
| 4515 | U8 *namestr; |
| 4516 | SV** othersvp; |
| 4517 | HV* otherhv; |
| 4518 | STRLEN otherbits; |
| 4519 | SV **otherbitssvp, *other; |
| 4520 | U8 *nl; |
| 4521 | |
| 4522 | const U8 opc = *x++; |
| 4523 | if (opc == '\n') |
| 4524 | continue; |
| 4525 | |
| 4526 | nl = (U8*)memchr(x, '\n', xend - x); |
| 4527 | |
| 4528 | if (opc != '-' && opc != '+' && opc != '!' && opc != '&') { |
| 4529 | if (nl) { |
| 4530 | x = nl + 1; /* 1 is length of "\n" */ |
| 4531 | continue; |
| 4532 | } |
| 4533 | else { |
| 4534 | x = xend; /* to EXTRAS' end at which \n is not found */ |
| 4535 | break; |
| 4536 | } |
| 4537 | } |
| 4538 | |
| 4539 | namestr = x; |
| 4540 | if (nl) { |
| 4541 | namelen = nl - namestr; |
| 4542 | x = nl + 1; |
| 4543 | } |
| 4544 | else { |
| 4545 | namelen = xend - namestr; |
| 4546 | x = xend; |
| 4547 | } |
| 4548 | |
| 4549 | othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE); |
| 4550 | otherhv = MUTABLE_HV(SvRV(*othersvp)); |
| 4551 | otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE); |
| 4552 | otherbits = (STRLEN)SvUV(*otherbitssvp); |
| 4553 | |
| 4554 | if (bits != otherbits || bits != 1) { |
| 4555 | Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean " |
| 4556 | "properties, bits=%"UVuf", otherbits=%"UVuf, |
| 4557 | (UV)bits, (UV)otherbits); |
| 4558 | } |
| 4559 | |
| 4560 | /* The "other" swatch must be destroyed after. */ |
| 4561 | other = _swash_to_invlist((SV *)*othersvp); |
| 4562 | |
| 4563 | /* End of code copied from swatch_get() */ |
| 4564 | switch (opc) { |
| 4565 | case '+': |
| 4566 | _invlist_union(invlist, other, &invlist); |
| 4567 | break; |
| 4568 | case '!': |
| 4569 | _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist); |
| 4570 | break; |
| 4571 | case '-': |
| 4572 | _invlist_subtract(invlist, other, &invlist); |
| 4573 | break; |
| 4574 | case '&': |
| 4575 | _invlist_intersection(invlist, other, &invlist); |
| 4576 | break; |
| 4577 | default: |
| 4578 | break; |
| 4579 | } |
| 4580 | sv_free(other); /* through with it! */ |
| 4581 | } |
| 4582 | |
| 4583 | SvREADONLY_on(invlist); |
| 4584 | return invlist; |
| 4585 | } |
| 4586 | |
| 4587 | SV* |
| 4588 | Perl__get_swash_invlist(pTHX_ SV* const swash) |
| 4589 | { |
| 4590 | SV** ptr; |
| 4591 | |
| 4592 | PERL_ARGS_ASSERT__GET_SWASH_INVLIST; |
| 4593 | |
| 4594 | if (! SvROK(swash)) { |
| 4595 | return NULL; |
| 4596 | } |
| 4597 | |
| 4598 | /* If it really isn't a hash, it isn't really swash; must be an inversion |
| 4599 | * list */ |
| 4600 | if (SvTYPE(SvRV(swash)) != SVt_PVHV) { |
| 4601 | return SvRV(swash); |
| 4602 | } |
| 4603 | |
| 4604 | ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE); |
| 4605 | if (! ptr) { |
| 4606 | return NULL; |
| 4607 | } |
| 4608 | |
| 4609 | return *ptr; |
| 4610 | } |
| 4611 | |
| 4612 | bool |
| 4613 | Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len) |
| 4614 | { |
| 4615 | /* May change: warns if surrogates, non-character code points, or |
| 4616 | * non-Unicode code points are in s which has length len bytes. Returns |
| 4617 | * TRUE if none found; FALSE otherwise. The only other validity check is |
| 4618 | * to make sure that this won't exceed the string's length. |
| 4619 | * |
| 4620 | * Code points above the platform's C<IV_MAX> will raise a deprecation |
| 4621 | * warning, unless those are turned off. */ |
| 4622 | |
| 4623 | const U8* const e = s + len; |
| 4624 | bool ok = TRUE; |
| 4625 | |
| 4626 | PERL_ARGS_ASSERT_CHECK_UTF8_PRINT; |
| 4627 | |
| 4628 | while (s < e) { |
| 4629 | if (UTF8SKIP(s) > len) { |
| 4630 | Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), |
| 4631 | "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print"); |
| 4632 | return FALSE; |
| 4633 | } |
| 4634 | if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) { |
| 4635 | STRLEN char_len; |
| 4636 | if (UNLIKELY(UTF8_IS_SUPER(s, e))) { |
| 4637 | if ( ckWARN_d(WARN_NON_UNICODE) |
| 4638 | || ( ckWARN_d(WARN_DEPRECATED) |
| 4639 | #ifndef UV_IS_QUAD |
| 4640 | && UNLIKELY(is_utf8_cp_above_31_bits(s, e)) |
| 4641 | #else /* Below is 64-bit words */ |
| 4642 | /* 2**63 and up meet these conditions provided we have |
| 4643 | * a 64-bit word. */ |
| 4644 | # ifdef EBCDIC |
| 4645 | && *s == 0xFE |
| 4646 | && NATIVE_UTF8_TO_I8(s[1]) >= 0xA8 |
| 4647 | # else |
| 4648 | && *s == 0xFF |
| 4649 | /* s[1] being above 0x80 overflows */ |
| 4650 | && s[2] >= 0x88 |
| 4651 | # endif |
| 4652 | #endif |
| 4653 | )) { |
| 4654 | /* A side effect of this function will be to warn */ |
| 4655 | (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_SUPER); |
| 4656 | ok = FALSE; |
| 4657 | } |
| 4658 | } |
| 4659 | else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) { |
| 4660 | if (ckWARN_d(WARN_SURROGATE)) { |
| 4661 | /* This has a different warning than the one the called |
| 4662 | * function would output, so can't just call it, unlike we |
| 4663 | * do for the non-chars and above-unicodes */ |
| 4664 | UV uv = utf8_to_uvchr_buf(s, e, &char_len); |
| 4665 | Perl_warner(aTHX_ packWARN(WARN_SURROGATE), |
| 4666 | "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv); |
| 4667 | ok = FALSE; |
| 4668 | } |
| 4669 | } |
| 4670 | else if (UNLIKELY(UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) { |
| 4671 | /* A side effect of this function will be to warn */ |
| 4672 | (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_NONCHAR); |
| 4673 | ok = FALSE; |
| 4674 | } |
| 4675 | } |
| 4676 | s += UTF8SKIP(s); |
| 4677 | } |
| 4678 | |
| 4679 | return ok; |
| 4680 | } |
| 4681 | |
| 4682 | /* |
| 4683 | =for apidoc pv_uni_display |
| 4684 | |
| 4685 | Build to the scalar C<dsv> a displayable version of the string C<spv>, |
| 4686 | length C<len>, the displayable version being at most C<pvlim> bytes long |
| 4687 | (if longer, the rest is truncated and C<"..."> will be appended). |
| 4688 | |
| 4689 | The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display |
| 4690 | C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH> |
| 4691 | to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">) |
| 4692 | (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">). |
| 4693 | C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both |
| 4694 | C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on. |
| 4695 | |
| 4696 | The pointer to the PV of the C<dsv> is returned. |
| 4697 | |
| 4698 | See also L</sv_uni_display>. |
| 4699 | |
| 4700 | =cut */ |
| 4701 | char * |
| 4702 | Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags) |
| 4703 | { |
| 4704 | int truncated = 0; |
| 4705 | const char *s, *e; |
| 4706 | |
| 4707 | PERL_ARGS_ASSERT_PV_UNI_DISPLAY; |
| 4708 | |
| 4709 | SvPVCLEAR(dsv); |
| 4710 | SvUTF8_off(dsv); |
| 4711 | for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) { |
| 4712 | UV u; |
| 4713 | /* This serves double duty as a flag and a character to print after |
| 4714 | a \ when flags & UNI_DISPLAY_BACKSLASH is true. |
| 4715 | */ |
| 4716 | char ok = 0; |
| 4717 | |
| 4718 | if (pvlim && SvCUR(dsv) >= pvlim) { |
| 4719 | truncated++; |
| 4720 | break; |
| 4721 | } |
| 4722 | u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0); |
| 4723 | if (u < 256) { |
| 4724 | const unsigned char c = (unsigned char)u & 0xFF; |
| 4725 | if (flags & UNI_DISPLAY_BACKSLASH) { |
| 4726 | switch (c) { |
| 4727 | case '\n': |
| 4728 | ok = 'n'; break; |
| 4729 | case '\r': |
| 4730 | ok = 'r'; break; |
| 4731 | case '\t': |
| 4732 | ok = 't'; break; |
| 4733 | case '\f': |
| 4734 | ok = 'f'; break; |
| 4735 | case '\a': |
| 4736 | ok = 'a'; break; |
| 4737 | case '\\': |
| 4738 | ok = '\\'; break; |
| 4739 | default: break; |
| 4740 | } |
| 4741 | if (ok) { |
| 4742 | const char string = ok; |
| 4743 | sv_catpvs(dsv, "\\"); |
| 4744 | sv_catpvn(dsv, &string, 1); |
| 4745 | } |
| 4746 | } |
| 4747 | /* isPRINT() is the locale-blind version. */ |
| 4748 | if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) { |
| 4749 | const char string = c; |
| 4750 | sv_catpvn(dsv, &string, 1); |
| 4751 | ok = 1; |
| 4752 | } |
| 4753 | } |
| 4754 | if (!ok) |
| 4755 | Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u); |
| 4756 | } |
| 4757 | if (truncated) |
| 4758 | sv_catpvs(dsv, "..."); |
| 4759 | |
| 4760 | return SvPVX(dsv); |
| 4761 | } |
| 4762 | |
| 4763 | /* |
| 4764 | =for apidoc sv_uni_display |
| 4765 | |
| 4766 | Build to the scalar C<dsv> a displayable version of the scalar C<sv>, |
| 4767 | the displayable version being at most C<pvlim> bytes long |
| 4768 | (if longer, the rest is truncated and "..." will be appended). |
| 4769 | |
| 4770 | The C<flags> argument is as in L</pv_uni_display>(). |
| 4771 | |
| 4772 | The pointer to the PV of the C<dsv> is returned. |
| 4773 | |
| 4774 | =cut |
| 4775 | */ |
| 4776 | char * |
| 4777 | Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags) |
| 4778 | { |
| 4779 | const char * const ptr = |
| 4780 | isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv); |
| 4781 | |
| 4782 | PERL_ARGS_ASSERT_SV_UNI_DISPLAY; |
| 4783 | |
| 4784 | return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr, |
| 4785 | SvCUR(ssv), pvlim, flags); |
| 4786 | } |
| 4787 | |
| 4788 | /* |
| 4789 | =for apidoc foldEQ_utf8 |
| 4790 | |
| 4791 | Returns true if the leading portions of the strings C<s1> and C<s2> (either or both |
| 4792 | of which may be in UTF-8) are the same case-insensitively; false otherwise. |
| 4793 | How far into the strings to compare is determined by other input parameters. |
| 4794 | |
| 4795 | If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode; |
| 4796 | otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2> |
| 4797 | with respect to C<s2>. |
| 4798 | |
| 4799 | If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold |
| 4800 | equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The |
| 4801 | scan will not be considered to be a match unless the goal is reached, and |
| 4802 | scanning won't continue past that goal. Correspondingly for C<l2> with respect to |
| 4803 | C<s2>. |
| 4804 | |
| 4805 | If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is |
| 4806 | considered an end pointer to the position 1 byte past the maximum point |
| 4807 | in C<s1> beyond which scanning will not continue under any circumstances. |
| 4808 | (This routine assumes that UTF-8 encoded input strings are not malformed; |
| 4809 | malformed input can cause it to read past C<pe1>). |
| 4810 | This means that if both C<l1> and C<pe1> are specified, and C<pe1> |
| 4811 | is less than C<s1>+C<l1>, the match will never be successful because it can |
| 4812 | never |
| 4813 | get as far as its goal (and in fact is asserted against). Correspondingly for |
| 4814 | C<pe2> with respect to C<s2>. |
| 4815 | |
| 4816 | At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and |
| 4817 | C<l2> must be non-zero), and if both do, both have to be |
| 4818 | reached for a successful match. Also, if the fold of a character is multiple |
| 4819 | characters, all of them must be matched (see tr21 reference below for |
| 4820 | 'folding'). |
| 4821 | |
| 4822 | Upon a successful match, if C<pe1> is non-C<NULL>, |
| 4823 | it will be set to point to the beginning of the I<next> character of C<s1> |
| 4824 | beyond what was matched. Correspondingly for C<pe2> and C<s2>. |
| 4825 | |
| 4826 | For case-insensitiveness, the "casefolding" of Unicode is used |
| 4827 | instead of upper/lowercasing both the characters, see |
| 4828 | L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings). |
| 4829 | |
| 4830 | =cut */ |
| 4831 | |
| 4832 | /* A flags parameter has been added which may change, and hence isn't |
| 4833 | * externally documented. Currently it is: |
| 4834 | * 0 for as-documented above |
| 4835 | * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an |
| 4836 | ASCII one, to not match |
| 4837 | * FOLDEQ_LOCALE is set iff the rules from the current underlying |
| 4838 | * locale are to be used. |
| 4839 | * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this |
| 4840 | * routine. This allows that step to be skipped. |
| 4841 | * Currently, this requires s1 to be encoded as UTF-8 |
| 4842 | * (u1 must be true), which is asserted for. |
| 4843 | * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may |
| 4844 | * cross certain boundaries. Hence, the caller should |
| 4845 | * let this function do the folding instead of |
| 4846 | * pre-folding. This code contains an assertion to |
| 4847 | * that effect. However, if the caller knows what |
| 4848 | * it's doing, it can pass this flag to indicate that, |
| 4849 | * and the assertion is skipped. |
| 4850 | * FOLDEQ_S2_ALREADY_FOLDED Similarly. |
| 4851 | * FOLDEQ_S2_FOLDS_SANE |
| 4852 | */ |
| 4853 | I32 |
| 4854 | 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) |
| 4855 | { |
| 4856 | const U8 *p1 = (const U8*)s1; /* Point to current char */ |
| 4857 | const U8 *p2 = (const U8*)s2; |
| 4858 | const U8 *g1 = NULL; /* goal for s1 */ |
| 4859 | const U8 *g2 = NULL; |
| 4860 | const U8 *e1 = NULL; /* Don't scan s1 past this */ |
| 4861 | U8 *f1 = NULL; /* Point to current folded */ |
| 4862 | const U8 *e2 = NULL; |
| 4863 | U8 *f2 = NULL; |
| 4864 | STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */ |
| 4865 | U8 foldbuf1[UTF8_MAXBYTES_CASE+1]; |
| 4866 | U8 foldbuf2[UTF8_MAXBYTES_CASE+1]; |
| 4867 | U8 flags_for_folder = FOLD_FLAGS_FULL; |
| 4868 | |
| 4869 | PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS; |
| 4870 | |
| 4871 | assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE)) |
| 4872 | && (((flags & FOLDEQ_S1_ALREADY_FOLDED) |
| 4873 | && !(flags & FOLDEQ_S1_FOLDS_SANE)) |
| 4874 | || ((flags & FOLDEQ_S2_ALREADY_FOLDED) |
| 4875 | && !(flags & FOLDEQ_S2_FOLDS_SANE))))); |
| 4876 | /* The algorithm is to trial the folds without regard to the flags on |
| 4877 | * the first line of the above assert(), and then see if the result |
| 4878 | * violates them. This means that the inputs can't be pre-folded to a |
| 4879 | * violating result, hence the assert. This could be changed, with the |
| 4880 | * addition of extra tests here for the already-folded case, which would |
| 4881 | * slow it down. That cost is more than any possible gain for when these |
| 4882 | * flags are specified, as the flags indicate /il or /iaa matching which |
| 4883 | * is less common than /iu, and I (khw) also believe that real-world /il |
| 4884 | * and /iaa matches are most likely to involve code points 0-255, and this |
| 4885 | * function only under rare conditions gets called for 0-255. */ |
| 4886 | |
| 4887 | if (flags & FOLDEQ_LOCALE) { |
| 4888 | if (IN_UTF8_CTYPE_LOCALE) { |
| 4889 | flags &= ~FOLDEQ_LOCALE; |
| 4890 | } |
| 4891 | else { |
| 4892 | flags_for_folder |= FOLD_FLAGS_LOCALE; |
| 4893 | } |
| 4894 | } |
| 4895 | |
| 4896 | if (pe1) { |
| 4897 | e1 = *(U8**)pe1; |
| 4898 | } |
| 4899 | |
| 4900 | if (l1) { |
| 4901 | g1 = (const U8*)s1 + l1; |
| 4902 | } |
| 4903 | |
| 4904 | if (pe2) { |
| 4905 | e2 = *(U8**)pe2; |
| 4906 | } |
| 4907 | |
| 4908 | if (l2) { |
| 4909 | g2 = (const U8*)s2 + l2; |
| 4910 | } |
| 4911 | |
| 4912 | /* Must have at least one goal */ |
| 4913 | assert(g1 || g2); |
| 4914 | |
| 4915 | if (g1) { |
| 4916 | |
| 4917 | /* Will never match if goal is out-of-bounds */ |
| 4918 | assert(! e1 || e1 >= g1); |
| 4919 | |
| 4920 | /* Here, there isn't an end pointer, or it is beyond the goal. We |
| 4921 | * only go as far as the goal */ |
| 4922 | e1 = g1; |
| 4923 | } |
| 4924 | else { |
| 4925 | assert(e1); /* Must have an end for looking at s1 */ |
| 4926 | } |
| 4927 | |
| 4928 | /* Same for goal for s2 */ |
| 4929 | if (g2) { |
| 4930 | assert(! e2 || e2 >= g2); |
| 4931 | e2 = g2; |
| 4932 | } |
| 4933 | else { |
| 4934 | assert(e2); |
| 4935 | } |
| 4936 | |
| 4937 | /* If both operands are already folded, we could just do a memEQ on the |
| 4938 | * whole strings at once, but it would be better if the caller realized |
| 4939 | * this and didn't even call us */ |
| 4940 | |
| 4941 | /* Look through both strings, a character at a time */ |
| 4942 | while (p1 < e1 && p2 < e2) { |
| 4943 | |
| 4944 | /* If at the beginning of a new character in s1, get its fold to use |
| 4945 | * and the length of the fold. */ |
| 4946 | if (n1 == 0) { |
| 4947 | if (flags & FOLDEQ_S1_ALREADY_FOLDED) { |
| 4948 | f1 = (U8 *) p1; |
| 4949 | assert(u1); |
| 4950 | n1 = UTF8SKIP(f1); |
| 4951 | } |
| 4952 | else { |
| 4953 | if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) { |
| 4954 | |
| 4955 | /* We have to forbid mixing ASCII with non-ASCII if the |
| 4956 | * flags so indicate. And, we can short circuit having to |
| 4957 | * call the general functions for this common ASCII case, |
| 4958 | * all of whose non-locale folds are also ASCII, and hence |
| 4959 | * UTF-8 invariants, so the UTF8ness of the strings is not |
| 4960 | * relevant. */ |
| 4961 | if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) { |
| 4962 | return 0; |
| 4963 | } |
| 4964 | n1 = 1; |
| 4965 | *foldbuf1 = toFOLD(*p1); |
| 4966 | } |
| 4967 | else if (u1) { |
| 4968 | _to_utf8_fold_flags(p1, foldbuf1, &n1, flags_for_folder); |
| 4969 | } |
| 4970 | else { /* Not UTF-8, get UTF-8 fold */ |
| 4971 | _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder); |
| 4972 | } |
| 4973 | f1 = foldbuf1; |
| 4974 | } |
| 4975 | } |
| 4976 | |
| 4977 | if (n2 == 0) { /* Same for s2 */ |
| 4978 | if (flags & FOLDEQ_S2_ALREADY_FOLDED) { |
| 4979 | f2 = (U8 *) p2; |
| 4980 | assert(u2); |
| 4981 | n2 = UTF8SKIP(f2); |
| 4982 | } |
| 4983 | else { |
| 4984 | if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) { |
| 4985 | if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) { |
| 4986 | return 0; |
| 4987 | } |
| 4988 | n2 = 1; |
| 4989 | *foldbuf2 = toFOLD(*p2); |
| 4990 | } |
| 4991 | else if (u2) { |
| 4992 | _to_utf8_fold_flags(p2, foldbuf2, &n2, flags_for_folder); |
| 4993 | } |
| 4994 | else { |
| 4995 | _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder); |
| 4996 | } |
| 4997 | f2 = foldbuf2; |
| 4998 | } |
| 4999 | } |
| 5000 | |
| 5001 | /* Here f1 and f2 point to the beginning of the strings to compare. |
| 5002 | * These strings are the folds of the next character from each input |
| 5003 | * string, stored in UTF-8. */ |
| 5004 | |
| 5005 | /* While there is more to look for in both folds, see if they |
| 5006 | * continue to match */ |
| 5007 | while (n1 && n2) { |
| 5008 | U8 fold_length = UTF8SKIP(f1); |
| 5009 | if (fold_length != UTF8SKIP(f2) |
| 5010 | || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE |
| 5011 | function call for single |
| 5012 | byte */ |
| 5013 | || memNE((char*)f1, (char*)f2, fold_length)) |
| 5014 | { |
| 5015 | return 0; /* mismatch */ |
| 5016 | } |
| 5017 | |
| 5018 | /* Here, they matched, advance past them */ |
| 5019 | n1 -= fold_length; |
| 5020 | f1 += fold_length; |
| 5021 | n2 -= fold_length; |
| 5022 | f2 += fold_length; |
| 5023 | } |
| 5024 | |
| 5025 | /* When reach the end of any fold, advance the input past it */ |
| 5026 | if (n1 == 0) { |
| 5027 | p1 += u1 ? UTF8SKIP(p1) : 1; |
| 5028 | } |
| 5029 | if (n2 == 0) { |
| 5030 | p2 += u2 ? UTF8SKIP(p2) : 1; |
| 5031 | } |
| 5032 | } /* End of loop through both strings */ |
| 5033 | |
| 5034 | /* A match is defined by each scan that specified an explicit length |
| 5035 | * reaching its final goal, and the other not having matched a partial |
| 5036 | * character (which can happen when the fold of a character is more than one |
| 5037 | * character). */ |
| 5038 | if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) { |
| 5039 | return 0; |
| 5040 | } |
| 5041 | |
| 5042 | /* Successful match. Set output pointers */ |
| 5043 | if (pe1) { |
| 5044 | *pe1 = (char*)p1; |
| 5045 | } |
| 5046 | if (pe2) { |
| 5047 | *pe2 = (char*)p2; |
| 5048 | } |
| 5049 | return 1; |
| 5050 | } |
| 5051 | |
| 5052 | /* XXX The next two functions should likely be moved to mathoms.c once all |
| 5053 | * occurrences of them are removed from the core; some cpan-upstream modules |
| 5054 | * still use them */ |
| 5055 | |
| 5056 | U8 * |
| 5057 | Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv) |
| 5058 | { |
| 5059 | PERL_ARGS_ASSERT_UVUNI_TO_UTF8; |
| 5060 | |
| 5061 | return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0); |
| 5062 | } |
| 5063 | |
| 5064 | /* |
| 5065 | =for apidoc utf8n_to_uvuni |
| 5066 | |
| 5067 | Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>. |
| 5068 | |
| 5069 | This function was useful for code that wanted to handle both EBCDIC and |
| 5070 | ASCII platforms with Unicode properties, but starting in Perl v5.20, the |
| 5071 | distinctions between the platforms have mostly been made invisible to most |
| 5072 | code, so this function is quite unlikely to be what you want. If you do need |
| 5073 | this precise functionality, use instead |
| 5074 | C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> |
| 5075 | or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>. |
| 5076 | |
| 5077 | =cut |
| 5078 | */ |
| 5079 | |
| 5080 | UV |
| 5081 | Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags) |
| 5082 | { |
| 5083 | PERL_ARGS_ASSERT_UTF8N_TO_UVUNI; |
| 5084 | |
| 5085 | return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags)); |
| 5086 | } |
| 5087 | |
| 5088 | /* |
| 5089 | =for apidoc uvuni_to_utf8_flags |
| 5090 | |
| 5091 | Instead you almost certainly want to use L</uvchr_to_utf8> or |
| 5092 | L</uvchr_to_utf8_flags>. |
| 5093 | |
| 5094 | This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>, |
| 5095 | which itself, while not deprecated, should be used only in isolated |
| 5096 | circumstances. These functions were useful for code that wanted to handle |
| 5097 | both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl |
| 5098 | v5.20, the distinctions between the platforms have mostly been made invisible |
| 5099 | to most code, so this function is quite unlikely to be what you want. |
| 5100 | |
| 5101 | =cut |
| 5102 | */ |
| 5103 | |
| 5104 | U8 * |
| 5105 | Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags) |
| 5106 | { |
| 5107 | PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS; |
| 5108 | |
| 5109 | return uvoffuni_to_utf8_flags(d, uv, flags); |
| 5110 | } |
| 5111 | |
| 5112 | /* |
| 5113 | * ex: set ts=8 sts=4 sw=4 et: |
| 5114 | */ |