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 UV Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen);
41 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
44 static const char unees[] =
45 "Malformed UTF-8 character (unexpected end of string)";
48 =head1 Unicode Support
50 This file contains various utility functions for manipulating UTF8-encoded
51 strings. For the uninitiated, this is a method of representing arbitrary
52 Unicode characters as a variable number of bytes, in such a way that
53 characters in the ASCII range are unmodified, and a zero byte never appears
54 within non-zero characters.
60 =for apidoc is_ascii_string
62 Returns true if the first C<len> bytes of the string C<s> are the same whether
63 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
64 is, if they are invariant. On ASCII-ish machines, only ASCII characters
65 fit this definition, hence the function's name.
67 If C<len> is 0, it will be calculated using C<strlen(s)>.
69 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
75 Perl_is_ascii_string(const U8 *s, STRLEN len)
77 const U8* const send = s + (len ? len : strlen((const char *)s));
80 PERL_ARGS_ASSERT_IS_ASCII_STRING;
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
91 =for apidoc uvuni_to_utf8_flags
93 Adds the UTF-8 representation of the code point C<uv> to the end
94 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
95 bytes available. The return value is the pointer to the byte after the
96 end of the new character. In other words,
98 d = uvuni_to_utf8_flags(d, uv, flags);
102 d = uvuni_to_utf8(d, uv);
104 (which is equivalent to)
106 d = uvuni_to_utf8_flags(d, uv, 0);
108 This is the recommended Unicode-aware way of saying
112 This function will convert to UTF-8 (and not warn) even code points that aren't
113 legal Unicode or are problematic, unless C<flags> contains one or more of the
116 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
117 the function will raise a warning, provided UTF8 warnings are enabled. If instead
118 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
119 If both flags are set, the function will both warn and return NULL.
121 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
122 affect how the function handles a Unicode non-character. And, likewise for the
123 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
124 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
125 even less portable) can be warned and/or disallowed even if other above-Unicode
126 code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
129 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
130 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
138 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
140 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
142 /* The first problematic code point is the first surrogate */
143 if (uv >= UNICODE_SURROGATE_FIRST
144 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
146 if (UNICODE_IS_SURROGATE(uv)) {
147 if (flags & UNICODE_WARN_SURROGATE) {
148 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
149 "UTF-16 surrogate U+%04"UVXf, uv);
151 if (flags & UNICODE_DISALLOW_SURROGATE) {
155 else if (UNICODE_IS_SUPER(uv)) {
156 if (flags & UNICODE_WARN_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
159 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
160 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
162 if (flags & UNICODE_DISALLOW_SUPER
163 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
168 else if (UNICODE_IS_NONCHAR(uv)) {
169 if (flags & UNICODE_WARN_NONCHAR) {
170 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
171 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
174 if (flags & UNICODE_DISALLOW_NONCHAR) {
179 if (UNI_IS_INVARIANT(uv)) {
180 *d++ = (U8)UTF_TO_NATIVE(uv);
185 STRLEN len = UNISKIP(uv);
188 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
189 uv >>= UTF_ACCUMULATION_SHIFT;
191 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
194 #else /* Non loop style */
196 *d++ = (U8)(( uv >> 6) | 0xc0);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
201 *d++ = (U8)(( uv >> 12) | 0xe0);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
207 *d++ = (U8)(( uv >> 18) | 0xf0);
208 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
209 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
210 *d++ = (U8)(( uv & 0x3f) | 0x80);
213 if (uv < 0x4000000) {
214 *d++ = (U8)(( uv >> 24) | 0xf8);
215 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
218 *d++ = (U8)(( uv & 0x3f) | 0x80);
221 if (uv < 0x80000000) {
222 *d++ = (U8)(( uv >> 30) | 0xfc);
223 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
227 *d++ = (U8)(( uv & 0x3f) | 0x80);
231 if (uv < UTF8_QUAD_MAX)
234 *d++ = 0xfe; /* Can't match U+FEFF! */
235 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
239 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
240 *d++ = (U8)(( uv & 0x3f) | 0x80);
245 *d++ = 0xff; /* Can't match U+FFFE! */
246 *d++ = 0x80; /* 6 Reserved bits */
247 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
248 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
257 *d++ = (U8)(( uv & 0x3f) | 0x80);
261 #endif /* Loop style */
266 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
267 character. Note that an INVARIANT (i.e. ASCII) character is a valid
268 UTF-8 character. The number of bytes in the UTF-8 character
269 will be returned if it is valid, otherwise 0.
271 This is the "slow" version as opposed to the "fast" version which is
272 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
273 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
274 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
275 you should use the _slow(). In practice this means that the _slow()
276 will be used very rarely, since the maximum Unicode code point (as of
277 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
278 the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
282 PERL_STATIC_INLINE STRLEN
283 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
285 dTHX; /* The function called below requires thread context */
289 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
291 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
293 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
297 =for apidoc is_utf8_char_buf
299 Returns the number of bytes that comprise the first UTF-8 encoded character in
300 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
301 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
304 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
305 machines) is a valid UTF-8 character.
310 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
315 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
317 if (buf_end <= buf) {
322 if (len > UTF8SKIP(buf)) {
327 if (IS_UTF8_CHAR_FAST(len))
328 return IS_UTF8_CHAR(buf, len) ? len : 0;
329 #endif /* #ifdef IS_UTF8_CHAR */
330 return is_utf8_char_slow(buf, len);
334 =for apidoc is_utf8_char
338 Tests if some arbitrary number of bytes begins in a valid UTF-8
339 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
340 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
341 character will be returned if it is valid, otherwise 0.
343 This function is deprecated due to the possibility that malformed input could
344 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
350 Perl_is_utf8_char(const U8 *s)
352 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
354 /* Assumes we have enough space, which is why this is deprecated */
355 return is_utf8_char_buf(s, s + UTF8SKIP(s));
360 =for apidoc is_utf8_string
362 Returns true if the first C<len> bytes of string C<s> form a valid
363 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
364 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
365 terminating NUL byte). Note that all characters being ASCII constitute 'a
368 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
374 Perl_is_utf8_string(const U8 *s, STRLEN len)
376 const U8* const send = s + (len ? len : strlen((const char *)s));
379 PERL_ARGS_ASSERT_IS_UTF8_STRING;
382 /* Inline the easy bits of is_utf8_char() here for speed... */
383 if (UTF8_IS_INVARIANT(*x)) {
387 /* ... and call is_utf8_char() only if really needed. */
388 const STRLEN c = UTF8SKIP(x);
389 const U8* const next_char_ptr = x + c;
391 if (next_char_ptr > send) {
395 if (IS_UTF8_CHAR_FAST(c)) {
396 if (!IS_UTF8_CHAR(x, c))
399 else if (! is_utf8_char_slow(x, c)) {
410 Implemented as a macro in utf8.h
412 =for apidoc is_utf8_string_loc
414 Like L</is_utf8_string> but stores the location of the failure (in the
415 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
416 "utf8ness success") in the C<ep>.
418 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
420 =for apidoc is_utf8_string_loclen
422 Like L</is_utf8_string>() but stores the location of the failure (in the
423 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
424 "utf8ness success") in the C<ep>, and the number of UTF-8
425 encoded characters in the C<el>.
427 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
433 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
435 const U8* const send = s + (len ? len : strlen((const char *)s));
440 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
443 const U8* next_char_ptr;
445 /* Inline the easy bits of is_utf8_char() here for speed... */
446 if (UTF8_IS_INVARIANT(*x))
447 next_char_ptr = x + 1;
449 /* ... and call is_utf8_char() only if really needed. */
451 next_char_ptr = c + x;
452 if (next_char_ptr > send) {
455 if (IS_UTF8_CHAR_FAST(c)) {
456 if (!IS_UTF8_CHAR(x, c))
459 c = is_utf8_char_slow(x, c);
478 =for apidoc utf8n_to_uvuni
480 Bottom level UTF-8 decode routine.
481 Returns the code point value of the first character in the string C<s>,
482 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
483 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
484 the length, in bytes, of that character.
486 The value of C<flags> determines the behavior when C<s> does not point to a
487 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
488 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
489 next possible position in C<s> that could begin a non-malformed character.
490 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
492 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
493 individual types of malformations, such as the sequence being overlong (that
494 is, when there is a shorter sequence that can express the same code point;
495 overlong sequences are expressly forbidden in the UTF-8 standard due to
496 potential security issues). Another malformation example is the first byte of
497 a character not being a legal first byte. See F<utf8.h> for the list of such
498 flags. For allowed 0 length strings, this function returns 0; for allowed
499 overlong sequences, the computed code point is returned; for all other allowed
500 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
501 determinable reasonable value.
503 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
504 flags) malformation is found. If this flag is set, the routine assumes that
505 the caller will raise a warning, and this function will silently just set
506 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
508 Note that this API requires disambiguation between successful decoding a NUL
509 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
510 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
511 first byte of C<s> is 0 as well. If so, the input was a NUL; if not, the input
514 Certain code points are considered problematic. These are Unicode surrogates,
515 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
516 By default these are considered regular code points, but certain situations
517 warrant special handling for them. If C<flags> contains
518 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
519 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
520 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
521 maximum) can be set to disallow these categories individually.
523 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
524 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
525 for their respective categories, but otherwise the code points are considered
526 valid (not malformations). To get a category to both be treated as a
527 malformation and raise a warning, specify both the WARN and DISALLOW flags.
528 (But note that warnings are not raised if lexically disabled nor if
529 UTF8_CHECK_ONLY is also specified.)
531 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
532 the others that are above the Unicode legal maximum. There are several
533 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
534 not representable at all on EBCDIC platforms, and the original UTF-8
535 specification never went above this number (the current 0x10FFFF limit was
536 imposed later). (The smaller ones, those that fit into 32 bits, are
537 representable by a UV on ASCII platforms, but not by an IV, which means that
538 the number of operations that can be performed on them is quite restricted.)
539 The UTF-8 encoding on ASCII platforms for these large code points begins with a
540 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
541 be treated as malformations, while allowing smaller above-Unicode code points.
542 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
543 including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
544 the other WARN flags, but applies just to these code points.
546 All other code points corresponding to Unicode characters, including private
547 use and those yet to be assigned, are never considered malformed and never
550 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
556 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
559 const U8 * const s0 = s;
560 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
565 UV outlier_ret = 0; /* return value when input is in error or problematic
567 UV pack_warn = 0; /* Save result of packWARN() for later */
568 bool unexpected_non_continuation = FALSE;
569 bool overflowed = FALSE;
570 bool do_overlong_test = TRUE; /* May have to skip this test */
572 const char* const malformed_text = "Malformed UTF-8 character";
574 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
576 /* The order of malformation tests here is important. We should consume as
577 * few bytes as possible in order to not skip any valid character. This is
578 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
579 * http://unicode.org/reports/tr36 for more discussion as to why. For
580 * example, once we've done a UTF8SKIP, we can tell the expected number of
581 * bytes, and could fail right off the bat if the input parameters indicate
582 * that there are too few available. But it could be that just that first
583 * byte is garbled, and the intended character occupies fewer bytes. If we
584 * blindly assumed that the first byte is correct, and skipped based on
585 * that number, we could skip over a valid input character. So instead, we
586 * always examine the sequence byte-by-byte.
588 * We also should not consume too few bytes, otherwise someone could inject
589 * things. For example, an input could be deliberately designed to
590 * overflow, and if this code bailed out immediately upon discovering that,
591 * returning to the caller *retlen pointing to the very next byte (one
592 * which is actually part of of the overflowing sequence), that could look
593 * legitimate to the caller, which could discard the initial partial
594 * sequence and process the rest, inappropriately */
596 /* Zero length strings, if allowed, of necessity are zero */
597 if (UNLIKELY(curlen == 0)) {
602 if (flags & UTF8_ALLOW_EMPTY) {
605 if (! (flags & UTF8_CHECK_ONLY)) {
606 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
611 expectlen = UTF8SKIP(s);
613 /* A well-formed UTF-8 character, as the vast majority of calls to this
614 * function will be for, has this expected length. For efficiency, set
615 * things up here to return it. It will be overriden only in those rare
616 * cases where a malformation is found */
621 /* An invariant is trivially well-formed */
622 if (UTF8_IS_INVARIANT(uv)) {
623 return (UV) (NATIVE_TO_UTF(*s));
626 /* A continuation character can't start a valid sequence */
627 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
628 if (flags & UTF8_ALLOW_CONTINUATION) {
632 return UNICODE_REPLACEMENT;
635 if (! (flags & UTF8_CHECK_ONLY)) {
636 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
643 uv = NATIVE_TO_UTF(uv);
646 /* Here is not a continuation byte, nor an invariant. The only thing left
647 * is a start byte (possibly for an overlong) */
649 /* Remove the leading bits that indicate the number of bytes in the
650 * character's whole UTF-8 sequence, leaving just the bits that are part of
652 uv &= UTF_START_MASK(expectlen);
654 /* Now, loop through the remaining bytes in the character's sequence,
655 * accumulating each into the working value as we go. Be sure to not look
656 * past the end of the input string */
657 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
659 for (s = s0 + 1; s < send; s++) {
660 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
661 #ifndef EBCDIC /* Can't overflow in EBCDIC */
662 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
664 /* The original implementors viewed this malformation as more
665 * serious than the others (though I, khw, don't understand
666 * why, since other malformations also give very very wrong
667 * results), so there is no way to turn off checking for it.
668 * Set a flag, but keep going in the loop, so that we absorb
669 * the rest of the bytes that comprise the character. */
671 overflow_byte = *s; /* Save for warning message's use */
674 uv = UTF8_ACCUMULATE(uv, *s);
677 /* Here, found a non-continuation before processing all expected
678 * bytes. This byte begins a new character, so quit, even if
679 * allowing this malformation. */
680 unexpected_non_continuation = TRUE;
683 } /* End of loop through the character's bytes */
685 /* Save how many bytes were actually in the character */
688 /* The loop above finds two types of malformations: non-continuation and/or
689 * overflow. The non-continuation malformation is really a too-short
690 * malformation, as it means that the current character ended before it was
691 * expected to (being terminated prematurely by the beginning of the next
692 * character, whereas in the too-short malformation there just are too few
693 * bytes available to hold the character. In both cases, the check below
694 * that we have found the expected number of bytes would fail if executed.)
695 * Thus the non-continuation malformation is really unnecessary, being a
696 * subset of the too-short malformation. But there may be existing
697 * applications that are expecting the non-continuation type, so we retain
698 * it, and return it in preference to the too-short malformation. (If this
699 * code were being written from scratch, the two types might be collapsed
700 * into one.) I, khw, am also giving priority to returning the
701 * non-continuation and too-short malformations over overflow when multiple
702 * ones are present. I don't know of any real reason to prefer one over
703 * the other, except that it seems to me that multiple-byte errors trumps
704 * errors from a single byte */
705 if (UNLIKELY(unexpected_non_continuation)) {
706 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
707 if (! (flags & UTF8_CHECK_ONLY)) {
709 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
712 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));
717 uv = UNICODE_REPLACEMENT;
719 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
720 * as what the original expectations were. */
721 do_overlong_test = FALSE;
726 else if (UNLIKELY(curlen < expectlen)) {
727 if (! (flags & UTF8_ALLOW_SHORT)) {
728 if (! (flags & UTF8_CHECK_ONLY)) {
729 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));
733 uv = UNICODE_REPLACEMENT;
734 do_overlong_test = FALSE;
740 #ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
741 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
742 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
744 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
745 * generation of the sv, since no warnings are raised under CHECK */
746 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
747 && ckWARN_d(WARN_UTF8))
749 /* This message is deliberately not of the same syntax as the other
750 * messages for malformations, for backwards compatibility in the
751 * unlikely event that code is relying on its precise earlier text
753 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
754 pack_warn = packWARN(WARN_UTF8);
756 if (flags & UTF8_DISALLOW_FE_FF) {
760 if (UNLIKELY(overflowed)) {
762 /* If the first byte is FF, it will overflow a 32-bit word. If the
763 * first byte is FE, it will overflow a signed 32-bit word. The
764 * above preserves backward compatibility, since its message was used
765 * in earlier versions of this code in preference to overflow */
766 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
772 && expectlen > (STRLEN)UNISKIP(uv)
773 && ! (flags & UTF8_ALLOW_LONG))
775 /* The overlong malformation has lower precedence than the others.
776 * Note that if this malformation is allowed, we return the actual
777 * value, instead of the replacement character. This is because this
778 * value is actually well-defined. */
779 if (! (flags & UTF8_CHECK_ONLY)) {
780 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));
785 /* Here, the input is considered to be well-formed , but could be a
786 * problematic code point that is not allowed by the input parameters. */
787 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
788 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
789 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
791 if (UNICODE_IS_SURROGATE(uv)) {
792 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
793 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
795 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
796 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
798 if (flags & UTF8_DISALLOW_SURROGATE) {
802 else if ((uv > PERL_UNICODE_MAX)) {
803 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
804 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
806 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
807 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
809 if (flags & UTF8_DISALLOW_SUPER) {
813 else if (UNICODE_IS_NONCHAR(uv)) {
814 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
815 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
817 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
818 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
820 if (flags & UTF8_DISALLOW_NONCHAR) {
830 /* Here, this is not considered a malformed character, so drop through
836 /* There are three cases which get to beyond this point. In all 3 cases:
837 * <sv> if not null points to a string to print as a warning.
838 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
840 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
841 * This is done by initializing it to 0, and changing it only
844 * 1) The input is valid but problematic, and to be warned about. The
845 * return value is the resultant code point; <*retlen> is set to
846 * <curlen>, the number of bytes that comprise the code point.
847 * <pack_warn> contains the result of packWARN() for the warning
848 * types. The entry point for this case is the label <do_warn>;
849 * 2) The input is a valid code point but disallowed by the parameters to
850 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
851 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
852 * comprise the code point. <pack_warn> contains the result of
853 * packWARN() for the warning types. The entry point for this case is
854 * the label <disallowed>.
855 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
856 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
857 * bytes that comprise the malformation. All such malformations are
858 * assumed to be warning type <utf8>. The entry point for this case
859 * is the label <malformed>.
864 if (sv && ckWARN_d(WARN_UTF8)) {
865 pack_warn = packWARN(WARN_UTF8);
870 if (flags & UTF8_CHECK_ONLY) {
872 *retlen = ((STRLEN) -1);
878 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
879 if warnings are to be raised. */
880 const char * const string = SvPVX_const(sv);
883 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
885 Perl_warner(aTHX_ pack_warn, "%s", string);
896 =for apidoc utf8_to_uvchr_buf
898 Returns the native code point of the first character in the string C<s> which
899 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
900 C<*retlen> will be set to the length, in bytes, of that character.
902 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
903 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
904 NULL) to -1. If those warnings are off, the computed value, if well-defined
905 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
906 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
907 the next possible position in C<s> that could begin a non-malformed character.
908 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is
916 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
918 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
922 return utf8n_to_uvchr(s, send - s, retlen,
923 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
926 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
927 * there are no malformations in the input UTF-8 string C<s>. surrogates,
928 * non-character code points, and non-Unicode code points are allowed. A macro
929 * in utf8.h is used to normally avoid this function wrapper */
932 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
934 const UV uv = valid_utf8_to_uvuni(s, retlen);
936 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
938 return UNI_TO_NATIVE(uv);
942 =for apidoc utf8_to_uvchr
946 Returns the native code point of the first character in the string C<s>
947 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
948 length, in bytes, of that character.
950 Some, but not all, UTF-8 malformations are detected, and in fact, some
951 malformed input could cause reading beyond the end of the input buffer, which
952 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
954 If C<s> points to one of the detected malformations, and UTF8 warnings are
955 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
956 NULL) to -1. If those warnings are off, the computed value if well-defined (or
957 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
958 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
959 next possible position in C<s> that could begin a non-malformed character.
960 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
966 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
968 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
970 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
974 =for apidoc utf8_to_uvuni_buf
976 Returns the Unicode code point of the first character in the string C<s> which
977 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
978 C<retlen> will be set to the length, in bytes, of that character.
980 This function should only be used when the returned UV is considered
981 an index into the Unicode semantic tables (e.g. swashes).
983 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
984 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
985 NULL) to -1. If those warnings are off, the computed value if well-defined (or
986 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
987 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
988 next possible position in C<s> that could begin a non-malformed character.
989 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
995 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
997 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1001 /* Call the low level routine asking for checks */
1002 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1003 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1006 /* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1007 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1008 * non-character code points, and non-Unicode code points are allowed */
1011 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1013 UV expectlen = UTF8SKIP(s);
1014 const U8* send = s + expectlen;
1015 UV uv = NATIVE_TO_UTF(*s);
1017 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1020 *retlen = expectlen;
1023 /* An invariant is trivially returned */
1024 if (expectlen == 1) {
1028 /* Remove the leading bits that indicate the number of bytes, leaving just
1029 * the bits that are part of the value */
1030 uv &= UTF_START_MASK(expectlen);
1032 /* Now, loop through the remaining bytes, accumulating each into the
1033 * working total as we go. (I khw tried unrolling the loop for up to 4
1034 * bytes, but there was no performance improvement) */
1035 for (++s; s < send; s++) {
1036 uv = UTF8_ACCUMULATE(uv, *s);
1043 =for apidoc utf8_to_uvuni
1047 Returns the Unicode code point of the first character in the string C<s>
1048 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1049 length, in bytes, of that character.
1051 This function should only be used when the returned UV is considered
1052 an index into the Unicode semantic tables (e.g. swashes).
1054 Some, but not all, UTF-8 malformations are detected, and in fact, some
1055 malformed input could cause reading beyond the end of the input buffer, which
1056 is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
1058 If C<s> points to one of the detected malformations, and UTF8 warnings are
1059 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1060 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1061 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1062 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1063 next possible position in C<s> that could begin a non-malformed character.
1064 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
1070 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1072 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1074 return valid_utf8_to_uvuni(s, retlen);
1078 =for apidoc utf8_length
1080 Return the length of the UTF-8 char encoded string C<s> in characters.
1081 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1082 up past C<e>, croaks.
1088 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1093 PERL_ARGS_ASSERT_UTF8_LENGTH;
1095 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1096 * the bitops (especially ~) can create illegal UTF-8.
1097 * In other words: in Perl UTF-8 is not just for Unicode. */
1100 goto warn_and_return;
1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1111 "%s in %s", unees, OP_DESC(PL_op));
1113 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1120 =for apidoc utf8_distance
1122 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1125 WARNING: use only if you *know* that the pointers point inside the
1132 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1134 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1136 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1140 =for apidoc utf8_hop
1142 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1143 forward or backward.
1145 WARNING: do not use the following unless you *know* C<off> is within
1146 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1147 on the first byte of character or just after the last byte of a character.
1153 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1155 PERL_ARGS_ASSERT_UTF8_HOP;
1157 PERL_UNUSED_CONTEXT;
1158 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1159 * the bitops (especially ~) can create illegal UTF-8.
1160 * In other words: in Perl UTF-8 is not just for Unicode. */
1169 while (UTF8_IS_CONTINUATION(*s))
1177 =for apidoc bytes_cmp_utf8
1179 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1180 sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
1181 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1182 if the first string is greater than the second string.
1184 -1 or +1 is returned if the shorter string was identical to the start of the
1185 longer string. -2 or +2 is returned if the was a difference between characters
1192 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1194 const U8 *const bend = b + blen;
1195 const U8 *const uend = u + ulen;
1197 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1199 PERL_UNUSED_CONTEXT;
1201 while (b < bend && u < uend) {
1203 if (!UTF8_IS_INVARIANT(c)) {
1204 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1207 if (UTF8_IS_CONTINUATION(c1)) {
1208 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
1210 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1211 "Malformed UTF-8 character "
1212 "(unexpected non-continuation byte 0x%02x"
1213 ", immediately after start byte 0x%02x)"
1214 /* Dear diag.t, it's in the pod. */
1216 PL_op ? " in " : "",
1217 PL_op ? OP_DESC(PL_op) : "");
1222 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1223 "%s in %s", unees, OP_DESC(PL_op));
1225 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1226 return -2; /* Really want to return undef :-) */
1233 return *b < c ? -2 : +2;
1238 if (b == bend && u == uend)
1241 return b < bend ? +1 : -1;
1245 =for apidoc utf8_to_bytes
1247 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1248 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1249 updates C<len> to contain the new length.
1250 Returns zero on failure, setting C<len> to -1.
1252 If you need a copy of the string, see L</bytes_from_utf8>.
1258 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1260 U8 * const save = s;
1261 U8 * const send = s + *len;
1264 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1266 /* ensure valid UTF-8 and chars < 256 before updating string */
1270 if (!UTF8_IS_INVARIANT(c) &&
1271 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1272 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
1273 *len = ((STRLEN) -1);
1281 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
1290 =for apidoc bytes_from_utf8
1292 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1293 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1294 the newly-created string, and updates C<len> to contain the new
1295 length. Returns the original string if no conversion occurs, C<len>
1296 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1297 0 if C<s> is converted or consisted entirely of characters that are invariant
1298 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1304 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1307 const U8 *start = s;
1311 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1313 PERL_UNUSED_CONTEXT;
1317 /* ensure valid UTF-8 and chars < 256 before converting string */
1318 for (send = s + *len; s < send;) {
1320 if (!UTF8_IS_INVARIANT(c)) {
1321 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1322 (c = *s++) && UTF8_IS_CONTINUATION(c))
1331 Newx(d, (*len) - count + 1, U8);
1332 s = start; start = d;
1335 if (!UTF8_IS_INVARIANT(c)) {
1336 /* Then it is two-byte encoded */
1337 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
1347 =for apidoc bytes_to_utf8
1349 Converts a string C<s> of length C<len> bytes from the native encoding into
1351 Returns a pointer to the newly-created string, and sets C<len> to
1352 reflect the new length in bytes.
1354 A NUL character will be written after the end of the string.
1356 If you want to convert to UTF-8 from encodings other than
1357 the native (Latin1 or EBCDIC),
1358 see L</sv_recode_to_utf8>().
1363 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1364 likewise need duplication. */
1367 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1369 const U8 * const send = s + (*len);
1373 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1374 PERL_UNUSED_CONTEXT;
1376 Newx(d, (*len) * 2 + 1, U8);
1380 const UV uv = NATIVE_TO_ASCII(*s++);
1381 if (UNI_IS_INVARIANT(uv))
1382 *d++ = (U8)UTF_TO_NATIVE(uv);
1384 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1385 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
1394 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1396 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1397 * We optimize for native, for obvious reasons. */
1400 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1405 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1408 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1413 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1417 *d++ = UNI_TO_NATIVE(uv);
1424 *d++ = (U8)(( uv >> 6) | 0xc0);
1425 *d++ = (U8)(( uv & 0x3f) | 0x80);
1428 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
1430 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1432 UV low = (p[0] << 8) + p[1];
1434 if (low < 0xdc00 || low > 0xdfff)
1435 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1436 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1438 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1439 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1442 *d++ = (U8)(( uv >> 12) | 0xe0);
1443 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1444 *d++ = (U8)(( uv & 0x3f) | 0x80);
1448 *d++ = (U8)(( uv >> 18) | 0xf0);
1449 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1450 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1451 *d++ = (U8)(( uv & 0x3f) | 0x80);
1455 *newlen = d - dstart;
1459 /* Note: this one is slightly destructive of the source. */
1462 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1465 U8* const send = s + bytelen;
1467 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1470 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1474 const U8 tmp = s[0];
1479 return utf16_to_utf8(p, d, bytelen, newlen);
1483 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1485 U8 tmpbuf[UTF8_MAXBYTES+1];
1486 uvchr_to_utf8(tmpbuf, c);
1487 return _is_utf8_FOO(classnum, tmpbuf);
1490 /* for now these are all defined (inefficiently) in terms of the utf8 versions.
1491 * Note that the macros in handy.h that call these short-circuit calling them
1492 * for Latin-1 range inputs */
1495 Perl_is_uni_alnum(pTHX_ UV c)
1497 U8 tmpbuf[UTF8_MAXBYTES+1];
1498 uvchr_to_utf8(tmpbuf, c);
1499 return _is_utf8_FOO(_CC_WORDCHAR, tmpbuf);
1503 Perl_is_uni_alnumc(pTHX_ UV c)
1505 U8 tmpbuf[UTF8_MAXBYTES+1];
1506 uvchr_to_utf8(tmpbuf, c);
1507 return _is_utf8_FOO(_CC_ALPHANUMERIC, tmpbuf);
1510 bool /* Internal function so we can deprecate the external one, and call
1511 this one from other deprecated functions in this file */
1512 S_is_utf8_idfirst(pTHX_ const U8 *p)
1518 /* is_utf8_idstart would be more logical. */
1519 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1523 Perl_is_uni_idfirst(pTHX_ UV c)
1525 U8 tmpbuf[UTF8_MAXBYTES+1];
1526 uvchr_to_utf8(tmpbuf, c);
1527 return S_is_utf8_idfirst(aTHX_ tmpbuf);
1531 Perl__is_uni_perl_idcont(pTHX_ UV c)
1533 U8 tmpbuf[UTF8_MAXBYTES+1];
1534 uvchr_to_utf8(tmpbuf, c);
1535 return _is_utf8_perl_idcont(tmpbuf);
1539 Perl__is_uni_perl_idstart(pTHX_ UV c)
1541 U8 tmpbuf[UTF8_MAXBYTES+1];
1542 uvchr_to_utf8(tmpbuf, c);
1543 return _is_utf8_perl_idstart(tmpbuf);
1547 Perl_is_uni_alpha(pTHX_ UV c)
1549 U8 tmpbuf[UTF8_MAXBYTES+1];
1550 uvchr_to_utf8(tmpbuf, c);
1551 return _is_utf8_FOO(_CC_ALPHA, tmpbuf);
1555 Perl_is_uni_ascii(pTHX_ UV c)
1561 Perl_is_uni_blank(pTHX_ UV c)
1563 return isBLANK_uni(c);
1567 Perl_is_uni_space(pTHX_ UV c)
1569 return isSPACE_uni(c);
1573 Perl_is_uni_digit(pTHX_ UV c)
1575 U8 tmpbuf[UTF8_MAXBYTES+1];
1576 uvchr_to_utf8(tmpbuf, c);
1577 return _is_utf8_FOO(_CC_DIGIT, tmpbuf);
1581 Perl_is_uni_upper(pTHX_ UV c)
1583 U8 tmpbuf[UTF8_MAXBYTES+1];
1584 uvchr_to_utf8(tmpbuf, c);
1585 return _is_utf8_FOO(_CC_UPPER, tmpbuf);
1589 Perl_is_uni_lower(pTHX_ UV c)
1591 U8 tmpbuf[UTF8_MAXBYTES+1];
1592 uvchr_to_utf8(tmpbuf, c);
1593 return _is_utf8_FOO(_CC_LOWER, tmpbuf);
1597 Perl_is_uni_cntrl(pTHX_ UV c)
1599 return isCNTRL_L1(c);
1603 Perl_is_uni_graph(pTHX_ UV c)
1605 U8 tmpbuf[UTF8_MAXBYTES+1];
1606 uvchr_to_utf8(tmpbuf, c);
1607 return _is_utf8_FOO(_CC_GRAPH, tmpbuf);
1611 Perl_is_uni_print(pTHX_ UV c)
1613 U8 tmpbuf[UTF8_MAXBYTES+1];
1614 uvchr_to_utf8(tmpbuf, c);
1615 return _is_utf8_FOO(_CC_PRINT, tmpbuf);
1619 Perl_is_uni_punct(pTHX_ UV c)
1621 U8 tmpbuf[UTF8_MAXBYTES+1];
1622 uvchr_to_utf8(tmpbuf, c);
1623 return _is_utf8_FOO(_CC_PUNCT, tmpbuf);
1627 Perl_is_uni_xdigit(pTHX_ UV c)
1629 return isXDIGIT_uni(c);
1633 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1635 /* We have the latin1-range values compiled into the core, so just use
1636 * those, converting the result to utf8. The only difference between upper
1637 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1638 * either "SS" or "Ss". Which one to use is passed into the routine in
1639 * 'S_or_s' to avoid a test */
1641 UV converted = toUPPER_LATIN1_MOD(c);
1643 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1645 assert(S_or_s == 'S' || S_or_s == 's');
1647 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1648 characters in this range */
1649 *p = (U8) converted;
1654 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1655 * which it maps to one of them, so as to only have to have one check for
1656 * it in the main case */
1657 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1659 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1660 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1663 converted = GREEK_CAPITAL_LETTER_MU;
1665 case LATIN_SMALL_LETTER_SHARP_S:
1671 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1672 assert(0); /* NOTREACHED */
1676 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1677 *p = UTF8_TWO_BYTE_LO(converted);
1683 /* Call the function to convert a UTF-8 encoded character to the specified case.
1684 * Note that there may be more than one character in the result.
1685 * INP is a pointer to the first byte of the input character
1686 * OUTP will be set to the first byte of the string of changed characters. It
1687 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1688 * LENP will be set to the length in bytes of the string of changed characters
1690 * The functions return the ordinal of the first character in the string of OUTP */
1691 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1692 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1693 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
1695 /* This additionally has the input parameter SPECIALS, which if non-zero will
1696 * cause this to use the SPECIALS hash for folding (meaning get full case
1697 * folding); otherwise, when zero, this implies a simple case fold */
1698 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
1701 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1705 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1706 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1707 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1708 * the changed version may be longer than the original character.
1710 * The ordinal of the first character of the changed version is returned
1711 * (but note, as explained above, that there may be more.) */
1713 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1716 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1719 uvchr_to_utf8(p, c);
1720 return CALL_UPPER_CASE(p, p, lenp);
1724 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1728 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1731 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1734 uvchr_to_utf8(p, c);
1735 return CALL_TITLE_CASE(p, p, lenp);
1739 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1741 /* We have the latin1-range values compiled into the core, so just use
1742 * those, converting the result to utf8. Since the result is always just
1743 * one character, we allow <p> to be NULL */
1745 U8 converted = toLOWER_LATIN1(c);
1748 if (UNI_IS_INVARIANT(converted)) {
1753 *p = UTF8_TWO_BYTE_HI(converted);
1754 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1762 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1766 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1769 return to_lower_latin1((U8) c, p, lenp);
1772 uvchr_to_utf8(p, c);
1773 return CALL_LOWER_CASE(p, p, lenp);
1777 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
1779 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
1784 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1786 if (c == MICRO_SIGN) {
1787 converted = GREEK_SMALL_LETTER_MU;
1789 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1795 else { /* In this range the fold of all other characters is their lower
1797 converted = toLOWER_LATIN1(c);
1800 if (UNI_IS_INVARIANT(converted)) {
1801 *p = (U8) converted;
1805 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1806 *p = UTF8_TWO_BYTE_LO(converted);
1814 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
1817 /* Not currently externally documented, and subject to change
1818 * <flags> bits meanings:
1819 * FOLD_FLAGS_FULL iff full folding is to be used;
1820 * FOLD_FLAGS_LOCALE iff in locale
1821 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1824 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1827 UV result = _to_fold_latin1((U8) c, p, lenp,
1828 cBOOL(((flags & FOLD_FLAGS_FULL)
1829 /* If ASCII-safe, don't allow full folding,
1830 * as that could include SHARP S => ss;
1831 * otherwise there is no crossing of
1832 * ascii/non-ascii in the latin1 range */
1833 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1834 /* It is illegal for the fold to cross the 255/256 boundary under
1835 * locale; in this case return the original */
1836 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1841 /* If no special needs, just use the macro */
1842 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1843 uvchr_to_utf8(p, c);
1844 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1846 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1847 the special flags. */
1848 U8 utf8_c[UTF8_MAXBYTES + 1];
1849 uvchr_to_utf8(utf8_c, c);
1850 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1855 Perl_is_uni_alnum_lc(pTHX_ UV c)
1858 return isALNUM_LC(UNI_TO_NATIVE(c));
1860 return _is_uni_FOO(_CC_WORDCHAR, c);
1864 Perl_is_uni_alnumc_lc(pTHX_ UV c)
1867 return isALPHANUMERIC_LC(UNI_TO_NATIVE(c));
1869 return _is_uni_FOO(_CC_ALPHANUMERIC, c);
1873 Perl_is_uni_idfirst_lc(pTHX_ UV c)
1876 return isIDFIRST_LC(UNI_TO_NATIVE(c));
1878 return _is_uni_perl_idstart(c);
1882 Perl_is_uni_alpha_lc(pTHX_ UV c)
1885 return isALPHA_LC(UNI_TO_NATIVE(c));
1887 return _is_uni_FOO(_CC_ALPHA, c);
1891 Perl_is_uni_ascii_lc(pTHX_ UV c)
1894 return isASCII_LC(UNI_TO_NATIVE(c));
1900 Perl_is_uni_blank_lc(pTHX_ UV c)
1903 return isBLANK_LC(UNI_TO_NATIVE(c));
1905 return isBLANK_uni(c);
1909 Perl_is_uni_space_lc(pTHX_ UV c)
1912 return isSPACE_LC(UNI_TO_NATIVE(c));
1914 return isSPACE_uni(c);
1918 Perl_is_uni_digit_lc(pTHX_ UV c)
1921 return isDIGIT_LC(UNI_TO_NATIVE(c));
1923 return _is_uni_FOO(_CC_DIGIT, c);
1927 Perl_is_uni_upper_lc(pTHX_ UV c)
1930 return isUPPER_LC(UNI_TO_NATIVE(c));
1932 return _is_uni_FOO(_CC_UPPER, c);
1936 Perl_is_uni_lower_lc(pTHX_ UV c)
1939 return isLOWER_LC(UNI_TO_NATIVE(c));
1941 return _is_uni_FOO(_CC_LOWER, c);
1945 Perl_is_uni_cntrl_lc(pTHX_ UV c)
1948 return isCNTRL_LC(UNI_TO_NATIVE(c));
1954 Perl_is_uni_graph_lc(pTHX_ UV c)
1957 return isGRAPH_LC(UNI_TO_NATIVE(c));
1959 return _is_uni_FOO(_CC_GRAPH, c);
1963 Perl_is_uni_print_lc(pTHX_ UV c)
1966 return isPRINT_LC(UNI_TO_NATIVE(c));
1968 return _is_uni_FOO(_CC_PRINT, c);
1972 Perl_is_uni_punct_lc(pTHX_ UV c)
1975 return isPUNCT_LC(UNI_TO_NATIVE(c));
1977 return _is_uni_FOO(_CC_PUNCT, c);
1981 Perl_is_uni_xdigit_lc(pTHX_ UV c)
1984 return isXDIGIT_LC(UNI_TO_NATIVE(c));
1986 return isXDIGIT_uni(c);
1990 Perl_to_uni_upper_lc(pTHX_ U32 c)
1992 /* XXX returns only the first character -- do not use XXX */
1993 /* XXX no locale support yet */
1995 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1996 return (U32)to_uni_upper(c, tmpbuf, &len);
2000 Perl_to_uni_title_lc(pTHX_ U32 c)
2002 /* XXX returns only the first character XXX -- do not use XXX */
2003 /* XXX no locale support yet */
2005 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2006 return (U32)to_uni_title(c, tmpbuf, &len);
2010 Perl_to_uni_lower_lc(pTHX_ U32 c)
2012 /* XXX returns only the first character -- do not use XXX */
2013 /* XXX no locale support yet */
2015 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2016 return (U32)to_uni_lower(c, tmpbuf, &len);
2019 PERL_STATIC_INLINE bool
2020 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2021 const char *const swashname)
2023 /* returns a boolean giving whether or not the UTF8-encoded character that
2024 * starts at <p> is in the swash indicated by <swashname>. <swash>
2025 * contains a pointer to where the swash indicated by <swashname>
2026 * is to be stored; which this routine will do, so that future calls will
2027 * look at <*swash> and only generate a swash if it is not null
2029 * Note that it is assumed that the buffer length of <p> is enough to
2030 * contain all the bytes that comprise the character. Thus, <*p> should
2031 * have been checked before this call for mal-formedness enough to assure
2036 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2038 /* The API should have included a length for the UTF-8 character in <p>,
2039 * but it doesn't. We therefore assume that p has been validated at least
2040 * as far as there being enough bytes available in it to accommodate the
2041 * character without reading beyond the end, and pass that number on to the
2042 * validating routine */
2043 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
2044 if (ckWARN_d(WARN_UTF8)) {
2045 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
2046 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
2047 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
2048 what the malformation is */
2049 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
2055 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2056 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
2059 return swash_fetch(*swash, p, TRUE) != 0;
2063 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2067 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2069 assert(classnum < _FIRST_NON_SWASH_CC);
2071 return is_utf8_common(p, &PL_utf8_swash_ptrs[classnum], swash_property_names[classnum]);
2075 Perl_is_utf8_alnum(pTHX_ const U8 *p)
2079 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
2081 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
2082 * descendant of isalnum(3), in other words, it doesn't
2083 * contain the '_'. --jhi */
2084 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_WORDCHAR], "IsWord");
2088 Perl_is_utf8_alnumc(pTHX_ const U8 *p)
2092 PERL_ARGS_ASSERT_IS_UTF8_ALNUMC;
2094 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHANUMERIC], "IsAlnum");
2098 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
2102 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
2104 return S_is_utf8_idfirst(aTHX_ p);
2108 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
2112 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
2116 /* is_utf8_idstart would be more logical. */
2117 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2121 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
2125 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
2127 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2131 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2135 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2137 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont");
2142 Perl_is_utf8_idcont(pTHX_ const U8 *p)
2146 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2148 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
2152 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2156 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2158 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2162 Perl_is_utf8_alpha(pTHX_ const U8 *p)
2166 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2168 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHA], "IsAlpha");
2172 Perl_is_utf8_ascii(pTHX_ const U8 *p)
2176 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2178 /* ASCII characters are the same whether in utf8 or not. So the macro
2179 * works on both utf8 and non-utf8 representations. */
2184 Perl_is_utf8_blank(pTHX_ const U8 *p)
2188 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2190 return isBLANK_utf8(p);
2194 Perl_is_utf8_space(pTHX_ const U8 *p)
2198 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2200 return isSPACE_utf8(p);
2204 Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2208 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2210 /* Only true if is an ASCII space-like character, and ASCII is invariant
2211 * under utf8, so can just use the macro */
2212 return isSPACE_A(*p);
2216 Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2220 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2222 /* Only true if is an ASCII word character, and ASCII is invariant
2223 * under utf8, so can just use the macro */
2224 return isWORDCHAR_A(*p);
2228 Perl_is_utf8_digit(pTHX_ const U8 *p)
2232 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2234 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_DIGIT], "IsDigit");
2238 Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2242 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2244 /* Only true if is an ASCII digit character, and ASCII is invariant
2245 * under utf8, so can just use the macro */
2246 return isDIGIT_A(*p);
2250 Perl_is_utf8_upper(pTHX_ const U8 *p)
2254 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2256 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_UPPER], "IsUppercase");
2260 Perl_is_utf8_lower(pTHX_ const U8 *p)
2264 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2266 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_LOWER], "IsLowercase");
2270 Perl_is_utf8_cntrl(pTHX_ const U8 *p)
2274 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2276 return isCNTRL_utf8(p);
2280 Perl_is_utf8_graph(pTHX_ const U8 *p)
2284 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2286 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_GRAPH], "IsGraph");
2290 Perl_is_utf8_print(pTHX_ const U8 *p)
2294 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2296 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PRINT], "IsPrint");
2300 Perl_is_utf8_punct(pTHX_ const U8 *p)
2304 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2306 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PUNCT], "IsPunct");
2310 Perl_is_utf8_xdigit(pTHX_ const U8 *p)
2314 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2316 return is_XDIGIT_utf8(p);
2320 Perl__is_utf8_mark(pTHX_ const U8 *p)
2324 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2326 return is_utf8_common(p, &PL_utf8_mark, "IsM");
2331 Perl_is_utf8_mark(pTHX_ const U8 *p)
2335 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2337 return _is_utf8_mark(p);
2341 =for apidoc to_utf8_case
2343 The C<p> contains the pointer to the UTF-8 string encoding
2344 the character that is being converted. This routine assumes that the character
2345 at C<p> is well-formed.
2347 The C<ustrp> is a pointer to the character buffer to put the
2348 conversion result to. The C<lenp> is a pointer to the length
2351 The C<swashp> is a pointer to the swash to use.
2353 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2354 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
2355 but not always, a multicharacter mapping), is tried first.
2357 The C<special> is a string like "utf8::ToSpecLower", which means the
2358 hash %utf8::ToSpecLower. The access to the hash is through
2359 Perl_to_utf8_case().
2361 The C<normal> is a string like "ToLower" which means the swash
2367 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2368 SV **swashp, const char *normal, const char *special)
2371 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2373 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
2374 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2375 * are necessary in EBCDIC, they are redundant no-ops
2376 * in ASCII-ish platforms, and hopefully optimized away. */
2377 const UV uv1 = NATIVE_TO_UNI(uv0);
2379 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2381 /* Note that swash_fetch() doesn't output warnings for these because it
2382 * assumes we will */
2383 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2384 if (uv1 <= UNICODE_SURROGATE_LAST) {
2385 if (ckWARN_d(WARN_SURROGATE)) {
2386 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2387 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2388 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2391 else if (UNICODE_IS_SUPER(uv1)) {
2392 if (ckWARN_d(WARN_NON_UNICODE)) {
2393 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2394 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2395 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2399 /* Note that non-characters are perfectly legal, so no warning should
2403 uvuni_to_utf8(tmpbuf, uv1);
2405 if (!*swashp) /* load on-demand */
2406 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2409 /* It might be "special" (sometimes, but not always,
2410 * a multicharacter mapping) */
2411 HV * const hv = get_hv(special, 0);
2415 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2419 s = SvPV_const(*svp, len);
2421 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2424 /* If we have EBCDIC we need to remap the characters
2425 * since any characters in the low 256 are Unicode
2426 * code points, not EBCDIC. */
2427 U8 *t = (U8*)s, *tend = t + len, *d;
2434 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2436 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2445 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2450 Copy(tmpbuf, ustrp, len, U8);
2452 Copy(s, ustrp, len, U8);
2458 if (!len && *swashp) {
2459 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
2462 /* It was "normal" (a single character mapping). */
2463 const UV uv3 = UNI_TO_NATIVE(uv2);
2464 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2472 return valid_utf8_to_uvchr(ustrp, 0);
2475 /* Here, there was no mapping defined, which means that the code point maps
2476 * to itself. Return the inputs */
2478 if (p != ustrp) { /* Don't copy onto itself */
2479 Copy(p, ustrp, len, U8);
2490 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2492 /* This is called when changing the case of a utf8-encoded character above
2493 * the Latin1 range, and the operation is in locale. If the result
2494 * contains a character that crosses the 255/256 boundary, disallow the
2495 * change, and return the original code point. See L<perlfunc/lc> for why;
2497 * p points to the original string whose case was changed; assumed
2498 * by this routine to be well-formed
2499 * result the code point of the first character in the changed-case string
2500 * ustrp points to the changed-case string (<result> represents its first char)
2501 * lenp points to the length of <ustrp> */
2503 UV original; /* To store the first code point of <p> */
2505 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2507 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2509 /* We know immediately if the first character in the string crosses the
2510 * boundary, so can skip */
2513 /* Look at every character in the result; if any cross the
2514 * boundary, the whole thing is disallowed */
2515 U8* s = ustrp + UTF8SKIP(ustrp);
2516 U8* e = ustrp + *lenp;
2518 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2525 /* Here, no characters crossed, result is ok as-is */
2531 /* Failed, have to return the original */
2532 original = valid_utf8_to_uvchr(p, lenp);
2533 Copy(p, ustrp, *lenp, char);
2538 =for apidoc to_utf8_upper
2540 Convert the UTF-8 encoded character at C<p> to its uppercase version and
2541 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2542 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2543 the uppercase version may be longer than the original character.
2545 The first character of the uppercased version is returned
2546 (but note, as explained above, that there may be more.)
2548 The character at C<p> is assumed by this routine to be well-formed.
2552 /* Not currently externally documented, and subject to change:
2553 * <flags> is set iff locale semantics are to be used for code points < 256
2554 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2555 * were used in the calculation; otherwise unchanged. */
2558 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2564 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2566 if (UTF8_IS_INVARIANT(*p)) {
2568 result = toUPPER_LC(*p);
2571 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2574 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2576 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2579 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2583 else { /* utf8, ord above 255 */
2584 result = CALL_UPPER_CASE(p, ustrp, lenp);
2587 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2592 /* Here, used locale rules. Convert back to utf8 */
2593 if (UTF8_IS_INVARIANT(result)) {
2594 *ustrp = (U8) result;
2598 *ustrp = UTF8_EIGHT_BIT_HI(result);
2599 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2604 *tainted_ptr = TRUE;
2610 =for apidoc to_utf8_title
2612 Convert the UTF-8 encoded character at C<p> to its titlecase version and
2613 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2614 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2615 titlecase version may be longer than the original character.
2617 The first character of the titlecased version is returned
2618 (but note, as explained above, that there may be more.)
2620 The character at C<p> is assumed by this routine to be well-formed.
2624 /* Not currently externally documented, and subject to change:
2625 * <flags> is set iff locale semantics are to be used for code points < 256
2626 * Since titlecase is not defined in POSIX, uppercase is used instead
2628 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2629 * were used in the calculation; otherwise unchanged. */
2632 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2638 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2640 if (UTF8_IS_INVARIANT(*p)) {
2642 result = toUPPER_LC(*p);
2645 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2648 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2650 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2653 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2657 else { /* utf8, ord above 255 */
2658 result = CALL_TITLE_CASE(p, ustrp, lenp);
2661 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2666 /* Here, used locale rules. Convert back to utf8 */
2667 if (UTF8_IS_INVARIANT(result)) {
2668 *ustrp = (U8) result;
2672 *ustrp = UTF8_EIGHT_BIT_HI(result);
2673 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2678 *tainted_ptr = TRUE;
2684 =for apidoc to_utf8_lower
2686 Convert the UTF-8 encoded character at C<p> to its lowercase version and
2687 store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2688 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2689 lowercase version may be longer than the original character.
2691 The first character of the lowercased version is returned
2692 (but note, as explained above, that there may be more.)
2694 The character at C<p> is assumed by this routine to be well-formed.
2698 /* Not currently externally documented, and subject to change:
2699 * <flags> is set iff locale semantics are to be used for code points < 256
2700 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2701 * were used in the calculation; otherwise unchanged. */
2704 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2710 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2712 if (UTF8_IS_INVARIANT(*p)) {
2714 result = toLOWER_LC(*p);
2717 return to_lower_latin1(*p, ustrp, lenp);
2720 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2722 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2725 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2729 else { /* utf8, ord above 255 */
2730 result = CALL_LOWER_CASE(p, ustrp, lenp);
2733 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2739 /* Here, used locale rules. Convert back to utf8 */
2740 if (UTF8_IS_INVARIANT(result)) {
2741 *ustrp = (U8) result;
2745 *ustrp = UTF8_EIGHT_BIT_HI(result);
2746 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2751 *tainted_ptr = TRUE;
2757 =for apidoc to_utf8_fold
2759 Convert the UTF-8 encoded character at C<p> to its foldcase version and
2760 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2761 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2762 foldcase version may be longer than the original character (up to
2765 The first character of the foldcased version is returned
2766 (but note, as explained above, that there may be more.)
2768 The character at C<p> is assumed by this routine to be well-formed.
2772 /* Not currently externally documented, and subject to change,
2774 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2775 * points < 256. Since foldcase is not defined in
2776 * POSIX, lowercase is used instead
2777 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2778 * otherwise simple folds
2779 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2781 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2782 * were used in the calculation; otherwise unchanged. */
2785 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2791 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2793 /* These are mutually exclusive */
2794 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2796 assert(p != ustrp); /* Otherwise overwrites */
2798 if (UTF8_IS_INVARIANT(*p)) {
2799 if (flags & FOLD_FLAGS_LOCALE) {
2800 result = toLOWER_LC(*p);
2803 return _to_fold_latin1(*p, ustrp, lenp,
2804 cBOOL(flags & FOLD_FLAGS_FULL));
2807 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2808 if (flags & FOLD_FLAGS_LOCALE) {
2809 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2812 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2814 cBOOL((flags & FOLD_FLAGS_FULL
2815 /* If ASCII safe, don't allow full
2816 * folding, as that could include SHARP
2817 * S => ss; otherwise there is no
2818 * crossing of ascii/non-ascii in the
2820 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
2823 else { /* utf8, ord above 255 */
2824 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2826 if ((flags & FOLD_FLAGS_LOCALE)) {
2827 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2829 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2833 /* This is called when changing the case of a utf8-encoded
2834 * character above the Latin1 range, and the result should not
2835 * contain an ASCII character. */
2837 UV original; /* To store the first code point of <p> */
2839 /* Look at every character in the result; if any cross the
2840 * boundary, the whole thing is disallowed */
2842 U8* e = ustrp + *lenp;
2845 /* Crossed, have to return the original */
2846 original = valid_utf8_to_uvchr(p, lenp);
2847 Copy(p, ustrp, *lenp, char);
2853 /* Here, no characters crossed, result is ok as-is */
2858 /* Here, used locale rules. Convert back to utf8 */
2859 if (UTF8_IS_INVARIANT(result)) {
2860 *ustrp = (U8) result;
2864 *ustrp = UTF8_EIGHT_BIT_HI(result);
2865 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2870 *tainted_ptr = TRUE;
2876 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2877 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2878 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2882 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2884 PERL_ARGS_ASSERT_SWASH_INIT;
2886 /* Returns a copy of a swash initiated by the called function. This is the
2887 * public interface, and returning a copy prevents others from doing
2888 * mischief on the original */
2890 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2894 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2896 /* Initialize and return a swash, creating it if necessary. It does this
2897 * by calling utf8_heavy.pl in the general case. The returned value may be
2898 * the swash's inversion list instead if the input parameters allow it.
2899 * Which is returned should be immaterial to callers, as the only
2900 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2901 * and swash_to_invlist() handle both these transparently.
2903 * This interface should only be used by functions that won't destroy or
2904 * adversely change the swash, as doing so affects all other uses of the
2905 * swash in the program; the general public should use 'Perl_swash_init'
2908 * pkg is the name of the package that <name> should be in.
2909 * name is the name of the swash to find. Typically it is a Unicode
2910 * property name, including user-defined ones
2911 * listsv is a string to initialize the swash with. It must be of the form
2912 * documented as the subroutine return value in
2913 * L<perlunicode/User-Defined Character Properties>
2914 * minbits is the number of bits required to represent each data element.
2915 * It is '1' for binary properties.
2916 * none I (khw) do not understand this one, but it is used only in tr///.
2917 * invlist is an inversion list to initialize the swash with (or NULL)
2918 * flags_p if non-NULL is the address of various input and output flag bits
2919 * to the routine, as follows: ('I' means is input to the routine;
2920 * 'O' means output from the routine. Only flags marked O are
2921 * meaningful on return.)
2922 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2923 * came from a user-defined property. (I O)
2924 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2925 * when the swash cannot be located, to simply return NULL. (I)
2926 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2927 * return of an inversion list instead of a swash hash if this routine
2928 * thinks that would result in faster execution of swash_fetch() later
2931 * Thus there are three possible inputs to find the swash: <name>,
2932 * <listsv>, and <invlist>. At least one must be specified. The result
2933 * will be the union of the specified ones, although <listsv>'s various
2934 * actions can intersect, etc. what <name> gives.
2936 * <invlist> is only valid for binary properties */
2939 SV* retval = &PL_sv_undef;
2940 HV* swash_hv = NULL;
2941 const int invlist_swash_boundary =
2942 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2943 ? 512 /* Based on some benchmarking, but not extensive, see commit
2945 : -1; /* Never return just an inversion list */
2947 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2948 assert(! invlist || minbits == 1);
2950 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2952 if (listsv != &PL_sv_undef || strNE(name, "")) {
2954 const size_t pkg_len = strlen(pkg);
2955 const size_t name_len = strlen(name);
2956 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2960 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2962 PUSHSTACKi(PERLSI_MAGIC);
2966 /* We might get here via a subroutine signature which uses a utf8
2967 * parameter name, at which point PL_subname will have been set
2968 * but not yet used. */
2969 save_item(PL_subname);
2970 if (PL_parser && PL_parser->error_count)
2971 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2972 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2973 if (!method) { /* demand load utf8 */
2975 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2976 GvSV(PL_errgv) = NULL;
2977 /* It is assumed that callers of this routine are not passing in
2978 * any user derived data. */
2979 /* Need to do this after save_re_context() as it will set
2980 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2981 * in Perl_magic_get). Even line to create errsv_save can turn on
2983 #ifndef NO_TAINT_SUPPORT
2984 SAVEBOOL(TAINT_get);
2987 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2990 /* Not ERRSV, as there is no need to vivify a scalar we are
2991 about to discard. */
2992 SV * const errsv = GvSV(PL_errgv);
2993 if (!SvTRUE(errsv)) {
2994 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2995 SvREFCNT_dec(errsv);
3003 mPUSHp(pkg, pkg_len);
3004 mPUSHp(name, name_len);
3009 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3010 GvSV(PL_errgv) = NULL;
3011 /* If we already have a pointer to the method, no need to use
3012 * call_method() to repeat the lookup. */
3014 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3015 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3017 retval = *PL_stack_sp--;
3018 SvREFCNT_inc(retval);
3021 /* Not ERRSV. See above. */
3022 SV * const errsv = GvSV(PL_errgv);
3023 if (!SvTRUE(errsv)) {
3024 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3025 SvREFCNT_dec(errsv);
3030 if (IN_PERL_COMPILETIME) {
3031 CopHINTS_set(PL_curcop, PL_hints);
3033 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3036 /* If caller wants to handle missing properties, let them */
3037 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3041 "Can't find Unicode property definition \"%"SVf"\"",
3043 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3045 } /* End of calling the module to find the swash */
3047 /* If this operation fetched a swash, and we will need it later, get it */
3048 if (retval != &PL_sv_undef
3049 && (minbits == 1 || (flags_p
3051 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3053 swash_hv = MUTABLE_HV(SvRV(retval));
3055 /* If we don't already know that there is a user-defined component to
3056 * this swash, and the user has indicated they wish to know if there is
3057 * one (by passing <flags_p>), find out */
3058 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3059 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3060 if (user_defined && SvUV(*user_defined)) {
3061 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3066 /* Make sure there is an inversion list for binary properties */
3068 SV** swash_invlistsvp = NULL;
3069 SV* swash_invlist = NULL;
3070 bool invlist_in_swash_is_valid = FALSE;
3071 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3072 an unclaimed reference count */
3074 /* If this operation fetched a swash, get its already existing
3075 * inversion list, or create one for it */
3078 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3079 if (swash_invlistsvp) {
3080 swash_invlist = *swash_invlistsvp;
3081 invlist_in_swash_is_valid = TRUE;
3084 swash_invlist = _swash_to_invlist(retval);
3085 swash_invlist_unclaimed = TRUE;
3089 /* If an inversion list was passed in, have to include it */
3092 /* Any fetched swash will by now have an inversion list in it;
3093 * otherwise <swash_invlist> will be NULL, indicating that we
3094 * didn't fetch a swash */
3095 if (swash_invlist) {
3097 /* Add the passed-in inversion list, which invalidates the one
3098 * already stored in the swash */
3099 invlist_in_swash_is_valid = FALSE;
3100 _invlist_union(invlist, swash_invlist, &swash_invlist);
3104 /* Here, there is no swash already. Set up a minimal one, if
3105 * we are going to return a swash */
3106 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3108 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3110 swash_invlist = invlist;
3114 /* Here, we have computed the union of all the passed-in data. It may
3115 * be that there was an inversion list in the swash which didn't get
3116 * touched; otherwise save the one computed one */
3117 if (! invlist_in_swash_is_valid
3118 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3120 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3122 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3124 /* We just stole a reference count. */
3125 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3126 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3129 /* Use the inversion list stand-alone if small enough */
3130 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3131 SvREFCNT_dec(retval);
3132 if (!swash_invlist_unclaimed)
3133 SvREFCNT_inc_simple_void_NN(swash_invlist);
3134 retval = newRV_noinc(swash_invlist);
3142 /* This API is wrong for special case conversions since we may need to
3143 * return several Unicode characters for a single Unicode character
3144 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3145 * the lower-level routine, and it is similarly broken for returning
3146 * multiple values. --jhi
3147 * For those, you should use to_utf8_case() instead */
3148 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3151 * Returns the value of property/mapping C<swash> for the first character
3152 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3153 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3154 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3156 * A "swash" is a hash which contains initially the keys/values set up by
3157 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3158 * property for all possible code points. Things are stored in a compact form
3159 * (see utf8_heavy.pl) so that calculation is required to find the actual
3160 * property value for a given code point. As code points are looked up, new
3161 * key/value pairs are added to the hash, so that the calculation doesn't have
3162 * to ever be re-done. Further, each calculation is done, not just for the
3163 * desired one, but for a whole block of code points adjacent to that one.
3164 * For binary properties on ASCII machines, the block is usually for 64 code
3165 * points, starting with a code point evenly divisible by 64. Thus if the
3166 * property value for code point 257 is requested, the code goes out and
3167 * calculates the property values for all 64 code points between 256 and 319,
3168 * and stores these as a single 64-bit long bit vector, called a "swatch",
3169 * under the key for code point 256. The key is the UTF-8 encoding for code
3170 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3171 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3172 * for code point 258 is then requested, this code realizes that it would be
3173 * stored under the key for 256, and would find that value and extract the
3174 * relevant bit, offset from 256.
3176 * Non-binary properties are stored in as many bits as necessary to represent
3177 * their values (32 currently, though the code is more general than that), not
3178 * as single bits, but the principal is the same: the value for each key is a
3179 * vector that encompasses the property values for all code points whose UTF-8
3180 * representations are represented by the key. That is, for all code points
3181 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3185 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3188 HV *const hv = MUTABLE_HV(SvRV(swash));
3193 const U8 *tmps = NULL;
3197 const UV c = NATIVE_TO_ASCII(*ptr);
3199 PERL_ARGS_ASSERT_SWASH_FETCH;
3201 /* If it really isn't a hash, it isn't really swash; must be an inversion
3203 if (SvTYPE(hv) != SVt_PVHV) {
3204 return _invlist_contains_cp((SV*)hv,
3206 ? valid_utf8_to_uvchr(ptr, NULL)
3210 /* Convert to utf8 if not already */
3211 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
3212 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3213 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3216 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
3217 * then the "swatch" is a vec() for all the chars which start
3219 * So the key in the hash (klen) is length of encoded char -1
3221 klen = UTF8SKIP(ptr) - 1;
3225 /* If char is invariant then swatch is for all the invariant chars
3226 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
3228 needents = UTF_CONTINUATION_MARK;
3229 off = NATIVE_TO_UTF(ptr[klen]);
3232 /* If char is encoded then swatch is for the prefix */
3233 needents = (1 << UTF_ACCUMULATION_SHIFT);
3234 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3238 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3239 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3240 * it's nothing to sniff at.) Pity we usually come through at least
3241 * two function calls to get here...
3243 * NB: this code assumes that swatches are never modified, once generated!
3246 if (hv == PL_last_swash_hv &&
3247 klen == PL_last_swash_klen &&
3248 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3250 tmps = PL_last_swash_tmps;
3251 slen = PL_last_swash_slen;
3254 /* Try our second-level swatch cache, kept in a hash. */
3255 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3257 /* If not cached, generate it via swatch_get */
3258 if (!svp || !SvPOK(*svp)
3259 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
3260 /* We use utf8n_to_uvuni() as we want an index into
3261 Unicode tables, not a native character number.
3263 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
3265 0 : UTF8_ALLOW_ANY);
3266 swatch = swatch_get(swash,
3267 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
3268 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
3271 if (IN_PERL_COMPILETIME)
3272 CopHINTS_set(PL_curcop, PL_hints);
3274 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3276 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3277 || (slen << 3) < needents)
3278 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3279 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3280 svp, tmps, (UV)slen, (UV)needents);
3283 PL_last_swash_hv = hv;
3284 assert(klen <= sizeof(PL_last_swash_key));
3285 PL_last_swash_klen = (U8)klen;
3286 /* FIXME change interpvar.h? */
3287 PL_last_swash_tmps = (U8 *) tmps;
3288 PL_last_swash_slen = slen;
3290 Copy(ptr, PL_last_swash_key, klen, U8);
3293 switch ((int)((slen << 3) / needents)) {
3295 bit = 1 << (off & 7);
3297 return (tmps[off] & bit) != 0;
3302 return (tmps[off] << 8) + tmps[off + 1] ;
3305 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3307 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3308 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
3309 NORETURN_FUNCTION_END;
3312 /* Read a single line of the main body of the swash input text. These are of
3315 * where each number is hex. The first two numbers form the minimum and
3316 * maximum of a range, and the third is the value associated with the range.
3317 * Not all swashes should have a third number
3319 * On input: l points to the beginning of the line to be examined; it points
3320 * to somewhere in the string of the whole input text, and is
3321 * terminated by a \n or the null string terminator.
3322 * lend points to the null terminator of that string
3323 * wants_value is non-zero if the swash expects a third number
3324 * typestr is the name of the swash's mapping, like 'ToLower'
3325 * On output: *min, *max, and *val are set to the values read from the line.
3326 * returns a pointer just beyond the line examined. If there was no
3327 * valid min number on the line, returns lend+1
3331 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3332 const bool wants_value, const U8* const typestr)
3334 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3335 STRLEN numlen; /* Length of the number */
3336 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3337 | PERL_SCAN_DISALLOW_PREFIX
3338 | PERL_SCAN_SILENT_NON_PORTABLE;
3340 /* nl points to the next \n in the scan */
3341 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3343 /* Get the first number on the line: the range minimum */
3345 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3346 if (numlen) /* If found a hex number, position past it */
3348 else if (nl) { /* Else, go handle next line, if any */
3349 return nl + 1; /* 1 is length of "\n" */
3351 else { /* Else, no next line */
3352 return lend + 1; /* to LIST's end at which \n is not found */
3355 /* The max range value follows, separated by a BLANK */
3358 flags = PERL_SCAN_SILENT_ILLDIGIT
3359 | PERL_SCAN_DISALLOW_PREFIX
3360 | PERL_SCAN_SILENT_NON_PORTABLE;
3362 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3365 else /* If no value here, it is a single element range */
3368 /* Non-binary tables have a third entry: what the first element of the
3374 /* The ToLc, etc table mappings are not in hex, and must be
3375 * corrected by adding the code point to them */
3377 char *after_strtol = (char *) lend;
3378 *val = Strtol((char *)l, &after_strtol, 10);
3379 l = (U8 *) after_strtol;
3381 else { /* Other tables are in hex, and are the correct result
3383 flags = PERL_SCAN_SILENT_ILLDIGIT
3384 | PERL_SCAN_DISALLOW_PREFIX
3385 | PERL_SCAN_SILENT_NON_PORTABLE;
3387 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3397 /* diag_listed_as: To%s: illegal mapping '%s' */
3398 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3404 *val = 0; /* bits == 1, then any val should be ignored */
3406 else { /* Nothing following range min, should be single element with no
3412 /* diag_listed_as: To%s: illegal mapping '%s' */
3413 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3417 *val = 0; /* bits == 1, then val should be ignored */
3420 /* Position to next line if any, or EOF */
3430 * Returns a swatch (a bit vector string) for a code point sequence
3431 * that starts from the value C<start> and comprises the number C<span>.
3432 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3433 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3436 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3439 U8 *l, *lend, *x, *xend, *s, *send;
3440 STRLEN lcur, xcur, scur;
3441 HV *const hv = MUTABLE_HV(SvRV(swash));
3442 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3444 SV** listsvp = NULL; /* The string containing the main body of the table */
3445 SV** extssvp = NULL;
3446 SV** invert_it_svp = NULL;
3449 STRLEN octets; /* if bits == 1, then octets == 0 */
3451 UV end = start + span;
3453 if (invlistsvp == NULL) {
3454 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3455 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3456 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3457 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3458 listsvp = hv_fetchs(hv, "LIST", FALSE);
3459 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3461 bits = SvUV(*bitssvp);
3462 none = SvUV(*nonesvp);
3463 typestr = (U8*)SvPV_nolen(*typesvp);
3469 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3471 PERL_ARGS_ASSERT_SWATCH_GET;
3473 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3474 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3478 /* If overflowed, use the max possible */
3484 /* create and initialize $swatch */
3485 scur = octets ? (span * octets) : (span + 7) / 8;
3486 swatch = newSV(scur);
3488 s = (U8*)SvPVX(swatch);
3489 if (octets && none) {
3490 const U8* const e = s + scur;
3493 *s++ = (U8)(none & 0xff);
3494 else if (bits == 16) {
3495 *s++ = (U8)((none >> 8) & 0xff);
3496 *s++ = (U8)( none & 0xff);
3498 else if (bits == 32) {
3499 *s++ = (U8)((none >> 24) & 0xff);
3500 *s++ = (U8)((none >> 16) & 0xff);
3501 *s++ = (U8)((none >> 8) & 0xff);
3502 *s++ = (U8)( none & 0xff);
3508 (void)memzero((U8*)s, scur + 1);
3510 SvCUR_set(swatch, scur);
3511 s = (U8*)SvPVX(swatch);
3513 if (invlistsvp) { /* If has an inversion list set up use that */
3514 _invlist_populate_swatch(*invlistsvp, start, end, s);
3518 /* read $swash->{LIST} */
3519 l = (U8*)SvPV(*listsvp, lcur);
3522 UV min, max, val, upper;
3523 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3524 cBOOL(octets), typestr);
3529 /* If looking for something beyond this range, go try the next one */
3533 /* <end> is generally 1 beyond where we want to set things, but at the
3534 * platform's infinity, where we can't go any higher, we want to
3535 * include the code point at <end> */
3538 : (max != UV_MAX || end != UV_MAX)
3545 if (!none || val < none) {
3550 for (key = min; key <= upper; key++) {
3552 /* offset must be non-negative (start <= min <= key < end) */
3553 offset = octets * (key - start);
3555 s[offset] = (U8)(val & 0xff);
3556 else if (bits == 16) {
3557 s[offset ] = (U8)((val >> 8) & 0xff);
3558 s[offset + 1] = (U8)( val & 0xff);
3560 else if (bits == 32) {
3561 s[offset ] = (U8)((val >> 24) & 0xff);
3562 s[offset + 1] = (U8)((val >> 16) & 0xff);
3563 s[offset + 2] = (U8)((val >> 8) & 0xff);
3564 s[offset + 3] = (U8)( val & 0xff);
3567 if (!none || val < none)
3571 else { /* bits == 1, then val should be ignored */
3576 for (key = min; key <= upper; key++) {
3577 const STRLEN offset = (STRLEN)(key - start);
3578 s[offset >> 3] |= 1 << (offset & 7);
3583 /* Invert if the data says it should be. Assumes that bits == 1 */
3584 if (invert_it_svp && SvUV(*invert_it_svp)) {
3586 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3587 * be 0, and their inversion should also be 0, as we don't succeed any
3588 * Unicode property matches for non-Unicode code points */
3589 if (start <= PERL_UNICODE_MAX) {
3591 /* The code below assumes that we never cross the
3592 * Unicode/above-Unicode boundary in a range, as otherwise we would
3593 * have to figure out where to stop flipping the bits. Since this
3594 * boundary is divisible by a large power of 2, and swatches comes
3595 * in small powers of 2, this should be a valid assumption */
3596 assert(start + span - 1 <= PERL_UNICODE_MAX);
3606 /* read $swash->{EXTRAS}
3607 * This code also copied to swash_to_invlist() below */
3608 x = (U8*)SvPV(*extssvp, xcur);
3616 SV **otherbitssvp, *other;
3620 const U8 opc = *x++;
3624 nl = (U8*)memchr(x, '\n', xend - x);
3626 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3628 x = nl + 1; /* 1 is length of "\n" */
3632 x = xend; /* to EXTRAS' end at which \n is not found */
3639 namelen = nl - namestr;
3643 namelen = xend - namestr;
3647 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3648 otherhv = MUTABLE_HV(SvRV(*othersvp));
3649 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3650 otherbits = (STRLEN)SvUV(*otherbitssvp);
3651 if (bits < otherbits)
3652 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3653 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3655 /* The "other" swatch must be destroyed after. */
3656 other = swatch_get(*othersvp, start, span);
3657 o = (U8*)SvPV(other, olen);
3660 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3662 s = (U8*)SvPV(swatch, slen);
3663 if (bits == 1 && otherbits == 1) {
3665 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3666 "mismatch, slen=%"UVuf", olen=%"UVuf,
3667 (UV)slen, (UV)olen);
3691 STRLEN otheroctets = otherbits >> 3;
3693 U8* const send = s + slen;
3698 if (otherbits == 1) {
3699 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3703 STRLEN vlen = otheroctets;
3711 if (opc == '+' && otherval)
3712 NOOP; /* replace with otherval */
3713 else if (opc == '!' && !otherval)
3715 else if (opc == '-' && otherval)
3717 else if (opc == '&' && !otherval)
3720 s += octets; /* no replacement */
3725 *s++ = (U8)( otherval & 0xff);
3726 else if (bits == 16) {
3727 *s++ = (U8)((otherval >> 8) & 0xff);
3728 *s++ = (U8)( otherval & 0xff);
3730 else if (bits == 32) {
3731 *s++ = (U8)((otherval >> 24) & 0xff);
3732 *s++ = (U8)((otherval >> 16) & 0xff);
3733 *s++ = (U8)((otherval >> 8) & 0xff);
3734 *s++ = (U8)( otherval & 0xff);
3738 sv_free(other); /* through with it! */
3744 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3747 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3748 * Can't be used on a property that is subject to user override, as it
3749 * relies on the value of SPECIALS in the swash which would be set by
3750 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3751 * for overridden properties
3753 * Returns a hash which is the inversion and closure of a swash mapping.
3754 * For example, consider the input lines:
3759 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3760 * 006C. The value for each key is an array. For 006C, the array would
3761 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3762 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3764 * Essentially, for any code point, it gives all the code points that map to
3765 * it, or the list of 'froms' for that point.
3767 * Currently it ignores any additions or deletions from other swashes,
3768 * looking at just the main body of the swash, and if there are SPECIALS
3769 * in the swash, at that hash
3771 * The specials hash can be extra code points, and most likely consists of
3772 * maps from single code points to multiple ones (each expressed as a string
3773 * of utf8 characters). This function currently returns only 1-1 mappings.
3774 * However consider this possible input in the specials hash:
3775 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3776 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3778 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3779 * currently handle. But it also means that FB05 and FB06 are equivalent in
3780 * a 1-1 mapping which we should handle, and this relationship may not be in
3781 * the main table. Therefore this function examines all the multi-char
3782 * sequences and adds the 1-1 mappings that come out of that. */
3786 HV *const hv = MUTABLE_HV(SvRV(swash));
3788 /* The string containing the main body of the table. This will have its
3789 * assertion fail if the swash has been converted to its inversion list */
3790 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3792 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3793 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3794 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3795 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3796 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3797 const STRLEN bits = SvUV(*bitssvp);
3798 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3799 const UV none = SvUV(*nonesvp);
3800 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3804 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3806 /* Must have at least 8 bits to get the mappings */
3807 if (bits != 8 && bits != 16 && bits != 32) {
3808 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3812 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3813 mapping to more than one character */
3815 /* Construct an inverse mapping hash for the specials */
3816 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3817 HV * specials_inverse = newHV();
3818 char *char_from; /* the lhs of the map */
3819 I32 from_len; /* its byte length */
3820 char *char_to; /* the rhs of the map */
3821 I32 to_len; /* its byte length */
3822 SV *sv_to; /* and in a sv */
3823 AV* from_list; /* list of things that map to each 'to' */
3825 hv_iterinit(specials_hv);
3827 /* The keys are the characters (in utf8) that map to the corresponding
3828 * utf8 string value. Iterate through the list creating the inverse
3830 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3832 if (! SvPOK(sv_to)) {
3833 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3834 "unexpectedly is not a string, flags=%lu",
3835 (unsigned long)SvFLAGS(sv_to));
3837 /*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)));*/
3839 /* Each key in the inverse list is a mapped-to value, and the key's
3840 * hash value is a list of the strings (each in utf8) that map to
3841 * it. Those strings are all one character long */
3842 if ((listp = hv_fetch(specials_inverse,
3846 from_list = (AV*) *listp;
3848 else { /* No entry yet for it: create one */
3849 from_list = newAV();
3850 if (! hv_store(specials_inverse,
3853 (SV*) from_list, 0))
3855 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3859 /* Here have the list associated with this 'to' (perhaps newly
3860 * created and empty). Just add to it. Note that we ASSUME that
3861 * the input is guaranteed to not have duplications, so we don't
3862 * check for that. Duplications just slow down execution time. */
3863 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3866 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3867 * it looking for cases like the FB05/FB06 examples above. There would
3868 * be an entry in the hash like
3869 * 'st' => [ FB05, FB06 ]
3870 * In this example we will create two lists that get stored in the
3871 * returned hash, 'ret':
3872 * FB05 => [ FB05, FB06 ]
3873 * FB06 => [ FB05, FB06 ]
3875 * Note that there is nothing to do if the array only has one element.
3876 * (In the normal 1-1 case handled below, we don't have to worry about
3877 * two lists, as everything gets tied to the single list that is
3878 * generated for the single character 'to'. But here, we are omitting
3879 * that list, ('st' in the example), so must have multiple lists.) */
3880 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3881 &char_to, &to_len)))
3883 if (av_len(from_list) > 0) {
3886 /* We iterate over all combinations of i,j to place each code
3887 * point on each list */
3888 for (i = 0; i <= av_len(from_list); i++) {
3890 AV* i_list = newAV();
3891 SV** entryp = av_fetch(from_list, i, FALSE);
3892 if (entryp == NULL) {
3893 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3895 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3896 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3898 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3899 (SV*) i_list, FALSE))
3901 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3904 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3905 for (j = 0; j <= av_len(from_list); j++) {
3906 entryp = av_fetch(from_list, j, FALSE);
3907 if (entryp == NULL) {
3908 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3911 /* When i==j this adds itself to the list */
3912 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3913 (U8*) SvPVX(*entryp),
3914 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3916 /*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));*/
3921 SvREFCNT_dec(specials_inverse); /* done with it */
3922 } /* End of specials */
3924 /* read $swash->{LIST} */
3925 l = (U8*)SvPV(*listsvp, lcur);
3928 /* Go through each input line */
3932 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3933 cBOOL(octets), typestr);
3938 /* Each element in the range is to be inverted */
3939 for (inverse = min; inverse <= max; inverse++) {
3943 bool found_key = FALSE;
3944 bool found_inverse = FALSE;
3946 /* The key is the inverse mapping */
3947 char key[UTF8_MAXBYTES+1];
3948 char* key_end = (char *) uvuni_to_utf8((U8*) key, val);
3949 STRLEN key_len = key_end - key;
3951 /* Get the list for the map */
3952 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3953 list = (AV*) *listp;
3955 else { /* No entry yet for it: create one */
3957 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3958 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3962 /* Look through list to see if this inverse mapping already is
3963 * listed, or if there is a mapping to itself already */
3964 for (i = 0; i <= av_len(list); i++) {
3965 SV** entryp = av_fetch(list, i, FALSE);
3967 if (entryp == NULL) {
3968 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3971 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3972 if (SvUV(entry) == val) {
3975 if (SvUV(entry) == inverse) {
3976 found_inverse = TRUE;
3979 /* No need to continue searching if found everything we are
3981 if (found_key && found_inverse) {
3986 /* Make sure there is a mapping to itself on the list */
3988 av_push(list, newSVuv(val));
3989 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3993 /* Simply add the value to the list */
3994 if (! found_inverse) {
3995 av_push(list, newSVuv(inverse));
3996 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3999 /* swatch_get() increments the value of val for each element in the
4000 * range. That makes more compact tables possible. You can
4001 * express the capitalization, for example, of all consecutive
4002 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4003 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4004 * and it's not documented; it appears to be used only in
4005 * implementing tr//; I copied the semantics from swatch_get(), just
4007 if (!none || val < none) {
4017 Perl__swash_to_invlist(pTHX_ SV* const swash)
4020 /* Subject to change or removal. For use only in one place in regcomp.c.
4021 * Ownership is given to one reference count in the returned SV* */
4026 HV *const hv = MUTABLE_HV(SvRV(swash));
4027 UV elements = 0; /* Number of elements in the inversion list */
4037 STRLEN octets; /* if bits == 1, then octets == 0 */
4043 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4045 /* If not a hash, it must be the swash's inversion list instead */
4046 if (SvTYPE(hv) != SVt_PVHV) {
4047 return SvREFCNT_inc_simple_NN((SV*) hv);
4050 /* The string containing the main body of the table */
4051 listsvp = hv_fetchs(hv, "LIST", FALSE);
4052 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4053 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4054 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4055 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4057 typestr = (U8*)SvPV_nolen(*typesvp);
4058 bits = SvUV(*bitssvp);
4059 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4061 /* read $swash->{LIST} */
4062 if (SvPOK(*listsvp)) {
4063 l = (U8*)SvPV(*listsvp, lcur);
4066 /* LIST legitimately doesn't contain a string during compilation phases
4067 * of Perl itself, before the Unicode tables are generated. In this
4068 * case, just fake things up by creating an empty list */
4075 /* Scan the input to count the number of lines to preallocate array size
4076 * based on worst possible case, which is each line in the input creates 2
4077 * elements in the inversion list: 1) the beginning of a range in the list;
4078 * 2) the beginning of a range not in the list. */
4079 while ((loc = (strchr(loc, '\n'))) != NULL) {
4084 /* If the ending is somehow corrupt and isn't a new line, add another
4085 * element for the final range that isn't in the inversion list */
4086 if (! (*lend == '\n'
4087 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4092 invlist = _new_invlist(elements);
4094 /* Now go through the input again, adding each range to the list */
4097 UV val; /* Not used by this function */
4099 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
4100 cBOOL(octets), typestr);
4106 invlist = _add_range_to_invlist(invlist, start, end);
4109 /* Invert if the data says it should be */
4110 if (invert_it_svp && SvUV(*invert_it_svp)) {
4111 _invlist_invert_prop(invlist);
4114 /* This code is copied from swatch_get()
4115 * read $swash->{EXTRAS} */
4116 x = (U8*)SvPV(*extssvp, xcur);
4124 SV **otherbitssvp, *other;
4127 const U8 opc = *x++;
4131 nl = (U8*)memchr(x, '\n', xend - x);
4133 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4135 x = nl + 1; /* 1 is length of "\n" */
4139 x = xend; /* to EXTRAS' end at which \n is not found */
4146 namelen = nl - namestr;
4150 namelen = xend - namestr;
4154 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4155 otherhv = MUTABLE_HV(SvRV(*othersvp));
4156 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4157 otherbits = (STRLEN)SvUV(*otherbitssvp);
4159 if (bits != otherbits || bits != 1) {
4160 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4161 "properties, bits=%"UVuf", otherbits=%"UVuf,
4162 (UV)bits, (UV)otherbits);
4165 /* The "other" swatch must be destroyed after. */
4166 other = _swash_to_invlist((SV *)*othersvp);
4168 /* End of code copied from swatch_get() */
4171 _invlist_union(invlist, other, &invlist);
4174 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4177 _invlist_subtract(invlist, other, &invlist);
4180 _invlist_intersection(invlist, other, &invlist);
4185 sv_free(other); /* through with it! */
4192 Perl__get_swash_invlist(pTHX_ SV* const swash)
4196 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4198 if (! SvROK(swash)) {
4202 /* If it really isn't a hash, it isn't really swash; must be an inversion
4204 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4208 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4217 =for apidoc uvchr_to_utf8
4219 Adds the UTF-8 representation of the Native code point C<uv> to the end
4220 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
4221 bytes available. The return value is the pointer to the byte after the
4222 end of the new character. In other words,
4224 d = uvchr_to_utf8(d, uv);
4226 is the recommended wide native character-aware way of saying
4233 /* On ASCII machines this is normally a macro but we want a
4234 real function in case XS code wants it
4237 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
4239 PERL_ARGS_ASSERT_UVCHR_TO_UTF8;
4241 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), 0);
4245 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4247 PERL_ARGS_ASSERT_UVCHR_TO_UTF8_FLAGS;
4249 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), flags);
4253 =for apidoc utf8n_to_uvchr
4255 Returns the native character value of the first character in the string
4257 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
4258 length, in bytes, of that character.
4260 C<length> and C<flags> are the same as L</utf8n_to_uvuni>().
4264 /* On ASCII machines this is normally a macro but we want
4265 a real function in case XS code wants it
4268 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen,
4271 const UV uv = Perl_utf8n_to_uvuni(aTHX_ s, curlen, retlen, flags);
4273 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
4275 return UNI_TO_NATIVE(uv);
4279 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4281 /* May change: warns if surrogates, non-character code points, or
4282 * non-Unicode code points are in s which has length len bytes. Returns
4283 * TRUE if none found; FALSE otherwise. The only other validity check is
4284 * to make sure that this won't exceed the string's length */
4286 const U8* const e = s + len;
4289 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4292 if (UTF8SKIP(s) > len) {
4293 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4294 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4297 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
4299 if (UTF8_IS_SUPER(s)) {
4300 if (ckWARN_d(WARN_NON_UNICODE)) {
4301 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4302 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
4303 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
4307 else if (UTF8_IS_SURROGATE(s)) {
4308 if (ckWARN_d(WARN_SURROGATE)) {
4309 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4310 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4311 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
4316 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
4317 && (ckWARN_d(WARN_NONCHAR)))
4319 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4320 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
4321 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
4332 =for apidoc pv_uni_display
4334 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4335 length C<len>, the displayable version being at most C<pvlim> bytes long
4336 (if longer, the rest is truncated and "..." will be appended).
4338 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
4339 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
4340 to display the \\[nrfta\\] as the backslashed versions (like '\n')
4341 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
4342 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
4343 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
4345 The pointer to the PV of the C<dsv> is returned.
4349 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4354 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4358 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4360 /* This serves double duty as a flag and a character to print after
4361 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4365 if (pvlim && SvCUR(dsv) >= pvlim) {
4369 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4371 const unsigned char c = (unsigned char)u & 0xFF;
4372 if (flags & UNI_DISPLAY_BACKSLASH) {
4389 const char string = ok;
4390 sv_catpvs(dsv, "\\");
4391 sv_catpvn(dsv, &string, 1);
4394 /* isPRINT() is the locale-blind version. */
4395 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4396 const char string = c;
4397 sv_catpvn(dsv, &string, 1);
4402 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4405 sv_catpvs(dsv, "...");
4411 =for apidoc sv_uni_display
4413 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4414 the displayable version being at most C<pvlim> bytes long
4415 (if longer, the rest is truncated and "..." will be appended).
4417 The C<flags> argument is as in L</pv_uni_display>().
4419 The pointer to the PV of the C<dsv> is returned.
4424 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4426 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4428 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)SvPVX_const(ssv),
4429 SvCUR(ssv), pvlim, flags);
4433 =for apidoc foldEQ_utf8
4435 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4436 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4437 How far into the strings to compare is determined by other input parameters.
4439 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4440 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4441 with respect to C<s2>.
4443 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4444 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4445 scan will not be considered to be a match unless the goal is reached, and
4446 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4449 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4450 considered an end pointer to the position 1 byte past the maximum point
4451 in C<s1> beyond which scanning will not continue under any circumstances.
4452 (This routine assumes that UTF-8 encoded input strings are not malformed;
4453 malformed input can cause it to read past C<pe1>).
4454 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4455 is less than C<s1>+C<l1>, the match will never be successful because it can
4457 get as far as its goal (and in fact is asserted against). Correspondingly for
4458 C<pe2> with respect to C<s2>.
4460 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4461 C<l2> must be non-zero), and if both do, both have to be
4462 reached for a successful match. Also, if the fold of a character is multiple
4463 characters, all of them must be matched (see tr21 reference below for
4466 Upon a successful match, if C<pe1> is non-NULL,
4467 it will be set to point to the beginning of the I<next> character of C<s1>
4468 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4470 For case-insensitiveness, the "casefolding" of Unicode is used
4471 instead of upper/lowercasing both the characters, see
4472 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4476 /* A flags parameter has been added which may change, and hence isn't
4477 * externally documented. Currently it is:
4478 * 0 for as-documented above
4479 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4480 ASCII one, to not match
4481 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
4482 * points below 256; unicode rules for above 255; and
4483 * folds that cross those boundaries are disallowed,
4484 * like the NOMIX_ASCII option
4485 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4486 * routine. This allows that step to be skipped.
4487 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4490 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, const char *s2, char **pe2, UV l2, bool u2, U32 flags)
4493 const U8 *p1 = (const U8*)s1; /* Point to current char */
4494 const U8 *p2 = (const U8*)s2;
4495 const U8 *g1 = NULL; /* goal for s1 */
4496 const U8 *g2 = NULL;
4497 const U8 *e1 = NULL; /* Don't scan s1 past this */
4498 U8 *f1 = NULL; /* Point to current folded */
4499 const U8 *e2 = NULL;
4501 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4502 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4503 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4505 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4507 /* The algorithm requires that input with the flags on the first line of
4508 * the assert not be pre-folded. */
4509 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
4510 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4517 g1 = (const U8*)s1 + l1;
4525 g2 = (const U8*)s2 + l2;
4528 /* Must have at least one goal */
4533 /* Will never match if goal is out-of-bounds */
4534 assert(! e1 || e1 >= g1);
4536 /* Here, there isn't an end pointer, or it is beyond the goal. We
4537 * only go as far as the goal */
4541 assert(e1); /* Must have an end for looking at s1 */
4544 /* Same for goal for s2 */
4546 assert(! e2 || e2 >= g2);
4553 /* If both operands are already folded, we could just do a memEQ on the
4554 * whole strings at once, but it would be better if the caller realized
4555 * this and didn't even call us */
4557 /* Look through both strings, a character at a time */
4558 while (p1 < e1 && p2 < e2) {
4560 /* If at the beginning of a new character in s1, get its fold to use
4561 * and the length of the fold. (exception: locale rules just get the
4562 * character to a single byte) */
4564 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4569 /* If in locale matching, we use two sets of rules, depending
4570 * on if the code point is above or below 255. Here, we test
4571 * for and handle locale rules */
4572 if ((flags & FOLDEQ_UTF8_LOCALE)
4573 && (! u1 || UTF8_IS_INVARIANT(*p1)
4574 || UTF8_IS_DOWNGRADEABLE_START(*p1)))
4576 /* There is no mixing of code points above and below 255. */
4577 if (u2 && (! UTF8_IS_INVARIANT(*p2)
4578 && ! UTF8_IS_DOWNGRADEABLE_START(*p2)))
4583 /* We handle locale rules by converting, if necessary, the
4584 * code point to a single byte. */
4585 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4589 *foldbuf1 = TWO_BYTE_UTF8_TO_UNI(*p1, *(p1 + 1));
4593 else if (isASCII(*p1)) { /* Note, that here won't be both
4594 ASCII and using locale rules */
4596 /* If trying to mix non- with ASCII, and not supposed to,
4598 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4602 *foldbuf1 = toLOWER(*p1); /* Folds in the ASCII range are
4606 to_utf8_fold(p1, foldbuf1, &n1);
4608 else { /* Not utf8, get utf8 fold */
4609 to_uni_fold(NATIVE_TO_UNI(*p1), foldbuf1, &n1);
4615 if (n2 == 0) { /* Same for s2 */
4616 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4621 if ((flags & FOLDEQ_UTF8_LOCALE)
4622 && (! u2 || UTF8_IS_INVARIANT(*p2) || UTF8_IS_DOWNGRADEABLE_START(*p2)))
4624 /* Here, the next char in s2 is < 256. We've already
4625 * worked on s1, and if it isn't also < 256, can't match */
4626 if (u1 && (! UTF8_IS_INVARIANT(*p1)
4627 && ! UTF8_IS_DOWNGRADEABLE_START(*p1)))
4631 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4635 *foldbuf2 = TWO_BYTE_UTF8_TO_UNI(*p2, *(p2 + 1));
4638 /* Use another function to handle locale rules. We've made
4639 * sure that both characters to compare are single bytes */
4640 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4645 else if (isASCII(*p2)) {
4646 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4650 *foldbuf2 = toLOWER(*p2);
4653 to_utf8_fold(p2, foldbuf2, &n2);
4656 to_uni_fold(NATIVE_TO_UNI(*p2), foldbuf2, &n2);
4662 /* Here f1 and f2 point to the beginning of the strings to compare.
4663 * These strings are the folds of the next character from each input
4664 * string, stored in utf8. */
4666 /* While there is more to look for in both folds, see if they
4667 * continue to match */
4669 U8 fold_length = UTF8SKIP(f1);
4670 if (fold_length != UTF8SKIP(f2)
4671 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4672 function call for single
4674 || memNE((char*)f1, (char*)f2, fold_length))
4676 return 0; /* mismatch */
4679 /* Here, they matched, advance past them */
4686 /* When reach the end of any fold, advance the input past it */
4688 p1 += u1 ? UTF8SKIP(p1) : 1;
4691 p2 += u2 ? UTF8SKIP(p2) : 1;
4693 } /* End of loop through both strings */
4695 /* A match is defined by each scan that specified an explicit length
4696 * reaching its final goal, and the other not having matched a partial
4697 * character (which can happen when the fold of a character is more than one
4699 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4703 /* Successful match. Set output pointers */
4715 * c-indentation-style: bsd
4717 * indent-tabs-mode: nil
4720 * ex: set ts=8 sts=4 sw=4 et: