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 Unicode 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 where uv is a code point expressed in Latin-1 or above, not the platform's
113 native character set. B<Almost all code should instead use L</uvchr_to_utf8>
114 or L</uvchr_to_utf8_flags>>.
116 This function will convert to UTF-8 (and not warn) even code points that aren't
117 legal Unicode or are problematic, unless C<flags> contains one or more of the
120 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
121 the function will raise a warning, provided UTF8 warnings are enabled. If instead
122 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
123 If both flags are set, the function will both warn and return NULL.
125 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
126 affect how the function handles a Unicode non-character. And likewise, the
127 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, affect the handling of
129 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
130 even less portable) can be warned and/or disallowed even if other above-Unicode
131 code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
134 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
135 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
143 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
145 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
147 /* The first problematic code point is the first surrogate */
148 if (uv >= UNICODE_SURROGATE_FIRST
149 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
151 if (UNICODE_IS_SURROGATE(uv)) {
152 if (flags & UNICODE_WARN_SURROGATE) {
153 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
154 "UTF-16 surrogate U+%04"UVXf, uv);
156 if (flags & UNICODE_DISALLOW_SURROGATE) {
160 else if (UNICODE_IS_SUPER(uv)) {
161 if (flags & UNICODE_WARN_SUPER
162 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
164 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
165 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
167 if (flags & UNICODE_DISALLOW_SUPER
168 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
173 else if (UNICODE_IS_NONCHAR(uv)) {
174 if (flags & UNICODE_WARN_NONCHAR) {
175 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
176 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
179 if (flags & UNICODE_DISALLOW_NONCHAR) {
184 if (UNI_IS_INVARIANT(uv)) {
185 *d++ = (U8)UTF_TO_NATIVE(uv);
190 STRLEN len = UNISKIP(uv);
193 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
194 uv >>= UTF_ACCUMULATION_SHIFT;
196 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
199 #else /* Non loop style */
201 *d++ = (U8)(( uv >> 6) | 0xc0);
202 *d++ = (U8)(( uv & 0x3f) | 0x80);
206 *d++ = (U8)(( uv >> 12) | 0xe0);
207 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
208 *d++ = (U8)(( uv & 0x3f) | 0x80);
212 *d++ = (U8)(( uv >> 18) | 0xf0);
213 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
215 *d++ = (U8)(( uv & 0x3f) | 0x80);
218 if (uv < 0x4000000) {
219 *d++ = (U8)(( uv >> 24) | 0xf8);
220 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
221 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
222 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
223 *d++ = (U8)(( uv & 0x3f) | 0x80);
226 if (uv < 0x80000000) {
227 *d++ = (U8)(( uv >> 30) | 0xfc);
228 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
232 *d++ = (U8)(( uv & 0x3f) | 0x80);
236 if (uv < UTF8_QUAD_MAX)
239 *d++ = 0xfe; /* Can't match U+FEFF! */
240 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
241 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
242 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
243 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
244 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
245 *d++ = (U8)(( uv & 0x3f) | 0x80);
250 *d++ = 0xff; /* Can't match U+FFFE! */
251 *d++ = 0x80; /* 6 Reserved bits */
252 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
253 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
257 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
258 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
259 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
260 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
261 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
262 *d++ = (U8)(( uv & 0x3f) | 0x80);
266 #endif /* Non loop style */
271 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
272 character. Note that an INVARIANT (i.e. ASCII) character is a valid
273 UTF-8 character. The number of bytes in the UTF-8 character
274 will be returned if it is valid, otherwise 0.
276 This is the "slow" version as opposed to the "fast" version which is
277 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
278 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
279 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
280 you should use the _slow(). In practice this means that the _slow()
281 will be used very rarely, since the maximum Unicode code point (as of
282 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
283 the "Perl extended UTF-8" (e.g, the infamous 'v-strings') will encode into
287 PERL_STATIC_INLINE STRLEN
288 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
290 dTHX; /* The function called below requires thread context */
294 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
296 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
298 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
302 =for apidoc is_utf8_char_buf
304 Returns the number of bytes that comprise the first UTF-8 encoded character in
305 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
306 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
309 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
310 machines) is a valid UTF-8 character.
315 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
320 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
322 if (buf_end <= buf) {
327 if (len > UTF8SKIP(buf)) {
332 if (IS_UTF8_CHAR_FAST(len))
333 return IS_UTF8_CHAR(buf, len) ? len : 0;
334 #endif /* #ifdef IS_UTF8_CHAR */
335 return is_utf8_char_slow(buf, len);
339 =for apidoc is_utf8_char
341 Tests if some arbitrary number of bytes begins in a valid UTF-8
342 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
343 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
344 character will be returned if it is valid, otherwise 0.
346 This function is deprecated due to the possibility that malformed input could
347 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
353 Perl_is_utf8_char(const U8 *s)
355 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
357 /* Assumes we have enough space, which is why this is deprecated */
358 return is_utf8_char_buf(s, s + UTF8SKIP(s));
363 =for apidoc is_utf8_string
365 Returns true if the first C<len> bytes of string C<s> form a valid
366 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
367 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
368 terminating NUL byte). Note that all characters being ASCII constitute 'a
371 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
377 Perl_is_utf8_string(const U8 *s, STRLEN len)
379 const U8* const send = s + (len ? len : strlen((const char *)s));
382 PERL_ARGS_ASSERT_IS_UTF8_STRING;
385 /* Inline the easy bits of is_utf8_char() here for speed... */
386 if (UTF8_IS_INVARIANT(*x)) {
390 /* ... and call is_utf8_char() only if really needed. */
391 const STRLEN c = UTF8SKIP(x);
392 const U8* const next_char_ptr = x + c;
394 if (next_char_ptr > send) {
398 if (IS_UTF8_CHAR_FAST(c)) {
399 if (!IS_UTF8_CHAR(x, c))
402 else if (! is_utf8_char_slow(x, c)) {
413 Implemented as a macro in utf8.h
415 =for apidoc is_utf8_string_loc
417 Like L</is_utf8_string> but stores the location of the failure (in the
418 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
419 "utf8ness success") in the C<ep>.
421 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
423 =for apidoc is_utf8_string_loclen
425 Like L</is_utf8_string>() but stores the location of the failure (in the
426 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
427 "utf8ness success") in the C<ep>, and the number of UTF-8
428 encoded characters in the C<el>.
430 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
436 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
438 const U8* const send = s + (len ? len : strlen((const char *)s));
443 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
446 const U8* next_char_ptr;
448 /* Inline the easy bits of is_utf8_char() here for speed... */
449 if (UTF8_IS_INVARIANT(*x))
450 next_char_ptr = x + 1;
452 /* ... and call is_utf8_char() only if really needed. */
454 next_char_ptr = c + x;
455 if (next_char_ptr > send) {
458 if (IS_UTF8_CHAR_FAST(c)) {
459 if (!IS_UTF8_CHAR(x, c))
462 c = is_utf8_char_slow(x, c);
481 =for apidoc utf8n_to_uvuni
483 Bottom level UTF-8 decode routine.
484 Returns the code point value of the first character in the string C<s>,
485 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
486 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
487 the length, in bytes, of that character.
489 The value of C<flags> determines the behavior when C<s> does not point to a
490 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
491 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
492 next possible position in C<s> that could begin a non-malformed character.
493 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
495 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
496 individual types of malformations, such as the sequence being overlong (that
497 is, when there is a shorter sequence that can express the same code point;
498 overlong sequences are expressly forbidden in the UTF-8 standard due to
499 potential security issues). Another malformation example is the first byte of
500 a character not being a legal first byte. See F<utf8.h> for the list of such
501 flags. For allowed 0 length strings, this function returns 0; for allowed
502 overlong sequences, the computed code point is returned; for all other allowed
503 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
504 determinable reasonable value.
506 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
507 flags) malformation is found. If this flag is set, the routine assumes that
508 the caller will raise a warning, and this function will silently just set
509 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
511 Note that this API requires disambiguation between successful decoding a NUL
512 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
513 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
514 first byte of C<s> is 0 as well. If so, the input was a NUL; if not, the input
517 Certain code points are considered problematic. These are Unicode surrogates,
518 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
519 By default these are considered regular code points, but certain situations
520 warrant special handling for them. If C<flags> contains
521 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
522 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
523 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
524 maximum) can be set to disallow these categories individually.
526 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
527 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
528 for their respective categories, but otherwise the code points are considered
529 valid (not malformations). To get a category to both be treated as a
530 malformation and raise a warning, specify both the WARN and DISALLOW flags.
531 (But note that warnings are not raised if lexically disabled nor if
532 UTF8_CHECK_ONLY is also specified.)
534 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
535 the others that are above the Unicode legal maximum. There are several
536 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
537 not representable at all on EBCDIC platforms, and the original UTF-8
538 specification never went above this number (the current 0x10FFFF limit was
539 imposed later). (The smaller ones, those that fit into 32 bits, are
540 representable by a UV on ASCII platforms, but not by an IV, which means that
541 the number of operations that can be performed on them is quite restricted.)
542 The UTF-8 encoding on ASCII platforms for these large code points begins with a
543 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
544 be treated as malformations, while allowing smaller above-Unicode code points.
545 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
546 including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
547 the other WARN flags, but applies just to these code points.
549 All other code points corresponding to Unicode characters, including private
550 use and those yet to be assigned, are never considered malformed and never
553 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
559 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
562 const U8 * const s0 = s;
563 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
568 UV outlier_ret = 0; /* return value when input is in error or problematic
570 UV pack_warn = 0; /* Save result of packWARN() for later */
571 bool unexpected_non_continuation = FALSE;
572 bool overflowed = FALSE;
573 bool do_overlong_test = TRUE; /* May have to skip this test */
575 const char* const malformed_text = "Malformed UTF-8 character";
577 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
579 /* The order of malformation tests here is important. We should consume as
580 * few bytes as possible in order to not skip any valid character. This is
581 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
582 * http://unicode.org/reports/tr36 for more discussion as to why. For
583 * example, once we've done a UTF8SKIP, we can tell the expected number of
584 * bytes, and could fail right off the bat if the input parameters indicate
585 * that there are too few available. But it could be that just that first
586 * byte is garbled, and the intended character occupies fewer bytes. If we
587 * blindly assumed that the first byte is correct, and skipped based on
588 * that number, we could skip over a valid input character. So instead, we
589 * always examine the sequence byte-by-byte.
591 * We also should not consume too few bytes, otherwise someone could inject
592 * things. For example, an input could be deliberately designed to
593 * overflow, and if this code bailed out immediately upon discovering that,
594 * returning to the caller *retlen pointing to the very next byte (one
595 * which is actually part of of the overflowing sequence), that could look
596 * legitimate to the caller, which could discard the initial partial
597 * sequence and process the rest, inappropriately */
599 /* Zero length strings, if allowed, of necessity are zero */
600 if (UNLIKELY(curlen == 0)) {
605 if (flags & UTF8_ALLOW_EMPTY) {
608 if (! (flags & UTF8_CHECK_ONLY)) {
609 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
614 expectlen = UTF8SKIP(s);
616 /* A well-formed UTF-8 character, as the vast majority of calls to this
617 * function will be for, has this expected length. For efficiency, set
618 * things up here to return it. It will be overriden only in those rare
619 * cases where a malformation is found */
624 /* An invariant is trivially well-formed */
625 if (UTF8_IS_INVARIANT(uv)) {
626 return (UV) (NATIVE_TO_UTF(*s));
629 /* A continuation character can't start a valid sequence */
630 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
631 if (flags & UTF8_ALLOW_CONTINUATION) {
635 return UNICODE_REPLACEMENT;
638 if (! (flags & UTF8_CHECK_ONLY)) {
639 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
646 uv = NATIVE_TO_UTF(uv);
649 /* Here is not a continuation byte, nor an invariant. The only thing left
650 * is a start byte (possibly for an overlong) */
652 /* Remove the leading bits that indicate the number of bytes in the
653 * character's whole UTF-8 sequence, leaving just the bits that are part of
655 uv &= UTF_START_MASK(expectlen);
657 /* Now, loop through the remaining bytes in the character's sequence,
658 * accumulating each into the working value as we go. Be sure to not look
659 * past the end of the input string */
660 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
662 for (s = s0 + 1; s < send; s++) {
663 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
664 #ifndef EBCDIC /* Can't overflow in EBCDIC */
665 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
667 /* The original implementors viewed this malformation as more
668 * serious than the others (though I, khw, don't understand
669 * why, since other malformations also give very very wrong
670 * results), so there is no way to turn off checking for it.
671 * Set a flag, but keep going in the loop, so that we absorb
672 * the rest of the bytes that comprise the character. */
674 overflow_byte = *s; /* Save for warning message's use */
677 uv = UTF8_ACCUMULATE(uv, *s);
680 /* Here, found a non-continuation before processing all expected
681 * bytes. This byte begins a new character, so quit, even if
682 * allowing this malformation. */
683 unexpected_non_continuation = TRUE;
686 } /* End of loop through the character's bytes */
688 /* Save how many bytes were actually in the character */
691 /* The loop above finds two types of malformations: non-continuation and/or
692 * overflow. The non-continuation malformation is really a too-short
693 * malformation, as it means that the current character ended before it was
694 * expected to (being terminated prematurely by the beginning of the next
695 * character, whereas in the too-short malformation there just are too few
696 * bytes available to hold the character. In both cases, the check below
697 * that we have found the expected number of bytes would fail if executed.)
698 * Thus the non-continuation malformation is really unnecessary, being a
699 * subset of the too-short malformation. But there may be existing
700 * applications that are expecting the non-continuation type, so we retain
701 * it, and return it in preference to the too-short malformation. (If this
702 * code were being written from scratch, the two types might be collapsed
703 * into one.) I, khw, am also giving priority to returning the
704 * non-continuation and too-short malformations over overflow when multiple
705 * ones are present. I don't know of any real reason to prefer one over
706 * the other, except that it seems to me that multiple-byte errors trumps
707 * errors from a single byte */
708 if (UNLIKELY(unexpected_non_continuation)) {
709 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
710 if (! (flags & UTF8_CHECK_ONLY)) {
712 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
715 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));
720 uv = UNICODE_REPLACEMENT;
722 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
723 * as what the original expectations were. */
724 do_overlong_test = FALSE;
729 else if (UNLIKELY(curlen < expectlen)) {
730 if (! (flags & UTF8_ALLOW_SHORT)) {
731 if (! (flags & UTF8_CHECK_ONLY)) {
732 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));
736 uv = UNICODE_REPLACEMENT;
737 do_overlong_test = FALSE;
743 #ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
744 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
745 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
747 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
748 * generation of the sv, since no warnings are raised under CHECK */
749 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
750 && ckWARN_d(WARN_UTF8))
752 /* This message is deliberately not of the same syntax as the other
753 * messages for malformations, for backwards compatibility in the
754 * unlikely event that code is relying on its precise earlier text
756 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
757 pack_warn = packWARN(WARN_UTF8);
759 if (flags & UTF8_DISALLOW_FE_FF) {
763 if (UNLIKELY(overflowed)) {
765 /* If the first byte is FF, it will overflow a 32-bit word. If the
766 * first byte is FE, it will overflow a signed 32-bit word. The
767 * above preserves backward compatibility, since its message was used
768 * in earlier versions of this code in preference to overflow */
769 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
775 && expectlen > (STRLEN)UNISKIP(uv)
776 && ! (flags & UTF8_ALLOW_LONG))
778 /* The overlong malformation has lower precedence than the others.
779 * Note that if this malformation is allowed, we return the actual
780 * value, instead of the replacement character. This is because this
781 * value is actually well-defined. */
782 if (! (flags & UTF8_CHECK_ONLY)) {
783 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));
788 /* Here, the input is considered to be well-formed , but could be a
789 * problematic code point that is not allowed by the input parameters. */
790 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
791 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
792 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
794 if (UNICODE_IS_SURROGATE(uv)) {
795 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
796 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
798 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
799 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
801 if (flags & UTF8_DISALLOW_SURROGATE) {
805 else if ((uv > PERL_UNICODE_MAX)) {
806 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
807 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
809 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
810 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
812 if (flags & UTF8_DISALLOW_SUPER) {
816 else if (UNICODE_IS_NONCHAR(uv)) {
817 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
818 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
820 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
821 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
823 if (flags & UTF8_DISALLOW_NONCHAR) {
833 /* Here, this is not considered a malformed character, so drop through
839 /* There are three cases which get to beyond this point. In all 3 cases:
840 * <sv> if not null points to a string to print as a warning.
841 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
843 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
844 * This is done by initializing it to 0, and changing it only
847 * 1) The input is valid but problematic, and to be warned about. The
848 * return value is the resultant code point; <*retlen> is set to
849 * <curlen>, the number of bytes that comprise the code point.
850 * <pack_warn> contains the result of packWARN() for the warning
851 * types. The entry point for this case is the label <do_warn>;
852 * 2) The input is a valid code point but disallowed by the parameters to
853 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
854 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
855 * comprise the code point. <pack_warn> contains the result of
856 * packWARN() for the warning types. The entry point for this case is
857 * the label <disallowed>.
858 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
859 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
860 * bytes that comprise the malformation. All such malformations are
861 * assumed to be warning type <utf8>. The entry point for this case
862 * is the label <malformed>.
867 if (sv && ckWARN_d(WARN_UTF8)) {
868 pack_warn = packWARN(WARN_UTF8);
873 if (flags & UTF8_CHECK_ONLY) {
875 *retlen = ((STRLEN) -1);
881 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
882 if warnings are to be raised. */
883 const char * const string = SvPVX_const(sv);
886 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
888 Perl_warner(aTHX_ pack_warn, "%s", string);
899 =for apidoc utf8_to_uvchr_buf
901 Returns the native code point of the first character in the string C<s> which
902 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
903 C<*retlen> will be set to the length, in bytes, of that character.
905 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
906 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
907 NULL) to -1. If those warnings are off, the computed value, if well-defined
908 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
909 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
910 the next possible position in C<s> that could begin a non-malformed character.
911 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is
919 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
921 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
925 return utf8n_to_uvchr(s, send - s, retlen,
926 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
929 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
930 * there are no malformations in the input UTF-8 string C<s>. surrogates,
931 * non-character code points, and non-Unicode code points are allowed. A macro
932 * in utf8.h is used to normally avoid this function wrapper */
935 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
937 const UV uv = valid_utf8_to_uvuni(s, retlen);
939 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
941 return UNI_TO_NATIVE(uv);
945 =for apidoc utf8_to_uvchr
947 Returns the native code point of the first character in the string C<s>
948 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
949 length, in bytes, of that character.
951 Some, but not all, UTF-8 malformations are detected, and in fact, some
952 malformed input could cause reading beyond the end of the input buffer, which
953 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
955 If C<s> points to one of the detected malformations, and UTF8 warnings are
956 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
957 NULL) to -1. If those warnings are off, the computed value if well-defined (or
958 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
959 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
960 next possible position in C<s> that could begin a non-malformed character.
961 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
967 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
969 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
971 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
975 =for apidoc utf8_to_uvuni_buf
977 Returns the Unicode code point of the first character in the string C<s> which
978 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
979 C<retlen> will be set to the length, in bytes, of that character.
981 This function should only be used when the returned UV is considered
982 an index into the Unicode semantic tables (e.g. swashes).
984 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
985 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
986 NULL) to -1. If those warnings are off, the computed value if well-defined (or
987 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
988 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
989 next possible position in C<s> that could begin a non-malformed character.
990 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
996 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
998 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1002 /* Call the low level routine asking for checks */
1003 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1004 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1007 /* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1008 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1009 * non-character code points, and non-Unicode code points are allowed */
1012 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1014 UV expectlen = UTF8SKIP(s);
1015 const U8* send = s + expectlen;
1016 UV uv = NATIVE_TO_UTF(*s);
1018 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1021 *retlen = expectlen;
1024 /* An invariant is trivially returned */
1025 if (expectlen == 1) {
1029 /* Remove the leading bits that indicate the number of bytes, leaving just
1030 * the bits that are part of the value */
1031 uv &= UTF_START_MASK(expectlen);
1033 /* Now, loop through the remaining bytes, accumulating each into the
1034 * working total as we go. (I khw tried unrolling the loop for up to 4
1035 * bytes, but there was no performance improvement) */
1036 for (++s; s < send; s++) {
1037 uv = UTF8_ACCUMULATE(uv, *s);
1044 =for apidoc utf8_to_uvuni
1046 Returns the Unicode code point of the first character in the string C<s>
1047 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1048 length, in bytes, of that character.
1050 This function should only be used when the returned UV is considered
1051 an index into the Unicode semantic tables (e.g. swashes).
1053 Some, but not all, UTF-8 malformations are detected, and in fact, some
1054 malformed input could cause reading beyond the end of the input buffer, which
1055 is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
1057 If C<s> points to one of the detected malformations, and UTF8 warnings are
1058 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1059 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1060 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1061 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1062 next possible position in C<s> that could begin a non-malformed character.
1063 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
1069 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1071 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1073 return valid_utf8_to_uvuni(s, retlen);
1077 =for apidoc utf8_length
1079 Return the length of the UTF-8 char encoded string C<s> in characters.
1080 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1081 up past C<e>, croaks.
1087 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1092 PERL_ARGS_ASSERT_UTF8_LENGTH;
1094 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1095 * the bitops (especially ~) can create illegal UTF-8.
1096 * In other words: in Perl UTF-8 is not just for Unicode. */
1099 goto warn_and_return;
1109 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1110 "%s in %s", unees, OP_DESC(PL_op));
1112 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1119 =for apidoc utf8_distance
1121 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1124 WARNING: use only if you *know* that the pointers point inside the
1131 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1133 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1135 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1139 =for apidoc utf8_hop
1141 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1142 forward or backward.
1144 WARNING: do not use the following unless you *know* C<off> is within
1145 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1146 on the first byte of character or just after the last byte of a character.
1152 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1154 PERL_ARGS_ASSERT_UTF8_HOP;
1156 PERL_UNUSED_CONTEXT;
1157 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1158 * the bitops (especially ~) can create illegal UTF-8.
1159 * In other words: in Perl UTF-8 is not just for Unicode. */
1168 while (UTF8_IS_CONTINUATION(*s))
1176 =for apidoc bytes_cmp_utf8
1178 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1179 sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
1180 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1181 if the first string is greater than the second string.
1183 -1 or +1 is returned if the shorter string was identical to the start of the
1184 longer string. -2 or +2 is returned if the was a difference between characters
1191 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1193 const U8 *const bend = b + blen;
1194 const U8 *const uend = u + ulen;
1196 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1198 PERL_UNUSED_CONTEXT;
1200 while (b < bend && u < uend) {
1202 if (!UTF8_IS_INVARIANT(c)) {
1203 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1206 if (UTF8_IS_CONTINUATION(c1)) {
1207 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
1209 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1210 "Malformed UTF-8 character "
1211 "(unexpected non-continuation byte 0x%02x"
1212 ", immediately after start byte 0x%02x)"
1213 /* Dear diag.t, it's in the pod. */
1215 PL_op ? " in " : "",
1216 PL_op ? OP_DESC(PL_op) : "");
1221 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1222 "%s in %s", unees, OP_DESC(PL_op));
1224 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1225 return -2; /* Really want to return undef :-) */
1232 return *b < c ? -2 : +2;
1237 if (b == bend && u == uend)
1240 return b < bend ? +1 : -1;
1244 =for apidoc utf8_to_bytes
1246 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1247 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1248 updates C<len> to contain the new length.
1249 Returns zero on failure, setting C<len> to -1.
1251 If you need a copy of the string, see L</bytes_from_utf8>.
1257 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1259 U8 * const save = s;
1260 U8 * const send = s + *len;
1263 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1265 /* ensure valid UTF-8 and chars < 256 before updating string */
1269 if (!UTF8_IS_INVARIANT(c) &&
1270 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1271 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
1272 *len = ((STRLEN) -1);
1280 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
1289 =for apidoc bytes_from_utf8
1291 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1292 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1293 the newly-created string, and updates C<len> to contain the new
1294 length. Returns the original string if no conversion occurs, C<len>
1295 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1296 0 if C<s> is converted or consisted entirely of characters that are invariant
1297 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1303 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1306 const U8 *start = s;
1310 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1312 PERL_UNUSED_CONTEXT;
1316 /* ensure valid UTF-8 and chars < 256 before converting string */
1317 for (send = s + *len; s < send;) {
1319 if (!UTF8_IS_INVARIANT(c)) {
1320 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1321 (c = *s++) && UTF8_IS_CONTINUATION(c))
1330 Newx(d, (*len) - count + 1, U8);
1331 s = start; start = d;
1334 if (!UTF8_IS_INVARIANT(c)) {
1335 /* Then it is two-byte encoded */
1336 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
1346 =for apidoc bytes_to_utf8
1348 Converts a string C<s> of length C<len> bytes from the native encoding into
1350 Returns a pointer to the newly-created string, and sets C<len> to
1351 reflect the new length in bytes.
1353 A NUL character will be written after the end of the string.
1355 If you want to convert to UTF-8 from encodings other than
1356 the native (Latin1 or EBCDIC),
1357 see L</sv_recode_to_utf8>().
1362 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1363 likewise need duplication. */
1366 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1368 const U8 * const send = s + (*len);
1372 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1373 PERL_UNUSED_CONTEXT;
1375 Newx(d, (*len) * 2 + 1, U8);
1379 const UV uv = NATIVE_TO_ASCII(*s++);
1380 if (UNI_IS_INVARIANT(uv))
1381 *d++ = (U8)UTF_TO_NATIVE(uv);
1383 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1384 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
1393 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1395 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1396 * We optimize for native, for obvious reasons. */
1399 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1404 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1407 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1412 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1416 *d++ = UNI_TO_NATIVE(uv);
1423 *d++ = (U8)(( uv >> 6) | 0xc0);
1424 *d++ = (U8)(( uv & 0x3f) | 0x80);
1427 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1428 #define LAST_HIGH_SURROGATE 0xDBFF
1429 #define FIRST_LOW_SURROGATE 0xDC00
1430 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1431 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1433 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1435 UV low = (p[0] << 8) + p[1];
1437 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1438 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1439 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1440 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1442 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1443 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1446 *d++ = (U8)(( uv >> 12) | 0xe0);
1447 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1448 *d++ = (U8)(( uv & 0x3f) | 0x80);
1452 *d++ = (U8)(( uv >> 18) | 0xf0);
1453 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1454 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1455 *d++ = (U8)(( uv & 0x3f) | 0x80);
1459 *newlen = d - dstart;
1463 /* Note: this one is slightly destructive of the source. */
1466 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1469 U8* const send = s + bytelen;
1471 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1474 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1478 const U8 tmp = s[0];
1483 return utf16_to_utf8(p, d, bytelen, newlen);
1487 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1489 U8 tmpbuf[UTF8_MAXBYTES+1];
1490 uvchr_to_utf8(tmpbuf, c);
1491 return _is_utf8_FOO(classnum, tmpbuf);
1494 /* for now these are all defined (inefficiently) in terms of the utf8 versions.
1495 * Note that the macros in handy.h that call these short-circuit calling them
1496 * for Latin-1 range inputs */
1499 Perl_is_uni_alnum(pTHX_ UV c)
1501 U8 tmpbuf[UTF8_MAXBYTES+1];
1502 uvchr_to_utf8(tmpbuf, c);
1503 return _is_utf8_FOO(_CC_WORDCHAR, tmpbuf);
1507 Perl_is_uni_alnumc(pTHX_ UV c)
1509 U8 tmpbuf[UTF8_MAXBYTES+1];
1510 uvchr_to_utf8(tmpbuf, c);
1511 return _is_utf8_FOO(_CC_ALPHANUMERIC, tmpbuf);
1514 /* Internal function so we can deprecate the external one, and call
1515 this one from other deprecated functions in this file */
1517 PERL_STATIC_INLINE bool
1518 S_is_utf8_idfirst(pTHX_ const U8 *p)
1524 /* is_utf8_idstart would be more logical. */
1525 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1529 Perl_is_uni_idfirst(pTHX_ UV c)
1531 U8 tmpbuf[UTF8_MAXBYTES+1];
1532 uvchr_to_utf8(tmpbuf, c);
1533 return S_is_utf8_idfirst(aTHX_ tmpbuf);
1537 Perl__is_uni_perl_idcont(pTHX_ UV c)
1539 U8 tmpbuf[UTF8_MAXBYTES+1];
1540 uvchr_to_utf8(tmpbuf, c);
1541 return _is_utf8_perl_idcont(tmpbuf);
1545 Perl__is_uni_perl_idstart(pTHX_ UV c)
1547 U8 tmpbuf[UTF8_MAXBYTES+1];
1548 uvchr_to_utf8(tmpbuf, c);
1549 return _is_utf8_perl_idstart(tmpbuf);
1553 Perl_is_uni_alpha(pTHX_ UV c)
1555 U8 tmpbuf[UTF8_MAXBYTES+1];
1556 uvchr_to_utf8(tmpbuf, c);
1557 return _is_utf8_FOO(_CC_ALPHA, tmpbuf);
1561 Perl_is_uni_ascii(pTHX_ UV c)
1567 Perl_is_uni_blank(pTHX_ UV c)
1569 return isBLANK_uni(c);
1573 Perl_is_uni_space(pTHX_ UV c)
1575 return isSPACE_uni(c);
1579 Perl_is_uni_digit(pTHX_ UV c)
1581 U8 tmpbuf[UTF8_MAXBYTES+1];
1582 uvchr_to_utf8(tmpbuf, c);
1583 return _is_utf8_FOO(_CC_DIGIT, tmpbuf);
1587 Perl_is_uni_upper(pTHX_ UV c)
1589 U8 tmpbuf[UTF8_MAXBYTES+1];
1590 uvchr_to_utf8(tmpbuf, c);
1591 return _is_utf8_FOO(_CC_UPPER, tmpbuf);
1595 Perl_is_uni_lower(pTHX_ UV c)
1597 U8 tmpbuf[UTF8_MAXBYTES+1];
1598 uvchr_to_utf8(tmpbuf, c);
1599 return _is_utf8_FOO(_CC_LOWER, tmpbuf);
1603 Perl_is_uni_cntrl(pTHX_ UV c)
1605 return isCNTRL_L1(c);
1609 Perl_is_uni_graph(pTHX_ UV c)
1611 U8 tmpbuf[UTF8_MAXBYTES+1];
1612 uvchr_to_utf8(tmpbuf, c);
1613 return _is_utf8_FOO(_CC_GRAPH, tmpbuf);
1617 Perl_is_uni_print(pTHX_ UV c)
1619 U8 tmpbuf[UTF8_MAXBYTES+1];
1620 uvchr_to_utf8(tmpbuf, c);
1621 return _is_utf8_FOO(_CC_PRINT, tmpbuf);
1625 Perl_is_uni_punct(pTHX_ UV c)
1627 U8 tmpbuf[UTF8_MAXBYTES+1];
1628 uvchr_to_utf8(tmpbuf, c);
1629 return _is_utf8_FOO(_CC_PUNCT, tmpbuf);
1633 Perl_is_uni_xdigit(pTHX_ UV c)
1635 return isXDIGIT_uni(c);
1639 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1641 /* We have the latin1-range values compiled into the core, so just use
1642 * those, converting the result to utf8. The only difference between upper
1643 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1644 * either "SS" or "Ss". Which one to use is passed into the routine in
1645 * 'S_or_s' to avoid a test */
1647 UV converted = toUPPER_LATIN1_MOD(c);
1649 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1651 assert(S_or_s == 'S' || S_or_s == 's');
1653 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1654 characters in this range */
1655 *p = (U8) converted;
1660 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1661 * which it maps to one of them, so as to only have to have one check for
1662 * it in the main case */
1663 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1665 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1666 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1669 converted = GREEK_CAPITAL_LETTER_MU;
1671 case LATIN_SMALL_LETTER_SHARP_S:
1677 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1678 assert(0); /* NOTREACHED */
1682 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1683 *p = UTF8_TWO_BYTE_LO(converted);
1689 /* Call the function to convert a UTF-8 encoded character to the specified case.
1690 * Note that there may be more than one character in the result.
1691 * INP is a pointer to the first byte of the input character
1692 * OUTP will be set to the first byte of the string of changed characters. It
1693 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1694 * LENP will be set to the length in bytes of the string of changed characters
1696 * The functions return the ordinal of the first character in the string of OUTP */
1697 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1698 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1699 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
1701 /* This additionally has the input parameter SPECIALS, which if non-zero will
1702 * cause this to use the SPECIALS hash for folding (meaning get full case
1703 * folding); otherwise, when zero, this implies a simple case fold */
1704 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
1707 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1711 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1712 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1713 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1714 * the changed version may be longer than the original character.
1716 * The ordinal of the first character of the changed version is returned
1717 * (but note, as explained above, that there may be more.) */
1719 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1722 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1725 uvchr_to_utf8(p, c);
1726 return CALL_UPPER_CASE(p, p, lenp);
1730 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1734 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1737 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1740 uvchr_to_utf8(p, c);
1741 return CALL_TITLE_CASE(p, p, lenp);
1745 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1747 /* We have the latin1-range values compiled into the core, so just use
1748 * those, converting the result to utf8. Since the result is always just
1749 * one character, we allow <p> to be NULL */
1751 U8 converted = toLOWER_LATIN1(c);
1754 if (UNI_IS_INVARIANT(converted)) {
1759 *p = UTF8_TWO_BYTE_HI(converted);
1760 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1768 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1772 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1775 return to_lower_latin1((U8) c, p, lenp);
1778 uvchr_to_utf8(p, c);
1779 return CALL_LOWER_CASE(p, p, lenp);
1783 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1785 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1786 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1787 * FOLD_FLAGS_FULL iff full folding is to be used;
1789 * Not to be used for locale folds
1794 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1796 assert (! (flags & FOLD_FLAGS_LOCALE));
1798 if (c == MICRO_SIGN) {
1799 converted = GREEK_SMALL_LETTER_MU;
1801 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1803 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1804 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1805 * under those circumstances. */
1806 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1807 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1808 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1810 return LATIN_SMALL_LETTER_LONG_S;
1819 else { /* In this range the fold of all other characters is their lower
1821 converted = toLOWER_LATIN1(c);
1824 if (UNI_IS_INVARIANT(converted)) {
1825 *p = (U8) converted;
1829 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1830 *p = UTF8_TWO_BYTE_LO(converted);
1838 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
1841 /* Not currently externally documented, and subject to change
1842 * <flags> bits meanings:
1843 * FOLD_FLAGS_FULL iff full folding is to be used;
1844 * FOLD_FLAGS_LOCALE iff in locale
1845 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1848 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1851 UV result = _to_fold_latin1((U8) c, p, lenp,
1852 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1853 /* It is illegal for the fold to cross the 255/256 boundary under
1854 * locale; in this case return the original */
1855 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1860 /* If no special needs, just use the macro */
1861 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1862 uvchr_to_utf8(p, c);
1863 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1865 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1866 the special flags. */
1867 U8 utf8_c[UTF8_MAXBYTES + 1];
1868 uvchr_to_utf8(utf8_c, c);
1869 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1874 Perl_is_uni_alnum_lc(pTHX_ UV c)
1877 return isALNUM_LC(UNI_TO_NATIVE(c));
1879 return _is_uni_FOO(_CC_WORDCHAR, c);
1883 Perl_is_uni_alnumc_lc(pTHX_ UV c)
1886 return isALPHANUMERIC_LC(UNI_TO_NATIVE(c));
1888 return _is_uni_FOO(_CC_ALPHANUMERIC, c);
1892 Perl_is_uni_idfirst_lc(pTHX_ UV c)
1895 return isIDFIRST_LC(UNI_TO_NATIVE(c));
1897 return _is_uni_perl_idstart(c);
1901 Perl_is_uni_alpha_lc(pTHX_ UV c)
1904 return isALPHA_LC(UNI_TO_NATIVE(c));
1906 return _is_uni_FOO(_CC_ALPHA, c);
1910 Perl_is_uni_ascii_lc(pTHX_ UV c)
1913 return isASCII_LC(UNI_TO_NATIVE(c));
1919 Perl_is_uni_blank_lc(pTHX_ UV c)
1922 return isBLANK_LC(UNI_TO_NATIVE(c));
1924 return isBLANK_uni(c);
1928 Perl_is_uni_space_lc(pTHX_ UV c)
1931 return isSPACE_LC(UNI_TO_NATIVE(c));
1933 return isSPACE_uni(c);
1937 Perl_is_uni_digit_lc(pTHX_ UV c)
1940 return isDIGIT_LC(UNI_TO_NATIVE(c));
1942 return _is_uni_FOO(_CC_DIGIT, c);
1946 Perl_is_uni_upper_lc(pTHX_ UV c)
1949 return isUPPER_LC(UNI_TO_NATIVE(c));
1951 return _is_uni_FOO(_CC_UPPER, c);
1955 Perl_is_uni_lower_lc(pTHX_ UV c)
1958 return isLOWER_LC(UNI_TO_NATIVE(c));
1960 return _is_uni_FOO(_CC_LOWER, c);
1964 Perl_is_uni_cntrl_lc(pTHX_ UV c)
1967 return isCNTRL_LC(UNI_TO_NATIVE(c));
1973 Perl_is_uni_graph_lc(pTHX_ UV c)
1976 return isGRAPH_LC(UNI_TO_NATIVE(c));
1978 return _is_uni_FOO(_CC_GRAPH, c);
1982 Perl_is_uni_print_lc(pTHX_ UV c)
1985 return isPRINT_LC(UNI_TO_NATIVE(c));
1987 return _is_uni_FOO(_CC_PRINT, c);
1991 Perl_is_uni_punct_lc(pTHX_ UV c)
1994 return isPUNCT_LC(UNI_TO_NATIVE(c));
1996 return _is_uni_FOO(_CC_PUNCT, c);
2000 Perl_is_uni_xdigit_lc(pTHX_ UV c)
2003 return isXDIGIT_LC(UNI_TO_NATIVE(c));
2005 return isXDIGIT_uni(c);
2009 Perl_to_uni_upper_lc(pTHX_ U32 c)
2011 /* XXX returns only the first character -- do not use XXX */
2012 /* XXX no locale support yet */
2014 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2015 return (U32)to_uni_upper(c, tmpbuf, &len);
2019 Perl_to_uni_title_lc(pTHX_ U32 c)
2021 /* XXX returns only the first character XXX -- do not use XXX */
2022 /* XXX no locale support yet */
2024 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2025 return (U32)to_uni_title(c, tmpbuf, &len);
2029 Perl_to_uni_lower_lc(pTHX_ U32 c)
2031 /* XXX returns only the first character -- do not use XXX */
2032 /* XXX no locale support yet */
2034 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2035 return (U32)to_uni_lower(c, tmpbuf, &len);
2038 PERL_STATIC_INLINE bool
2039 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2040 const char *const swashname)
2042 /* returns a boolean giving whether or not the UTF8-encoded character that
2043 * starts at <p> is in the swash indicated by <swashname>. <swash>
2044 * contains a pointer to where the swash indicated by <swashname>
2045 * is to be stored; which this routine will do, so that future calls will
2046 * look at <*swash> and only generate a swash if it is not null
2048 * Note that it is assumed that the buffer length of <p> is enough to
2049 * contain all the bytes that comprise the character. Thus, <*p> should
2050 * have been checked before this call for mal-formedness enough to assure
2055 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2057 /* The API should have included a length for the UTF-8 character in <p>,
2058 * but it doesn't. We therefore assume that p has been validated at least
2059 * as far as there being enough bytes available in it to accommodate the
2060 * character without reading beyond the end, and pass that number on to the
2061 * validating routine */
2062 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
2063 if (ckWARN_d(WARN_UTF8)) {
2064 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
2065 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
2066 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
2067 what the malformation is */
2068 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
2074 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2075 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
2078 return swash_fetch(*swash, p, TRUE) != 0;
2082 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2086 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2088 assert(classnum < _FIRST_NON_SWASH_CC);
2090 return is_utf8_common(p, &PL_utf8_swash_ptrs[classnum], swash_property_names[classnum]);
2094 Perl_is_utf8_alnum(pTHX_ const U8 *p)
2098 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
2100 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
2101 * descendant of isalnum(3), in other words, it doesn't
2102 * contain the '_'. --jhi */
2103 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_WORDCHAR], "IsWord");
2107 Perl_is_utf8_alnumc(pTHX_ const U8 *p)
2111 PERL_ARGS_ASSERT_IS_UTF8_ALNUMC;
2113 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHANUMERIC], "IsAlnum");
2117 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
2121 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
2123 return S_is_utf8_idfirst(aTHX_ p);
2127 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
2131 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
2135 /* is_utf8_idstart would be more logical. */
2136 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2140 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
2144 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
2146 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2150 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2154 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2156 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont");
2161 Perl_is_utf8_idcont(pTHX_ const U8 *p)
2165 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2167 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
2171 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2175 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2177 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2181 Perl_is_utf8_alpha(pTHX_ const U8 *p)
2185 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2187 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHA], "IsAlpha");
2191 Perl_is_utf8_ascii(pTHX_ const U8 *p)
2195 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2197 /* ASCII characters are the same whether in utf8 or not. So the macro
2198 * works on both utf8 and non-utf8 representations. */
2203 Perl_is_utf8_blank(pTHX_ const U8 *p)
2207 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2209 return isBLANK_utf8(p);
2213 Perl_is_utf8_space(pTHX_ const U8 *p)
2217 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2219 return isSPACE_utf8(p);
2223 Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2227 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2229 /* Only true if is an ASCII space-like character, and ASCII is invariant
2230 * under utf8, so can just use the macro */
2231 return isSPACE_A(*p);
2235 Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2239 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2241 /* Only true if is an ASCII word character, and ASCII is invariant
2242 * under utf8, so can just use the macro */
2243 return isWORDCHAR_A(*p);
2247 Perl_is_utf8_digit(pTHX_ const U8 *p)
2251 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2253 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_DIGIT], "IsDigit");
2257 Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2261 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2263 /* Only true if is an ASCII digit character, and ASCII is invariant
2264 * under utf8, so can just use the macro */
2265 return isDIGIT_A(*p);
2269 Perl_is_utf8_upper(pTHX_ const U8 *p)
2273 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2275 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_UPPER], "IsUppercase");
2279 Perl_is_utf8_lower(pTHX_ const U8 *p)
2283 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2285 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_LOWER], "IsLowercase");
2289 Perl_is_utf8_cntrl(pTHX_ const U8 *p)
2293 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2295 return isCNTRL_utf8(p);
2299 Perl_is_utf8_graph(pTHX_ const U8 *p)
2303 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2305 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_GRAPH], "IsGraph");
2309 Perl_is_utf8_print(pTHX_ const U8 *p)
2313 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2315 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PRINT], "IsPrint");
2319 Perl_is_utf8_punct(pTHX_ const U8 *p)
2323 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2325 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PUNCT], "IsPunct");
2329 Perl_is_utf8_xdigit(pTHX_ const U8 *p)
2333 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2335 return is_XDIGIT_utf8(p);
2339 Perl__is_utf8_mark(pTHX_ const U8 *p)
2343 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2345 return is_utf8_common(p, &PL_utf8_mark, "IsM");
2350 Perl_is_utf8_mark(pTHX_ const U8 *p)
2354 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2356 return _is_utf8_mark(p);
2360 =for apidoc to_utf8_case
2362 The C<p> contains the pointer to the UTF-8 string encoding
2363 the character that is being converted. This routine assumes that the character
2364 at C<p> is well-formed.
2366 The C<ustrp> is a pointer to the character buffer to put the
2367 conversion result to. The C<lenp> is a pointer to the length
2370 The C<swashp> is a pointer to the swash to use.
2372 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2373 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
2374 but not always, a multicharacter mapping), is tried first.
2376 The C<special> is a string like "utf8::ToSpecLower", which means the
2377 hash %utf8::ToSpecLower. The access to the hash is through
2378 Perl_to_utf8_case().
2380 The C<normal> is a string like "ToLower" which means the swash
2386 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2387 SV **swashp, const char *normal, const char *special)
2390 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2392 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
2393 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2394 * are necessary in EBCDIC, they are redundant no-ops
2395 * in ASCII-ish platforms, and hopefully optimized away. */
2396 const UV uv1 = NATIVE_TO_UNI(uv0);
2398 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2400 /* Note that swash_fetch() doesn't output warnings for these because it
2401 * assumes we will */
2402 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2403 if (uv1 <= UNICODE_SURROGATE_LAST) {
2404 if (ckWARN_d(WARN_SURROGATE)) {
2405 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2406 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2407 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2410 else if (UNICODE_IS_SUPER(uv1)) {
2411 if (ckWARN_d(WARN_NON_UNICODE)) {
2412 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2413 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2414 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2418 /* Note that non-characters are perfectly legal, so no warning should
2422 uvuni_to_utf8(tmpbuf, uv1);
2424 if (!*swashp) /* load on-demand */
2425 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2428 /* It might be "special" (sometimes, but not always,
2429 * a multicharacter mapping) */
2430 HV * const hv = get_hv(special, 0);
2434 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2438 s = SvPV_const(*svp, len);
2440 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2443 /* If we have EBCDIC we need to remap the characters
2444 * since any characters in the low 256 are Unicode
2445 * code points, not EBCDIC. */
2446 U8 *t = (U8*)s, *tend = t + len, *d;
2453 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2455 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2464 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2469 Copy(tmpbuf, ustrp, len, U8);
2471 Copy(s, ustrp, len, U8);
2477 if (!len && *swashp) {
2478 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
2481 /* It was "normal" (a single character mapping). */
2482 const UV uv3 = UNI_TO_NATIVE(uv2);
2483 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2491 return valid_utf8_to_uvchr(ustrp, 0);
2494 /* Here, there was no mapping defined, which means that the code point maps
2495 * to itself. Return the inputs */
2497 if (p != ustrp) { /* Don't copy onto itself */
2498 Copy(p, ustrp, len, U8);
2509 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2511 /* This is called when changing the case of a utf8-encoded character above
2512 * the Latin1 range, and the operation is in locale. If the result
2513 * contains a character that crosses the 255/256 boundary, disallow the
2514 * change, and return the original code point. See L<perlfunc/lc> for why;
2516 * p points to the original string whose case was changed; assumed
2517 * by this routine to be well-formed
2518 * result the code point of the first character in the changed-case string
2519 * ustrp points to the changed-case string (<result> represents its first char)
2520 * lenp points to the length of <ustrp> */
2522 UV original; /* To store the first code point of <p> */
2524 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2526 assert(UTF8_IS_ABOVE_LATIN1(*p));
2528 /* We know immediately if the first character in the string crosses the
2529 * boundary, so can skip */
2532 /* Look at every character in the result; if any cross the
2533 * boundary, the whole thing is disallowed */
2534 U8* s = ustrp + UTF8SKIP(ustrp);
2535 U8* e = ustrp + *lenp;
2537 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2543 /* Here, no characters crossed, result is ok as-is */
2549 /* Failed, have to return the original */
2550 original = valid_utf8_to_uvchr(p, lenp);
2551 Copy(p, ustrp, *lenp, char);
2556 =for apidoc to_utf8_upper
2558 Instead use L</toUPPER_utf8>.
2562 /* Not currently externally documented, and subject to change:
2563 * <flags> is set iff locale semantics are to be used for code points < 256
2564 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2565 * were used in the calculation; otherwise unchanged. */
2568 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2574 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2576 if (UTF8_IS_INVARIANT(*p)) {
2578 result = toUPPER_LC(*p);
2581 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2584 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2586 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2589 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2593 else { /* utf8, ord above 255 */
2594 result = CALL_UPPER_CASE(p, ustrp, lenp);
2597 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2602 /* Here, used locale rules. Convert back to utf8 */
2603 if (UTF8_IS_INVARIANT(result)) {
2604 *ustrp = (U8) result;
2608 *ustrp = UTF8_EIGHT_BIT_HI(result);
2609 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2614 *tainted_ptr = TRUE;
2620 =for apidoc to_utf8_title
2622 Instead use L</toTITLE_utf8>.
2626 /* Not currently externally documented, and subject to change:
2627 * <flags> is set iff locale semantics are to be used for code points < 256
2628 * Since titlecase is not defined in POSIX, uppercase is used instead
2630 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2631 * were used in the calculation; otherwise unchanged. */
2634 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2640 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2642 if (UTF8_IS_INVARIANT(*p)) {
2644 result = toUPPER_LC(*p);
2647 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2650 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2652 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2655 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2659 else { /* utf8, ord above 255 */
2660 result = CALL_TITLE_CASE(p, ustrp, lenp);
2663 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2668 /* Here, used locale rules. Convert back to utf8 */
2669 if (UTF8_IS_INVARIANT(result)) {
2670 *ustrp = (U8) result;
2674 *ustrp = UTF8_EIGHT_BIT_HI(result);
2675 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2680 *tainted_ptr = TRUE;
2686 =for apidoc to_utf8_lower
2688 Instead use L</toLOWER_utf8>.
2692 /* Not currently externally documented, and subject to change:
2693 * <flags> is set iff locale semantics are to be used for code points < 256
2694 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2695 * were used in the calculation; otherwise unchanged. */
2698 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2704 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2706 if (UTF8_IS_INVARIANT(*p)) {
2708 result = toLOWER_LC(*p);
2711 return to_lower_latin1(*p, ustrp, lenp);
2714 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2716 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2719 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2723 else { /* utf8, ord above 255 */
2724 result = CALL_LOWER_CASE(p, ustrp, lenp);
2727 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2733 /* Here, used locale rules. Convert back to utf8 */
2734 if (UTF8_IS_INVARIANT(result)) {
2735 *ustrp = (U8) result;
2739 *ustrp = UTF8_EIGHT_BIT_HI(result);
2740 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2745 *tainted_ptr = TRUE;
2751 =for apidoc to_utf8_fold
2753 Instead use L</toFOLD_utf8>.
2757 /* Not currently externally documented, and subject to change,
2759 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2760 * points < 256. Since foldcase is not defined in
2761 * POSIX, lowercase is used instead
2762 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2763 * otherwise simple folds
2764 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2766 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2767 * were used in the calculation; otherwise unchanged. */
2770 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2776 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2778 /* These are mutually exclusive */
2779 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2781 assert(p != ustrp); /* Otherwise overwrites */
2783 if (UTF8_IS_INVARIANT(*p)) {
2784 if (flags & FOLD_FLAGS_LOCALE) {
2785 result = toFOLD_LC(*p);
2788 return _to_fold_latin1(*p, ustrp, lenp,
2789 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2792 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2793 if (flags & FOLD_FLAGS_LOCALE) {
2794 result = toFOLD_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2797 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2799 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2802 else { /* utf8, ord above 255 */
2803 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2805 if (flags & FOLD_FLAGS_LOCALE) {
2807 /* Special case this character, as what normally gets returned
2808 * under locale doesn't work */
2809 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2810 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2811 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2815 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2817 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2821 /* This is called when changing the case of a utf8-encoded
2822 * character above the Latin1 range, and the result should not
2823 * contain an ASCII character. */
2825 UV original; /* To store the first code point of <p> */
2827 /* Look at every character in the result; if any cross the
2828 * boundary, the whole thing is disallowed */
2830 U8* e = ustrp + *lenp;
2833 /* Crossed, have to return the original */
2834 original = valid_utf8_to_uvchr(p, lenp);
2836 /* But in this one instance, there is an alternative we can
2837 * return that is valid */
2838 if (original == LATIN_CAPITAL_LETTER_SHARP_S) {
2841 Copy(p, ustrp, *lenp, char);
2847 /* Here, no characters crossed, result is ok as-is */
2852 /* Here, used locale rules. Convert back to utf8 */
2853 if (UTF8_IS_INVARIANT(result)) {
2854 *ustrp = (U8) result;
2858 *ustrp = UTF8_EIGHT_BIT_HI(result);
2859 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2864 *tainted_ptr = TRUE;
2869 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2870 * folds to a string of two of these characters. By returning this
2871 * instead, then, e.g.,
2872 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2875 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2876 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2878 return LATIN_SMALL_LETTER_LONG_S;
2882 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2883 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2884 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2888 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2890 PERL_ARGS_ASSERT_SWASH_INIT;
2892 /* Returns a copy of a swash initiated by the called function. This is the
2893 * public interface, and returning a copy prevents others from doing
2894 * mischief on the original */
2896 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2900 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2902 /* Initialize and return a swash, creating it if necessary. It does this
2903 * by calling utf8_heavy.pl in the general case. The returned value may be
2904 * the swash's inversion list instead if the input parameters allow it.
2905 * Which is returned should be immaterial to callers, as the only
2906 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2907 * and swash_to_invlist() handle both these transparently.
2909 * This interface should only be used by functions that won't destroy or
2910 * adversely change the swash, as doing so affects all other uses of the
2911 * swash in the program; the general public should use 'Perl_swash_init'
2914 * pkg is the name of the package that <name> should be in.
2915 * name is the name of the swash to find. Typically it is a Unicode
2916 * property name, including user-defined ones
2917 * listsv is a string to initialize the swash with. It must be of the form
2918 * documented as the subroutine return value in
2919 * L<perlunicode/User-Defined Character Properties>
2920 * minbits is the number of bits required to represent each data element.
2921 * It is '1' for binary properties.
2922 * none I (khw) do not understand this one, but it is used only in tr///.
2923 * invlist is an inversion list to initialize the swash with (or NULL)
2924 * flags_p if non-NULL is the address of various input and output flag bits
2925 * to the routine, as follows: ('I' means is input to the routine;
2926 * 'O' means output from the routine. Only flags marked O are
2927 * meaningful on return.)
2928 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2929 * came from a user-defined property. (I O)
2930 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2931 * when the swash cannot be located, to simply return NULL. (I)
2932 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2933 * return of an inversion list instead of a swash hash if this routine
2934 * thinks that would result in faster execution of swash_fetch() later
2937 * Thus there are three possible inputs to find the swash: <name>,
2938 * <listsv>, and <invlist>. At least one must be specified. The result
2939 * will be the union of the specified ones, although <listsv>'s various
2940 * actions can intersect, etc. what <name> gives.
2942 * <invlist> is only valid for binary properties */
2945 SV* retval = &PL_sv_undef;
2946 HV* swash_hv = NULL;
2947 const int invlist_swash_boundary =
2948 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2949 ? 512 /* Based on some benchmarking, but not extensive, see commit
2951 : -1; /* Never return just an inversion list */
2953 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2954 assert(! invlist || minbits == 1);
2956 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2958 if (listsv != &PL_sv_undef || strNE(name, "")) {
2960 const size_t pkg_len = strlen(pkg);
2961 const size_t name_len = strlen(name);
2962 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2966 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2968 PUSHSTACKi(PERLSI_MAGIC);
2972 /* We might get here via a subroutine signature which uses a utf8
2973 * parameter name, at which point PL_subname will have been set
2974 * but not yet used. */
2975 save_item(PL_subname);
2976 if (PL_parser && PL_parser->error_count)
2977 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2978 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2979 if (!method) { /* demand load utf8 */
2981 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2982 GvSV(PL_errgv) = NULL;
2983 /* It is assumed that callers of this routine are not passing in
2984 * any user derived data. */
2985 /* Need to do this after save_re_context() as it will set
2986 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2987 * in Perl_magic_get). Even line to create errsv_save can turn on
2989 #ifndef NO_TAINT_SUPPORT
2990 SAVEBOOL(TAINT_get);
2993 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2996 /* Not ERRSV, as there is no need to vivify a scalar we are
2997 about to discard. */
2998 SV * const errsv = GvSV(PL_errgv);
2999 if (!SvTRUE(errsv)) {
3000 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3001 SvREFCNT_dec(errsv);
3009 mPUSHp(pkg, pkg_len);
3010 mPUSHp(name, name_len);
3015 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3016 GvSV(PL_errgv) = NULL;
3017 /* If we already have a pointer to the method, no need to use
3018 * call_method() to repeat the lookup. */
3020 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3021 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3023 retval = *PL_stack_sp--;
3024 SvREFCNT_inc(retval);
3027 /* Not ERRSV. See above. */
3028 SV * const errsv = GvSV(PL_errgv);
3029 if (!SvTRUE(errsv)) {
3030 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3031 SvREFCNT_dec(errsv);
3036 if (IN_PERL_COMPILETIME) {
3037 CopHINTS_set(PL_curcop, PL_hints);
3039 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3042 /* If caller wants to handle missing properties, let them */
3043 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3047 "Can't find Unicode property definition \"%"SVf"\"",
3049 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3051 } /* End of calling the module to find the swash */
3053 /* If this operation fetched a swash, and we will need it later, get it */
3054 if (retval != &PL_sv_undef
3055 && (minbits == 1 || (flags_p
3057 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3059 swash_hv = MUTABLE_HV(SvRV(retval));
3061 /* If we don't already know that there is a user-defined component to
3062 * this swash, and the user has indicated they wish to know if there is
3063 * one (by passing <flags_p>), find out */
3064 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3065 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3066 if (user_defined && SvUV(*user_defined)) {
3067 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3072 /* Make sure there is an inversion list for binary properties */
3074 SV** swash_invlistsvp = NULL;
3075 SV* swash_invlist = NULL;
3076 bool invlist_in_swash_is_valid = FALSE;
3077 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3078 an unclaimed reference count */
3080 /* If this operation fetched a swash, get its already existing
3081 * inversion list, or create one for it */
3084 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3085 if (swash_invlistsvp) {
3086 swash_invlist = *swash_invlistsvp;
3087 invlist_in_swash_is_valid = TRUE;
3090 swash_invlist = _swash_to_invlist(retval);
3091 swash_invlist_unclaimed = TRUE;
3095 /* If an inversion list was passed in, have to include it */
3098 /* Any fetched swash will by now have an inversion list in it;
3099 * otherwise <swash_invlist> will be NULL, indicating that we
3100 * didn't fetch a swash */
3101 if (swash_invlist) {
3103 /* Add the passed-in inversion list, which invalidates the one
3104 * already stored in the swash */
3105 invlist_in_swash_is_valid = FALSE;
3106 _invlist_union(invlist, swash_invlist, &swash_invlist);
3110 /* Here, there is no swash already. Set up a minimal one, if
3111 * we are going to return a swash */
3112 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3114 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3116 swash_invlist = invlist;
3120 /* Here, we have computed the union of all the passed-in data. It may
3121 * be that there was an inversion list in the swash which didn't get
3122 * touched; otherwise save the one computed one */
3123 if (! invlist_in_swash_is_valid
3124 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3126 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3128 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3130 /* We just stole a reference count. */
3131 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3132 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3135 /* Use the inversion list stand-alone if small enough */
3136 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3137 SvREFCNT_dec(retval);
3138 if (!swash_invlist_unclaimed)
3139 SvREFCNT_inc_simple_void_NN(swash_invlist);
3140 retval = newRV_noinc(swash_invlist);
3148 /* This API is wrong for special case conversions since we may need to
3149 * return several Unicode characters for a single Unicode character
3150 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3151 * the lower-level routine, and it is similarly broken for returning
3152 * multiple values. --jhi
3153 * For those, you should use to_utf8_case() instead */
3154 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3157 * Returns the value of property/mapping C<swash> for the first character
3158 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3159 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3160 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3162 * A "swash" is a hash which contains initially the keys/values set up by
3163 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3164 * property for all possible code points. Things are stored in a compact form
3165 * (see utf8_heavy.pl) so that calculation is required to find the actual
3166 * property value for a given code point. As code points are looked up, new
3167 * key/value pairs are added to the hash, so that the calculation doesn't have
3168 * to ever be re-done. Further, each calculation is done, not just for the
3169 * desired one, but for a whole block of code points adjacent to that one.
3170 * For binary properties on ASCII machines, the block is usually for 64 code
3171 * points, starting with a code point evenly divisible by 64. Thus if the
3172 * property value for code point 257 is requested, the code goes out and
3173 * calculates the property values for all 64 code points between 256 and 319,
3174 * and stores these as a single 64-bit long bit vector, called a "swatch",
3175 * under the key for code point 256. The key is the UTF-8 encoding for code
3176 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3177 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3178 * for code point 258 is then requested, this code realizes that it would be
3179 * stored under the key for 256, and would find that value and extract the
3180 * relevant bit, offset from 256.
3182 * Non-binary properties are stored in as many bits as necessary to represent
3183 * their values (32 currently, though the code is more general than that), not
3184 * as single bits, but the principal is the same: the value for each key is a
3185 * vector that encompasses the property values for all code points whose UTF-8
3186 * representations are represented by the key. That is, for all code points
3187 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3191 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3194 HV *const hv = MUTABLE_HV(SvRV(swash));
3199 const U8 *tmps = NULL;
3203 const UV c = NATIVE_TO_ASCII(*ptr);
3205 PERL_ARGS_ASSERT_SWASH_FETCH;
3207 /* If it really isn't a hash, it isn't really swash; must be an inversion
3209 if (SvTYPE(hv) != SVt_PVHV) {
3210 return _invlist_contains_cp((SV*)hv,
3212 ? valid_utf8_to_uvchr(ptr, NULL)
3216 /* Convert to utf8 if not already */
3217 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
3218 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3219 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3222 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
3223 * then the "swatch" is a vec() for all the chars which start
3225 * So the key in the hash (klen) is length of encoded char -1
3227 klen = UTF8SKIP(ptr) - 1;
3230 /* If char is invariant then swatch is for all the invariant chars
3231 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
3233 needents = UTF_CONTINUATION_MARK;
3234 off = NATIVE_TO_UTF(ptr[klen]);
3237 /* If char is encoded then swatch is for the prefix */
3238 needents = (1 << UTF_ACCUMULATION_SHIFT);
3239 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3243 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3244 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3245 * it's nothing to sniff at.) Pity we usually come through at least
3246 * two function calls to get here...
3248 * NB: this code assumes that swatches are never modified, once generated!
3251 if (hv == PL_last_swash_hv &&
3252 klen == PL_last_swash_klen &&
3253 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3255 tmps = PL_last_swash_tmps;
3256 slen = PL_last_swash_slen;
3259 /* Try our second-level swatch cache, kept in a hash. */
3260 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3262 /* If not cached, generate it via swatch_get */
3263 if (!svp || !SvPOK(*svp)
3264 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
3265 /* We use utf8n_to_uvuni() as we want an index into
3266 Unicode tables, not a native character number.
3268 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
3270 0 : UTF8_ALLOW_ANY);
3271 swatch = swatch_get(swash,
3272 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
3273 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
3276 if (IN_PERL_COMPILETIME)
3277 CopHINTS_set(PL_curcop, PL_hints);
3279 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3281 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3282 || (slen << 3) < needents)
3283 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3284 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3285 svp, tmps, (UV)slen, (UV)needents);
3288 PL_last_swash_hv = hv;
3289 assert(klen <= sizeof(PL_last_swash_key));
3290 PL_last_swash_klen = (U8)klen;
3291 /* FIXME change interpvar.h? */
3292 PL_last_swash_tmps = (U8 *) tmps;
3293 PL_last_swash_slen = slen;
3295 Copy(ptr, PL_last_swash_key, klen, U8);
3298 switch ((int)((slen << 3) / needents)) {
3300 bit = 1 << (off & 7);
3302 return (tmps[off] & bit) != 0;
3307 return (tmps[off] << 8) + tmps[off + 1] ;
3310 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3312 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3313 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
3314 NORETURN_FUNCTION_END;
3317 /* Read a single line of the main body of the swash input text. These are of
3320 * where each number is hex. The first two numbers form the minimum and
3321 * maximum of a range, and the third is the value associated with the range.
3322 * Not all swashes should have a third number
3324 * On input: l points to the beginning of the line to be examined; it points
3325 * to somewhere in the string of the whole input text, and is
3326 * terminated by a \n or the null string terminator.
3327 * lend points to the null terminator of that string
3328 * wants_value is non-zero if the swash expects a third number
3329 * typestr is the name of the swash's mapping, like 'ToLower'
3330 * On output: *min, *max, and *val are set to the values read from the line.
3331 * returns a pointer just beyond the line examined. If there was no
3332 * valid min number on the line, returns lend+1
3336 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3337 const bool wants_value, const U8* const typestr)
3339 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3340 STRLEN numlen; /* Length of the number */
3341 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3342 | PERL_SCAN_DISALLOW_PREFIX
3343 | PERL_SCAN_SILENT_NON_PORTABLE;
3345 /* nl points to the next \n in the scan */
3346 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3348 /* Get the first number on the line: the range minimum */
3350 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3351 if (numlen) /* If found a hex number, position past it */
3353 else if (nl) { /* Else, go handle next line, if any */
3354 return nl + 1; /* 1 is length of "\n" */
3356 else { /* Else, no next line */
3357 return lend + 1; /* to LIST's end at which \n is not found */
3360 /* The max range value follows, separated by a BLANK */
3363 flags = PERL_SCAN_SILENT_ILLDIGIT
3364 | PERL_SCAN_DISALLOW_PREFIX
3365 | PERL_SCAN_SILENT_NON_PORTABLE;
3367 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3370 else /* If no value here, it is a single element range */
3373 /* Non-binary tables have a third entry: what the first element of the
3379 /* The ToLc, etc table mappings are not in hex, and must be
3380 * corrected by adding the code point to them */
3382 char *after_strtol = (char *) lend;
3383 *val = Strtol((char *)l, &after_strtol, 10);
3384 l = (U8 *) after_strtol;
3386 else { /* Other tables are in hex, and are the correct result
3388 flags = PERL_SCAN_SILENT_ILLDIGIT
3389 | PERL_SCAN_DISALLOW_PREFIX
3390 | PERL_SCAN_SILENT_NON_PORTABLE;
3392 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3402 /* diag_listed_as: To%s: illegal mapping '%s' */
3403 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3409 *val = 0; /* bits == 1, then any val should be ignored */
3411 else { /* Nothing following range min, should be single element with no
3417 /* diag_listed_as: To%s: illegal mapping '%s' */
3418 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3422 *val = 0; /* bits == 1, then val should be ignored */
3425 /* Position to next line if any, or EOF */
3435 * Returns a swatch (a bit vector string) for a code point sequence
3436 * that starts from the value C<start> and comprises the number C<span>.
3437 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3438 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3441 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3444 U8 *l, *lend, *x, *xend, *s, *send;
3445 STRLEN lcur, xcur, scur;
3446 HV *const hv = MUTABLE_HV(SvRV(swash));
3447 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3449 SV** listsvp = NULL; /* The string containing the main body of the table */
3450 SV** extssvp = NULL;
3451 SV** invert_it_svp = NULL;
3454 STRLEN octets; /* if bits == 1, then octets == 0 */
3456 UV end = start + span;
3458 if (invlistsvp == NULL) {
3459 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3460 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3461 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3462 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3463 listsvp = hv_fetchs(hv, "LIST", FALSE);
3464 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3466 bits = SvUV(*bitssvp);
3467 none = SvUV(*nonesvp);
3468 typestr = (U8*)SvPV_nolen(*typesvp);
3474 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3476 PERL_ARGS_ASSERT_SWATCH_GET;
3478 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3479 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3483 /* If overflowed, use the max possible */
3489 /* create and initialize $swatch */
3490 scur = octets ? (span * octets) : (span + 7) / 8;
3491 swatch = newSV(scur);
3493 s = (U8*)SvPVX(swatch);
3494 if (octets && none) {
3495 const U8* const e = s + scur;
3498 *s++ = (U8)(none & 0xff);
3499 else if (bits == 16) {
3500 *s++ = (U8)((none >> 8) & 0xff);
3501 *s++ = (U8)( none & 0xff);
3503 else if (bits == 32) {
3504 *s++ = (U8)((none >> 24) & 0xff);
3505 *s++ = (U8)((none >> 16) & 0xff);
3506 *s++ = (U8)((none >> 8) & 0xff);
3507 *s++ = (U8)( none & 0xff);
3513 (void)memzero((U8*)s, scur + 1);
3515 SvCUR_set(swatch, scur);
3516 s = (U8*)SvPVX(swatch);
3518 if (invlistsvp) { /* If has an inversion list set up use that */
3519 _invlist_populate_swatch(*invlistsvp, start, end, s);
3523 /* read $swash->{LIST} */
3524 l = (U8*)SvPV(*listsvp, lcur);
3527 UV min, max, val, upper;
3528 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3529 cBOOL(octets), typestr);
3534 /* If looking for something beyond this range, go try the next one */
3538 /* <end> is generally 1 beyond where we want to set things, but at the
3539 * platform's infinity, where we can't go any higher, we want to
3540 * include the code point at <end> */
3543 : (max != UV_MAX || end != UV_MAX)
3550 if (!none || val < none) {
3555 for (key = min; key <= upper; key++) {
3557 /* offset must be non-negative (start <= min <= key < end) */
3558 offset = octets * (key - start);
3560 s[offset] = (U8)(val & 0xff);
3561 else if (bits == 16) {
3562 s[offset ] = (U8)((val >> 8) & 0xff);
3563 s[offset + 1] = (U8)( val & 0xff);
3565 else if (bits == 32) {
3566 s[offset ] = (U8)((val >> 24) & 0xff);
3567 s[offset + 1] = (U8)((val >> 16) & 0xff);
3568 s[offset + 2] = (U8)((val >> 8) & 0xff);
3569 s[offset + 3] = (U8)( val & 0xff);
3572 if (!none || val < none)
3576 else { /* bits == 1, then val should be ignored */
3581 for (key = min; key <= upper; key++) {
3582 const STRLEN offset = (STRLEN)(key - start);
3583 s[offset >> 3] |= 1 << (offset & 7);
3588 /* Invert if the data says it should be. Assumes that bits == 1 */
3589 if (invert_it_svp && SvUV(*invert_it_svp)) {
3591 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3592 * be 0, and their inversion should also be 0, as we don't succeed any
3593 * Unicode property matches for non-Unicode code points */
3594 if (start <= PERL_UNICODE_MAX) {
3596 /* The code below assumes that we never cross the
3597 * Unicode/above-Unicode boundary in a range, as otherwise we would
3598 * have to figure out where to stop flipping the bits. Since this
3599 * boundary is divisible by a large power of 2, and swatches comes
3600 * in small powers of 2, this should be a valid assumption */
3601 assert(start + span - 1 <= PERL_UNICODE_MAX);
3611 /* read $swash->{EXTRAS}
3612 * This code also copied to swash_to_invlist() below */
3613 x = (U8*)SvPV(*extssvp, xcur);
3621 SV **otherbitssvp, *other;
3625 const U8 opc = *x++;
3629 nl = (U8*)memchr(x, '\n', xend - x);
3631 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3633 x = nl + 1; /* 1 is length of "\n" */
3637 x = xend; /* to EXTRAS' end at which \n is not found */
3644 namelen = nl - namestr;
3648 namelen = xend - namestr;
3652 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3653 otherhv = MUTABLE_HV(SvRV(*othersvp));
3654 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3655 otherbits = (STRLEN)SvUV(*otherbitssvp);
3656 if (bits < otherbits)
3657 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3658 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3660 /* The "other" swatch must be destroyed after. */
3661 other = swatch_get(*othersvp, start, span);
3662 o = (U8*)SvPV(other, olen);
3665 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3667 s = (U8*)SvPV(swatch, slen);
3668 if (bits == 1 && otherbits == 1) {
3670 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3671 "mismatch, slen=%"UVuf", olen=%"UVuf,
3672 (UV)slen, (UV)olen);
3696 STRLEN otheroctets = otherbits >> 3;
3698 U8* const send = s + slen;
3703 if (otherbits == 1) {
3704 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3708 STRLEN vlen = otheroctets;
3716 if (opc == '+' && otherval)
3717 NOOP; /* replace with otherval */
3718 else if (opc == '!' && !otherval)
3720 else if (opc == '-' && otherval)
3722 else if (opc == '&' && !otherval)
3725 s += octets; /* no replacement */
3730 *s++ = (U8)( otherval & 0xff);
3731 else if (bits == 16) {
3732 *s++ = (U8)((otherval >> 8) & 0xff);
3733 *s++ = (U8)( otherval & 0xff);
3735 else if (bits == 32) {
3736 *s++ = (U8)((otherval >> 24) & 0xff);
3737 *s++ = (U8)((otherval >> 16) & 0xff);
3738 *s++ = (U8)((otherval >> 8) & 0xff);
3739 *s++ = (U8)( otherval & 0xff);
3743 sv_free(other); /* through with it! */
3749 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3752 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3753 * Can't be used on a property that is subject to user override, as it
3754 * relies on the value of SPECIALS in the swash which would be set by
3755 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3756 * for overridden properties
3758 * Returns a hash which is the inversion and closure of a swash mapping.
3759 * For example, consider the input lines:
3764 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3765 * 006C. The value for each key is an array. For 006C, the array would
3766 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3767 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3769 * Essentially, for any code point, it gives all the code points that map to
3770 * it, or the list of 'froms' for that point.
3772 * Currently it ignores any additions or deletions from other swashes,
3773 * looking at just the main body of the swash, and if there are SPECIALS
3774 * in the swash, at that hash
3776 * The specials hash can be extra code points, and most likely consists of
3777 * maps from single code points to multiple ones (each expressed as a string
3778 * of utf8 characters). This function currently returns only 1-1 mappings.
3779 * However consider this possible input in the specials hash:
3780 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3781 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3783 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3784 * currently handle. But it also means that FB05 and FB06 are equivalent in
3785 * a 1-1 mapping which we should handle, and this relationship may not be in
3786 * the main table. Therefore this function examines all the multi-char
3787 * sequences and adds the 1-1 mappings that come out of that. */
3791 HV *const hv = MUTABLE_HV(SvRV(swash));
3793 /* The string containing the main body of the table. This will have its
3794 * assertion fail if the swash has been converted to its inversion list */
3795 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3797 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3798 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3799 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3800 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3801 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3802 const STRLEN bits = SvUV(*bitssvp);
3803 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3804 const UV none = SvUV(*nonesvp);
3805 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3809 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3811 /* Must have at least 8 bits to get the mappings */
3812 if (bits != 8 && bits != 16 && bits != 32) {
3813 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3817 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3818 mapping to more than one character */
3820 /* Construct an inverse mapping hash for the specials */
3821 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3822 HV * specials_inverse = newHV();
3823 char *char_from; /* the lhs of the map */
3824 I32 from_len; /* its byte length */
3825 char *char_to; /* the rhs of the map */
3826 I32 to_len; /* its byte length */
3827 SV *sv_to; /* and in a sv */
3828 AV* from_list; /* list of things that map to each 'to' */
3830 hv_iterinit(specials_hv);
3832 /* The keys are the characters (in utf8) that map to the corresponding
3833 * utf8 string value. Iterate through the list creating the inverse
3835 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3837 if (! SvPOK(sv_to)) {
3838 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3839 "unexpectedly is not a string, flags=%lu",
3840 (unsigned long)SvFLAGS(sv_to));
3842 /*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)));*/
3844 /* Each key in the inverse list is a mapped-to value, and the key's
3845 * hash value is a list of the strings (each in utf8) that map to
3846 * it. Those strings are all one character long */
3847 if ((listp = hv_fetch(specials_inverse,
3851 from_list = (AV*) *listp;
3853 else { /* No entry yet for it: create one */
3854 from_list = newAV();
3855 if (! hv_store(specials_inverse,
3858 (SV*) from_list, 0))
3860 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3864 /* Here have the list associated with this 'to' (perhaps newly
3865 * created and empty). Just add to it. Note that we ASSUME that
3866 * the input is guaranteed to not have duplications, so we don't
3867 * check for that. Duplications just slow down execution time. */
3868 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3871 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3872 * it looking for cases like the FB05/FB06 examples above. There would
3873 * be an entry in the hash like
3874 * 'st' => [ FB05, FB06 ]
3875 * In this example we will create two lists that get stored in the
3876 * returned hash, 'ret':
3877 * FB05 => [ FB05, FB06 ]
3878 * FB06 => [ FB05, FB06 ]
3880 * Note that there is nothing to do if the array only has one element.
3881 * (In the normal 1-1 case handled below, we don't have to worry about
3882 * two lists, as everything gets tied to the single list that is
3883 * generated for the single character 'to'. But here, we are omitting
3884 * that list, ('st' in the example), so must have multiple lists.) */
3885 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3886 &char_to, &to_len)))
3888 if (av_len(from_list) > 0) {
3891 /* We iterate over all combinations of i,j to place each code
3892 * point on each list */
3893 for (i = 0; i <= av_len(from_list); i++) {
3895 AV* i_list = newAV();
3896 SV** entryp = av_fetch(from_list, i, FALSE);
3897 if (entryp == NULL) {
3898 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3900 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3901 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3903 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3904 (SV*) i_list, FALSE))
3906 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3909 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3910 for (j = 0; j <= av_len(from_list); j++) {
3911 entryp = av_fetch(from_list, j, FALSE);
3912 if (entryp == NULL) {
3913 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3916 /* When i==j this adds itself to the list */
3917 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3918 (U8*) SvPVX(*entryp),
3919 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3921 /*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));*/
3926 SvREFCNT_dec(specials_inverse); /* done with it */
3927 } /* End of specials */
3929 /* read $swash->{LIST} */
3930 l = (U8*)SvPV(*listsvp, lcur);
3933 /* Go through each input line */
3937 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3938 cBOOL(octets), typestr);
3943 /* Each element in the range is to be inverted */
3944 for (inverse = min; inverse <= max; inverse++) {
3948 bool found_key = FALSE;
3949 bool found_inverse = FALSE;
3951 /* The key is the inverse mapping */
3952 char key[UTF8_MAXBYTES+1];
3953 char* key_end = (char *) uvuni_to_utf8((U8*) key, val);
3954 STRLEN key_len = key_end - key;
3956 /* Get the list for the map */
3957 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3958 list = (AV*) *listp;
3960 else { /* No entry yet for it: create one */
3962 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3963 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3967 /* Look through list to see if this inverse mapping already is
3968 * listed, or if there is a mapping to itself already */
3969 for (i = 0; i <= av_len(list); i++) {
3970 SV** entryp = av_fetch(list, i, FALSE);
3972 if (entryp == NULL) {
3973 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3976 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3977 if (SvUV(entry) == val) {
3980 if (SvUV(entry) == inverse) {
3981 found_inverse = TRUE;
3984 /* No need to continue searching if found everything we are
3986 if (found_key && found_inverse) {
3991 /* Make sure there is a mapping to itself on the list */
3993 av_push(list, newSVuv(val));
3994 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3998 /* Simply add the value to the list */
3999 if (! found_inverse) {
4000 av_push(list, newSVuv(inverse));
4001 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
4004 /* swatch_get() increments the value of val for each element in the
4005 * range. That makes more compact tables possible. You can
4006 * express the capitalization, for example, of all consecutive
4007 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4008 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4009 * and it's not documented; it appears to be used only in
4010 * implementing tr//; I copied the semantics from swatch_get(), just
4012 if (!none || val < none) {
4022 Perl__swash_to_invlist(pTHX_ SV* const swash)
4025 /* Subject to change or removal. For use only in one place in regcomp.c.
4026 * Ownership is given to one reference count in the returned SV* */
4031 HV *const hv = MUTABLE_HV(SvRV(swash));
4032 UV elements = 0; /* Number of elements in the inversion list */
4042 STRLEN octets; /* if bits == 1, then octets == 0 */
4048 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4050 /* If not a hash, it must be the swash's inversion list instead */
4051 if (SvTYPE(hv) != SVt_PVHV) {
4052 return SvREFCNT_inc_simple_NN((SV*) hv);
4055 /* The string containing the main body of the table */
4056 listsvp = hv_fetchs(hv, "LIST", FALSE);
4057 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4058 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4059 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4060 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4062 typestr = (U8*)SvPV_nolen(*typesvp);
4063 bits = SvUV(*bitssvp);
4064 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4066 /* read $swash->{LIST} */
4067 if (SvPOK(*listsvp)) {
4068 l = (U8*)SvPV(*listsvp, lcur);
4071 /* LIST legitimately doesn't contain a string during compilation phases
4072 * of Perl itself, before the Unicode tables are generated. In this
4073 * case, just fake things up by creating an empty list */
4080 /* Scan the input to count the number of lines to preallocate array size
4081 * based on worst possible case, which is each line in the input creates 2
4082 * elements in the inversion list: 1) the beginning of a range in the list;
4083 * 2) the beginning of a range not in the list. */
4084 while ((loc = (strchr(loc, '\n'))) != NULL) {
4089 /* If the ending is somehow corrupt and isn't a new line, add another
4090 * element for the final range that isn't in the inversion list */
4091 if (! (*lend == '\n'
4092 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4097 invlist = _new_invlist(elements);
4099 /* Now go through the input again, adding each range to the list */
4102 UV val; /* Not used by this function */
4104 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
4105 cBOOL(octets), typestr);
4111 invlist = _add_range_to_invlist(invlist, start, end);
4114 /* Invert if the data says it should be */
4115 if (invert_it_svp && SvUV(*invert_it_svp)) {
4116 _invlist_invert_prop(invlist);
4119 /* This code is copied from swatch_get()
4120 * read $swash->{EXTRAS} */
4121 x = (U8*)SvPV(*extssvp, xcur);
4129 SV **otherbitssvp, *other;
4132 const U8 opc = *x++;
4136 nl = (U8*)memchr(x, '\n', xend - x);
4138 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4140 x = nl + 1; /* 1 is length of "\n" */
4144 x = xend; /* to EXTRAS' end at which \n is not found */
4151 namelen = nl - namestr;
4155 namelen = xend - namestr;
4159 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4160 otherhv = MUTABLE_HV(SvRV(*othersvp));
4161 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4162 otherbits = (STRLEN)SvUV(*otherbitssvp);
4164 if (bits != otherbits || bits != 1) {
4165 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4166 "properties, bits=%"UVuf", otherbits=%"UVuf,
4167 (UV)bits, (UV)otherbits);
4170 /* The "other" swatch must be destroyed after. */
4171 other = _swash_to_invlist((SV *)*othersvp);
4173 /* End of code copied from swatch_get() */
4176 _invlist_union(invlist, other, &invlist);
4179 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4182 _invlist_subtract(invlist, other, &invlist);
4185 _invlist_intersection(invlist, other, &invlist);
4190 sv_free(other); /* through with it! */
4197 Perl__get_swash_invlist(pTHX_ SV* const swash)
4201 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4203 if (! SvROK(swash)) {
4207 /* If it really isn't a hash, it isn't really swash; must be an inversion
4209 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4213 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4222 =for apidoc uvchr_to_utf8
4224 Adds the UTF-8 representation of the Native code point C<uv> to the end
4225 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
4226 bytes available. The return value is the pointer to the byte after the
4227 end of the new character. In other words,
4229 d = uvchr_to_utf8(d, uv);
4231 is the recommended wide native character-aware way of saying
4238 /* On ASCII machines this is normally a macro but we want a
4239 real function in case XS code wants it
4242 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
4244 PERL_ARGS_ASSERT_UVCHR_TO_UTF8;
4246 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), 0);
4250 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4252 PERL_ARGS_ASSERT_UVCHR_TO_UTF8_FLAGS;
4254 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), flags);
4258 =for apidoc utf8n_to_uvchr
4260 Returns the native character value of the first character in the string
4262 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
4263 length, in bytes, of that character.
4265 C<length> and C<flags> are the same as L</utf8n_to_uvuni>().
4269 /* On ASCII machines this is normally a macro but we want
4270 a real function in case XS code wants it
4273 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen,
4276 const UV uv = Perl_utf8n_to_uvuni(aTHX_ s, curlen, retlen, flags);
4278 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
4280 return UNI_TO_NATIVE(uv);
4284 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4286 /* May change: warns if surrogates, non-character code points, or
4287 * non-Unicode code points are in s which has length len bytes. Returns
4288 * TRUE if none found; FALSE otherwise. The only other validity check is
4289 * to make sure that this won't exceed the string's length */
4291 const U8* const e = s + len;
4294 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4297 if (UTF8SKIP(s) > len) {
4298 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4299 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4302 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
4304 if (UTF8_IS_SUPER(s)) {
4305 if (ckWARN_d(WARN_NON_UNICODE)) {
4306 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4307 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
4308 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
4312 else if (UTF8_IS_SURROGATE(s)) {
4313 if (ckWARN_d(WARN_SURROGATE)) {
4314 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4315 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4316 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
4321 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
4322 && (ckWARN_d(WARN_NONCHAR)))
4324 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4325 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
4326 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
4337 =for apidoc pv_uni_display
4339 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4340 length C<len>, the displayable version being at most C<pvlim> bytes long
4341 (if longer, the rest is truncated and "..." will be appended).
4343 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
4344 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
4345 to display the \\[nrfta\\] as the backslashed versions (like '\n')
4346 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
4347 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
4348 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
4350 The pointer to the PV of the C<dsv> is returned.
4354 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4359 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4363 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4365 /* This serves double duty as a flag and a character to print after
4366 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4370 if (pvlim && SvCUR(dsv) >= pvlim) {
4374 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4376 const unsigned char c = (unsigned char)u & 0xFF;
4377 if (flags & UNI_DISPLAY_BACKSLASH) {
4394 const char string = ok;
4395 sv_catpvs(dsv, "\\");
4396 sv_catpvn(dsv, &string, 1);
4399 /* isPRINT() is the locale-blind version. */
4400 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4401 const char string = c;
4402 sv_catpvn(dsv, &string, 1);
4407 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4410 sv_catpvs(dsv, "...");
4416 =for apidoc sv_uni_display
4418 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4419 the displayable version being at most C<pvlim> bytes long
4420 (if longer, the rest is truncated and "..." will be appended).
4422 The C<flags> argument is as in L</pv_uni_display>().
4424 The pointer to the PV of the C<dsv> is returned.
4429 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4431 const char * const ptr =
4432 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4434 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4436 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4437 SvCUR(ssv), pvlim, flags);
4441 =for apidoc foldEQ_utf8
4443 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4444 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4445 How far into the strings to compare is determined by other input parameters.
4447 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4448 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4449 with respect to C<s2>.
4451 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4452 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4453 scan will not be considered to be a match unless the goal is reached, and
4454 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4457 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4458 considered an end pointer to the position 1 byte past the maximum point
4459 in C<s1> beyond which scanning will not continue under any circumstances.
4460 (This routine assumes that UTF-8 encoded input strings are not malformed;
4461 malformed input can cause it to read past C<pe1>).
4462 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4463 is less than C<s1>+C<l1>, the match will never be successful because it can
4465 get as far as its goal (and in fact is asserted against). Correspondingly for
4466 C<pe2> with respect to C<s2>.
4468 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4469 C<l2> must be non-zero), and if both do, both have to be
4470 reached for a successful match. Also, if the fold of a character is multiple
4471 characters, all of them must be matched (see tr21 reference below for
4474 Upon a successful match, if C<pe1> is non-NULL,
4475 it will be set to point to the beginning of the I<next> character of C<s1>
4476 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4478 For case-insensitiveness, the "casefolding" of Unicode is used
4479 instead of upper/lowercasing both the characters, see
4480 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4484 /* A flags parameter has been added which may change, and hence isn't
4485 * externally documented. Currently it is:
4486 * 0 for as-documented above
4487 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4488 ASCII one, to not match
4489 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
4490 * points below 256; unicode rules for above 255; and
4491 * folds that cross those boundaries are disallowed,
4492 * like the NOMIX_ASCII option
4493 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4494 * routine. This allows that step to be skipped.
4495 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4498 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)
4501 const U8 *p1 = (const U8*)s1; /* Point to current char */
4502 const U8 *p2 = (const U8*)s2;
4503 const U8 *g1 = NULL; /* goal for s1 */
4504 const U8 *g2 = NULL;
4505 const U8 *e1 = NULL; /* Don't scan s1 past this */
4506 U8 *f1 = NULL; /* Point to current folded */
4507 const U8 *e2 = NULL;
4509 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4510 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4511 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4513 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4515 /* The algorithm requires that input with the flags on the first line of
4516 * the assert not be pre-folded. */
4517 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
4518 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4525 g1 = (const U8*)s1 + l1;
4533 g2 = (const U8*)s2 + l2;
4536 /* Must have at least one goal */
4541 /* Will never match if goal is out-of-bounds */
4542 assert(! e1 || e1 >= g1);
4544 /* Here, there isn't an end pointer, or it is beyond the goal. We
4545 * only go as far as the goal */
4549 assert(e1); /* Must have an end for looking at s1 */
4552 /* Same for goal for s2 */
4554 assert(! e2 || e2 >= g2);
4561 /* If both operands are already folded, we could just do a memEQ on the
4562 * whole strings at once, but it would be better if the caller realized
4563 * this and didn't even call us */
4565 /* Look through both strings, a character at a time */
4566 while (p1 < e1 && p2 < e2) {
4568 /* If at the beginning of a new character in s1, get its fold to use
4569 * and the length of the fold. (exception: locale rules just get the
4570 * character to a single byte) */
4572 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4577 /* If in locale matching, we use two sets of rules, depending
4578 * on if the code point is above or below 255. Here, we test
4579 * for and handle locale rules */
4580 if ((flags & FOLDEQ_UTF8_LOCALE)
4581 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
4583 /* There is no mixing of code points above and below 255. */
4584 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
4588 /* We handle locale rules by converting, if necessary, the
4589 * code point to a single byte. */
4590 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4594 *foldbuf1 = TWO_BYTE_UTF8_TO_UNI(*p1, *(p1 + 1));
4598 else if (isASCII(*p1)) { /* Note, that here won't be both
4599 ASCII and using locale rules */
4601 /* If trying to mix non- with ASCII, and not supposed to,
4603 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4607 *foldbuf1 = toFOLD(*p1);
4610 to_utf8_fold(p1, foldbuf1, &n1);
4612 else { /* Not utf8, get utf8 fold */
4613 to_uni_fold(NATIVE_TO_UNI(*p1), foldbuf1, &n1);
4619 if (n2 == 0) { /* Same for s2 */
4620 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4625 if ((flags & FOLDEQ_UTF8_LOCALE)
4626 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4628 /* Here, the next char in s2 is < 256. We've already
4629 * worked on s1, and if it isn't also < 256, can't match */
4630 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4633 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4637 *foldbuf2 = TWO_BYTE_UTF8_TO_UNI(*p2, *(p2 + 1));
4640 /* Use another function to handle locale rules. We've made
4641 * sure that both characters to compare are single bytes */
4642 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4647 else if (isASCII(*p2)) {
4648 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4652 *foldbuf2 = toFOLD(*p2);
4655 to_utf8_fold(p2, foldbuf2, &n2);
4658 to_uni_fold(NATIVE_TO_UNI(*p2), foldbuf2, &n2);
4664 /* Here f1 and f2 point to the beginning of the strings to compare.
4665 * These strings are the folds of the next character from each input
4666 * string, stored in utf8. */
4668 /* While there is more to look for in both folds, see if they
4669 * continue to match */
4671 U8 fold_length = UTF8SKIP(f1);
4672 if (fold_length != UTF8SKIP(f2)
4673 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4674 function call for single
4676 || memNE((char*)f1, (char*)f2, fold_length))
4678 return 0; /* mismatch */
4681 /* Here, they matched, advance past them */
4688 /* When reach the end of any fold, advance the input past it */
4690 p1 += u1 ? UTF8SKIP(p1) : 1;
4693 p2 += u2 ? UTF8SKIP(p2) : 1;
4695 } /* End of loop through both strings */
4697 /* A match is defined by each scan that specified an explicit length
4698 * reaching its final goal, and the other not having matched a partial
4699 * character (which can happen when the fold of a character is more than one
4701 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4705 /* Successful match. Set output pointers */
4717 * c-indentation-style: bsd
4719 * indent-tabs-mode: nil
4722 * ex: set ts=8 sts=4 sw=4 et: