3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
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.
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.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
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
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "inline_invlist.c"
37 /* Separate prototypes needed because in ASCII systems these are
38 * usually macros but they still are compiled as code, too. */
39 PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
40 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
43 static const char unees[] =
44 "Malformed UTF-8 character (unexpected end of string)";
47 =head1 Unicode Support
49 This file contains various utility functions for manipulating UTF8-encoded
50 strings. For the uninitiated, this is a method of representing arbitrary
51 Unicode characters as a variable number of bytes, in such a way that
52 characters in the ASCII range are unmodified, and a zero byte never appears
53 within non-zero characters.
59 =for apidoc is_ascii_string
61 Returns true if the first C<len> bytes of the string C<s> are the same whether
62 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
63 is, if they are invariant. On ASCII-ish machines, only ASCII characters
64 fit this definition, hence the function's name.
66 If C<len> is 0, it will be calculated using C<strlen(s)>.
68 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
74 Perl_is_ascii_string(const U8 *s, STRLEN len)
76 const U8* const send = s + (len ? len : strlen((const char *)s));
79 PERL_ARGS_ASSERT_IS_ASCII_STRING;
81 for (; x < send; ++x) {
82 if (!UTF8_IS_INVARIANT(*x))
90 =for apidoc uvuni_to_utf8_flags
92 Adds the UTF-8 representation of the code point C<uv> to the end
93 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
94 bytes available. The return value is the pointer to the byte after the
95 end of the new character. In other words,
97 d = uvuni_to_utf8_flags(d, uv, flags);
101 d = uvuni_to_utf8(d, uv);
103 (which is equivalent to)
105 d = uvuni_to_utf8_flags(d, uv, 0);
107 This is the recommended Unicode-aware way of saying
111 This function will convert to UTF-8 (and not warn) even code points that aren't
112 legal Unicode or are problematic, unless C<flags> contains one or more of the
115 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
116 the function will raise a warning, provided UTF8 warnings are enabled. If instead
117 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
118 If both flags are set, the function will both warn and return NULL.
120 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
121 affect how the function handles a Unicode non-character. And, likewise for the
122 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
123 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
124 even less portable) can be warned and/or disallowed even if other above-Unicode
125 code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
128 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
129 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
137 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
139 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
141 /* The first problematic code point is the first surrogate */
142 if (uv >= UNICODE_SURROGATE_FIRST
143 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
145 if (UNICODE_IS_SURROGATE(uv)) {
146 if (flags & UNICODE_WARN_SURROGATE) {
147 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
148 "UTF-16 surrogate U+%04"UVXf, uv);
150 if (flags & UNICODE_DISALLOW_SURROGATE) {
154 else if (UNICODE_IS_SUPER(uv)) {
155 if (flags & UNICODE_WARN_SUPER
156 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
158 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
159 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
161 if (flags & UNICODE_DISALLOW_SUPER
162 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
167 else if (UNICODE_IS_NONCHAR(uv)) {
168 if (flags & UNICODE_WARN_NONCHAR) {
169 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
170 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
173 if (flags & UNICODE_DISALLOW_NONCHAR) {
178 if (UNI_IS_INVARIANT(uv)) {
179 *d++ = (U8)UTF_TO_NATIVE(uv);
184 STRLEN len = UNISKIP(uv);
187 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
188 uv >>= UTF_ACCUMULATION_SHIFT;
190 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
193 #else /* Non loop style */
195 *d++ = (U8)(( uv >> 6) | 0xc0);
196 *d++ = (U8)(( uv & 0x3f) | 0x80);
200 *d++ = (U8)(( uv >> 12) | 0xe0);
201 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
202 *d++ = (U8)(( uv & 0x3f) | 0x80);
206 *d++ = (U8)(( uv >> 18) | 0xf0);
207 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
208 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
209 *d++ = (U8)(( uv & 0x3f) | 0x80);
212 if (uv < 0x4000000) {
213 *d++ = (U8)(( uv >> 24) | 0xf8);
214 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
217 *d++ = (U8)(( uv & 0x3f) | 0x80);
220 if (uv < 0x80000000) {
221 *d++ = (U8)(( uv >> 30) | 0xfc);
222 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
223 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
226 *d++ = (U8)(( uv & 0x3f) | 0x80);
230 if (uv < UTF8_QUAD_MAX)
233 *d++ = 0xfe; /* Can't match U+FEFF! */
234 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
235 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
239 *d++ = (U8)(( uv & 0x3f) | 0x80);
244 *d++ = 0xff; /* Can't match U+FFFE! */
245 *d++ = 0x80; /* 6 Reserved bits */
246 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
247 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
248 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
256 *d++ = (U8)(( uv & 0x3f) | 0x80);
260 #endif /* Loop style */
265 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
266 character. Note that an INVARIANT (i.e. ASCII) character is a valid
267 UTF-8 character. The number of bytes in the UTF-8 character
268 will be returned if it is valid, otherwise 0.
270 This is the "slow" version as opposed to the "fast" version which is
271 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
272 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
273 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
274 you should use the _slow(). In practice this means that the _slow()
275 will be used very rarely, since the maximum Unicode code point (as of
276 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
277 the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
282 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
284 dTHX; /* The function called below requires thread context */
288 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
290 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
292 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
296 =for apidoc is_utf8_char_buf
298 Returns the number of bytes that comprise the first UTF-8 encoded character in
299 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
300 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
303 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
304 machines) is a valid UTF-8 character.
309 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
314 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
316 if (buf_end <= buf) {
321 if (len > UTF8SKIP(buf)) {
326 if (IS_UTF8_CHAR_FAST(len))
327 return IS_UTF8_CHAR(buf, len) ? len : 0;
328 #endif /* #ifdef IS_UTF8_CHAR */
329 return is_utf8_char_slow(buf, len);
333 =for apidoc is_utf8_char
337 Tests if some arbitrary number of bytes begins in a valid UTF-8
338 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
339 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
340 character will be returned if it is valid, otherwise 0.
342 This function is deprecated due to the possibility that malformed input could
343 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
349 Perl_is_utf8_char(const U8 *s)
351 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
353 /* Assumes we have enough space, which is why this is deprecated */
354 return is_utf8_char_buf(s, s + UTF8SKIP(s));
359 =for apidoc is_utf8_string
361 Returns true if the first C<len> bytes of string C<s> form a valid
362 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
363 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
364 terminating NUL byte). Note that all characters being ASCII constitute 'a
367 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
373 Perl_is_utf8_string(const U8 *s, STRLEN len)
375 const U8* const send = s + (len ? len : strlen((const char *)s));
378 PERL_ARGS_ASSERT_IS_UTF8_STRING;
381 /* Inline the easy bits of is_utf8_char() here for speed... */
382 if (UTF8_IS_INVARIANT(*x)) {
385 else if (!UTF8_IS_START(*x))
388 /* ... and call is_utf8_char() only if really needed. */
389 const STRLEN c = UTF8SKIP(x);
390 const U8* const next_char_ptr = x + c;
392 if (next_char_ptr > send) {
396 if (IS_UTF8_CHAR_FAST(c)) {
397 if (!IS_UTF8_CHAR(x, c))
400 else if (! is_utf8_char_slow(x, c)) {
411 Implemented as a macro in utf8.h
413 =for apidoc is_utf8_string_loc
415 Like L</is_utf8_string> but stores the location of the failure (in the
416 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
417 "utf8ness success") in the C<ep>.
419 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
421 =for apidoc is_utf8_string_loclen
423 Like L</is_utf8_string>() but stores the location of the failure (in the
424 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
425 "utf8ness success") in the C<ep>, and the number of UTF-8
426 encoded characters in the C<el>.
428 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
434 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
436 const U8* const send = s + (len ? len : strlen((const char *)s));
441 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
444 const U8* next_char_ptr;
446 /* Inline the easy bits of is_utf8_char() here for speed... */
447 if (UTF8_IS_INVARIANT(*x))
448 next_char_ptr = x + 1;
449 else if (!UTF8_IS_START(*x))
452 /* ... and call is_utf8_char() only if really needed. */
454 next_char_ptr = c + x;
455 if (next_char_ptr > send) {
458 if (IS_UTF8_CHAR_FAST(c)) {
459 if (!IS_UTF8_CHAR(x, c))
462 c = is_utf8_char_slow(x, c);
481 =for apidoc utf8n_to_uvuni
483 Bottom level UTF-8 decode routine.
484 Returns the code point value of the first character in the string C<s>,
485 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
486 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
487 the length, in bytes, of that character.
489 The value of C<flags> determines the behavior when C<s> does not point to a
490 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
491 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
492 next possible position in C<s> that could begin a non-malformed character.
493 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
495 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
496 individual types of malformations, such as the sequence being overlong (that
497 is, when there is a shorter sequence that can express the same code point;
498 overlong sequences are expressly forbidden in the UTF-8 standard due to
499 potential security issues). Another malformation example is the first byte of
500 a character not being a legal first byte. See F<utf8.h> for the list of such
501 flags. For allowed 0 length strings, this function returns 0; for allowed
502 overlong sequences, the computed code point is returned; for all other allowed
503 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
504 determinable reasonable value.
506 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
507 flags) malformation is found. If this flag is set, the routine assumes that
508 the caller will raise a warning, and this function will silently just set
509 C<retlen> to C<-1> and return zero.
511 Certain code points are considered problematic. These are Unicode surrogates,
512 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
513 By default these are considered regular code points, but certain situations
514 warrant special handling for them. If C<flags> contains
515 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
516 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
517 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
518 maximum) can be set to disallow these categories individually.
520 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
521 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
522 for their respective categories, but otherwise the code points are considered
523 valid (not malformations). To get a category to both be treated as a
524 malformation and raise a warning, specify both the WARN and DISALLOW flags.
525 (But note that warnings are not raised if lexically disabled nor if
526 UTF8_CHECK_ONLY is also specified.)
528 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
529 the others that are above the Unicode legal maximum. There are several
530 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
531 not representable at all on EBCDIC platforms, and the original UTF-8
532 specification never went above this number (the current 0x10FFFF limit was
533 imposed later). (The smaller ones, those that fit into 32 bits, are
534 representable by a UV on ASCII platforms, but not by an IV, which means that
535 the number of operations that can be performed on them is quite restricted.)
536 The UTF-8 encoding on ASCII platforms for these large code points begins with a
537 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
538 be treated as malformations, while allowing smaller above-Unicode code points.
539 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
540 including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
541 the other WARN flags, but applies just to these code points.
543 All other code points corresponding to Unicode characters, including private
544 use and those yet to be assigned, are never considered malformed and never
547 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
553 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
556 const U8 * const s0 = s;
557 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
562 UV outlier_ret = 0; /* return value when input is in error or problematic
564 UV pack_warn = 0; /* Save result of packWARN() for later */
565 bool unexpected_non_continuation = FALSE;
566 bool overflowed = FALSE;
567 bool do_overlong_test = TRUE; /* May have to skip this test */
569 const char* const malformed_text = "Malformed UTF-8 character";
571 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
573 /* The order of malformation tests here is important. We should consume as
574 * few bytes as possible in order to not skip any valid character. This is
575 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
576 * http://unicode.org/reports/tr36 for more discussion as to why. For
577 * example, once we've done a UTF8SKIP, we can tell the expected number of
578 * bytes, and could fail right off the bat if the input parameters indicate
579 * that there are too few available. But it could be that just that first
580 * byte is garbled, and the intended character occupies fewer bytes. If we
581 * blindly assumed that the first byte is correct, and skipped based on
582 * that number, we could skip over a valid input character. So instead, we
583 * always examine the sequence byte-by-byte.
585 * We also should not consume too few bytes, otherwise someone could inject
586 * things. For example, an input could be deliberately designed to
587 * overflow, and if this code bailed out immediately upon discovering that,
588 * returning to the caller *retlen pointing to the very next byte (one
589 * which is actually part of of the overflowing sequence), that could look
590 * legitimate to the caller, which could discard the initial partial
591 * sequence and process the rest, inappropriately */
593 /* Zero length strings, if allowed, of necessity are zero */
594 if (UNLIKELY(curlen == 0)) {
599 if (flags & UTF8_ALLOW_EMPTY) {
602 if (! (flags & UTF8_CHECK_ONLY)) {
603 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
608 expectlen = UTF8SKIP(s);
610 /* A well-formed UTF-8 character, as the vast majority of calls to this
611 * function will be for, has this expected length. For efficiency, set
612 * things up here to return it. It will be overriden only in those rare
613 * cases where a malformation is found */
618 /* An invariant is trivially well-formed */
619 if (UTF8_IS_INVARIANT(uv)) {
620 return (UV) (NATIVE_TO_UTF(*s));
623 /* A continuation character can't start a valid sequence */
624 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
625 if (flags & UTF8_ALLOW_CONTINUATION) {
629 return UNICODE_REPLACEMENT;
632 if (! (flags & UTF8_CHECK_ONLY)) {
633 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
640 uv = NATIVE_TO_UTF(uv);
643 /* Here is not a continuation byte, nor an invariant. The only thing left
644 * is a start byte (possibly for an overlong) */
646 /* Remove the leading bits that indicate the number of bytes in the
647 * character's whole UTF-8 sequence, leaving just the bits that are part of
649 uv &= UTF_START_MASK(expectlen);
651 /* Now, loop through the remaining bytes in the character's sequence,
652 * accumulating each into the working value as we go. Be sure to not look
653 * past the end of the input string */
654 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
656 for (s = s0 + 1; s < send; s++) {
657 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
658 #ifndef EBCDIC /* Can't overflow in EBCDIC */
659 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
661 /* The original implementors viewed this malformation as more
662 * serious than the others (though I, khw, don't understand
663 * why, since other malformations also give very very wrong
664 * results), so there is no way to turn off checking for it.
665 * Set a flag, but keep going in the loop, so that we absorb
666 * the rest of the bytes that comprise the character. */
668 overflow_byte = *s; /* Save for warning message's use */
671 uv = UTF8_ACCUMULATE(uv, *s);
674 /* Here, found a non-continuation before processing all expected
675 * bytes. This byte begins a new character, so quit, even if
676 * allowing this malformation. */
677 unexpected_non_continuation = TRUE;
680 } /* End of loop through the character's bytes */
682 /* Save how many bytes were actually in the character */
685 /* The loop above finds two types of malformations: non-continuation and/or
686 * overflow. The non-continuation malformation is really a too-short
687 * malformation, as it means that the current character ended before it was
688 * expected to (being terminated prematurely by the beginning of the next
689 * character, whereas in the too-short malformation there just are too few
690 * bytes available to hold the character. In both cases, the check below
691 * that we have found the expected number of bytes would fail if executed.)
692 * Thus the non-continuation malformation is really unnecessary, being a
693 * subset of the too-short malformation. But there may be existing
694 * applications that are expecting the non-continuation type, so we retain
695 * it, and return it in preference to the too-short malformation. (If this
696 * code were being written from scratch, the two types might be collapsed
697 * into one.) I, khw, am also giving priority to returning the
698 * non-continuation and too-short malformations over overflow when multiple
699 * ones are present. I don't know of any real reason to prefer one over
700 * the other, except that it seems to me that multiple-byte errors trumps
701 * errors from a single byte */
702 if (UNLIKELY(unexpected_non_continuation)) {
703 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
704 if (! (flags & UTF8_CHECK_ONLY)) {
706 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
709 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, %d bytes after start byte 0x%02x, expected %d bytes)", malformed_text, *s, (int) curlen, *s0, (int)expectlen));
714 uv = UNICODE_REPLACEMENT;
716 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
717 * as what the original expectations were. */
718 do_overlong_test = FALSE;
723 else if (UNLIKELY(curlen < expectlen)) {
724 if (! (flags & UTF8_ALLOW_SHORT)) {
725 if (! (flags & UTF8_CHECK_ONLY)) {
726 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, *s0));
730 uv = UNICODE_REPLACEMENT;
731 do_overlong_test = FALSE;
737 #ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
738 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
739 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
741 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
742 * generation of the sv, since no warnings are raised under CHECK */
743 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
744 && ckWARN_d(WARN_UTF8))
746 /* This message is deliberately not of the same syntax as the other
747 * messages for malformations, for backwards compatibility in the
748 * unlikely event that code is relying on its precise earlier text
750 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
751 pack_warn = packWARN(WARN_UTF8);
753 if (flags & UTF8_DISALLOW_FE_FF) {
757 if (UNLIKELY(overflowed)) {
759 /* If the first byte is FF, it will overflow a 32-bit word. If the
760 * first byte is FE, it will overflow a signed 32-bit word. The
761 * above preserves backward compatibility, since its message was used
762 * in earlier versions of this code in preference to overflow */
763 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
769 && expectlen > (STRLEN)UNISKIP(uv)
770 && ! (flags & UTF8_ALLOW_LONG))
772 /* The overlong malformation has lower precedence than the others.
773 * Note that if this malformation is allowed, we return the actual
774 * value, instead of the replacement character. This is because this
775 * value is actually well-defined. */
776 if (! (flags & UTF8_CHECK_ONLY)) {
777 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), *s0));
782 /* Here, the input is considered to be well-formed , but could be a
783 * problematic code point that is not allowed by the input parameters. */
784 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
785 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
786 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
788 if (UNICODE_IS_SURROGATE(uv)) {
789 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
790 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
792 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
793 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
795 if (flags & UTF8_DISALLOW_SURROGATE) {
799 else if ((uv > PERL_UNICODE_MAX)) {
800 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
801 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
803 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
804 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
806 if (flags & UTF8_DISALLOW_SUPER) {
810 else if (UNICODE_IS_NONCHAR(uv)) {
811 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
812 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
814 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
815 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
817 if (flags & UTF8_DISALLOW_NONCHAR) {
827 /* Here, this is not considered a malformed character, so drop through
833 /* There are three cases which get to beyond this point. In all 3 cases:
834 * <sv> if not null points to a string to print as a warning.
835 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
837 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
838 * This is done by initializing it to 0, and changing it only
841 * 1) The input is valid but problematic, and to be warned about. The
842 * return value is the resultant code point; <*retlen> is set to
843 * <curlen>, the number of bytes that comprise the code point.
844 * <pack_warn> contains the result of packWARN() for the warning
845 * types. The entry point for this case is the label <do_warn>;
846 * 2) The input is a valid code point but disallowed by the parameters to
847 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
848 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
849 * comprise the code point. <pack_warn> contains the result of
850 * packWARN() for the warning types. The entry point for this case is
851 * the label <disallowed>.
852 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
853 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
854 * bytes that comprise the malformation. All such malformations are
855 * assumed to be warning type <utf8>. The entry point for this case
856 * is the label <malformed>.
861 if (sv && ckWARN_d(WARN_UTF8)) {
862 pack_warn = packWARN(WARN_UTF8);
867 if (flags & UTF8_CHECK_ONLY) {
869 *retlen = ((STRLEN) -1);
875 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
876 if warnings are to be raised. */
877 const char * const string = SvPVX_const(sv);
880 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
882 Perl_warner(aTHX_ pack_warn, "%s", string);
893 =for apidoc utf8_to_uvchr_buf
895 Returns the native code point of the first character in the string C<s> which
896 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
897 C<*retlen> will be set to the length, in bytes, of that character.
899 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
900 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
901 NULL) to -1. If those warnings are off, the computed value if well-defined (or
902 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
903 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
904 next possible position in C<s> that could begin a non-malformed character.
905 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
912 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
914 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
918 return utf8n_to_uvchr(s, send - s, retlen,
919 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
922 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
923 * there are no malformations in the input UTF-8 string C<s>. surrogates,
924 * non-character code points, and non-Unicode code points are allowed */
927 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
929 const UV uv = valid_utf8_to_uvuni(s, retlen);
931 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
933 return UNI_TO_NATIVE(uv);
937 =for apidoc utf8_to_uvchr
941 Returns the native code point of the first character in the string C<s>
942 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
943 length, in bytes, of that character.
945 Some, but not all, UTF-8 malformations are detected, and in fact, some
946 malformed input could cause reading beyond the end of the input buffer, which
947 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
949 If C<s> points to one of the detected malformations, and UTF8 warnings are
950 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
951 NULL) to -1. If those warnings are off, the computed value if well-defined (or
952 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
953 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
954 next possible position in C<s> that could begin a non-malformed character.
955 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
961 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
963 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
965 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
969 =for apidoc utf8_to_uvuni_buf
971 Returns the Unicode code point of the first character in the string C<s> which
972 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
973 C<retlen> will be set to the length, in bytes, of that character.
975 This function should only be used when the returned UV is considered
976 an index into the Unicode semantic tables (e.g. swashes).
978 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
979 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
980 NULL) to -1. If those warnings are off, the computed value if well-defined (or
981 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
982 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
983 next possible position in C<s> that could begin a non-malformed character.
984 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
990 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
992 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
996 /* Call the low level routine asking for checks */
997 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
998 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1001 /* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1002 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1003 * non-character code points, and non-Unicode code points are allowed */
1006 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1008 UV expectlen = UTF8SKIP(s);
1009 const U8* send = s + expectlen;
1010 UV uv = NATIVE_TO_UTF(*s);
1012 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1015 *retlen = expectlen;
1018 /* An invariant is trivially returned */
1019 if (expectlen == 1) {
1023 /* Remove the leading bits that indicate the number of bytes, leaving just
1024 * the bits that are part of the value */
1025 uv &= UTF_START_MASK(expectlen);
1027 /* Now, loop through the remaining bytes, accumulating each into the
1028 * working total as we go */
1029 for (++s; s < send; s++) {
1030 uv = UTF8_ACCUMULATE(uv, *s);
1037 =for apidoc utf8_to_uvuni
1041 Returns the Unicode code point of the first character in the string C<s>
1042 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1043 length, in bytes, of that character.
1045 This function should only be used when the returned UV is considered
1046 an index into the Unicode semantic tables (e.g. swashes).
1048 Some, but not all, UTF-8 malformations are detected, and in fact, some
1049 malformed input could cause reading beyond the end of the input buffer, which
1050 is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
1052 If C<s> points to one of the detected malformations, and UTF8 warnings are
1053 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1054 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1055 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1056 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1057 next possible position in C<s> that could begin a non-malformed character.
1058 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
1064 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1066 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1068 return valid_utf8_to_uvuni(s, retlen);
1072 =for apidoc utf8_length
1074 Return the length of the UTF-8 char encoded string C<s> in characters.
1075 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1076 up past C<e>, croaks.
1082 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1087 PERL_ARGS_ASSERT_UTF8_LENGTH;
1089 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1090 * the bitops (especially ~) can create illegal UTF-8.
1091 * In other words: in Perl UTF-8 is not just for Unicode. */
1094 goto warn_and_return;
1096 if (!UTF8_IS_INVARIANT(*s))
1107 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1108 "%s in %s", unees, OP_DESC(PL_op));
1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1117 =for apidoc utf8_distance
1119 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1122 WARNING: use only if you *know* that the pointers point inside the
1129 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1131 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1133 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1137 =for apidoc utf8_hop
1139 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1140 forward or backward.
1142 WARNING: do not use the following unless you *know* C<off> is within
1143 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1144 on the first byte of character or just after the last byte of a character.
1150 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1152 PERL_ARGS_ASSERT_UTF8_HOP;
1154 PERL_UNUSED_CONTEXT;
1155 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1156 * the bitops (especially ~) can create illegal UTF-8.
1157 * In other words: in Perl UTF-8 is not just for Unicode. */
1166 while (UTF8_IS_CONTINUATION(*s))
1174 =for apidoc bytes_cmp_utf8
1176 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1177 sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
1178 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1179 if the first string is greater than the second string.
1181 -1 or +1 is returned if the shorter string was identical to the start of the
1182 longer string. -2 or +2 is returned if the was a difference between characters
1189 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1191 const U8 *const bend = b + blen;
1192 const U8 *const uend = u + ulen;
1194 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1196 PERL_UNUSED_CONTEXT;
1198 while (b < bend && u < uend) {
1200 if (!UTF8_IS_INVARIANT(c)) {
1201 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1204 if (UTF8_IS_CONTINUATION(c1)) {
1205 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
1207 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1208 "Malformed UTF-8 character "
1209 "(unexpected non-continuation byte 0x%02x"
1210 ", immediately after start byte 0x%02x)"
1211 /* Dear diag.t, it's in the pod. */
1213 PL_op ? " in " : "",
1214 PL_op ? OP_DESC(PL_op) : "");
1219 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1220 "%s in %s", unees, OP_DESC(PL_op));
1222 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1223 return -2; /* Really want to return undef :-) */
1230 return *b < c ? -2 : +2;
1235 if (b == bend && u == uend)
1238 return b < bend ? +1 : -1;
1242 =for apidoc utf8_to_bytes
1244 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1245 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1246 updates C<len> to contain the new length.
1247 Returns zero on failure, setting C<len> to -1.
1249 If you need a copy of the string, see L</bytes_from_utf8>.
1255 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1257 U8 * const save = s;
1258 U8 * const send = s + *len;
1261 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1263 /* ensure valid UTF-8 and chars < 256 before updating string */
1267 if (!UTF8_IS_INVARIANT(c) &&
1268 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1269 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
1270 *len = ((STRLEN) -1);
1278 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
1287 =for apidoc bytes_from_utf8
1289 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1290 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1291 the newly-created string, and updates C<len> to contain the new
1292 length. Returns the original string if no conversion occurs, C<len>
1293 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1294 0 if C<s> is converted or consisted entirely of characters that are invariant
1295 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1301 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1304 const U8 *start = s;
1308 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1310 PERL_UNUSED_CONTEXT;
1314 /* ensure valid UTF-8 and chars < 256 before converting string */
1315 for (send = s + *len; s < send;) {
1317 if (!UTF8_IS_INVARIANT(c)) {
1318 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1319 (c = *s++) && UTF8_IS_CONTINUATION(c))
1328 Newx(d, (*len) - count + 1, U8);
1329 s = start; start = d;
1332 if (!UTF8_IS_INVARIANT(c)) {
1333 /* Then it is two-byte encoded */
1334 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
1344 =for apidoc bytes_to_utf8
1346 Converts a string C<s> of length C<len> bytes from the native encoding into
1348 Returns a pointer to the newly-created string, and sets C<len> to
1349 reflect the new length in bytes.
1351 A NUL character will be written after the end of the string.
1353 If you want to convert to UTF-8 from encodings other than
1354 the native (Latin1 or EBCDIC),
1355 see L</sv_recode_to_utf8>().
1360 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1361 likewise need duplication. */
1364 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1366 const U8 * const send = s + (*len);
1370 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1371 PERL_UNUSED_CONTEXT;
1373 Newx(d, (*len) * 2 + 1, U8);
1377 const UV uv = NATIVE_TO_ASCII(*s++);
1378 if (UNI_IS_INVARIANT(uv))
1379 *d++ = (U8)UTF_TO_NATIVE(uv);
1381 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1382 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
1391 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1393 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1394 * We optimize for native, for obvious reasons. */
1397 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1402 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1405 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1410 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1414 *d++ = UNI_TO_NATIVE(uv);
1421 *d++ = (U8)(( uv >> 6) | 0xc0);
1422 *d++ = (U8)(( uv & 0x3f) | 0x80);
1425 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
1427 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1429 UV low = (p[0] << 8) + p[1];
1431 if (low < 0xdc00 || low > 0xdfff)
1432 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1433 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1435 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1436 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1439 *d++ = (U8)(( uv >> 12) | 0xe0);
1440 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1441 *d++ = (U8)(( uv & 0x3f) | 0x80);
1445 *d++ = (U8)(( uv >> 18) | 0xf0);
1446 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1447 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1448 *d++ = (U8)(( uv & 0x3f) | 0x80);
1452 *newlen = d - dstart;
1456 /* Note: this one is slightly destructive of the source. */
1459 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1462 U8* const send = s + bytelen;
1464 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1467 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1471 const U8 tmp = s[0];
1476 return utf16_to_utf8(p, d, bytelen, newlen);
1479 /* for now these are all defined (inefficiently) in terms of the utf8 versions.
1480 * Note that the macros in handy.h that call these short-circuit calling them
1481 * for Latin-1 range inputs */
1484 Perl_is_uni_alnum(pTHX_ UV c)
1486 U8 tmpbuf[UTF8_MAXBYTES+1];
1487 uvchr_to_utf8(tmpbuf, c);
1488 return is_utf8_alnum(tmpbuf);
1492 Perl_is_uni_idfirst(pTHX_ UV c)
1494 U8 tmpbuf[UTF8_MAXBYTES+1];
1495 uvchr_to_utf8(tmpbuf, c);
1496 return is_utf8_idfirst(tmpbuf);
1500 Perl_is_uni_alpha(pTHX_ UV c)
1502 U8 tmpbuf[UTF8_MAXBYTES+1];
1503 uvchr_to_utf8(tmpbuf, c);
1504 return is_utf8_alpha(tmpbuf);
1508 Perl_is_uni_ascii(pTHX_ UV c)
1514 Perl_is_uni_blank(pTHX_ UV c)
1516 U8 tmpbuf[UTF8_MAXBYTES+1];
1517 uvchr_to_utf8(tmpbuf, c);
1518 return is_utf8_blank(tmpbuf);
1522 Perl_is_uni_space(pTHX_ UV c)
1524 U8 tmpbuf[UTF8_MAXBYTES+1];
1525 uvchr_to_utf8(tmpbuf, c);
1526 return is_utf8_space(tmpbuf);
1530 Perl_is_uni_digit(pTHX_ UV c)
1532 U8 tmpbuf[UTF8_MAXBYTES+1];
1533 uvchr_to_utf8(tmpbuf, c);
1534 return is_utf8_digit(tmpbuf);
1538 Perl_is_uni_upper(pTHX_ UV c)
1540 U8 tmpbuf[UTF8_MAXBYTES+1];
1541 uvchr_to_utf8(tmpbuf, c);
1542 return is_utf8_upper(tmpbuf);
1546 Perl_is_uni_lower(pTHX_ UV c)
1548 U8 tmpbuf[UTF8_MAXBYTES+1];
1549 uvchr_to_utf8(tmpbuf, c);
1550 return is_utf8_lower(tmpbuf);
1554 Perl_is_uni_cntrl(pTHX_ UV c)
1556 return isCNTRL_L1(c);
1560 Perl_is_uni_graph(pTHX_ UV c)
1562 U8 tmpbuf[UTF8_MAXBYTES+1];
1563 uvchr_to_utf8(tmpbuf, c);
1564 return is_utf8_graph(tmpbuf);
1568 Perl_is_uni_print(pTHX_ UV c)
1570 U8 tmpbuf[UTF8_MAXBYTES+1];
1571 uvchr_to_utf8(tmpbuf, c);
1572 return is_utf8_print(tmpbuf);
1576 Perl_is_uni_punct(pTHX_ UV c)
1578 U8 tmpbuf[UTF8_MAXBYTES+1];
1579 uvchr_to_utf8(tmpbuf, c);
1580 return is_utf8_punct(tmpbuf);
1584 Perl_is_uni_xdigit(pTHX_ UV c)
1586 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1587 uvchr_to_utf8(tmpbuf, c);
1588 return is_utf8_xdigit(tmpbuf);
1592 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1594 /* We have the latin1-range values compiled into the core, so just use
1595 * those, converting the result to utf8. The only difference between upper
1596 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1597 * either "SS" or "Ss". Which one to use is passed into the routine in
1598 * 'S_or_s' to avoid a test */
1600 UV converted = toUPPER_LATIN1_MOD(c);
1602 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1604 assert(S_or_s == 'S' || S_or_s == 's');
1606 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1607 characters in this range */
1608 *p = (U8) converted;
1613 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1614 * which it maps to one of them, so as to only have to have one check for
1615 * it in the main case */
1616 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1618 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1619 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1622 converted = GREEK_CAPITAL_LETTER_MU;
1624 case LATIN_SMALL_LETTER_SHARP_S:
1630 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1631 assert(0); /* NOTREACHED */
1635 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1636 *p = UTF8_TWO_BYTE_LO(converted);
1642 /* Call the function to convert a UTF-8 encoded character to the specified case.
1643 * Note that there may be more than one character in the result.
1644 * INP is a pointer to the first byte of the input character
1645 * OUTP will be set to the first byte of the string of changed characters. It
1646 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1647 * LENP will be set to the length in bytes of the string of changed characters
1649 * The functions return the ordinal of the first character in the string of OUTP */
1650 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1651 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1652 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
1654 /* This additionally has the input parameter SPECIALS, which if non-zero will
1655 * cause this to use the SPECIALS hash for folding (meaning get full case
1656 * folding); otherwise, when zero, this implies a simple case fold */
1657 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
1660 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1664 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1665 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1666 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1667 * the changed version may be longer than the original character.
1669 * The ordinal of the first character of the changed version is returned
1670 * (but note, as explained above, that there may be more.) */
1672 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1675 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1678 uvchr_to_utf8(p, c);
1679 return CALL_UPPER_CASE(p, p, lenp);
1683 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1687 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1690 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1693 uvchr_to_utf8(p, c);
1694 return CALL_TITLE_CASE(p, p, lenp);
1698 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1700 /* We have the latin1-range values compiled into the core, so just use
1701 * those, converting the result to utf8. Since the result is always just
1702 * one character, we allow <p> to be NULL */
1704 U8 converted = toLOWER_LATIN1(c);
1707 if (UNI_IS_INVARIANT(converted)) {
1712 *p = UTF8_TWO_BYTE_HI(converted);
1713 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1721 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1725 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1728 return to_lower_latin1((U8) c, p, lenp);
1731 uvchr_to_utf8(p, c);
1732 return CALL_LOWER_CASE(p, p, lenp);
1736 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
1738 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
1743 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1745 if (c == MICRO_SIGN) {
1746 converted = GREEK_SMALL_LETTER_MU;
1748 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1754 else { /* In this range the fold of all other characters is their lower
1756 converted = toLOWER_LATIN1(c);
1759 if (UNI_IS_INVARIANT(converted)) {
1760 *p = (U8) converted;
1764 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1765 *p = UTF8_TWO_BYTE_LO(converted);
1773 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
1776 /* Not currently externally documented, and subject to change
1777 * <flags> bits meanings:
1778 * FOLD_FLAGS_FULL iff full folding is to be used;
1779 * FOLD_FLAGS_LOCALE iff in locale
1780 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1783 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1786 UV result = _to_fold_latin1((U8) c, p, lenp,
1787 cBOOL(((flags & FOLD_FLAGS_FULL)
1788 /* If ASCII-safe, don't allow full folding,
1789 * as that could include SHARP S => ss;
1790 * otherwise there is no crossing of
1791 * ascii/non-ascii in the latin1 range */
1792 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1793 /* It is illegal for the fold to cross the 255/256 boundary under
1794 * locale; in this case return the original */
1795 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1800 /* If no special needs, just use the macro */
1801 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1802 uvchr_to_utf8(p, c);
1803 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1805 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1806 the special flags. */
1807 U8 utf8_c[UTF8_MAXBYTES + 1];
1808 uvchr_to_utf8(utf8_c, c);
1809 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1813 /* for now these all assume no locale info available for Unicode > 255; and
1814 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1815 * called instead, so that these don't get called for < 255 */
1818 Perl_is_uni_alnum_lc(pTHX_ UV c)
1820 return is_uni_alnum(c); /* XXX no locale support yet */
1824 Perl_is_uni_idfirst_lc(pTHX_ UV c)
1826 return is_uni_idfirst(c); /* XXX no locale support yet */
1830 Perl_is_uni_alpha_lc(pTHX_ UV c)
1832 return is_uni_alpha(c); /* XXX no locale support yet */
1836 Perl_is_uni_ascii_lc(pTHX_ UV c)
1838 return is_uni_ascii(c); /* XXX no locale support yet */
1842 Perl_is_uni_blank_lc(pTHX_ UV c)
1844 return is_uni_blank(c); /* XXX no locale support yet */
1848 Perl_is_uni_space_lc(pTHX_ UV c)
1850 return is_uni_space(c); /* XXX no locale support yet */
1854 Perl_is_uni_digit_lc(pTHX_ UV c)
1856 return is_uni_digit(c); /* XXX no locale support yet */
1860 Perl_is_uni_upper_lc(pTHX_ UV c)
1862 return is_uni_upper(c); /* XXX no locale support yet */
1866 Perl_is_uni_lower_lc(pTHX_ UV c)
1868 return is_uni_lower(c); /* XXX no locale support yet */
1872 Perl_is_uni_cntrl_lc(pTHX_ UV c)
1874 return is_uni_cntrl(c); /* XXX no locale support yet */
1878 Perl_is_uni_graph_lc(pTHX_ UV c)
1880 return is_uni_graph(c); /* XXX no locale support yet */
1884 Perl_is_uni_print_lc(pTHX_ UV c)
1886 return is_uni_print(c); /* XXX no locale support yet */
1890 Perl_is_uni_punct_lc(pTHX_ UV c)
1892 return is_uni_punct(c); /* XXX no locale support yet */
1896 Perl_is_uni_xdigit_lc(pTHX_ UV c)
1898 return is_uni_xdigit(c); /* XXX no locale support yet */
1902 Perl_to_uni_upper_lc(pTHX_ U32 c)
1904 /* XXX returns only the first character -- do not use XXX */
1905 /* XXX no locale support yet */
1907 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1908 return (U32)to_uni_upper(c, tmpbuf, &len);
1912 Perl_to_uni_title_lc(pTHX_ U32 c)
1914 /* XXX returns only the first character XXX -- do not use XXX */
1915 /* XXX no locale support yet */
1917 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1918 return (U32)to_uni_title(c, tmpbuf, &len);
1922 Perl_to_uni_lower_lc(pTHX_ U32 c)
1924 /* XXX returns only the first character -- do not use XXX */
1925 /* XXX no locale support yet */
1927 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1928 return (U32)to_uni_lower(c, tmpbuf, &len);
1932 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1933 const char *const swashname)
1935 /* returns a boolean giving whether or not the UTF8-encoded character that
1936 * starts at <p> is in the swash indicated by <swashname>. <swash>
1937 * contains a pointer to where the swash indicated by <swashname>
1938 * is to be stored; which this routine will do, so that future calls will
1939 * look at <*swash> and only generate a swash if it is not null
1941 * Note that it is assumed that the buffer length of <p> is enough to
1942 * contain all the bytes that comprise the character. Thus, <*p> should
1943 * have been checked before this call for mal-formedness enough to assure
1948 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1950 /* The API should have included a length for the UTF-8 character in <p>,
1951 * but it doesn't. We therefor assume that p has been validated at least
1952 * as far as there being enough bytes available in it to accommodate the
1953 * character without reading beyond the end, and pass that number on to the
1954 * validating routine */
1955 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
1958 *swash = swash_init("utf8", swashname, &PL_sv_undef, 1, 0);
1959 return swash_fetch(*swash, p, TRUE) != 0;
1963 Perl_is_utf8_alnum(pTHX_ const U8 *p)
1967 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1969 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1970 * descendant of isalnum(3), in other words, it doesn't
1971 * contain the '_'. --jhi */
1972 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
1976 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
1980 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1984 /* is_utf8_idstart would be more logical. */
1985 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1989 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1993 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1997 /* is_utf8_idstart would be more logical. */
1998 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2002 Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2006 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2008 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2012 Perl_is_utf8_idcont(pTHX_ const U8 *p)
2016 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2018 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
2022 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2026 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2028 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2032 Perl_is_utf8_alpha(pTHX_ const U8 *p)
2036 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2038 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
2042 Perl_is_utf8_ascii(pTHX_ const U8 *p)
2046 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2048 /* ASCII characters are the same whether in utf8 or not. So the macro
2049 * works on both utf8 and non-utf8 representations. */
2054 Perl_is_utf8_blank(pTHX_ const U8 *p)
2058 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2060 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2064 Perl_is_utf8_space(pTHX_ const U8 *p)
2068 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2070 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
2074 Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2078 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2080 /* Only true if is an ASCII space-like character, and ASCII is invariant
2081 * under utf8, so can just use the macro */
2082 return isSPACE_A(*p);
2086 Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2090 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2092 /* Only true if is an ASCII word character, and ASCII is invariant
2093 * under utf8, so can just use the macro */
2094 return isWORDCHAR_A(*p);
2098 Perl_is_utf8_digit(pTHX_ const U8 *p)
2102 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2104 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
2108 Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2112 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2114 /* Only true if is an ASCII digit character, and ASCII is invariant
2115 * under utf8, so can just use the macro */
2116 return isDIGIT_A(*p);
2120 Perl_is_utf8_upper(pTHX_ const U8 *p)
2124 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2126 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
2130 Perl_is_utf8_lower(pTHX_ const U8 *p)
2134 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2136 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
2140 Perl_is_utf8_cntrl(pTHX_ const U8 *p)
2144 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2147 return isCNTRL_A(*p);
2150 /* All controls are in Latin1 */
2151 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2154 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2158 Perl_is_utf8_graph(pTHX_ const U8 *p)
2162 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2164 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
2168 Perl_is_utf8_print(pTHX_ const U8 *p)
2172 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2174 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
2178 Perl_is_utf8_punct(pTHX_ const U8 *p)
2182 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2184 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
2188 Perl_is_utf8_xdigit(pTHX_ const U8 *p)
2192 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2194 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
2198 Perl_is_utf8_mark(pTHX_ const U8 *p)
2202 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2204 return is_utf8_common(p, &PL_utf8_mark, "IsM");
2208 Perl_is_utf8_X_begin(pTHX_ const U8 *p)
2212 PERL_ARGS_ASSERT_IS_UTF8_X_BEGIN;
2214 return is_utf8_common(p, &PL_utf8_X_begin, "_X_Begin");
2218 Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2222 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2224 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2228 Perl_is_utf8_X_prepend(pTHX_ const U8 *p)
2232 PERL_ARGS_ASSERT_IS_UTF8_X_PREPEND;
2234 return is_utf8_common(p, &PL_utf8_X_prepend, "_X_GCB_Prepend");
2238 Perl_is_utf8_X_non_hangul(pTHX_ const U8 *p)
2242 PERL_ARGS_ASSERT_IS_UTF8_X_NON_HANGUL;
2244 return is_utf8_common(p, &PL_utf8_X_non_hangul, "_X_HST_Not_Applicable");
2248 Perl_is_utf8_X_L(pTHX_ const U8 *p)
2252 PERL_ARGS_ASSERT_IS_UTF8_X_L;
2254 return is_utf8_common(p, &PL_utf8_X_L, "_X_GCB_L");
2257 /* These constants are for finding GCB=LV and GCB=LVT. These are for the
2258 * pre-composed Hangul syllables, which are all in a contiguous block and
2259 * arranged there in such a way so as to facilitate alorithmic determination of
2260 * their characteristics. As such, they don't need a swash, but can be
2261 * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
2263 #define SBASE 0xAC00 /* Start of block */
2264 #define SCount 11172 /* Length of block */
2267 #if 0 /* This routine is not currently used */
2269 Perl_is_utf8_X_LV(pTHX_ const U8 *p)
2271 /* Unlike most other similarly named routines here, this does not create a
2272 * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
2276 UV cp = valid_utf8_to_uvchr(p, NULL);
2278 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
2280 /* The earliest Unicode releases did not have these precomposed Hangul
2281 * syllables. Set to point to undef in that case, so will return false on
2283 if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
2284 PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
2285 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
2286 SvREFCNT_dec(PL_utf8_X_LV);
2287 PL_utf8_X_LV = &PL_sv_undef;
2291 return (PL_utf8_X_LV != &PL_sv_undef
2292 && cp >= SBASE && cp < SBASE + SCount
2293 && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
2298 Perl_is_utf8_X_LVT(pTHX_ const U8 *p)
2300 /* Unlike most other similarly named routines here, this does not create a
2301 * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
2305 UV cp = valid_utf8_to_uvchr(p, NULL);
2307 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
2309 /* The earliest Unicode releases did not have these precomposed Hangul
2310 * syllables. Set to point to undef in that case, so will return false on
2312 if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
2313 PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
2314 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
2315 SvREFCNT_dec(PL_utf8_X_LVT);
2316 PL_utf8_X_LVT = &PL_sv_undef;
2320 return (PL_utf8_X_LVT != &PL_sv_undef
2321 && cp >= SBASE && cp < SBASE + SCount
2322 && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
2326 Perl_is_utf8_X_T(pTHX_ const U8 *p)
2330 PERL_ARGS_ASSERT_IS_UTF8_X_T;
2332 return is_utf8_common(p, &PL_utf8_X_T, "_X_GCB_T");
2336 Perl_is_utf8_X_V(pTHX_ const U8 *p)
2340 PERL_ARGS_ASSERT_IS_UTF8_X_V;
2342 return is_utf8_common(p, &PL_utf8_X_V, "_X_GCB_V");
2346 Perl_is_utf8_X_LV_LVT_V(pTHX_ const U8 *p)
2350 PERL_ARGS_ASSERT_IS_UTF8_X_LV_LVT_V;
2352 return is_utf8_common(p, &PL_utf8_X_LV_LVT_V, "_X_LV_LVT_V");
2356 Perl__is_utf8_quotemeta(pTHX_ const U8 *p)
2358 /* For exclusive use of pp_quotemeta() */
2362 PERL_ARGS_ASSERT__IS_UTF8_QUOTEMETA;
2364 return is_utf8_common(p, &PL_utf8_quotemeta, "_Perl_Quotemeta");
2368 =for apidoc to_utf8_case
2370 The C<p> contains the pointer to the UTF-8 string encoding
2371 the character that is being converted. This routine assumes that the character
2372 at C<p> is well-formed.
2374 The C<ustrp> is a pointer to the character buffer to put the
2375 conversion result to. The C<lenp> is a pointer to the length
2378 The C<swashp> is a pointer to the swash to use.
2380 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2381 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
2382 but not always, a multicharacter mapping), is tried first.
2384 The C<special> is a string like "utf8::ToSpecLower", which means the
2385 hash %utf8::ToSpecLower. The access to the hash is through
2386 Perl_to_utf8_case().
2388 The C<normal> is a string like "ToLower" which means the swash
2394 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2395 SV **swashp, const char *normal, const char *special)
2398 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2400 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
2401 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2402 * are necessary in EBCDIC, they are redundant no-ops
2403 * in ASCII-ish platforms, and hopefully optimized away. */
2404 const UV uv1 = NATIVE_TO_UNI(uv0);
2406 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2408 /* Note that swash_fetch() doesn't output warnings for these because it
2409 * assumes we will */
2410 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2411 if (uv1 <= UNICODE_SURROGATE_LAST) {
2412 if (ckWARN_d(WARN_SURROGATE)) {
2413 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2414 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2415 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2418 else if (UNICODE_IS_SUPER(uv1)) {
2419 if (ckWARN_d(WARN_NON_UNICODE)) {
2420 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2421 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2422 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2426 /* Note that non-characters are perfectly legal, so no warning should
2430 uvuni_to_utf8(tmpbuf, uv1);
2432 if (!*swashp) /* load on-demand */
2433 *swashp = swash_init("utf8", normal, &PL_sv_undef, 4, 0);
2436 /* It might be "special" (sometimes, but not always,
2437 * a multicharacter mapping) */
2438 HV * const hv = get_hv(special, 0);
2442 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2446 s = SvPV_const(*svp, len);
2448 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2451 /* If we have EBCDIC we need to remap the characters
2452 * since any characters in the low 256 are Unicode
2453 * code points, not EBCDIC. */
2454 U8 *t = (U8*)s, *tend = t + len, *d;
2461 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2463 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2472 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2477 Copy(tmpbuf, ustrp, len, U8);
2479 Copy(s, ustrp, len, U8);
2485 if (!len && *swashp) {
2486 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
2489 /* It was "normal" (a single character mapping). */
2490 const UV uv3 = UNI_TO_NATIVE(uv2);
2491 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2499 return valid_utf8_to_uvchr(ustrp, 0);
2502 /* Here, there was no mapping defined, which means that the code point maps
2503 * to itself. Return the inputs */
2505 Copy(p, ustrp, len, U8);
2515 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2517 /* This is called when changing the case of a utf8-encoded character above
2518 * the Latin1 range, and the operation is in locale. If the result
2519 * contains a character that crosses the 255/256 boundary, disallow the
2520 * change, and return the original code point. See L<perlfunc/lc> for why;
2522 * p points to the original string whose case was changed; assumed
2523 * by this routine to be well-formed
2524 * result the code point of the first character in the changed-case string
2525 * ustrp points to the changed-case string (<result> represents its first char)
2526 * lenp points to the length of <ustrp> */
2528 UV original; /* To store the first code point of <p> */
2530 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2532 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2534 /* We know immediately if the first character in the string crosses the
2535 * boundary, so can skip */
2538 /* Look at every character in the result; if any cross the
2539 * boundary, the whole thing is disallowed */
2540 U8* s = ustrp + UTF8SKIP(ustrp);
2541 U8* e = ustrp + *lenp;
2543 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2550 /* Here, no characters crossed, result is ok as-is */
2556 /* Failed, have to return the original */
2557 original = valid_utf8_to_uvchr(p, lenp);
2558 Copy(p, ustrp, *lenp, char);
2563 =for apidoc to_utf8_upper
2565 Convert the UTF-8 encoded character at C<p> to its uppercase version and
2566 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2567 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2568 the uppercase version may be longer than the original character.
2570 The first character of the uppercased version is returned
2571 (but note, as explained above, that there may be more.)
2573 The character at C<p> is assumed by this routine to be well-formed.
2577 /* Not currently externally documented, and subject to change:
2578 * <flags> is set iff locale semantics are to be used for code points < 256
2579 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2580 * were used in the calculation; otherwise unchanged. */
2583 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2589 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2591 if (UTF8_IS_INVARIANT(*p)) {
2593 result = toUPPER_LC(*p);
2596 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2599 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2601 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2604 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2608 else { /* utf8, ord above 255 */
2609 result = CALL_UPPER_CASE(p, ustrp, lenp);
2612 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2617 /* Here, used locale rules. Convert back to utf8 */
2618 if (UTF8_IS_INVARIANT(result)) {
2619 *ustrp = (U8) result;
2623 *ustrp = UTF8_EIGHT_BIT_HI(result);
2624 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2629 *tainted_ptr = TRUE;
2635 =for apidoc to_utf8_title
2637 Convert the UTF-8 encoded character at C<p> to its titlecase version and
2638 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2639 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2640 titlecase version may be longer than the original character.
2642 The first character of the titlecased version is returned
2643 (but note, as explained above, that there may be more.)
2645 The character at C<p> is assumed by this routine to be well-formed.
2649 /* Not currently externally documented, and subject to change:
2650 * <flags> is set iff locale semantics are to be used for code points < 256
2651 * Since titlecase is not defined in POSIX, uppercase is used instead
2653 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2654 * were used in the calculation; otherwise unchanged. */
2657 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2663 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2665 if (UTF8_IS_INVARIANT(*p)) {
2667 result = toUPPER_LC(*p);
2670 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2673 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2675 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2678 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2682 else { /* utf8, ord above 255 */
2683 result = CALL_TITLE_CASE(p, ustrp, lenp);
2686 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2691 /* Here, used locale rules. Convert back to utf8 */
2692 if (UTF8_IS_INVARIANT(result)) {
2693 *ustrp = (U8) result;
2697 *ustrp = UTF8_EIGHT_BIT_HI(result);
2698 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2703 *tainted_ptr = TRUE;
2709 =for apidoc to_utf8_lower
2711 Convert the UTF-8 encoded character at C<p> to its lowercase version and
2712 store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2713 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2714 lowercase version may be longer than the original character.
2716 The first character of the lowercased version is returned
2717 (but note, as explained above, that there may be more.)
2719 The character at C<p> is assumed by this routine to be well-formed.
2723 /* Not currently externally documented, and subject to change:
2724 * <flags> is set iff locale semantics are to be used for code points < 256
2725 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2726 * were used in the calculation; otherwise unchanged. */
2729 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2735 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2737 if (UTF8_IS_INVARIANT(*p)) {
2739 result = toLOWER_LC(*p);
2742 return to_lower_latin1(*p, ustrp, lenp);
2745 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2747 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2750 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2754 else { /* utf8, ord above 255 */
2755 result = CALL_LOWER_CASE(p, ustrp, lenp);
2758 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2764 /* Here, used locale rules. Convert back to utf8 */
2765 if (UTF8_IS_INVARIANT(result)) {
2766 *ustrp = (U8) result;
2770 *ustrp = UTF8_EIGHT_BIT_HI(result);
2771 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2776 *tainted_ptr = TRUE;
2782 =for apidoc to_utf8_fold
2784 Convert the UTF-8 encoded character at C<p> to its foldcase version and
2785 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2786 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2787 foldcase version may be longer than the original character (up to
2790 The first character of the foldcased version is returned
2791 (but note, as explained above, that there may be more.)
2793 The character at C<p> is assumed by this routine to be well-formed.
2797 /* Not currently externally documented, and subject to change,
2799 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2800 * points < 256. Since foldcase is not defined in
2801 * POSIX, lowercase is used instead
2802 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2803 * otherwise simple folds
2804 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2806 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2807 * were used in the calculation; otherwise unchanged. */
2810 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2816 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2818 /* These are mutually exclusive */
2819 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2821 assert(p != ustrp); /* Otherwise overwrites */
2823 if (UTF8_IS_INVARIANT(*p)) {
2824 if (flags & FOLD_FLAGS_LOCALE) {
2825 result = toLOWER_LC(*p);
2828 return _to_fold_latin1(*p, ustrp, lenp,
2829 cBOOL(flags & FOLD_FLAGS_FULL));
2832 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2833 if (flags & FOLD_FLAGS_LOCALE) {
2834 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2837 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2839 cBOOL((flags & FOLD_FLAGS_FULL
2840 /* If ASCII safe, don't allow full
2841 * folding, as that could include SHARP
2842 * S => ss; otherwise there is no
2843 * crossing of ascii/non-ascii in the
2845 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
2848 else { /* utf8, ord above 255 */
2849 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2851 if ((flags & FOLD_FLAGS_LOCALE)) {
2852 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2854 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2858 /* This is called when changing the case of a utf8-encoded
2859 * character above the Latin1 range, and the result should not
2860 * contain an ASCII character. */
2862 UV original; /* To store the first code point of <p> */
2864 /* Look at every character in the result; if any cross the
2865 * boundary, the whole thing is disallowed */
2867 U8* e = ustrp + *lenp;
2870 /* Crossed, have to return the original */
2871 original = valid_utf8_to_uvchr(p, lenp);
2872 Copy(p, ustrp, *lenp, char);
2878 /* Here, no characters crossed, result is ok as-is */
2883 /* Here, used locale rules. Convert back to utf8 */
2884 if (UTF8_IS_INVARIANT(result)) {
2885 *ustrp = (U8) result;
2889 *ustrp = UTF8_EIGHT_BIT_HI(result);
2890 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2895 *tainted_ptr = TRUE;
2901 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2902 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2903 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2907 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2909 PERL_ARGS_ASSERT_SWASH_INIT;
2911 /* Returns a copy of a swash initiated by the called function. This is the
2912 * public interface, and returning a copy prevents others from doing
2913 * mischief on the original */
2915 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, FALSE, NULL, NULL));
2919 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, bool return_if_undef, SV* invlist, U8* const flags_p)
2921 /* Initialize and return a swash, creating it if necessary. It does this
2922 * by calling utf8_heavy.pl in the general case.
2924 * This interface should only be used by functions that won't destroy or
2925 * adversely change the swash, as doing so affects all other uses of the
2926 * swash in the program; the general public should use 'Perl_swash_init'
2929 * pkg is the name of the package that <name> should be in.
2930 * name is the name of the swash to find. Typically it is a Unicode
2931 * property name, including user-defined ones
2932 * listsv is a string to initialize the swash with. It must be of the form
2933 * documented as the subroutine return value in
2934 * L<perlunicode/User-Defined Character Properties>
2935 * minbits is the number of bits required to represent each data element.
2936 * It is '1' for binary properties.
2937 * none I (khw) do not understand this one, but it is used only in tr///.
2938 * return_if_undef is TRUE if the routine shouldn't croak if it can't find
2939 * the requested property
2940 * invlist is an inversion list to initialize the swash with (or NULL)
2941 * flags_p if non-NULL is the address of various input and output flag bits
2942 * to the routine, as follows: ('I' means is input to the routine;
2943 * 'O' means output from the routine. Only flags marked O are
2944 * meaningful on return.)
2945 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2946 * came from a user-defined property. (I O)
2948 * Thus there are three possible inputs to find the swash: <name>,
2949 * <listsv>, and <invlist>. At least one must be specified. The result
2950 * will be the union of the specified ones, although <listsv>'s various
2951 * actions can intersect, etc. what <name> gives.
2953 * <invlist> is only valid for binary properties */
2956 SV* retval = &PL_sv_undef;
2957 HV* swash_hv = NULL;
2959 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2960 assert(! invlist || minbits == 1);
2962 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2964 if (listsv != &PL_sv_undef || strNE(name, "")) {
2966 const size_t pkg_len = strlen(pkg);
2967 const size_t name_len = strlen(name);
2968 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2972 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2974 PUSHSTACKi(PERLSI_MAGIC);
2978 if (PL_parser && PL_parser->error_count)
2979 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2980 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2981 if (!method) { /* demand load utf8 */
2983 errsv_save = newSVsv(ERRSV);
2984 /* It is assumed that callers of this routine are not passing in
2985 * any user derived data. */
2986 /* Need to do this after save_re_context() as it will set
2987 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2988 * in Perl_magic_get). Even line to create errsv_save can turn on
2990 SAVEBOOL(PL_tainted);
2992 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2995 sv_setsv(ERRSV, errsv_save);
2996 SvREFCNT_dec(errsv_save);
3002 mPUSHp(pkg, pkg_len);
3003 mPUSHp(name, name_len);
3008 errsv_save = newSVsv(ERRSV);
3009 /* If we already have a pointer to the method, no need to use
3010 * call_method() to repeat the lookup. */
3011 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
3012 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3014 retval = *PL_stack_sp--;
3015 SvREFCNT_inc(retval);
3018 sv_setsv(ERRSV, errsv_save);
3019 SvREFCNT_dec(errsv_save);
3022 if (IN_PERL_COMPILETIME) {
3023 CopHINTS_set(PL_curcop, PL_hints);
3025 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3028 /* If caller wants to handle missing properties, let them */
3029 if (return_if_undef) {
3033 "Can't find Unicode property definition \"%"SVf"\"",
3035 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3037 } /* End of calling the module to find the swash */
3039 /* If this operation fetched a swash, and we will need it later, get it */
3040 if (retval != &PL_sv_undef
3041 && (minbits == 1 || (flags_p
3043 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3045 swash_hv = MUTABLE_HV(SvRV(retval));
3047 /* If we don't already know that there is a user-defined component to
3048 * this swash, and the user has indicated they wish to know if there is
3049 * one (by passing <flags_p>), find out */
3050 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3051 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3052 if (user_defined && SvUV(*user_defined)) {
3053 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3058 /* Make sure there is an inversion list for binary properties */
3060 SV** swash_invlistsvp = NULL;
3061 SV* swash_invlist = NULL;
3062 bool invlist_in_swash_is_valid = FALSE;
3064 /* If this operation fetched a swash, get its already existing
3065 * inversion list, or create one for it */
3068 swash_invlistsvp = hv_fetchs(swash_hv, "I", FALSE);
3069 if (swash_invlistsvp) {
3070 swash_invlist = *swash_invlistsvp;
3071 invlist_in_swash_is_valid = TRUE;
3074 swash_invlist = _swash_to_invlist(retval);
3078 /* If an inversion list was passed in, have to include it */
3081 /* Any fetched swash will by now have an inversion list in it;
3082 * otherwise <swash_invlist> will be NULL, indicating that we
3083 * didn't fetch a swash */
3084 if (swash_invlist) {
3086 /* Add the passed-in inversion list, which invalidates the one
3087 * already stored in the swash */
3088 invlist_in_swash_is_valid = FALSE;
3089 _invlist_union(invlist, swash_invlist, &swash_invlist);
3093 /* Here, there is no swash already. Set up a minimal one */
3095 retval = newRV_inc(MUTABLE_SV(swash_hv));
3096 swash_invlist = invlist;
3100 /* Here, we have computed the union of all the passed-in data. It may
3101 * be that there was an inversion list in the swash which didn't get
3102 * touched; otherwise save the one computed one */
3103 if (! invlist_in_swash_is_valid) {
3104 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "INVLIST", swash_invlist))
3106 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3115 /* This API is wrong for special case conversions since we may need to
3116 * return several Unicode characters for a single Unicode character
3117 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3118 * the lower-level routine, and it is similarly broken for returning
3119 * multiple values. --jhi
3120 * For those, you should use to_utf8_case() instead */
3121 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3124 * Returns the value of property/mapping C<swash> for the first character
3125 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3126 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3127 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3129 * A "swash" is a hash which contains initially the keys/values set up by
3130 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3131 * property for all possible code points. Things are stored in a compact form
3132 * (see utf8_heavy.pl) so that calculation is required to find the actual
3133 * property value for a given code point. As code points are looked up, new
3134 * key/value pairs are added to the hash, so that the calculation doesn't have
3135 * to ever be re-done. Further, each calculation is done, not just for the
3136 * desired one, but for a whole block of code points adjacent to that one.
3137 * For binary properties on ASCII machines, the block is usually for 64 code
3138 * points, starting with a code point evenly divisible by 64. Thus if the
3139 * property value for code point 257 is requested, the code goes out and
3140 * calculates the property values for all 64 code points between 256 and 319,
3141 * and stores these as a single 64-bit long bit vector, called a "swatch",
3142 * under the key for code point 256. The key is the UTF-8 encoding for code
3143 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3144 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3145 * for code point 258 is then requested, this code realizes that it would be
3146 * stored under the key for 256, and would find that value and extract the
3147 * relevant bit, offset from 256.
3149 * Non-binary properties are stored in as many bits as necessary to represent
3150 * their values (32 currently, though the code is more general than that), not
3151 * as single bits, but the principal is the same: the value for each key is a
3152 * vector that encompasses the property values for all code points whose UTF-8
3153 * representations are represented by the key. That is, for all code points
3154 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3158 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3161 HV *const hv = MUTABLE_HV(SvRV(swash));
3166 const U8 *tmps = NULL;
3170 const UV c = NATIVE_TO_ASCII(*ptr);
3172 PERL_ARGS_ASSERT_SWASH_FETCH;
3174 /* Convert to utf8 if not already */
3175 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
3176 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3177 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3180 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
3181 * then the "swatch" is a vec() for all the chars which start
3183 * So the key in the hash (klen) is length of encoded char -1
3185 klen = UTF8SKIP(ptr) - 1;
3189 /* If char is invariant then swatch is for all the invariant chars
3190 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
3192 needents = UTF_CONTINUATION_MARK;
3193 off = NATIVE_TO_UTF(ptr[klen]);
3196 /* If char is encoded then swatch is for the prefix */
3197 needents = (1 << UTF_ACCUMULATION_SHIFT);
3198 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3202 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3203 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3204 * it's nothing to sniff at.) Pity we usually come through at least
3205 * two function calls to get here...
3207 * NB: this code assumes that swatches are never modified, once generated!
3210 if (hv == PL_last_swash_hv &&
3211 klen == PL_last_swash_klen &&
3212 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3214 tmps = PL_last_swash_tmps;
3215 slen = PL_last_swash_slen;
3218 /* Try our second-level swatch cache, kept in a hash. */
3219 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3221 /* If not cached, generate it via swatch_get */
3222 if (!svp || !SvPOK(*svp)
3223 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
3224 /* We use utf8n_to_uvuni() as we want an index into
3225 Unicode tables, not a native character number.
3227 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
3229 0 : UTF8_ALLOW_ANY);
3230 swatch = swatch_get(swash,
3231 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
3232 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
3235 if (IN_PERL_COMPILETIME)
3236 CopHINTS_set(PL_curcop, PL_hints);
3238 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3240 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3241 || (slen << 3) < needents)
3242 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3243 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3244 svp, tmps, (UV)slen, (UV)needents);
3247 PL_last_swash_hv = hv;
3248 assert(klen <= sizeof(PL_last_swash_key));
3249 PL_last_swash_klen = (U8)klen;
3250 /* FIXME change interpvar.h? */
3251 PL_last_swash_tmps = (U8 *) tmps;
3252 PL_last_swash_slen = slen;
3254 Copy(ptr, PL_last_swash_key, klen, U8);
3257 switch ((int)((slen << 3) / needents)) {
3259 bit = 1 << (off & 7);
3261 return (tmps[off] & bit) != 0;
3266 return (tmps[off] << 8) + tmps[off + 1] ;
3269 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3271 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3272 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
3273 NORETURN_FUNCTION_END;
3276 /* Read a single line of the main body of the swash input text. These are of
3279 * where each number is hex. The first two numbers form the minimum and
3280 * maximum of a range, and the third is the value associated with the range.
3281 * Not all swashes should have a third number
3283 * On input: l points to the beginning of the line to be examined; it points
3284 * to somewhere in the string of the whole input text, and is
3285 * terminated by a \n or the null string terminator.
3286 * lend points to the null terminator of that string
3287 * wants_value is non-zero if the swash expects a third number
3288 * typestr is the name of the swash's mapping, like 'ToLower'
3289 * On output: *min, *max, and *val are set to the values read from the line.
3290 * returns a pointer just beyond the line examined. If there was no
3291 * valid min number on the line, returns lend+1
3295 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3296 const bool wants_value, const U8* const typestr)
3298 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3299 STRLEN numlen; /* Length of the number */
3300 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3301 | PERL_SCAN_DISALLOW_PREFIX
3302 | PERL_SCAN_SILENT_NON_PORTABLE;
3304 /* nl points to the next \n in the scan */
3305 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3307 /* Get the first number on the line: the range minimum */
3309 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3310 if (numlen) /* If found a hex number, position past it */
3312 else if (nl) { /* Else, go handle next line, if any */
3313 return nl + 1; /* 1 is length of "\n" */
3315 else { /* Else, no next line */
3316 return lend + 1; /* to LIST's end at which \n is not found */
3319 /* The max range value follows, separated by a BLANK */
3322 flags = PERL_SCAN_SILENT_ILLDIGIT
3323 | PERL_SCAN_DISALLOW_PREFIX
3324 | PERL_SCAN_SILENT_NON_PORTABLE;
3326 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3329 else /* If no value here, it is a single element range */
3332 /* Non-binary tables have a third entry: what the first element of the
3338 /* The ToLc, etc table mappings are not in hex, and must be
3339 * corrected by adding the code point to them */
3341 char *after_strtol = (char *) lend;
3342 *val = Strtol((char *)l, &after_strtol, 10);
3343 l = (U8 *) after_strtol;
3345 else { /* Other tables are in hex, and are the correct result
3347 flags = PERL_SCAN_SILENT_ILLDIGIT
3348 | PERL_SCAN_DISALLOW_PREFIX
3349 | PERL_SCAN_SILENT_NON_PORTABLE;
3351 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3361 /* diag_listed_as: To%s: illegal mapping '%s' */
3362 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3368 *val = 0; /* bits == 1, then any val should be ignored */
3370 else { /* Nothing following range min, should be single element with no
3376 /* diag_listed_as: To%s: illegal mapping '%s' */
3377 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3381 *val = 0; /* bits == 1, then val should be ignored */
3384 /* Position to next line if any, or EOF */
3394 * Returns a swatch (a bit vector string) for a code point sequence
3395 * that starts from the value C<start> and comprises the number C<span>.
3396 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3397 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3400 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3403 U8 *l, *lend, *x, *xend, *s, *send;
3404 STRLEN lcur, xcur, scur;
3405 HV *const hv = MUTABLE_HV(SvRV(swash));
3406 SV** const invlistsvp = hv_fetchs(hv, "INVLIST", FALSE);
3408 SV** listsvp = NULL; /* The string containing the main body of the table */
3409 SV** extssvp = NULL;
3410 SV** invert_it_svp = NULL;
3413 STRLEN octets; /* if bits == 1, then octets == 0 */
3415 UV end = start + span;
3417 if (invlistsvp == NULL) {
3418 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3419 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3420 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3421 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3422 listsvp = hv_fetchs(hv, "LIST", FALSE);
3423 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3425 bits = SvUV(*bitssvp);
3426 none = SvUV(*nonesvp);
3427 typestr = (U8*)SvPV_nolen(*typesvp);
3433 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3435 PERL_ARGS_ASSERT_SWATCH_GET;
3437 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3438 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3442 /* If overflowed, use the max possible */
3448 /* create and initialize $swatch */
3449 scur = octets ? (span * octets) : (span + 7) / 8;
3450 swatch = newSV(scur);
3452 s = (U8*)SvPVX(swatch);
3453 if (octets && none) {
3454 const U8* const e = s + scur;
3457 *s++ = (U8)(none & 0xff);
3458 else if (bits == 16) {
3459 *s++ = (U8)((none >> 8) & 0xff);
3460 *s++ = (U8)( none & 0xff);
3462 else if (bits == 32) {
3463 *s++ = (U8)((none >> 24) & 0xff);
3464 *s++ = (U8)((none >> 16) & 0xff);
3465 *s++ = (U8)((none >> 8) & 0xff);
3466 *s++ = (U8)( none & 0xff);
3472 (void)memzero((U8*)s, scur + 1);
3474 SvCUR_set(swatch, scur);
3475 s = (U8*)SvPVX(swatch);
3477 if (invlistsvp) { /* If has an inversion list set up use that */
3478 _invlist_populate_swatch(*invlistsvp, start, end, s);
3482 /* read $swash->{LIST} */
3483 l = (U8*)SvPV(*listsvp, lcur);
3486 UV min, max, val, upper;
3487 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3488 cBOOL(octets), typestr);
3493 /* If looking for something beyond this range, go try the next one */
3497 /* <end> is generally 1 beyond where we want to set things, but at the
3498 * platform's infinity, where we can't go any higher, we want to
3499 * include the code point at <end> */
3502 : (max != UV_MAX || end != UV_MAX)
3509 if (!none || val < none) {
3514 for (key = min; key <= upper; key++) {
3516 /* offset must be non-negative (start <= min <= key < end) */
3517 offset = octets * (key - start);
3519 s[offset] = (U8)(val & 0xff);
3520 else if (bits == 16) {
3521 s[offset ] = (U8)((val >> 8) & 0xff);
3522 s[offset + 1] = (U8)( val & 0xff);
3524 else if (bits == 32) {
3525 s[offset ] = (U8)((val >> 24) & 0xff);
3526 s[offset + 1] = (U8)((val >> 16) & 0xff);
3527 s[offset + 2] = (U8)((val >> 8) & 0xff);
3528 s[offset + 3] = (U8)( val & 0xff);
3531 if (!none || val < none)
3535 else { /* bits == 1, then val should be ignored */
3540 for (key = min; key <= upper; key++) {
3541 const STRLEN offset = (STRLEN)(key - start);
3542 s[offset >> 3] |= 1 << (offset & 7);
3547 /* Invert if the data says it should be. Assumes that bits == 1 */
3548 if (invert_it_svp && SvUV(*invert_it_svp)) {
3550 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3551 * be 0, and their inversion should also be 0, as we don't succeed any
3552 * Unicode property matches for non-Unicode code points */
3553 if (start <= PERL_UNICODE_MAX) {
3555 /* The code below assumes that we never cross the
3556 * Unicode/above-Unicode boundary in a range, as otherwise we would
3557 * have to figure out where to stop flipping the bits. Since this
3558 * boundary is divisible by a large power of 2, and swatches comes
3559 * in small powers of 2, this should be a valid assumption */
3560 assert(start + span - 1 <= PERL_UNICODE_MAX);
3570 /* read $swash->{EXTRAS}
3571 * This code also copied to swash_to_invlist() below */
3572 x = (U8*)SvPV(*extssvp, xcur);
3580 SV **otherbitssvp, *other;
3584 const U8 opc = *x++;
3588 nl = (U8*)memchr(x, '\n', xend - x);
3590 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3592 x = nl + 1; /* 1 is length of "\n" */
3596 x = xend; /* to EXTRAS' end at which \n is not found */
3603 namelen = nl - namestr;
3607 namelen = xend - namestr;
3611 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3612 otherhv = MUTABLE_HV(SvRV(*othersvp));
3613 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3614 otherbits = (STRLEN)SvUV(*otherbitssvp);
3615 if (bits < otherbits)
3616 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3617 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3619 /* The "other" swatch must be destroyed after. */
3620 other = swatch_get(*othersvp, start, span);
3621 o = (U8*)SvPV(other, olen);
3624 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3626 s = (U8*)SvPV(swatch, slen);
3627 if (bits == 1 && otherbits == 1) {
3629 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3630 "mismatch, slen=%"UVuf", olen=%"UVuf,
3631 (UV)slen, (UV)olen);
3655 STRLEN otheroctets = otherbits >> 3;
3657 U8* const send = s + slen;
3662 if (otherbits == 1) {
3663 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3667 STRLEN vlen = otheroctets;
3675 if (opc == '+' && otherval)
3676 NOOP; /* replace with otherval */
3677 else if (opc == '!' && !otherval)
3679 else if (opc == '-' && otherval)
3681 else if (opc == '&' && !otherval)
3684 s += octets; /* no replacement */
3689 *s++ = (U8)( otherval & 0xff);
3690 else if (bits == 16) {
3691 *s++ = (U8)((otherval >> 8) & 0xff);
3692 *s++ = (U8)( otherval & 0xff);
3694 else if (bits == 32) {
3695 *s++ = (U8)((otherval >> 24) & 0xff);
3696 *s++ = (U8)((otherval >> 16) & 0xff);
3697 *s++ = (U8)((otherval >> 8) & 0xff);
3698 *s++ = (U8)( otherval & 0xff);
3702 sv_free(other); /* through with it! */
3708 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3711 /* Subject to change or removal. For use only in one place in regcomp.c.
3712 * Can't be used on a property that is subject to user override, as it
3713 * relies on the value of SPECIALS in the swash which would be set by
3714 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3715 * for overridden properties
3717 * Returns a hash which is the inversion and closure of a swash mapping.
3718 * For example, consider the input lines:
3723 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3724 * 006C. The value for each key is an array. For 006C, the array would
3725 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3726 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3728 * Essentially, for any code point, it gives all the code points that map to
3729 * it, or the list of 'froms' for that point.
3731 * Currently it ignores any additions or deletions from other swashes,
3732 * looking at just the main body of the swash, and if there are SPECIALS
3733 * in the swash, at that hash
3735 * The specials hash can be extra code points, and most likely consists of
3736 * maps from single code points to multiple ones (each expressed as a string
3737 * of utf8 characters). This function currently returns only 1-1 mappings.
3738 * However consider this possible input in the specials hash:
3739 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3740 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3742 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3743 * currently handle. But it also means that FB05 and FB06 are equivalent in
3744 * a 1-1 mapping which we should handle, and this relationship may not be in
3745 * the main table. Therefore this function examines all the multi-char
3746 * sequences and adds the 1-1 mappings that come out of that. */
3750 HV *const hv = MUTABLE_HV(SvRV(swash));
3752 /* The string containing the main body of the table */
3753 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3755 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3756 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3757 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3758 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3759 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3760 const STRLEN bits = SvUV(*bitssvp);
3761 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3762 const UV none = SvUV(*nonesvp);
3763 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3767 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3769 /* Must have at least 8 bits to get the mappings */
3770 if (bits != 8 && bits != 16 && bits != 32) {
3771 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3775 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3776 mapping to more than one character */
3778 /* Construct an inverse mapping hash for the specials */
3779 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3780 HV * specials_inverse = newHV();
3781 char *char_from; /* the lhs of the map */
3782 I32 from_len; /* its byte length */
3783 char *char_to; /* the rhs of the map */
3784 I32 to_len; /* its byte length */
3785 SV *sv_to; /* and in a sv */
3786 AV* from_list; /* list of things that map to each 'to' */
3788 hv_iterinit(specials_hv);
3790 /* The keys are the characters (in utf8) that map to the corresponding
3791 * utf8 string value. Iterate through the list creating the inverse
3793 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3795 if (! SvPOK(sv_to)) {
3796 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3797 "unexpectedly is not a string, flags=%lu",
3798 (unsigned long)SvFLAGS(sv_to));
3800 /*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)));*/
3802 /* Each key in the inverse list is a mapped-to value, and the key's
3803 * hash value is a list of the strings (each in utf8) that map to
3804 * it. Those strings are all one character long */
3805 if ((listp = hv_fetch(specials_inverse,
3809 from_list = (AV*) *listp;
3811 else { /* No entry yet for it: create one */
3812 from_list = newAV();
3813 if (! hv_store(specials_inverse,
3816 (SV*) from_list, 0))
3818 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3822 /* Here have the list associated with this 'to' (perhaps newly
3823 * created and empty). Just add to it. Note that we ASSUME that
3824 * the input is guaranteed to not have duplications, so we don't
3825 * check for that. Duplications just slow down execution time. */
3826 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3829 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3830 * it looking for cases like the FB05/FB06 examples above. There would
3831 * be an entry in the hash like
3832 * 'st' => [ FB05, FB06 ]
3833 * In this example we will create two lists that get stored in the
3834 * returned hash, 'ret':
3835 * FB05 => [ FB05, FB06 ]
3836 * FB06 => [ FB05, FB06 ]
3838 * Note that there is nothing to do if the array only has one element.
3839 * (In the normal 1-1 case handled below, we don't have to worry about
3840 * two lists, as everything gets tied to the single list that is
3841 * generated for the single character 'to'. But here, we are omitting
3842 * that list, ('st' in the example), so must have multiple lists.) */
3843 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3844 &char_to, &to_len)))
3846 if (av_len(from_list) > 0) {
3849 /* We iterate over all combinations of i,j to place each code
3850 * point on each list */
3851 for (i = 0; i <= av_len(from_list); i++) {
3853 AV* i_list = newAV();
3854 SV** entryp = av_fetch(from_list, i, FALSE);
3855 if (entryp == NULL) {
3856 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3858 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3859 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3861 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3862 (SV*) i_list, FALSE))
3864 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3867 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3868 for (j = 0; j <= av_len(from_list); j++) {
3869 entryp = av_fetch(from_list, j, FALSE);
3870 if (entryp == NULL) {
3871 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3874 /* When i==j this adds itself to the list */
3875 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3876 (U8*) SvPVX(*entryp),
3877 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3879 /*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));*/
3884 SvREFCNT_dec(specials_inverse); /* done with it */
3885 } /* End of specials */
3887 /* read $swash->{LIST} */
3888 l = (U8*)SvPV(*listsvp, lcur);
3891 /* Go through each input line */
3895 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3896 cBOOL(octets), typestr);
3901 /* Each element in the range is to be inverted */
3902 for (inverse = min; inverse <= max; inverse++) {
3906 bool found_key = FALSE;
3907 bool found_inverse = FALSE;
3909 /* The key is the inverse mapping */
3910 char key[UTF8_MAXBYTES+1];
3911 char* key_end = (char *) uvuni_to_utf8((U8*) key, val);
3912 STRLEN key_len = key_end - key;
3914 /* Get the list for the map */
3915 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3916 list = (AV*) *listp;
3918 else { /* No entry yet for it: create one */
3920 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3921 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3925 /* Look through list to see if this inverse mapping already is
3926 * listed, or if there is a mapping to itself already */
3927 for (i = 0; i <= av_len(list); i++) {
3928 SV** entryp = av_fetch(list, i, FALSE);
3930 if (entryp == NULL) {
3931 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3934 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3935 if (SvUV(entry) == val) {
3938 if (SvUV(entry) == inverse) {
3939 found_inverse = TRUE;
3942 /* No need to continue searching if found everything we are
3944 if (found_key && found_inverse) {
3949 /* Make sure there is a mapping to itself on the list */
3951 av_push(list, newSVuv(val));
3952 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3956 /* Simply add the value to the list */
3957 if (! found_inverse) {
3958 av_push(list, newSVuv(inverse));
3959 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3962 /* swatch_get() increments the value of val for each element in the
3963 * range. That makes more compact tables possible. You can
3964 * express the capitalization, for example, of all consecutive
3965 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3966 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3967 * and it's not documented; it appears to be used only in
3968 * implementing tr//; I copied the semantics from swatch_get(), just
3970 if (!none || val < none) {
3980 Perl__swash_to_invlist(pTHX_ SV* const swash)
3983 /* Subject to change or removal. For use only in one place in regcomp.c */
3988 HV *const hv = MUTABLE_HV(SvRV(swash));
3989 UV elements = 0; /* Number of elements in the inversion list */
3992 /* The string containing the main body of the table */
3993 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3994 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3995 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3996 SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3997 SV** const invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3999 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4000 const STRLEN bits = SvUV(*bitssvp);
4001 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4007 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4009 /* read $swash->{LIST} */
4010 if (SvPOK(*listsvp)) {
4011 l = (U8*)SvPV(*listsvp, lcur);
4014 /* LIST legitimately doesn't contain a string during compilation phases
4015 * of Perl itself, before the Unicode tables are generated. In this
4016 * case, just fake things up by creating an empty list */
4023 /* Scan the input to count the number of lines to preallocate array size
4024 * based on worst possible case, which is each line in the input creates 2
4025 * elements in the inversion list: 1) the beginning of a range in the list;
4026 * 2) the beginning of a range not in the list. */
4027 while ((loc = (strchr(loc, '\n'))) != NULL) {
4032 /* If the ending is somehow corrupt and isn't a new line, add another
4033 * element for the final range that isn't in the inversion list */
4034 if (! (*lend == '\n'
4035 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4040 invlist = _new_invlist(elements);
4042 /* Now go through the input again, adding each range to the list */
4045 UV val; /* Not used by this function */
4047 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
4048 cBOOL(octets), typestr);
4054 invlist = _add_range_to_invlist(invlist, start, end);
4057 /* Invert if the data says it should be */
4058 if (invert_it_svp && SvUV(*invert_it_svp)) {
4059 _invlist_invert_prop(invlist);
4062 /* This code is copied from swatch_get()
4063 * read $swash->{EXTRAS} */
4064 x = (U8*)SvPV(*extssvp, xcur);
4072 SV **otherbitssvp, *other;
4075 const U8 opc = *x++;
4079 nl = (U8*)memchr(x, '\n', xend - x);
4081 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4083 x = nl + 1; /* 1 is length of "\n" */
4087 x = xend; /* to EXTRAS' end at which \n is not found */
4094 namelen = nl - namestr;
4098 namelen = xend - namestr;
4102 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4103 otherhv = MUTABLE_HV(SvRV(*othersvp));
4104 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4105 otherbits = (STRLEN)SvUV(*otherbitssvp);
4107 if (bits != otherbits || bits != 1) {
4108 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4109 "properties, bits=%"UVuf", otherbits=%"UVuf,
4110 (UV)bits, (UV)otherbits);
4113 /* The "other" swatch must be destroyed after. */
4114 other = _swash_to_invlist((SV *)*othersvp);
4116 /* End of code copied from swatch_get() */
4119 _invlist_union(invlist, other, &invlist);
4122 _invlist_invert(other);
4123 _invlist_union(invlist, other, &invlist);
4126 _invlist_subtract(invlist, other, &invlist);
4129 _invlist_intersection(invlist, other, &invlist);
4134 sv_free(other); /* through with it! */
4141 Perl__get_swash_invlist(pTHX_ SV* const swash)
4143 SV** ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "INVLIST", FALSE);
4145 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4155 =for apidoc uvchr_to_utf8
4157 Adds the UTF-8 representation of the Native code point C<uv> to the end
4158 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
4159 bytes available. The return value is the pointer to the byte after the
4160 end of the new character. In other words,
4162 d = uvchr_to_utf8(d, uv);
4164 is the recommended wide native character-aware way of saying
4171 /* On ASCII machines this is normally a macro but we want a
4172 real function in case XS code wants it
4175 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
4177 PERL_ARGS_ASSERT_UVCHR_TO_UTF8;
4179 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), 0);
4183 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4185 PERL_ARGS_ASSERT_UVCHR_TO_UTF8_FLAGS;
4187 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), flags);
4191 =for apidoc utf8n_to_uvchr
4193 Returns the native character value of the first character in the string
4195 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
4196 length, in bytes, of that character.
4198 C<length> and C<flags> are the same as L</utf8n_to_uvuni>().
4202 /* On ASCII machines this is normally a macro but we want
4203 a real function in case XS code wants it
4206 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen,
4209 const UV uv = Perl_utf8n_to_uvuni(aTHX_ s, curlen, retlen, flags);
4211 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
4213 return UNI_TO_NATIVE(uv);
4217 Perl_check_utf8_print(pTHX_ register const U8* s, const STRLEN len)
4219 /* May change: warns if surrogates, non-character code points, or
4220 * non-Unicode code points are in s which has length len bytes. Returns
4221 * TRUE if none found; FALSE otherwise. The only other validity check is
4222 * to make sure that this won't exceed the string's length */
4224 const U8* const e = s + len;
4227 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4230 if (UTF8SKIP(s) > len) {
4231 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4232 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4235 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
4237 if (UTF8_IS_SUPER(s)) {
4238 if (ckWARN_d(WARN_NON_UNICODE)) {
4239 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4240 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
4241 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
4245 else if (UTF8_IS_SURROGATE(s)) {
4246 if (ckWARN_d(WARN_SURROGATE)) {
4247 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4248 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4249 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
4254 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
4255 && (ckWARN_d(WARN_NONCHAR)))
4257 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4258 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
4259 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
4270 =for apidoc pv_uni_display
4272 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4273 length C<len>, the displayable version being at most C<pvlim> bytes long
4274 (if longer, the rest is truncated and "..." will be appended).
4276 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
4277 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
4278 to display the \\[nrfta\\] as the backslashed versions (like '\n')
4279 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
4280 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
4281 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
4283 The pointer to the PV of the C<dsv> is returned.
4287 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4292 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4296 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4298 /* This serves double duty as a flag and a character to print after
4299 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4303 if (pvlim && SvCUR(dsv) >= pvlim) {
4307 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4309 const unsigned char c = (unsigned char)u & 0xFF;
4310 if (flags & UNI_DISPLAY_BACKSLASH) {
4327 const char string = ok;
4328 sv_catpvs(dsv, "\\");
4329 sv_catpvn(dsv, &string, 1);
4332 /* isPRINT() is the locale-blind version. */
4333 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4334 const char string = c;
4335 sv_catpvn(dsv, &string, 1);
4340 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4343 sv_catpvs(dsv, "...");
4349 =for apidoc sv_uni_display
4351 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4352 the displayable version being at most C<pvlim> bytes long
4353 (if longer, the rest is truncated and "..." will be appended).
4355 The C<flags> argument is as in L</pv_uni_display>().
4357 The pointer to the PV of the C<dsv> is returned.
4362 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4364 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4366 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)SvPVX_const(ssv),
4367 SvCUR(ssv), pvlim, flags);
4371 =for apidoc foldEQ_utf8
4373 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4374 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4375 How far into the strings to compare is determined by other input parameters.
4377 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4378 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4379 with respect to C<s2>.
4381 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4382 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4383 scan will not be considered to be a match unless the goal is reached, and
4384 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4387 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4388 considered an end pointer beyond which scanning of C<s1> will not continue under
4389 any circumstances. This means that if both C<l1> and C<pe1> are specified, and
4391 is less than C<s1>+C<l1>, the match will never be successful because it can
4393 get as far as its goal (and in fact is asserted against). Correspondingly for
4394 C<pe2> with respect to C<s2>.
4396 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4397 C<l2> must be non-zero), and if both do, both have to be
4398 reached for a successful match. Also, if the fold of a character is multiple
4399 characters, all of them must be matched (see tr21 reference below for
4402 Upon a successful match, if C<pe1> is non-NULL,
4403 it will be set to point to the beginning of the I<next> character of C<s1>
4404 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4406 For case-insensitiveness, the "casefolding" of Unicode is used
4407 instead of upper/lowercasing both the characters, see
4408 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4412 /* A flags parameter has been added which may change, and hence isn't
4413 * externally documented. Currently it is:
4414 * 0 for as-documented above
4415 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4416 ASCII one, to not match
4417 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
4418 * points below 256; unicode rules for above 255; and
4419 * folds that cross those boundaries are disallowed,
4420 * like the NOMIX_ASCII option
4421 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4422 * routine. This allows that step to be skipped.
4423 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4426 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, register UV l1, bool u1, const char *s2, char **pe2, register UV l2, bool u2, U32 flags)
4429 const U8 *p1 = (const U8*)s1; /* Point to current char */
4430 const U8 *p2 = (const U8*)s2;
4431 const U8 *g1 = NULL; /* goal for s1 */
4432 const U8 *g2 = NULL;
4433 const U8 *e1 = NULL; /* Don't scan s1 past this */
4434 U8 *f1 = NULL; /* Point to current folded */
4435 const U8 *e2 = NULL;
4437 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4438 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4439 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4441 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4443 /* The algorithm requires that input with the flags on the first line of
4444 * the assert not be pre-folded. */
4445 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
4446 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4453 g1 = (const U8*)s1 + l1;
4461 g2 = (const U8*)s2 + l2;
4464 /* Must have at least one goal */
4469 /* Will never match if goal is out-of-bounds */
4470 assert(! e1 || e1 >= g1);
4472 /* Here, there isn't an end pointer, or it is beyond the goal. We
4473 * only go as far as the goal */
4477 assert(e1); /* Must have an end for looking at s1 */
4480 /* Same for goal for s2 */
4482 assert(! e2 || e2 >= g2);
4489 /* If both operands are already folded, we could just do a memEQ on the
4490 * whole strings at once, but it would be better if the caller realized
4491 * this and didn't even call us */
4493 /* Look through both strings, a character at a time */
4494 while (p1 < e1 && p2 < e2) {
4496 /* If at the beginning of a new character in s1, get its fold to use
4497 * and the length of the fold. (exception: locale rules just get the
4498 * character to a single byte) */
4500 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4506 /* If in locale matching, we use two sets of rules, depending
4507 * on if the code point is above or below 255. Here, we test
4508 * for and handle locale rules */
4509 if ((flags & FOLDEQ_UTF8_LOCALE)
4510 && (! u1 || UTF8_IS_INVARIANT(*p1)
4511 || UTF8_IS_DOWNGRADEABLE_START(*p1)))
4513 /* There is no mixing of code points above and below 255. */
4514 if (u2 && (! UTF8_IS_INVARIANT(*p2)
4515 && ! UTF8_IS_DOWNGRADEABLE_START(*p2)))
4520 /* We handle locale rules by converting, if necessary, the
4521 * code point to a single byte. */
4522 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4526 *foldbuf1 = TWO_BYTE_UTF8_TO_UNI(*p1, *(p1 + 1));
4530 else if (isASCII(*p1)) { /* Note, that here won't be both
4531 ASCII and using locale rules */
4533 /* If trying to mix non- with ASCII, and not supposed to,
4535 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4539 *foldbuf1 = toLOWER(*p1); /* Folds in the ASCII range are
4543 to_utf8_fold(p1, foldbuf1, &n1);
4545 else { /* Not utf8, get utf8 fold */
4546 to_uni_fold(NATIVE_TO_UNI(*p1), foldbuf1, &n1);
4552 if (n2 == 0) { /* Same for s2 */
4553 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4558 if ((flags & FOLDEQ_UTF8_LOCALE)
4559 && (! u2 || UTF8_IS_INVARIANT(*p2) || UTF8_IS_DOWNGRADEABLE_START(*p2)))
4561 /* Here, the next char in s2 is < 256. We've already
4562 * worked on s1, and if it isn't also < 256, can't match */
4563 if (u1 && (! UTF8_IS_INVARIANT(*p1)
4564 && ! UTF8_IS_DOWNGRADEABLE_START(*p1)))
4568 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4572 *foldbuf2 = TWO_BYTE_UTF8_TO_UNI(*p2, *(p2 + 1));
4575 /* Use another function to handle locale rules. We've made
4576 * sure that both characters to compare are single bytes */
4577 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4582 else if (isASCII(*p2)) {
4583 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4587 *foldbuf2 = toLOWER(*p2);
4590 to_utf8_fold(p2, foldbuf2, &n2);
4593 to_uni_fold(NATIVE_TO_UNI(*p2), foldbuf2, &n2);
4599 /* Here f1 and f2 point to the beginning of the strings to compare.
4600 * These strings are the folds of the next character from each input
4601 * string, stored in utf8. */
4603 /* While there is more to look for in both folds, see if they
4604 * continue to match */
4606 U8 fold_length = UTF8SKIP(f1);
4607 if (fold_length != UTF8SKIP(f2)
4608 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4609 function call for single
4611 || memNE((char*)f1, (char*)f2, fold_length))
4613 return 0; /* mismatch */
4616 /* Here, they matched, advance past them */
4623 /* When reach the end of any fold, advance the input past it */
4625 p1 += u1 ? UTF8SKIP(p1) : 1;
4628 p2 += u2 ? UTF8SKIP(p2) : 1;
4630 } /* End of loop through both strings */
4632 /* A match is defined by each scan that specified an explicit length
4633 * reaching its final goal, and the other not having matched a partial
4634 * character (which can happen when the fold of a character is more than one
4636 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4640 /* Successful match. Set output pointers */
4652 * c-indentation-style: bsd
4654 * indent-tabs-mode: nil
4657 * ex: set ts=8 sts=4 sw=4 et: