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"
36 static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
40 =head1 Unicode Support
42 This file contains various utility functions for manipulating UTF8-encoded
43 strings. For the uninitiated, this is a method of representing arbitrary
44 Unicode characters as a variable number of bytes, in such a way that
45 characters in the ASCII range are unmodified, and a zero byte never appears
46 within non-zero characters.
52 =for apidoc is_ascii_string
54 Returns true if the first C<len> bytes of the string C<s> are the same whether
55 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
56 is, if they are invariant. On ASCII-ish machines, only ASCII characters
57 fit this definition, hence the function's name.
59 If C<len> is 0, it will be calculated using C<strlen(s)>.
61 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
67 Perl_is_ascii_string(const U8 *s, STRLEN len)
69 const U8* const send = s + (len ? len : strlen((const char *)s));
72 PERL_ARGS_ASSERT_IS_ASCII_STRING;
74 for (; x < send; ++x) {
75 if (!UTF8_IS_INVARIANT(*x))
83 =for apidoc uvoffuni_to_utf8_flags
85 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
86 Instead, B<Almost all code should use L</uvchr_to_utf8> or
87 L</uvchr_to_utf8_flags>>.
89 This function is like them, but the input is a strict Unicode
90 (as opposed to native) code point. Only in very rare circumstances should code
91 not be using the native code point.
93 For details, see the description for L</uvchr_to_utf8_flags>>.
99 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
101 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
103 if (UNI_IS_INVARIANT(uv)) {
104 *d++ = (U8) LATIN1_TO_NATIVE(uv);
108 /* The first problematic code point is the first surrogate */
109 if (uv >= UNICODE_SURROGATE_FIRST
110 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
112 if (UNICODE_IS_SURROGATE(uv)) {
113 if (flags & UNICODE_WARN_SURROGATE) {
114 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
115 "UTF-16 surrogate U+%04"UVXf, uv);
117 if (flags & UNICODE_DISALLOW_SURROGATE) {
121 else if (UNICODE_IS_SUPER(uv)) {
122 if (flags & UNICODE_WARN_SUPER
123 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
125 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
126 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
128 if (flags & UNICODE_DISALLOW_SUPER
129 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
134 else if (UNICODE_IS_NONCHAR(uv)) {
135 if (flags & UNICODE_WARN_NONCHAR) {
136 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
137 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
140 if (flags & UNICODE_DISALLOW_NONCHAR) {
148 STRLEN len = OFFUNISKIP(uv);
151 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
152 uv >>= UTF_ACCUMULATION_SHIFT;
154 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
157 #else /* Non loop style */
159 *d++ = (U8)(( uv >> 6) | 0xc0);
160 *d++ = (U8)(( uv & 0x3f) | 0x80);
164 *d++ = (U8)(( uv >> 12) | 0xe0);
165 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
166 *d++ = (U8)(( uv & 0x3f) | 0x80);
170 *d++ = (U8)(( uv >> 18) | 0xf0);
171 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
172 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
173 *d++ = (U8)(( uv & 0x3f) | 0x80);
176 if (uv < 0x4000000) {
177 *d++ = (U8)(( uv >> 24) | 0xf8);
178 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
179 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
180 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
181 *d++ = (U8)(( uv & 0x3f) | 0x80);
184 if (uv < 0x80000000) {
185 *d++ = (U8)(( uv >> 30) | 0xfc);
186 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
187 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
188 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
189 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
190 *d++ = (U8)(( uv & 0x3f) | 0x80);
194 if (uv < UTF8_QUAD_MAX)
197 *d++ = 0xfe; /* Can't match U+FEFF! */
198 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
199 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
208 *d++ = 0xff; /* Can't match U+FFFE! */
209 *d++ = 0x80; /* 6 Reserved bits */
210 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
211 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
212 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
220 *d++ = (U8)(( uv & 0x3f) | 0x80);
224 #endif /* Non loop style */
227 =for apidoc uvchr_to_utf8
229 Adds the UTF-8 representation of the native code point C<uv> to the end
230 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
231 bytes available. The return value is the pointer to the byte after the
232 end of the new character. In other words,
234 d = uvchr_to_utf8(d, uv);
236 is the recommended wide native character-aware way of saying
240 This function accepts any UV as input. To forbid or warn on non-Unicode code
241 points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
246 /* This is also a macro */
247 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
250 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
252 return uvchr_to_utf8(d, uv);
256 =for apidoc uvchr_to_utf8_flags
258 Adds the UTF-8 representation of the native code point C<uv> to the end
259 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
260 bytes available. The return value is the pointer to the byte after the
261 end of the new character. In other words,
263 d = uvchr_to_utf8_flags(d, uv, flags);
267 d = uvchr_to_utf8_flags(d, uv, 0);
269 This is the Unicode-aware way of saying
273 This function will convert to UTF-8 (and not warn) even code points that aren't
274 legal Unicode or are problematic, unless C<flags> contains one or more of the
277 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
278 the function will raise a warning, provided UTF8 warnings are enabled. If instead
279 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
280 If both flags are set, the function will both warn and return NULL.
282 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
283 affect how the function handles a Unicode non-character. And likewise, the
284 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, affect the handling of
286 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
287 even less portable) can be warned and/or disallowed even if other above-Unicode
288 code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
291 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
292 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
298 /* This is also a macro */
299 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
302 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
304 return uvchr_to_utf8_flags(d, uv, flags);
309 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
310 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC) character is a
311 valid UTF-8 character. The number of bytes in the UTF-8 character
312 will be returned if it is valid, otherwise 0.
314 This is the "slow" version as opposed to the "fast" version which is
315 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
316 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
317 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
318 you should use the _slow(). In practice this means that the _slow()
319 will be used very rarely, since the maximum Unicode code point (as of
320 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
321 the "Perl extended UTF-8" (e.g, the infamous 'v-strings') will encode into
325 PERL_STATIC_INLINE STRLEN
326 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
328 dTHX; /* The function called below requires thread context */
332 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
334 utf8n_to_uvchr(s, len, &actual_len, UTF8_CHECK_ONLY);
336 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
340 =for apidoc is_utf8_char_buf
342 Returns the number of bytes that comprise the first UTF-8 encoded character in
343 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
344 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
347 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
348 machines) is a valid UTF-8 character.
353 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
358 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
360 if (buf_end <= buf) {
365 if (len > UTF8SKIP(buf)) {
369 if (IS_UTF8_CHAR_FAST(len))
370 return IS_UTF8_CHAR(buf, len) ? len : 0;
371 return is_utf8_char_slow(buf, len);
375 =for apidoc is_utf8_char
377 Tests if some arbitrary number of bytes begins in a valid UTF-8
378 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
379 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
380 character will be returned if it is valid, otherwise 0.
382 This function is deprecated due to the possibility that malformed input could
383 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
389 Perl_is_utf8_char(const U8 *s)
391 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
393 /* Assumes we have enough space, which is why this is deprecated */
394 return is_utf8_char_buf(s, s + UTF8SKIP(s));
399 =for apidoc is_utf8_string
401 Returns true if the first C<len> bytes of string C<s> form a valid
402 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
403 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
404 terminating NUL byte). Note that all characters being ASCII constitute 'a
407 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
413 Perl_is_utf8_string(const U8 *s, STRLEN len)
415 const U8* const send = s + (len ? len : strlen((const char *)s));
418 PERL_ARGS_ASSERT_IS_UTF8_STRING;
421 /* Inline the easy bits of is_utf8_char() here for speed... */
422 if (UTF8_IS_INVARIANT(*x)) {
426 /* ... and call is_utf8_char() only if really needed. */
427 const STRLEN c = UTF8SKIP(x);
428 const U8* const next_char_ptr = x + c;
430 if (next_char_ptr > send) {
434 if (IS_UTF8_CHAR_FAST(c)) {
435 if (!IS_UTF8_CHAR(x, c))
438 else if (! is_utf8_char_slow(x, c)) {
449 Implemented as a macro in utf8.h
451 =for apidoc is_utf8_string_loc
453 Like L</is_utf8_string> but stores the location of the failure (in the
454 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
455 "utf8ness success") in the C<ep>.
457 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
459 =for apidoc is_utf8_string_loclen
461 Like L</is_utf8_string>() but stores the location of the failure (in the
462 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
463 "utf8ness success") in the C<ep>, and the number of UTF-8
464 encoded characters in the C<el>.
466 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
472 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
474 const U8* const send = s + (len ? len : strlen((const char *)s));
479 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
482 const U8* next_char_ptr;
484 /* Inline the easy bits of is_utf8_char() here for speed... */
485 if (UTF8_IS_INVARIANT(*x))
486 next_char_ptr = x + 1;
488 /* ... and call is_utf8_char() only if really needed. */
490 next_char_ptr = c + x;
491 if (next_char_ptr > send) {
494 if (IS_UTF8_CHAR_FAST(c)) {
495 if (!IS_UTF8_CHAR(x, c))
498 c = is_utf8_char_slow(x, c);
517 =for apidoc utf8n_to_uvchr
519 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
520 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
522 Bottom level UTF-8 decode routine.
523 Returns the native code point value of the first character in the string C<s>,
524 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
525 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
526 the length, in bytes, of that character.
528 The value of C<flags> determines the behavior when C<s> does not point to a
529 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
530 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
531 next possible position in C<s> that could begin a non-malformed character.
532 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
534 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
535 individual types of malformations, such as the sequence being overlong (that
536 is, when there is a shorter sequence that can express the same code point;
537 overlong sequences are expressly forbidden in the UTF-8 standard due to
538 potential security issues). Another malformation example is the first byte of
539 a character not being a legal first byte. See F<utf8.h> for the list of such
540 flags. For allowed 0 length strings, this function returns 0; for allowed
541 overlong sequences, the computed code point is returned; for all other allowed
542 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
543 determinable reasonable value.
545 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
546 flags) malformation is found. If this flag is set, the routine assumes that
547 the caller will raise a warning, and this function will silently just set
548 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
550 Note that this API requires disambiguation between successful decoding a NUL
551 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
552 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
553 first byte of C<s> is 0 as well. If so, the input was a NUL; if not, the input
556 Certain code points are considered problematic. These are Unicode surrogates,
557 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
558 By default these are considered regular code points, but certain situations
559 warrant special handling for them. If C<flags> contains
560 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
561 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
562 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
563 maximum) can be set to disallow these categories individually.
565 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
566 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
567 for their respective categories, but otherwise the code points are considered
568 valid (not malformations). To get a category to both be treated as a
569 malformation and raise a warning, specify both the WARN and DISALLOW flags.
570 (But note that warnings are not raised if lexically disabled nor if
571 UTF8_CHECK_ONLY is also specified.)
573 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
574 the others that are above the Unicode legal maximum. There are several
575 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
576 not representable at all on EBCDIC platforms, and the original UTF-8
577 specification never went above this number (the current 0x10FFFF limit was
578 imposed later). (The smaller ones, those that fit into 32 bits, are
579 representable by a UV on ASCII platforms, but not by an IV, which means that
580 the number of operations that can be performed on them is quite restricted.)
581 The UTF-8 encoding on ASCII platforms for these large code points begins with a
582 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
583 be treated as malformations, while allowing smaller above-Unicode code points.
584 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
585 including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
586 the other WARN flags, but applies just to these code points.
588 All other code points corresponding to Unicode characters, including private
589 use and those yet to be assigned, are never considered malformed and never
596 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
599 const U8 * const s0 = s;
600 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
605 UV outlier_ret = 0; /* return value when input is in error or problematic
607 UV pack_warn = 0; /* Save result of packWARN() for later */
608 bool unexpected_non_continuation = FALSE;
609 bool overflowed = FALSE;
610 bool do_overlong_test = TRUE; /* May have to skip this test */
612 const char* const malformed_text = "Malformed UTF-8 character";
614 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
616 /* The order of malformation tests here is important. We should consume as
617 * few bytes as possible in order to not skip any valid character. This is
618 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
619 * http://unicode.org/reports/tr36 for more discussion as to why. For
620 * example, once we've done a UTF8SKIP, we can tell the expected number of
621 * bytes, and could fail right off the bat if the input parameters indicate
622 * that there are too few available. But it could be that just that first
623 * byte is garbled, and the intended character occupies fewer bytes. If we
624 * blindly assumed that the first byte is correct, and skipped based on
625 * that number, we could skip over a valid input character. So instead, we
626 * always examine the sequence byte-by-byte.
628 * We also should not consume too few bytes, otherwise someone could inject
629 * things. For example, an input could be deliberately designed to
630 * overflow, and if this code bailed out immediately upon discovering that,
631 * returning to the caller C<*retlen> pointing to the very next byte (one
632 * which is actually part of of the overflowing sequence), that could look
633 * legitimate to the caller, which could discard the initial partial
634 * sequence and process the rest, inappropriately */
636 /* Zero length strings, if allowed, of necessity are zero */
637 if (UNLIKELY(curlen == 0)) {
642 if (flags & UTF8_ALLOW_EMPTY) {
645 if (! (flags & UTF8_CHECK_ONLY)) {
646 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
651 expectlen = UTF8SKIP(s);
653 /* A well-formed UTF-8 character, as the vast majority of calls to this
654 * function will be for, has this expected length. For efficiency, set
655 * things up here to return it. It will be overriden only in those rare
656 * cases where a malformation is found */
661 /* An invariant is trivially well-formed */
662 if (UTF8_IS_INVARIANT(uv)) {
666 /* A continuation character can't start a valid sequence */
667 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
668 if (flags & UTF8_ALLOW_CONTINUATION) {
672 return UNICODE_REPLACEMENT;
675 if (! (flags & UTF8_CHECK_ONLY)) {
676 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
682 /* Here is not a continuation byte, nor an invariant. The only thing left
683 * is a start byte (possibly for an overlong) */
686 uv = NATIVE_UTF8_TO_I8(uv);
689 /* Remove the leading bits that indicate the number of bytes in the
690 * character's whole UTF-8 sequence, leaving just the bits that are part of
692 uv &= UTF_START_MASK(expectlen);
694 /* Now, loop through the remaining bytes in the character's sequence,
695 * accumulating each into the working value as we go. Be sure to not look
696 * past the end of the input string */
697 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
699 for (s = s0 + 1; s < send; s++) {
700 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
701 #ifndef EBCDIC /* Can't overflow in EBCDIC */
702 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
704 /* The original implementors viewed this malformation as more
705 * serious than the others (though I, khw, don't understand
706 * why, since other malformations also give very very wrong
707 * results), so there is no way to turn off checking for it.
708 * Set a flag, but keep going in the loop, so that we absorb
709 * the rest of the bytes that comprise the character. */
711 overflow_byte = *s; /* Save for warning message's use */
714 uv = UTF8_ACCUMULATE(uv, *s);
717 /* Here, found a non-continuation before processing all expected
718 * bytes. This byte begins a new character, so quit, even if
719 * allowing this malformation. */
720 unexpected_non_continuation = TRUE;
723 } /* End of loop through the character's bytes */
725 /* Save how many bytes were actually in the character */
728 /* The loop above finds two types of malformations: non-continuation and/or
729 * overflow. The non-continuation malformation is really a too-short
730 * malformation, as it means that the current character ended before it was
731 * expected to (being terminated prematurely by the beginning of the next
732 * character, whereas in the too-short malformation there just are too few
733 * bytes available to hold the character. In both cases, the check below
734 * that we have found the expected number of bytes would fail if executed.)
735 * Thus the non-continuation malformation is really unnecessary, being a
736 * subset of the too-short malformation. But there may be existing
737 * applications that are expecting the non-continuation type, so we retain
738 * it, and return it in preference to the too-short malformation. (If this
739 * code were being written from scratch, the two types might be collapsed
740 * into one.) I, khw, am also giving priority to returning the
741 * non-continuation and too-short malformations over overflow when multiple
742 * ones are present. I don't know of any real reason to prefer one over
743 * the other, except that it seems to me that multiple-byte errors trumps
744 * errors from a single byte */
745 if (UNLIKELY(unexpected_non_continuation)) {
746 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
747 if (! (flags & UTF8_CHECK_ONLY)) {
749 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
752 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));
757 uv = UNICODE_REPLACEMENT;
759 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
760 * as what the original expectations were. */
761 do_overlong_test = FALSE;
766 else if (UNLIKELY(curlen < expectlen)) {
767 if (! (flags & UTF8_ALLOW_SHORT)) {
768 if (! (flags & UTF8_CHECK_ONLY)) {
769 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));
773 uv = UNICODE_REPLACEMENT;
774 do_overlong_test = FALSE;
780 #ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
781 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
782 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
784 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
785 * generation of the sv, since no warnings are raised under CHECK */
786 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
787 && ckWARN_d(WARN_UTF8))
789 /* This message is deliberately not of the same syntax as the other
790 * messages for malformations, for backwards compatibility in the
791 * unlikely event that code is relying on its precise earlier text
793 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
794 pack_warn = packWARN(WARN_UTF8);
796 if (flags & UTF8_DISALLOW_FE_FF) {
800 if (UNLIKELY(overflowed)) {
802 /* If the first byte is FF, it will overflow a 32-bit word. If the
803 * first byte is FE, it will overflow a signed 32-bit word. The
804 * above preserves backward compatibility, since its message was used
805 * in earlier versions of this code in preference to overflow */
806 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
812 && expectlen > (STRLEN) OFFUNISKIP(uv)
813 && ! (flags & UTF8_ALLOW_LONG))
815 /* The overlong malformation has lower precedence than the others.
816 * Note that if this malformation is allowed, we return the actual
817 * value, instead of the replacement character. This is because this
818 * value is actually well-defined. */
819 if (! (flags & UTF8_CHECK_ONLY)) {
820 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", OFFUNISKIP(uv), *s0));
825 /* Here, the input is considered to be well-formed , but could be a
826 * problematic code point that is not allowed by the input parameters. */
827 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
828 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
829 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
831 if (UNICODE_IS_SURROGATE(uv)) {
832 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
833 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
835 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
836 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
838 if (flags & UTF8_DISALLOW_SURROGATE) {
842 else if ((uv > PERL_UNICODE_MAX)) {
843 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
844 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
846 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
847 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
849 if (flags & UTF8_DISALLOW_SUPER) {
853 else if (UNICODE_IS_NONCHAR(uv)) {
854 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
855 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
857 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
858 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
860 if (flags & UTF8_DISALLOW_NONCHAR) {
866 outlier_ret = uv; /* Note we don't bother to convert to native,
867 as all the outlier code points are the same
868 in both ASCII and EBCDIC */
872 /* Here, this is not considered a malformed character, so drop through
876 return UNI_TO_NATIVE(uv);
878 /* There are three cases which get to beyond this point. In all 3 cases:
879 * <sv> if not null points to a string to print as a warning.
880 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
882 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
883 * This is done by initializing it to 0, and changing it only
886 * 1) The input is valid but problematic, and to be warned about. The
887 * return value is the resultant code point; <*retlen> is set to
888 * <curlen>, the number of bytes that comprise the code point.
889 * <pack_warn> contains the result of packWARN() for the warning
890 * types. The entry point for this case is the label <do_warn>;
891 * 2) The input is a valid code point but disallowed by the parameters to
892 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
893 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
894 * comprise the code point. <pack_warn> contains the result of
895 * packWARN() for the warning types. The entry point for this case is
896 * the label <disallowed>.
897 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
898 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
899 * bytes that comprise the malformation. All such malformations are
900 * assumed to be warning type <utf8>. The entry point for this case
901 * is the label <malformed>.
906 if (sv && ckWARN_d(WARN_UTF8)) {
907 pack_warn = packWARN(WARN_UTF8);
912 if (flags & UTF8_CHECK_ONLY) {
914 *retlen = ((STRLEN) -1);
920 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
921 if warnings are to be raised. */
922 const char * const string = SvPVX_const(sv);
925 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
927 Perl_warner(aTHX_ pack_warn, "%s", string);
938 =for apidoc utf8_to_uvchr_buf
940 Returns the native code point of the first character in the string C<s> which
941 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
942 C<*retlen> will be set to the length, in bytes, of that character.
944 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
945 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
946 NULL) to -1. If those warnings are off, the computed value, if well-defined
947 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
948 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
949 the next possible position in C<s> that could begin a non-malformed character.
950 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
958 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
962 return utf8n_to_uvchr(s, send - s, retlen,
963 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
966 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
967 * there are no malformations in the input UTF-8 string C<s>. surrogates,
968 * non-character code points, and non-Unicode code points are allowed. */
971 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
973 UV expectlen = UTF8SKIP(s);
974 const U8* send = s + expectlen;
977 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
983 /* An invariant is trivially returned */
984 if (expectlen == 1) {
989 uv = NATIVE_UTF8_TO_I8(uv);
992 /* Remove the leading bits that indicate the number of bytes, leaving just
993 * the bits that are part of the value */
994 uv &= UTF_START_MASK(expectlen);
996 /* Now, loop through the remaining bytes, accumulating each into the
997 * working total as we go. (I khw tried unrolling the loop for up to 4
998 * bytes, but there was no performance improvement) */
999 for (++s; s < send; s++) {
1000 uv = UTF8_ACCUMULATE(uv, *s);
1003 return UNI_TO_NATIVE(uv);
1008 =for apidoc utf8_to_uvchr
1010 Returns the native code point of the first character in the string C<s>
1011 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1012 length, in bytes, of that character.
1014 Some, but not all, UTF-8 malformations are detected, and in fact, some
1015 malformed input could cause reading beyond the end of the input buffer, which
1016 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
1018 If C<s> points to one of the detected malformations, and UTF8 warnings are
1019 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1020 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1021 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1022 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1023 next possible position in C<s> that could begin a non-malformed character.
1024 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1030 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
1032 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
1034 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
1038 =for apidoc utf8_to_uvuni_buf
1040 Only in very rare circumstances should code need to be dealing in Unicode
1041 (as opposed to native) code points. In those few cases, use
1042 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1044 Returns the Unicode (not-native) code point of the first character in the
1046 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1047 C<retlen> will be set to the length, in bytes, of that character.
1049 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1050 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1051 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1052 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1053 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1054 next possible position in C<s> that could begin a non-malformed character.
1055 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1061 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1063 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1067 /* Call the low level routine asking for checks */
1068 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1069 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
1073 * Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1074 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1075 * non-character code points, and non-Unicode code points are allowed */
1078 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1080 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1082 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
1086 =for apidoc utf8_to_uvuni
1088 Returns the Unicode code point of the first character in the string C<s>
1089 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1090 length, in bytes, of that character.
1092 Some, but not all, UTF-8 malformations are detected, and in fact, some
1093 malformed input could cause reading beyond the end of the input buffer, which
1094 is one reason why this function is deprecated. The other is that only in
1095 extremely limited circumstances should the Unicode versus native code point be
1096 of any interest to you. See L</utf8_to_uvuni_buf> for alternatives.
1098 If C<s> points to one of the detected malformations, and UTF8 warnings are
1099 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1100 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1101 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1102 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1103 next possible position in C<s> that could begin a non-malformed character.
1104 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1110 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1112 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1114 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
1118 =for apidoc utf8_length
1120 Return the length of the UTF-8 char encoded string C<s> in characters.
1121 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1122 up past C<e>, croaks.
1128 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1133 PERL_ARGS_ASSERT_UTF8_LENGTH;
1135 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1136 * the bitops (especially ~) can create illegal UTF-8.
1137 * In other words: in Perl UTF-8 is not just for Unicode. */
1140 goto warn_and_return;
1150 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1151 "%s in %s", unees, OP_DESC(PL_op));
1153 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1160 =for apidoc utf8_distance
1162 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1165 WARNING: use only if you *know* that the pointers point inside the
1172 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1174 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1176 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1180 =for apidoc utf8_hop
1182 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1183 forward or backward.
1185 WARNING: do not use the following unless you *know* C<off> is within
1186 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1187 on the first byte of character or just after the last byte of a character.
1193 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1195 PERL_ARGS_ASSERT_UTF8_HOP;
1197 PERL_UNUSED_CONTEXT;
1198 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1199 * the bitops (especially ~) can create illegal UTF-8.
1200 * In other words: in Perl UTF-8 is not just for Unicode. */
1209 while (UTF8_IS_CONTINUATION(*s))
1217 =for apidoc bytes_cmp_utf8
1219 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1220 sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
1221 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1222 if the first string is greater than the second string.
1224 -1 or +1 is returned if the shorter string was identical to the start of the
1225 longer string. -2 or +2 is returned if the was a difference between characters
1232 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1234 const U8 *const bend = b + blen;
1235 const U8 *const uend = u + ulen;
1237 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1239 PERL_UNUSED_CONTEXT;
1241 while (b < bend && u < uend) {
1243 if (!UTF8_IS_INVARIANT(c)) {
1244 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1247 if (UTF8_IS_CONTINUATION(c1)) {
1248 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1250 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1251 "Malformed UTF-8 character "
1252 "(unexpected non-continuation byte 0x%02x"
1253 ", immediately after start byte 0x%02x)"
1254 /* Dear diag.t, it's in the pod. */
1256 PL_op ? " in " : "",
1257 PL_op ? OP_DESC(PL_op) : "");
1262 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1263 "%s in %s", unees, OP_DESC(PL_op));
1265 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1266 return -2; /* Really want to return undef :-) */
1273 return *b < c ? -2 : +2;
1278 if (b == bend && u == uend)
1281 return b < bend ? +1 : -1;
1285 =for apidoc utf8_to_bytes
1287 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1288 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1289 updates C<len> to contain the new length.
1290 Returns zero on failure, setting C<len> to -1.
1292 If you need a copy of the string, see L</bytes_from_utf8>.
1298 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1300 U8 * const save = s;
1301 U8 * const send = s + *len;
1304 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1306 /* ensure valid UTF-8 and chars < 256 before updating string */
1308 if (! UTF8_IS_INVARIANT(*s)) {
1309 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1310 *len = ((STRLEN) -1);
1321 if (! UTF8_IS_INVARIANT(c)) {
1322 /* Then it is two-byte encoded */
1323 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1334 =for apidoc bytes_from_utf8
1336 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1337 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1338 the newly-created string, and updates C<len> to contain the new
1339 length. Returns the original string if no conversion occurs, C<len>
1340 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1341 0 if C<s> is converted or consisted entirely of characters that are invariant
1342 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1348 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1351 const U8 *start = s;
1355 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1357 PERL_UNUSED_CONTEXT;
1361 /* ensure valid UTF-8 and chars < 256 before converting string */
1362 for (send = s + *len; s < send;) {
1363 if (! UTF8_IS_INVARIANT(*s)) {
1364 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1375 Newx(d, (*len) - count + 1, U8);
1376 s = start; start = d;
1379 if (! UTF8_IS_INVARIANT(c)) {
1380 /* Then it is two-byte encoded */
1381 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1392 =for apidoc bytes_to_utf8
1394 Converts a string C<s> of length C<len> bytes from the native encoding into
1396 Returns a pointer to the newly-created string, and sets C<len> to
1397 reflect the new length in bytes.
1399 A NUL character will be written after the end of the string.
1401 If you want to convert to UTF-8 from encodings other than
1402 the native (Latin1 or EBCDIC),
1403 see L</sv_recode_to_utf8>().
1408 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1409 likewise need duplication. */
1412 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1414 const U8 * const send = s + (*len);
1418 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1419 PERL_UNUSED_CONTEXT;
1421 Newx(d, (*len) * 2 + 1, U8);
1425 append_utf8_from_native_byte(*s, &d);
1434 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1436 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1437 * We optimize for native, for obvious reasons. */
1440 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1445 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1448 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1453 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1455 if (UNI_IS_INVARIANT(uv)) {
1456 *d++ = LATIN1_TO_NATIVE((U8) uv);
1459 if (uv <= MAX_UTF8_TWO_BYTE) {
1460 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1461 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1464 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1465 #define LAST_HIGH_SURROGATE 0xDBFF
1466 #define FIRST_LOW_SURROGATE 0xDC00
1467 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1468 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1470 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1472 UV low = (p[0] << 8) + p[1];
1474 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1475 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1476 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1477 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1479 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1480 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1483 d = uvoffuni_to_utf8_flags(d, uv, 0);
1486 *d++ = (U8)(( uv >> 12) | 0xe0);
1487 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1488 *d++ = (U8)(( uv & 0x3f) | 0x80);
1492 *d++ = (U8)(( uv >> 18) | 0xf0);
1493 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1494 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1495 *d++ = (U8)(( uv & 0x3f) | 0x80);
1500 *newlen = d - dstart;
1504 /* Note: this one is slightly destructive of the source. */
1507 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1510 U8* const send = s + bytelen;
1512 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1515 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1519 const U8 tmp = s[0];
1524 return utf16_to_utf8(p, d, bytelen, newlen);
1528 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1530 U8 tmpbuf[UTF8_MAXBYTES+1];
1531 uvchr_to_utf8(tmpbuf, c);
1532 return _is_utf8_FOO(classnum, tmpbuf);
1535 /* for now these are all defined (inefficiently) in terms of the utf8 versions.
1536 * Note that the macros in handy.h that call these short-circuit calling them
1537 * for Latin-1 range inputs */
1540 Perl_is_uni_alnum(pTHX_ UV c)
1542 U8 tmpbuf[UTF8_MAXBYTES+1];
1543 uvchr_to_utf8(tmpbuf, c);
1544 return _is_utf8_FOO(_CC_WORDCHAR, tmpbuf);
1548 Perl_is_uni_alnumc(pTHX_ UV c)
1550 U8 tmpbuf[UTF8_MAXBYTES+1];
1551 uvchr_to_utf8(tmpbuf, c);
1552 return _is_utf8_FOO(_CC_ALPHANUMERIC, tmpbuf);
1555 /* Internal function so we can deprecate the external one, and call
1556 this one from other deprecated functions in this file */
1558 PERL_STATIC_INLINE bool
1559 S_is_utf8_idfirst(pTHX_ const U8 *p)
1565 /* is_utf8_idstart would be more logical. */
1566 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1570 Perl_is_uni_idfirst(pTHX_ UV c)
1572 U8 tmpbuf[UTF8_MAXBYTES+1];
1573 uvchr_to_utf8(tmpbuf, c);
1574 return S_is_utf8_idfirst(aTHX_ tmpbuf);
1578 Perl__is_uni_perl_idcont(pTHX_ UV c)
1580 U8 tmpbuf[UTF8_MAXBYTES+1];
1581 uvchr_to_utf8(tmpbuf, c);
1582 return _is_utf8_perl_idcont(tmpbuf);
1586 Perl__is_uni_perl_idstart(pTHX_ UV c)
1588 U8 tmpbuf[UTF8_MAXBYTES+1];
1589 uvchr_to_utf8(tmpbuf, c);
1590 return _is_utf8_perl_idstart(tmpbuf);
1594 Perl_is_uni_alpha(pTHX_ UV c)
1596 U8 tmpbuf[UTF8_MAXBYTES+1];
1597 uvchr_to_utf8(tmpbuf, c);
1598 return _is_utf8_FOO(_CC_ALPHA, tmpbuf);
1602 Perl_is_uni_ascii(pTHX_ UV c)
1608 Perl_is_uni_blank(pTHX_ UV c)
1610 return isBLANK_uni(c);
1614 Perl_is_uni_space(pTHX_ UV c)
1616 return isSPACE_uni(c);
1620 Perl_is_uni_digit(pTHX_ UV c)
1622 U8 tmpbuf[UTF8_MAXBYTES+1];
1623 uvchr_to_utf8(tmpbuf, c);
1624 return _is_utf8_FOO(_CC_DIGIT, tmpbuf);
1628 Perl_is_uni_upper(pTHX_ UV c)
1630 U8 tmpbuf[UTF8_MAXBYTES+1];
1631 uvchr_to_utf8(tmpbuf, c);
1632 return _is_utf8_FOO(_CC_UPPER, tmpbuf);
1636 Perl_is_uni_lower(pTHX_ UV c)
1638 U8 tmpbuf[UTF8_MAXBYTES+1];
1639 uvchr_to_utf8(tmpbuf, c);
1640 return _is_utf8_FOO(_CC_LOWER, tmpbuf);
1644 Perl_is_uni_cntrl(pTHX_ UV c)
1646 return isCNTRL_L1(c);
1650 Perl_is_uni_graph(pTHX_ UV c)
1652 U8 tmpbuf[UTF8_MAXBYTES+1];
1653 uvchr_to_utf8(tmpbuf, c);
1654 return _is_utf8_FOO(_CC_GRAPH, tmpbuf);
1658 Perl_is_uni_print(pTHX_ UV c)
1660 U8 tmpbuf[UTF8_MAXBYTES+1];
1661 uvchr_to_utf8(tmpbuf, c);
1662 return _is_utf8_FOO(_CC_PRINT, tmpbuf);
1666 Perl_is_uni_punct(pTHX_ UV c)
1668 U8 tmpbuf[UTF8_MAXBYTES+1];
1669 uvchr_to_utf8(tmpbuf, c);
1670 return _is_utf8_FOO(_CC_PUNCT, tmpbuf);
1674 Perl_is_uni_xdigit(pTHX_ UV c)
1676 return isXDIGIT_uni(c);
1680 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1682 /* We have the latin1-range values compiled into the core, so just use
1683 * those, converting the result to utf8. The only difference between upper
1684 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1685 * either "SS" or "Ss". Which one to use is passed into the routine in
1686 * 'S_or_s' to avoid a test */
1688 UV converted = toUPPER_LATIN1_MOD(c);
1690 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1692 assert(S_or_s == 'S' || S_or_s == 's');
1694 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1695 characters in this range */
1696 *p = (U8) converted;
1701 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1702 * which it maps to one of them, so as to only have to have one check for
1703 * it in the main case */
1704 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1706 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1707 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1710 converted = GREEK_CAPITAL_LETTER_MU;
1712 case LATIN_SMALL_LETTER_SHARP_S:
1718 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1719 assert(0); /* NOTREACHED */
1723 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1724 *p = UTF8_TWO_BYTE_LO(converted);
1730 /* Call the function to convert a UTF-8 encoded character to the specified case.
1731 * Note that there may be more than one character in the result.
1732 * INP is a pointer to the first byte of the input character
1733 * OUTP will be set to the first byte of the string of changed characters. It
1734 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1735 * LENP will be set to the length in bytes of the string of changed characters
1737 * The functions return the ordinal of the first character in the string of OUTP */
1738 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1739 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1740 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
1742 /* This additionally has the input parameter SPECIALS, which if non-zero will
1743 * cause this to use the SPECIALS hash for folding (meaning get full case
1744 * folding); otherwise, when zero, this implies a simple case fold */
1745 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
1748 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1752 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1753 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1754 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1755 * the changed version may be longer than the original character.
1757 * The ordinal of the first character of the changed version is returned
1758 * (but note, as explained above, that there may be more.) */
1760 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1763 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1766 uvchr_to_utf8(p, c);
1767 return CALL_UPPER_CASE(p, p, lenp);
1771 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1775 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1778 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1781 uvchr_to_utf8(p, c);
1782 return CALL_TITLE_CASE(p, p, lenp);
1786 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1788 /* We have the latin1-range values compiled into the core, so just use
1789 * those, converting the result to utf8. Since the result is always just
1790 * one character, we allow <p> to be NULL */
1792 U8 converted = toLOWER_LATIN1(c);
1795 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1800 *p = UTF8_TWO_BYTE_HI(converted);
1801 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1809 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1813 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1816 return to_lower_latin1((U8) c, p, lenp);
1819 uvchr_to_utf8(p, c);
1820 return CALL_LOWER_CASE(p, p, lenp);
1824 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1826 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1827 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1828 * FOLD_FLAGS_FULL iff full folding is to be used;
1830 * Not to be used for locale folds
1835 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1837 assert (! (flags & FOLD_FLAGS_LOCALE));
1839 if (c == MICRO_SIGN) {
1840 converted = GREEK_SMALL_LETTER_MU;
1842 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1844 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1845 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1846 * under those circumstances. */
1847 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1848 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1849 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1851 return LATIN_SMALL_LETTER_LONG_S;
1860 else { /* In this range the fold of all other characters is their lower
1862 converted = toLOWER_LATIN1(c);
1865 if (UVCHR_IS_INVARIANT(converted)) {
1866 *p = (U8) converted;
1870 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1871 *p = UTF8_TWO_BYTE_LO(converted);
1879 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
1882 /* Not currently externally documented, and subject to change
1883 * <flags> bits meanings:
1884 * FOLD_FLAGS_FULL iff full folding is to be used;
1885 * FOLD_FLAGS_LOCALE iff in locale
1886 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1889 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1892 UV result = _to_fold_latin1((U8) c, p, lenp,
1893 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1894 /* It is illegal for the fold to cross the 255/256 boundary under
1895 * locale; in this case return the original */
1896 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1901 /* If no special needs, just use the macro */
1902 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1903 uvchr_to_utf8(p, c);
1904 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1906 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1907 the special flags. */
1908 U8 utf8_c[UTF8_MAXBYTES + 1];
1909 uvchr_to_utf8(utf8_c, c);
1910 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1915 Perl_is_uni_alnum_lc(pTHX_ UV c)
1918 return isALNUM_LC(c);
1920 return _is_uni_FOO(_CC_WORDCHAR, c);
1924 Perl_is_uni_alnumc_lc(pTHX_ UV c)
1927 return isALPHANUMERIC_LC(c);
1929 return _is_uni_FOO(_CC_ALPHANUMERIC, c);
1933 Perl_is_uni_idfirst_lc(pTHX_ UV c)
1936 return isIDFIRST_LC(c);
1938 return _is_uni_perl_idstart(c);
1942 Perl_is_uni_alpha_lc(pTHX_ UV c)
1945 return isALPHA_LC(c);
1947 return _is_uni_FOO(_CC_ALPHA, c);
1951 Perl_is_uni_ascii_lc(pTHX_ UV c)
1954 return isASCII_LC(c);
1960 Perl_is_uni_blank_lc(pTHX_ UV c)
1963 return isBLANK_LC(c);
1965 return isBLANK_uni(c);
1969 Perl_is_uni_space_lc(pTHX_ UV c)
1972 return isSPACE_LC(c);
1974 return isSPACE_uni(c);
1978 Perl_is_uni_digit_lc(pTHX_ UV c)
1981 return isDIGIT_LC(c);
1983 return _is_uni_FOO(_CC_DIGIT, c);
1987 Perl_is_uni_upper_lc(pTHX_ UV c)
1990 return isUPPER_LC(c);
1992 return _is_uni_FOO(_CC_UPPER, c);
1996 Perl_is_uni_lower_lc(pTHX_ UV c)
1999 return isLOWER_LC(c);
2001 return _is_uni_FOO(_CC_LOWER, c);
2005 Perl_is_uni_cntrl_lc(pTHX_ UV c)
2008 return isCNTRL_LC(c);
2014 Perl_is_uni_graph_lc(pTHX_ UV c)
2017 return isGRAPH_LC(c);
2019 return _is_uni_FOO(_CC_GRAPH, c);
2023 Perl_is_uni_print_lc(pTHX_ UV c)
2026 return isPRINT_LC(c);
2028 return _is_uni_FOO(_CC_PRINT, c);
2032 Perl_is_uni_punct_lc(pTHX_ UV c)
2035 return isPUNCT_LC(c);
2037 return _is_uni_FOO(_CC_PUNCT, c);
2041 Perl_is_uni_xdigit_lc(pTHX_ UV c)
2044 return isXDIGIT_LC(c);
2046 return isXDIGIT_uni(c);
2050 Perl_to_uni_upper_lc(pTHX_ U32 c)
2052 /* XXX returns only the first character -- do not use XXX */
2053 /* XXX no locale support yet */
2055 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2056 return (U32)to_uni_upper(c, tmpbuf, &len);
2060 Perl_to_uni_title_lc(pTHX_ U32 c)
2062 /* XXX returns only the first character XXX -- do not use XXX */
2063 /* XXX no locale support yet */
2065 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2066 return (U32)to_uni_title(c, tmpbuf, &len);
2070 Perl_to_uni_lower_lc(pTHX_ U32 c)
2072 /* XXX returns only the first character -- do not use XXX */
2073 /* XXX no locale support yet */
2075 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2076 return (U32)to_uni_lower(c, tmpbuf, &len);
2079 PERL_STATIC_INLINE bool
2080 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2081 const char *const swashname)
2083 /* returns a boolean giving whether or not the UTF8-encoded character that
2084 * starts at <p> is in the swash indicated by <swashname>. <swash>
2085 * contains a pointer to where the swash indicated by <swashname>
2086 * is to be stored; which this routine will do, so that future calls will
2087 * look at <*swash> and only generate a swash if it is not null
2089 * Note that it is assumed that the buffer length of <p> is enough to
2090 * contain all the bytes that comprise the character. Thus, <*p> should
2091 * have been checked before this call for mal-formedness enough to assure
2096 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2098 /* The API should have included a length for the UTF-8 character in <p>,
2099 * but it doesn't. We therefore assume that p has been validated at least
2100 * as far as there being enough bytes available in it to accommodate the
2101 * character without reading beyond the end, and pass that number on to the
2102 * validating routine */
2103 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
2104 if (ckWARN_d(WARN_UTF8)) {
2105 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
2106 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
2107 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
2108 what the malformation is */
2109 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
2115 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2116 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
2119 return swash_fetch(*swash, p, TRUE) != 0;
2123 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2127 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2129 assert(classnum < _FIRST_NON_SWASH_CC);
2131 return is_utf8_common(p, &PL_utf8_swash_ptrs[classnum], swash_property_names[classnum]);
2135 Perl_is_utf8_alnum(pTHX_ const U8 *p)
2139 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
2141 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
2142 * descendant of isalnum(3), in other words, it doesn't
2143 * contain the '_'. --jhi */
2144 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_WORDCHAR], "IsWord");
2148 Perl_is_utf8_alnumc(pTHX_ const U8 *p)
2152 PERL_ARGS_ASSERT_IS_UTF8_ALNUMC;
2154 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHANUMERIC], "IsAlnum");
2158 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
2162 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
2164 return S_is_utf8_idfirst(aTHX_ p);
2168 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
2172 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
2176 /* is_utf8_idstart would be more logical. */
2177 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2181 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
2185 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
2187 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2191 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2195 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2197 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont");
2202 Perl_is_utf8_idcont(pTHX_ const U8 *p)
2206 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2208 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
2212 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2216 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2218 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2222 Perl_is_utf8_alpha(pTHX_ const U8 *p)
2226 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2228 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHA], "IsAlpha");
2232 Perl_is_utf8_ascii(pTHX_ const U8 *p)
2236 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2238 /* ASCII characters are the same whether in utf8 or not. So the macro
2239 * works on both utf8 and non-utf8 representations. */
2244 Perl_is_utf8_blank(pTHX_ const U8 *p)
2248 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2250 return isBLANK_utf8(p);
2254 Perl_is_utf8_space(pTHX_ const U8 *p)
2258 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2260 return isSPACE_utf8(p);
2264 Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2268 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2270 /* Only true if is an ASCII space-like character, and ASCII is invariant
2271 * under utf8, so can just use the macro */
2272 return isSPACE_A(*p);
2276 Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2280 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2282 /* Only true if is an ASCII word character, and ASCII is invariant
2283 * under utf8, so can just use the macro */
2284 return isWORDCHAR_A(*p);
2288 Perl_is_utf8_digit(pTHX_ const U8 *p)
2292 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2294 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_DIGIT], "IsDigit");
2298 Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2302 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2304 /* Only true if is an ASCII digit character, and ASCII is invariant
2305 * under utf8, so can just use the macro */
2306 return isDIGIT_A(*p);
2310 Perl_is_utf8_upper(pTHX_ const U8 *p)
2314 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2316 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_UPPER], "IsUppercase");
2320 Perl_is_utf8_lower(pTHX_ const U8 *p)
2324 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2326 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_LOWER], "IsLowercase");
2330 Perl_is_utf8_cntrl(pTHX_ const U8 *p)
2334 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2336 return isCNTRL_utf8(p);
2340 Perl_is_utf8_graph(pTHX_ const U8 *p)
2344 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2346 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_GRAPH], "IsGraph");
2350 Perl_is_utf8_print(pTHX_ const U8 *p)
2354 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2356 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PRINT], "IsPrint");
2360 Perl_is_utf8_punct(pTHX_ const U8 *p)
2364 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2366 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PUNCT], "IsPunct");
2370 Perl_is_utf8_xdigit(pTHX_ const U8 *p)
2374 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2376 return is_XDIGIT_utf8(p);
2380 Perl__is_utf8_mark(pTHX_ const U8 *p)
2384 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2386 return is_utf8_common(p, &PL_utf8_mark, "IsM");
2391 Perl_is_utf8_mark(pTHX_ const U8 *p)
2395 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2397 return _is_utf8_mark(p);
2401 =for apidoc to_utf8_case
2403 The C<p> contains the pointer to the UTF-8 string encoding
2404 the character that is being converted. This routine assumes that the character
2405 at C<p> is well-formed.
2407 The C<ustrp> is a pointer to the character buffer to put the
2408 conversion result to. The C<lenp> is a pointer to the length
2411 The C<swashp> is a pointer to the swash to use.
2413 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2414 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
2415 but not always, a multicharacter mapping), is tried first.
2417 The C<special> is a string like "utf8::ToSpecLower", which means the
2418 hash %utf8::ToSpecLower. The access to the hash is through
2419 Perl_to_utf8_case().
2421 The C<normal> is a string like "ToLower" which means the swash
2427 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2428 SV **swashp, const char *normal, const char *special)
2432 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
2434 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2436 /* Note that swash_fetch() doesn't output warnings for these because it
2437 * assumes we will */
2438 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2439 if (uv1 <= UNICODE_SURROGATE_LAST) {
2440 if (ckWARN_d(WARN_SURROGATE)) {
2441 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2442 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2443 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2446 else if (UNICODE_IS_SUPER(uv1)) {
2447 if (ckWARN_d(WARN_NON_UNICODE)) {
2448 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2449 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2450 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2454 /* Note that non-characters are perfectly legal, so no warning should
2458 if (!*swashp) /* load on-demand */
2459 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2462 /* It might be "special" (sometimes, but not always,
2463 * a multicharacter mapping) */
2464 HV * const hv = get_hv(special, 0);
2468 (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE)) &&
2472 s = SvPV_const(*svp, len);
2475 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2477 Copy(s, ustrp, len, U8);
2482 if (!len && *swashp) {
2483 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
2486 /* It was "normal" (a single character mapping). */
2487 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2495 return valid_utf8_to_uvchr(ustrp, 0);
2498 /* Here, there was no mapping defined, which means that the code point maps
2499 * to itself. Return the inputs */
2501 if (p != ustrp) { /* Don't copy onto itself */
2502 Copy(p, ustrp, len, U8);
2513 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2515 /* This is called when changing the case of a utf8-encoded character above
2516 * the Latin1 range, and the operation is in locale. If the result
2517 * contains a character that crosses the 255/256 boundary, disallow the
2518 * change, and return the original code point. See L<perlfunc/lc> for why;
2520 * p points to the original string whose case was changed; assumed
2521 * by this routine to be well-formed
2522 * result the code point of the first character in the changed-case string
2523 * ustrp points to the changed-case string (<result> represents its first char)
2524 * lenp points to the length of <ustrp> */
2526 UV original; /* To store the first code point of <p> */
2528 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2530 assert(UTF8_IS_ABOVE_LATIN1(*p));
2532 /* We know immediately if the first character in the string crosses the
2533 * boundary, so can skip */
2536 /* Look at every character in the result; if any cross the
2537 * boundary, the whole thing is disallowed */
2538 U8* s = ustrp + UTF8SKIP(ustrp);
2539 U8* e = ustrp + *lenp;
2541 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2547 /* Here, no characters crossed, result is ok as-is */
2553 /* Failed, have to return the original */
2554 original = valid_utf8_to_uvchr(p, lenp);
2555 Copy(p, ustrp, *lenp, char);
2560 =for apidoc to_utf8_upper
2562 Instead use L</toUPPER_utf8>.
2566 /* Not currently externally documented, and subject to change:
2567 * <flags> is set iff locale semantics are to be used for code points < 256
2568 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2569 * were used in the calculation; otherwise unchanged. */
2572 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2578 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2580 if (UTF8_IS_INVARIANT(*p)) {
2582 result = toUPPER_LC(*p);
2585 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2588 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2590 result = toUPPER_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
2593 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2597 else { /* utf8, ord above 255 */
2598 result = CALL_UPPER_CASE(p, ustrp, lenp);
2601 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2606 /* Here, used locale rules. Convert back to utf8 */
2607 if (UTF8_IS_INVARIANT(result)) {
2608 *ustrp = (U8) result;
2612 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2613 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2618 *tainted_ptr = TRUE;
2624 =for apidoc to_utf8_title
2626 Instead use L</toTITLE_utf8>.
2630 /* Not currently externally documented, and subject to change:
2631 * <flags> is set iff locale semantics are to be used for code points < 256
2632 * Since titlecase is not defined in POSIX, uppercase is used instead
2634 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2635 * were used in the calculation; otherwise unchanged. */
2638 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2644 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2646 if (UTF8_IS_INVARIANT(*p)) {
2648 result = toUPPER_LC(*p);
2651 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2654 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2656 result = toUPPER_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
2659 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2663 else { /* utf8, ord above 255 */
2664 result = CALL_TITLE_CASE(p, ustrp, lenp);
2667 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2672 /* Here, used locale rules. Convert back to utf8 */
2673 if (UTF8_IS_INVARIANT(result)) {
2674 *ustrp = (U8) result;
2678 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2679 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2684 *tainted_ptr = TRUE;
2690 =for apidoc to_utf8_lower
2692 Instead use L</toLOWER_utf8>.
2696 /* Not currently externally documented, and subject to change:
2697 * <flags> is set iff locale semantics are to be used for code points < 256
2698 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2699 * were used in the calculation; otherwise unchanged. */
2702 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2708 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2710 if (UTF8_IS_INVARIANT(*p)) {
2712 result = toLOWER_LC(*p);
2715 return to_lower_latin1(*p, ustrp, lenp);
2718 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2720 result = toLOWER_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
2723 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2727 else { /* utf8, ord above 255 */
2728 result = CALL_LOWER_CASE(p, ustrp, lenp);
2731 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2737 /* Here, used locale rules. Convert back to utf8 */
2738 if (UTF8_IS_INVARIANT(result)) {
2739 *ustrp = (U8) result;
2743 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2744 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2749 *tainted_ptr = TRUE;
2755 =for apidoc to_utf8_fold
2757 Instead use L</toFOLD_utf8>.
2761 /* Not currently externally documented, and subject to change,
2763 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2764 * points < 256. Since foldcase is not defined in
2765 * POSIX, lowercase is used instead
2766 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2767 * otherwise simple folds
2768 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2770 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2771 * were used in the calculation; otherwise unchanged. */
2774 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2780 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2782 /* These are mutually exclusive */
2783 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2785 assert(p != ustrp); /* Otherwise overwrites */
2787 if (UTF8_IS_INVARIANT(*p)) {
2788 if (flags & FOLD_FLAGS_LOCALE) {
2789 result = toFOLD_LC(*p);
2792 return _to_fold_latin1(*p, ustrp, lenp,
2793 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2796 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2797 if (flags & FOLD_FLAGS_LOCALE) {
2798 result = toFOLD_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
2801 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2803 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2806 else { /* utf8, ord above 255 */
2807 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2809 if (flags & FOLD_FLAGS_LOCALE) {
2811 /* Special case these characters, as what normally gets returned
2812 * under locale doesn't work */
2813 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2814 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2815 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2819 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2820 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2821 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2823 goto return_ligature_st;
2825 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2827 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2831 /* This is called when changing the case of a utf8-encoded
2832 * character above the ASCII range, and the result should not
2833 * contain an ASCII character. */
2835 UV original; /* To store the first code point of <p> */
2837 /* Look at every character in the result; if any cross the
2838 * boundary, the whole thing is disallowed */
2840 U8* e = ustrp + *lenp;
2843 /* Crossed, have to return the original */
2844 original = valid_utf8_to_uvchr(p, lenp);
2846 /* But in these instances, there is an alternative we can
2847 * return that is valid */
2848 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2849 || original == LATIN_SMALL_LETTER_SHARP_S)
2853 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2854 goto return_ligature_st;
2856 Copy(p, ustrp, *lenp, char);
2862 /* Here, no characters crossed, result is ok as-is */
2867 /* Here, used locale rules. Convert back to utf8 */
2868 if (UTF8_IS_INVARIANT(result)) {
2869 *ustrp = (U8) result;
2873 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2874 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2879 *tainted_ptr = TRUE;
2884 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2885 * folds to a string of two of these characters. By returning this
2886 * instead, then, e.g.,
2887 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2890 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2891 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2893 return LATIN_SMALL_LETTER_LONG_S;
2896 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2897 * have the other one fold to it */
2899 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2900 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2901 return LATIN_SMALL_LIGATURE_ST;
2905 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2906 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2907 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2911 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2913 PERL_ARGS_ASSERT_SWASH_INIT;
2915 /* Returns a copy of a swash initiated by the called function. This is the
2916 * public interface, and returning a copy prevents others from doing
2917 * mischief on the original */
2919 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2923 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2925 /* Initialize and return a swash, creating it if necessary. It does this
2926 * by calling utf8_heavy.pl in the general case. The returned value may be
2927 * the swash's inversion list instead if the input parameters allow it.
2928 * Which is returned should be immaterial to callers, as the only
2929 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2930 * and swash_to_invlist() handle both these transparently.
2932 * This interface should only be used by functions that won't destroy or
2933 * adversely change the swash, as doing so affects all other uses of the
2934 * swash in the program; the general public should use 'Perl_swash_init'
2937 * pkg is the name of the package that <name> should be in.
2938 * name is the name of the swash to find. Typically it is a Unicode
2939 * property name, including user-defined ones
2940 * listsv is a string to initialize the swash with. It must be of the form
2941 * documented as the subroutine return value in
2942 * L<perlunicode/User-Defined Character Properties>
2943 * minbits is the number of bits required to represent each data element.
2944 * It is '1' for binary properties.
2945 * none I (khw) do not understand this one, but it is used only in tr///.
2946 * invlist is an inversion list to initialize the swash with (or NULL)
2947 * flags_p if non-NULL is the address of various input and output flag bits
2948 * to the routine, as follows: ('I' means is input to the routine;
2949 * 'O' means output from the routine. Only flags marked O are
2950 * meaningful on return.)
2951 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2952 * came from a user-defined property. (I O)
2953 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2954 * when the swash cannot be located, to simply return NULL. (I)
2955 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2956 * return of an inversion list instead of a swash hash if this routine
2957 * thinks that would result in faster execution of swash_fetch() later
2960 * Thus there are three possible inputs to find the swash: <name>,
2961 * <listsv>, and <invlist>. At least one must be specified. The result
2962 * will be the union of the specified ones, although <listsv>'s various
2963 * actions can intersect, etc. what <name> gives.
2965 * <invlist> is only valid for binary properties */
2968 SV* retval = &PL_sv_undef;
2969 HV* swash_hv = NULL;
2970 const int invlist_swash_boundary =
2971 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2972 ? 512 /* Based on some benchmarking, but not extensive, see commit
2974 : -1; /* Never return just an inversion list */
2976 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2977 assert(! invlist || minbits == 1);
2979 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2981 if (listsv != &PL_sv_undef || strNE(name, "")) {
2983 const size_t pkg_len = strlen(pkg);
2984 const size_t name_len = strlen(name);
2985 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2989 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2991 PUSHSTACKi(PERLSI_MAGIC);
2995 /* We might get here via a subroutine signature which uses a utf8
2996 * parameter name, at which point PL_subname will have been set
2997 * but not yet used. */
2998 save_item(PL_subname);
2999 if (PL_parser && PL_parser->error_count)
3000 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3001 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3002 if (!method) { /* demand load utf8 */
3004 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3005 GvSV(PL_errgv) = NULL;
3006 /* It is assumed that callers of this routine are not passing in
3007 * any user derived data. */
3008 /* Need to do this after save_re_context() as it will set
3009 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3010 * in Perl_magic_get). Even line to create errsv_save can turn on
3012 #ifndef NO_TAINT_SUPPORT
3013 SAVEBOOL(TAINT_get);
3016 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3019 /* Not ERRSV, as there is no need to vivify a scalar we are
3020 about to discard. */
3021 SV * const errsv = GvSV(PL_errgv);
3022 if (!SvTRUE(errsv)) {
3023 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3024 SvREFCNT_dec(errsv);
3032 mPUSHp(pkg, pkg_len);
3033 mPUSHp(name, name_len);
3038 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3039 GvSV(PL_errgv) = NULL;
3040 /* If we already have a pointer to the method, no need to use
3041 * call_method() to repeat the lookup. */
3043 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3044 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3046 retval = *PL_stack_sp--;
3047 SvREFCNT_inc(retval);
3050 /* Not ERRSV. See above. */
3051 SV * const errsv = GvSV(PL_errgv);
3052 if (!SvTRUE(errsv)) {
3053 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3054 SvREFCNT_dec(errsv);
3059 if (IN_PERL_COMPILETIME) {
3060 CopHINTS_set(PL_curcop, PL_hints);
3062 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3065 /* If caller wants to handle missing properties, let them */
3066 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3070 "Can't find Unicode property definition \"%"SVf"\"",
3072 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3074 } /* End of calling the module to find the swash */
3076 /* If this operation fetched a swash, and we will need it later, get it */
3077 if (retval != &PL_sv_undef
3078 && (minbits == 1 || (flags_p
3080 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3082 swash_hv = MUTABLE_HV(SvRV(retval));
3084 /* If we don't already know that there is a user-defined component to
3085 * this swash, and the user has indicated they wish to know if there is
3086 * one (by passing <flags_p>), find out */
3087 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3088 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3089 if (user_defined && SvUV(*user_defined)) {
3090 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3095 /* Make sure there is an inversion list for binary properties */
3097 SV** swash_invlistsvp = NULL;
3098 SV* swash_invlist = NULL;
3099 bool invlist_in_swash_is_valid = FALSE;
3100 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3101 an unclaimed reference count */
3103 /* If this operation fetched a swash, get its already existing
3104 * inversion list, or create one for it */
3107 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3108 if (swash_invlistsvp) {
3109 swash_invlist = *swash_invlistsvp;
3110 invlist_in_swash_is_valid = TRUE;
3113 swash_invlist = _swash_to_invlist(retval);
3114 swash_invlist_unclaimed = TRUE;
3118 /* If an inversion list was passed in, have to include it */
3121 /* Any fetched swash will by now have an inversion list in it;
3122 * otherwise <swash_invlist> will be NULL, indicating that we
3123 * didn't fetch a swash */
3124 if (swash_invlist) {
3126 /* Add the passed-in inversion list, which invalidates the one
3127 * already stored in the swash */
3128 invlist_in_swash_is_valid = FALSE;
3129 _invlist_union(invlist, swash_invlist, &swash_invlist);
3133 /* Here, there is no swash already. Set up a minimal one, if
3134 * we are going to return a swash */
3135 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3137 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3139 swash_invlist = invlist;
3143 /* Here, we have computed the union of all the passed-in data. It may
3144 * be that there was an inversion list in the swash which didn't get
3145 * touched; otherwise save the one computed one */
3146 if (! invlist_in_swash_is_valid
3147 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3149 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3151 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3153 /* We just stole a reference count. */
3154 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3155 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3158 /* Use the inversion list stand-alone if small enough */
3159 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3160 SvREFCNT_dec(retval);
3161 if (!swash_invlist_unclaimed)
3162 SvREFCNT_inc_simple_void_NN(swash_invlist);
3163 retval = newRV_noinc(swash_invlist);
3171 /* This API is wrong for special case conversions since we may need to
3172 * return several Unicode characters for a single Unicode character
3173 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3174 * the lower-level routine, and it is similarly broken for returning
3175 * multiple values. --jhi
3176 * For those, you should use to_utf8_case() instead */
3177 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3180 * Returns the value of property/mapping C<swash> for the first character
3181 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3182 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
3183 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3185 * A "swash" is a hash which contains initially the keys/values set up by
3186 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3187 * property for all possible code points. Things are stored in a compact form
3188 * (see utf8_heavy.pl) so that calculation is required to find the actual
3189 * property value for a given code point. As code points are looked up, new
3190 * key/value pairs are added to the hash, so that the calculation doesn't have
3191 * to ever be re-done. Further, each calculation is done, not just for the
3192 * desired one, but for a whole block of code points adjacent to that one.
3193 * For binary properties on ASCII machines, the block is usually for 64 code
3194 * points, starting with a code point evenly divisible by 64. Thus if the
3195 * property value for code point 257 is requested, the code goes out and
3196 * calculates the property values for all 64 code points between 256 and 319,
3197 * and stores these as a single 64-bit long bit vector, called a "swatch",
3198 * under the key for code point 256. The key is the UTF-8 encoding for code
3199 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3200 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3201 * for code point 258 is then requested, this code realizes that it would be
3202 * stored under the key for 256, and would find that value and extract the
3203 * relevant bit, offset from 256.
3205 * Non-binary properties are stored in as many bits as necessary to represent
3206 * their values (32 currently, though the code is more general than that), not
3207 * as single bits, but the principal is the same: the value for each key is a
3208 * vector that encompasses the property values for all code points whose UTF-8
3209 * representations are represented by the key. That is, for all code points
3210 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3214 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3217 HV *const hv = MUTABLE_HV(SvRV(swash));
3222 const U8 *tmps = NULL;
3227 PERL_ARGS_ASSERT_SWASH_FETCH;
3229 /* If it really isn't a hash, it isn't really swash; must be an inversion
3231 if (SvTYPE(hv) != SVt_PVHV) {
3232 return _invlist_contains_cp((SV*)hv,
3234 ? valid_utf8_to_uvchr(ptr, NULL)
3238 /* We store the values in a "swatch" which is a vec() value in a swash
3239 * hash. Code points 0-255 are a single vec() stored with key length
3240 * (klen) 0. All other code points have a UTF-8 representation
3241 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3242 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3243 * length for them is the length of the encoded char - 1. ptr[klen] is the
3244 * final byte in the sequence representing the character */
3245 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3250 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3253 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
3256 klen = UTF8SKIP(ptr) - 1;
3258 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3259 * the vec is the final byte in the sequence. (In EBCDIC this is
3260 * converted to I8 to get consecutive values.) To help you visualize
3262 * Straight 1047 After final byte
3263 * UTF-8 UTF-EBCDIC I8 transform
3264 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3265 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3267 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3268 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3270 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3271 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3273 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3274 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3276 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3277 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3279 * (There are no discontinuities in the elided (...) entries.)
3280 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3281 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3282 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3283 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3284 * index into the vec() swatch (after subtracting 0x80, which we
3285 * actually do with an '&').
3286 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3287 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3288 * dicontinuities which go away by transforming it into I8, and we
3289 * effectively subtract 0xA0 to get the index. */
3290 needents = (1 << UTF_ACCUMULATION_SHIFT);
3291 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
3295 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3296 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3297 * it's nothing to sniff at.) Pity we usually come through at least
3298 * two function calls to get here...
3300 * NB: this code assumes that swatches are never modified, once generated!
3303 if (hv == PL_last_swash_hv &&
3304 klen == PL_last_swash_klen &&
3305 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3307 tmps = PL_last_swash_tmps;
3308 slen = PL_last_swash_slen;
3311 /* Try our second-level swatch cache, kept in a hash. */
3312 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3314 /* If not cached, generate it via swatch_get */
3315 if (!svp || !SvPOK(*svp)
3316 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3319 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3320 swatch = swatch_get(swash,
3321 code_point & ~((UV)needents - 1),
3324 else { /* For the first 256 code points, the swatch has a key of
3326 swatch = swatch_get(swash, 0, needents);
3329 if (IN_PERL_COMPILETIME)
3330 CopHINTS_set(PL_curcop, PL_hints);
3332 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3334 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3335 || (slen << 3) < needents)
3336 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3337 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3338 svp, tmps, (UV)slen, (UV)needents);
3341 PL_last_swash_hv = hv;
3342 assert(klen <= sizeof(PL_last_swash_key));
3343 PL_last_swash_klen = (U8)klen;
3344 /* FIXME change interpvar.h? */
3345 PL_last_swash_tmps = (U8 *) tmps;
3346 PL_last_swash_slen = slen;
3348 Copy(ptr, PL_last_swash_key, klen, U8);
3351 switch ((int)((slen << 3) / needents)) {
3353 bit = 1 << (off & 7);
3355 return (tmps[off] & bit) != 0;
3360 return (tmps[off] << 8) + tmps[off + 1] ;
3363 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3365 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3366 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
3367 NORETURN_FUNCTION_END;
3370 /* Read a single line of the main body of the swash input text. These are of
3373 * where each number is hex. The first two numbers form the minimum and
3374 * maximum of a range, and the third is the value associated with the range.
3375 * Not all swashes should have a third number
3377 * On input: l points to the beginning of the line to be examined; it points
3378 * to somewhere in the string of the whole input text, and is
3379 * terminated by a \n or the null string terminator.
3380 * lend points to the null terminator of that string
3381 * wants_value is non-zero if the swash expects a third number
3382 * typestr is the name of the swash's mapping, like 'ToLower'
3383 * On output: *min, *max, and *val are set to the values read from the line.
3384 * returns a pointer just beyond the line examined. If there was no
3385 * valid min number on the line, returns lend+1
3389 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3390 const bool wants_value, const U8* const typestr)
3392 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3393 STRLEN numlen; /* Length of the number */
3394 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3395 | PERL_SCAN_DISALLOW_PREFIX
3396 | PERL_SCAN_SILENT_NON_PORTABLE;
3398 /* nl points to the next \n in the scan */
3399 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3401 /* Get the first number on the line: the range minimum */
3403 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3404 if (numlen) /* If found a hex number, position past it */
3406 else if (nl) { /* Else, go handle next line, if any */
3407 return nl + 1; /* 1 is length of "\n" */
3409 else { /* Else, no next line */
3410 return lend + 1; /* to LIST's end at which \n is not found */
3413 /* The max range value follows, separated by a BLANK */
3416 flags = PERL_SCAN_SILENT_ILLDIGIT
3417 | PERL_SCAN_DISALLOW_PREFIX
3418 | PERL_SCAN_SILENT_NON_PORTABLE;
3420 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3423 else /* If no value here, it is a single element range */
3426 /* Non-binary tables have a third entry: what the first element of the
3427 * range maps to. The map for those currently read here is in hex */
3431 flags = PERL_SCAN_SILENT_ILLDIGIT
3432 | PERL_SCAN_DISALLOW_PREFIX
3433 | PERL_SCAN_SILENT_NON_PORTABLE;
3435 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3444 /* diag_listed_as: To%s: illegal mapping '%s' */
3445 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3451 *val = 0; /* bits == 1, then any val should be ignored */
3453 else { /* Nothing following range min, should be single element with no
3459 /* diag_listed_as: To%s: illegal mapping '%s' */
3460 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3464 *val = 0; /* bits == 1, then val should be ignored */
3467 /* Position to next line if any, or EOF */
3477 * Returns a swatch (a bit vector string) for a code point sequence
3478 * that starts from the value C<start> and comprises the number C<span>.
3479 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3480 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3483 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3486 U8 *l, *lend, *x, *xend, *s, *send;
3487 STRLEN lcur, xcur, scur;
3488 HV *const hv = MUTABLE_HV(SvRV(swash));
3489 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3491 SV** listsvp = NULL; /* The string containing the main body of the table */
3492 SV** extssvp = NULL;
3493 SV** invert_it_svp = NULL;
3496 STRLEN octets; /* if bits == 1, then octets == 0 */
3498 UV end = start + span;
3500 if (invlistsvp == NULL) {
3501 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3502 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3503 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3504 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3505 listsvp = hv_fetchs(hv, "LIST", FALSE);
3506 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3508 bits = SvUV(*bitssvp);
3509 none = SvUV(*nonesvp);
3510 typestr = (U8*)SvPV_nolen(*typesvp);
3516 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3518 PERL_ARGS_ASSERT_SWATCH_GET;
3520 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3521 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3525 /* If overflowed, use the max possible */
3531 /* create and initialize $swatch */
3532 scur = octets ? (span * octets) : (span + 7) / 8;
3533 swatch = newSV(scur);
3535 s = (U8*)SvPVX(swatch);
3536 if (octets && none) {
3537 const U8* const e = s + scur;
3540 *s++ = (U8)(none & 0xff);
3541 else if (bits == 16) {
3542 *s++ = (U8)((none >> 8) & 0xff);
3543 *s++ = (U8)( none & 0xff);
3545 else if (bits == 32) {
3546 *s++ = (U8)((none >> 24) & 0xff);
3547 *s++ = (U8)((none >> 16) & 0xff);
3548 *s++ = (U8)((none >> 8) & 0xff);
3549 *s++ = (U8)( none & 0xff);
3555 (void)memzero((U8*)s, scur + 1);
3557 SvCUR_set(swatch, scur);
3558 s = (U8*)SvPVX(swatch);
3560 if (invlistsvp) { /* If has an inversion list set up use that */
3561 _invlist_populate_swatch(*invlistsvp, start, end, s);
3565 /* read $swash->{LIST} */
3566 l = (U8*)SvPV(*listsvp, lcur);
3569 UV min, max, val, upper;
3570 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3571 cBOOL(octets), typestr);
3576 /* If looking for something beyond this range, go try the next one */
3580 /* <end> is generally 1 beyond where we want to set things, but at the
3581 * platform's infinity, where we can't go any higher, we want to
3582 * include the code point at <end> */
3585 : (max != UV_MAX || end != UV_MAX)
3592 if (!none || val < none) {
3597 for (key = min; key <= upper; key++) {
3599 /* offset must be non-negative (start <= min <= key < end) */
3600 offset = octets * (key - start);
3602 s[offset] = (U8)(val & 0xff);
3603 else if (bits == 16) {
3604 s[offset ] = (U8)((val >> 8) & 0xff);
3605 s[offset + 1] = (U8)( val & 0xff);
3607 else if (bits == 32) {
3608 s[offset ] = (U8)((val >> 24) & 0xff);
3609 s[offset + 1] = (U8)((val >> 16) & 0xff);
3610 s[offset + 2] = (U8)((val >> 8) & 0xff);
3611 s[offset + 3] = (U8)( val & 0xff);
3614 if (!none || val < none)
3618 else { /* bits == 1, then val should be ignored */
3623 for (key = min; key <= upper; key++) {
3624 const STRLEN offset = (STRLEN)(key - start);
3625 s[offset >> 3] |= 1 << (offset & 7);
3630 /* Invert if the data says it should be. Assumes that bits == 1 */
3631 if (invert_it_svp && SvUV(*invert_it_svp)) {
3633 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3634 * be 0, and their inversion should also be 0, as we don't succeed any
3635 * Unicode property matches for non-Unicode code points */
3636 if (start <= PERL_UNICODE_MAX) {
3638 /* The code below assumes that we never cross the
3639 * Unicode/above-Unicode boundary in a range, as otherwise we would
3640 * have to figure out where to stop flipping the bits. Since this
3641 * boundary is divisible by a large power of 2, and swatches comes
3642 * in small powers of 2, this should be a valid assumption */
3643 assert(start + span - 1 <= PERL_UNICODE_MAX);
3653 /* read $swash->{EXTRAS}
3654 * This code also copied to swash_to_invlist() below */
3655 x = (U8*)SvPV(*extssvp, xcur);
3663 SV **otherbitssvp, *other;
3667 const U8 opc = *x++;
3671 nl = (U8*)memchr(x, '\n', xend - x);
3673 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3675 x = nl + 1; /* 1 is length of "\n" */
3679 x = xend; /* to EXTRAS' end at which \n is not found */
3686 namelen = nl - namestr;
3690 namelen = xend - namestr;
3694 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3695 otherhv = MUTABLE_HV(SvRV(*othersvp));
3696 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3697 otherbits = (STRLEN)SvUV(*otherbitssvp);
3698 if (bits < otherbits)
3699 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3700 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3702 /* The "other" swatch must be destroyed after. */
3703 other = swatch_get(*othersvp, start, span);
3704 o = (U8*)SvPV(other, olen);
3707 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3709 s = (U8*)SvPV(swatch, slen);
3710 if (bits == 1 && otherbits == 1) {
3712 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3713 "mismatch, slen=%"UVuf", olen=%"UVuf,
3714 (UV)slen, (UV)olen);
3738 STRLEN otheroctets = otherbits >> 3;
3740 U8* const send = s + slen;
3745 if (otherbits == 1) {
3746 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3750 STRLEN vlen = otheroctets;
3758 if (opc == '+' && otherval)
3759 NOOP; /* replace with otherval */
3760 else if (opc == '!' && !otherval)
3762 else if (opc == '-' && otherval)
3764 else if (opc == '&' && !otherval)
3767 s += octets; /* no replacement */
3772 *s++ = (U8)( otherval & 0xff);
3773 else if (bits == 16) {
3774 *s++ = (U8)((otherval >> 8) & 0xff);
3775 *s++ = (U8)( otherval & 0xff);
3777 else if (bits == 32) {
3778 *s++ = (U8)((otherval >> 24) & 0xff);
3779 *s++ = (U8)((otherval >> 16) & 0xff);
3780 *s++ = (U8)((otherval >> 8) & 0xff);
3781 *s++ = (U8)( otherval & 0xff);
3785 sv_free(other); /* through with it! */
3791 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3794 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3795 * Can't be used on a property that is subject to user override, as it
3796 * relies on the value of SPECIALS in the swash which would be set by
3797 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3798 * for overridden properties
3800 * Returns a hash which is the inversion and closure of a swash mapping.
3801 * For example, consider the input lines:
3806 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3807 * 006C. The value for each key is an array. For 006C, the array would
3808 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3809 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3811 * Essentially, for any code point, it gives all the code points that map to
3812 * it, or the list of 'froms' for that point.
3814 * Currently it ignores any additions or deletions from other swashes,
3815 * looking at just the main body of the swash, and if there are SPECIALS
3816 * in the swash, at that hash
3818 * The specials hash can be extra code points, and most likely consists of
3819 * maps from single code points to multiple ones (each expressed as a string
3820 * of utf8 characters). This function currently returns only 1-1 mappings.
3821 * However consider this possible input in the specials hash:
3822 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3823 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3825 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3826 * currently handle. But it also means that FB05 and FB06 are equivalent in
3827 * a 1-1 mapping which we should handle, and this relationship may not be in
3828 * the main table. Therefore this function examines all the multi-char
3829 * sequences and adds the 1-1 mappings that come out of that. */
3833 HV *const hv = MUTABLE_HV(SvRV(swash));
3835 /* The string containing the main body of the table. This will have its
3836 * assertion fail if the swash has been converted to its inversion list */
3837 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3839 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3840 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3841 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3842 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3843 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3844 const STRLEN bits = SvUV(*bitssvp);
3845 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3846 const UV none = SvUV(*nonesvp);
3847 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3851 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3853 /* Must have at least 8 bits to get the mappings */
3854 if (bits != 8 && bits != 16 && bits != 32) {
3855 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3859 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3860 mapping to more than one character */
3862 /* Construct an inverse mapping hash for the specials */
3863 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3864 HV * specials_inverse = newHV();
3865 char *char_from; /* the lhs of the map */
3866 I32 from_len; /* its byte length */
3867 char *char_to; /* the rhs of the map */
3868 I32 to_len; /* its byte length */
3869 SV *sv_to; /* and in a sv */
3870 AV* from_list; /* list of things that map to each 'to' */
3872 hv_iterinit(specials_hv);
3874 /* The keys are the characters (in utf8) that map to the corresponding
3875 * utf8 string value. Iterate through the list creating the inverse
3877 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3879 if (! SvPOK(sv_to)) {
3880 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3881 "unexpectedly is not a string, flags=%lu",
3882 (unsigned long)SvFLAGS(sv_to));
3884 /*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)));*/
3886 /* Each key in the inverse list is a mapped-to value, and the key's
3887 * hash value is a list of the strings (each in utf8) that map to
3888 * it. Those strings are all one character long */
3889 if ((listp = hv_fetch(specials_inverse,
3893 from_list = (AV*) *listp;
3895 else { /* No entry yet for it: create one */
3896 from_list = newAV();
3897 if (! hv_store(specials_inverse,
3900 (SV*) from_list, 0))
3902 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3906 /* Here have the list associated with this 'to' (perhaps newly
3907 * created and empty). Just add to it. Note that we ASSUME that
3908 * the input is guaranteed to not have duplications, so we don't
3909 * check for that. Duplications just slow down execution time. */
3910 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3913 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3914 * it looking for cases like the FB05/FB06 examples above. There would
3915 * be an entry in the hash like
3916 * 'st' => [ FB05, FB06 ]
3917 * In this example we will create two lists that get stored in the
3918 * returned hash, 'ret':
3919 * FB05 => [ FB05, FB06 ]
3920 * FB06 => [ FB05, FB06 ]
3922 * Note that there is nothing to do if the array only has one element.
3923 * (In the normal 1-1 case handled below, we don't have to worry about
3924 * two lists, as everything gets tied to the single list that is
3925 * generated for the single character 'to'. But here, we are omitting
3926 * that list, ('st' in the example), so must have multiple lists.) */
3927 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3928 &char_to, &to_len)))
3930 if (av_len(from_list) > 0) {
3933 /* We iterate over all combinations of i,j to place each code
3934 * point on each list */
3935 for (i = 0; i <= av_len(from_list); i++) {
3937 AV* i_list = newAV();
3938 SV** entryp = av_fetch(from_list, i, FALSE);
3939 if (entryp == NULL) {
3940 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3942 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3943 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3945 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3946 (SV*) i_list, FALSE))
3948 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3951 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3952 for (j = 0; j <= av_len(from_list); j++) {
3953 entryp = av_fetch(from_list, j, FALSE);
3954 if (entryp == NULL) {
3955 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3958 /* When i==j this adds itself to the list */
3959 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3960 (U8*) SvPVX(*entryp),
3961 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3963 /*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));*/
3968 SvREFCNT_dec(specials_inverse); /* done with it */
3969 } /* End of specials */
3971 /* read $swash->{LIST} */
3972 l = (U8*)SvPV(*listsvp, lcur);
3975 /* Go through each input line */
3979 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3980 cBOOL(octets), typestr);
3985 /* Each element in the range is to be inverted */
3986 for (inverse = min; inverse <= max; inverse++) {
3990 bool found_key = FALSE;
3991 bool found_inverse = FALSE;
3993 /* The key is the inverse mapping */
3994 char key[UTF8_MAXBYTES+1];
3995 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3996 STRLEN key_len = key_end - key;
3998 /* Get the list for the map */
3999 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4000 list = (AV*) *listp;
4002 else { /* No entry yet for it: create one */
4004 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4005 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4009 /* Look through list to see if this inverse mapping already is
4010 * listed, or if there is a mapping to itself already */
4011 for (i = 0; i <= av_len(list); i++) {
4012 SV** entryp = av_fetch(list, i, FALSE);
4014 if (entryp == NULL) {
4015 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4018 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
4019 if (SvUV(entry) == val) {
4022 if (SvUV(entry) == inverse) {
4023 found_inverse = TRUE;
4026 /* No need to continue searching if found everything we are
4028 if (found_key && found_inverse) {
4033 /* Make sure there is a mapping to itself on the list */
4035 av_push(list, newSVuv(val));
4036 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
4040 /* Simply add the value to the list */
4041 if (! found_inverse) {
4042 av_push(list, newSVuv(inverse));
4043 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
4046 /* swatch_get() increments the value of val for each element in the
4047 * range. That makes more compact tables possible. You can
4048 * express the capitalization, for example, of all consecutive
4049 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4050 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4051 * and it's not documented; it appears to be used only in
4052 * implementing tr//; I copied the semantics from swatch_get(), just
4054 if (!none || val < none) {
4064 Perl__swash_to_invlist(pTHX_ SV* const swash)
4067 /* Subject to change or removal. For use only in one place in regcomp.c.
4068 * Ownership is given to one reference count in the returned SV* */
4073 HV *const hv = MUTABLE_HV(SvRV(swash));
4074 UV elements = 0; /* Number of elements in the inversion list */
4084 STRLEN octets; /* if bits == 1, then octets == 0 */
4090 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4092 /* If not a hash, it must be the swash's inversion list instead */
4093 if (SvTYPE(hv) != SVt_PVHV) {
4094 return SvREFCNT_inc_simple_NN((SV*) hv);
4097 /* The string containing the main body of the table */
4098 listsvp = hv_fetchs(hv, "LIST", FALSE);
4099 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4100 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4101 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4102 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4104 typestr = (U8*)SvPV_nolen(*typesvp);
4105 bits = SvUV(*bitssvp);
4106 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4108 /* read $swash->{LIST} */
4109 if (SvPOK(*listsvp)) {
4110 l = (U8*)SvPV(*listsvp, lcur);
4113 /* LIST legitimately doesn't contain a string during compilation phases
4114 * of Perl itself, before the Unicode tables are generated. In this
4115 * case, just fake things up by creating an empty list */
4122 /* Scan the input to count the number of lines to preallocate array size
4123 * based on worst possible case, which is each line in the input creates 2
4124 * elements in the inversion list: 1) the beginning of a range in the list;
4125 * 2) the beginning of a range not in the list. */
4126 while ((loc = (strchr(loc, '\n'))) != NULL) {
4131 /* If the ending is somehow corrupt and isn't a new line, add another
4132 * element for the final range that isn't in the inversion list */
4133 if (! (*lend == '\n'
4134 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4139 invlist = _new_invlist(elements);
4141 /* Now go through the input again, adding each range to the list */
4144 UV val; /* Not used by this function */
4146 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
4147 cBOOL(octets), typestr);
4153 invlist = _add_range_to_invlist(invlist, start, end);
4156 /* Invert if the data says it should be */
4157 if (invert_it_svp && SvUV(*invert_it_svp)) {
4158 _invlist_invert_prop(invlist);
4161 /* This code is copied from swatch_get()
4162 * read $swash->{EXTRAS} */
4163 x = (U8*)SvPV(*extssvp, xcur);
4171 SV **otherbitssvp, *other;
4174 const U8 opc = *x++;
4178 nl = (U8*)memchr(x, '\n', xend - x);
4180 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4182 x = nl + 1; /* 1 is length of "\n" */
4186 x = xend; /* to EXTRAS' end at which \n is not found */
4193 namelen = nl - namestr;
4197 namelen = xend - namestr;
4201 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4202 otherhv = MUTABLE_HV(SvRV(*othersvp));
4203 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4204 otherbits = (STRLEN)SvUV(*otherbitssvp);
4206 if (bits != otherbits || bits != 1) {
4207 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4208 "properties, bits=%"UVuf", otherbits=%"UVuf,
4209 (UV)bits, (UV)otherbits);
4212 /* The "other" swatch must be destroyed after. */
4213 other = _swash_to_invlist((SV *)*othersvp);
4215 /* End of code copied from swatch_get() */
4218 _invlist_union(invlist, other, &invlist);
4221 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4224 _invlist_subtract(invlist, other, &invlist);
4227 _invlist_intersection(invlist, other, &invlist);
4232 sv_free(other); /* through with it! */
4239 Perl__get_swash_invlist(pTHX_ SV* const swash)
4243 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4245 if (! SvROK(swash)) {
4249 /* If it really isn't a hash, it isn't really swash; must be an inversion
4251 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4255 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4264 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4266 /* May change: warns if surrogates, non-character code points, or
4267 * non-Unicode code points are in s which has length len bytes. Returns
4268 * TRUE if none found; FALSE otherwise. The only other validity check is
4269 * to make sure that this won't exceed the string's length */
4271 const U8* const e = s + len;
4274 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4277 if (UTF8SKIP(s) > len) {
4278 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4279 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4282 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
4284 if (UTF8_IS_SUPER(s)) {
4285 if (ckWARN_d(WARN_NON_UNICODE)) {
4286 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4287 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
4288 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
4292 else if (UTF8_IS_SURROGATE(s)) {
4293 if (ckWARN_d(WARN_SURROGATE)) {
4294 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4295 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4296 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
4301 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
4302 && (ckWARN_d(WARN_NONCHAR)))
4304 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4305 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
4306 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
4317 =for apidoc pv_uni_display
4319 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4320 length C<len>, the displayable version being at most C<pvlim> bytes long
4321 (if longer, the rest is truncated and "..." will be appended).
4323 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
4324 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
4325 to display the \\[nrfta\\] as the backslashed versions (like '\n')
4326 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
4327 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
4328 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
4330 The pointer to the PV of the C<dsv> is returned.
4334 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4339 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4343 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4345 /* This serves double duty as a flag and a character to print after
4346 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4350 if (pvlim && SvCUR(dsv) >= pvlim) {
4354 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4356 const unsigned char c = (unsigned char)u & 0xFF;
4357 if (flags & UNI_DISPLAY_BACKSLASH) {
4374 const char string = ok;
4375 sv_catpvs(dsv, "\\");
4376 sv_catpvn(dsv, &string, 1);
4379 /* isPRINT() is the locale-blind version. */
4380 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4381 const char string = c;
4382 sv_catpvn(dsv, &string, 1);
4387 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4390 sv_catpvs(dsv, "...");
4396 =for apidoc sv_uni_display
4398 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4399 the displayable version being at most C<pvlim> bytes long
4400 (if longer, the rest is truncated and "..." will be appended).
4402 The C<flags> argument is as in L</pv_uni_display>().
4404 The pointer to the PV of the C<dsv> is returned.
4409 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4411 const char * const ptr =
4412 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4414 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4416 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4417 SvCUR(ssv), pvlim, flags);
4421 =for apidoc foldEQ_utf8
4423 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4424 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4425 How far into the strings to compare is determined by other input parameters.
4427 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4428 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4429 with respect to C<s2>.
4431 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4432 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4433 scan will not be considered to be a match unless the goal is reached, and
4434 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4437 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4438 considered an end pointer to the position 1 byte past the maximum point
4439 in C<s1> beyond which scanning will not continue under any circumstances.
4440 (This routine assumes that UTF-8 encoded input strings are not malformed;
4441 malformed input can cause it to read past C<pe1>).
4442 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4443 is less than C<s1>+C<l1>, the match will never be successful because it can
4445 get as far as its goal (and in fact is asserted against). Correspondingly for
4446 C<pe2> with respect to C<s2>.
4448 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4449 C<l2> must be non-zero), and if both do, both have to be
4450 reached for a successful match. Also, if the fold of a character is multiple
4451 characters, all of them must be matched (see tr21 reference below for
4454 Upon a successful match, if C<pe1> is non-NULL,
4455 it will be set to point to the beginning of the I<next> character of C<s1>
4456 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4458 For case-insensitiveness, the "casefolding" of Unicode is used
4459 instead of upper/lowercasing both the characters, see
4460 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4464 /* A flags parameter has been added which may change, and hence isn't
4465 * externally documented. Currently it is:
4466 * 0 for as-documented above
4467 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4468 ASCII one, to not match
4469 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
4470 * points below 256; unicode rules for above 255; and
4471 * folds that cross those boundaries are disallowed,
4472 * like the NOMIX_ASCII option
4473 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4474 * routine. This allows that step to be skipped.
4475 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4478 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)
4481 const U8 *p1 = (const U8*)s1; /* Point to current char */
4482 const U8 *p2 = (const U8*)s2;
4483 const U8 *g1 = NULL; /* goal for s1 */
4484 const U8 *g2 = NULL;
4485 const U8 *e1 = NULL; /* Don't scan s1 past this */
4486 U8 *f1 = NULL; /* Point to current folded */
4487 const U8 *e2 = NULL;
4489 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4490 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4491 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4493 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4495 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
4496 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4497 /* The algorithm is to trial the folds without regard to the flags on
4498 * the first line of the above assert(), and then see if the result
4499 * violates them. This means that the inputs can't be pre-folded to a
4500 * violating result, hence the assert. This could be changed, with the
4501 * addition of extra tests here for the already-folded case, which would
4502 * slow it down. That cost is more than any possible gain for when these
4503 * flags are specified, as the flags indicate /il or /iaa matching which
4504 * is less common than /iu, and I (khw) also believe that real-world /il
4505 * and /iaa matches are most likely to involve code points 0-255, and this
4506 * function only under rare conditions gets called for 0-255. */
4513 g1 = (const U8*)s1 + l1;
4521 g2 = (const U8*)s2 + l2;
4524 /* Must have at least one goal */
4529 /* Will never match if goal is out-of-bounds */
4530 assert(! e1 || e1 >= g1);
4532 /* Here, there isn't an end pointer, or it is beyond the goal. We
4533 * only go as far as the goal */
4537 assert(e1); /* Must have an end for looking at s1 */
4540 /* Same for goal for s2 */
4542 assert(! e2 || e2 >= g2);
4549 /* If both operands are already folded, we could just do a memEQ on the
4550 * whole strings at once, but it would be better if the caller realized
4551 * this and didn't even call us */
4553 /* Look through both strings, a character at a time */
4554 while (p1 < e1 && p2 < e2) {
4556 /* If at the beginning of a new character in s1, get its fold to use
4557 * and the length of the fold. (exception: locale rules just get the
4558 * character to a single byte) */
4560 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4565 /* If in locale matching, we use two sets of rules, depending
4566 * on if the code point is above or below 255. Here, we test
4567 * for and handle locale rules */
4568 if ((flags & FOLDEQ_UTF8_LOCALE)
4569 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
4571 /* There is no mixing of code points above and below 255. */
4572 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
4576 /* We handle locale rules by converting, if necessary, the
4577 * code point to a single byte. */
4578 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4582 *foldbuf1 = TWO_BYTE_UTF8_TO_NATIVE(*p1, *(p1 + 1));
4586 else if (isASCII(*p1)) { /* Note, that here won't be both
4587 ASCII and using locale rules */
4589 /* If trying to mix non- with ASCII, and not supposed to,
4591 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4595 *foldbuf1 = toFOLD(*p1);
4598 to_utf8_fold(p1, foldbuf1, &n1);
4600 else { /* Not utf8, get utf8 fold */
4601 to_uni_fold(*p1, foldbuf1, &n1);
4607 if (n2 == 0) { /* Same for s2 */
4608 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4613 if ((flags & FOLDEQ_UTF8_LOCALE)
4614 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4616 /* Here, the next char in s2 is < 256. We've already
4617 * worked on s1, and if it isn't also < 256, can't match */
4618 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4621 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4625 *foldbuf2 = TWO_BYTE_UTF8_TO_NATIVE(*p2, *(p2 + 1));
4628 /* Use another function to handle locale rules. We've made
4629 * sure that both characters to compare are single bytes */
4630 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4635 else if (isASCII(*p2)) {
4636 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4640 *foldbuf2 = toFOLD(*p2);
4643 to_utf8_fold(p2, foldbuf2, &n2);
4646 to_uni_fold(*p2, foldbuf2, &n2);
4652 /* Here f1 and f2 point to the beginning of the strings to compare.
4653 * These strings are the folds of the next character from each input
4654 * string, stored in utf8. */
4656 /* While there is more to look for in both folds, see if they
4657 * continue to match */
4659 U8 fold_length = UTF8SKIP(f1);
4660 if (fold_length != UTF8SKIP(f2)
4661 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4662 function call for single
4664 || memNE((char*)f1, (char*)f2, fold_length))
4666 return 0; /* mismatch */
4669 /* Here, they matched, advance past them */
4676 /* When reach the end of any fold, advance the input past it */
4678 p1 += u1 ? UTF8SKIP(p1) : 1;
4681 p2 += u2 ? UTF8SKIP(p2) : 1;
4683 } /* End of loop through both strings */
4685 /* A match is defined by each scan that specified an explicit length
4686 * reaching its final goal, and the other not having matched a partial
4687 * character (which can happen when the fold of a character is more than one
4689 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4693 /* Successful match. Set output pointers */
4703 /* XXX The next four functions should likely be moved to mathoms.c once all
4704 * occurrences of them are removed from the core; some cpan-upstream modules
4708 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4710 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4712 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4716 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4718 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4720 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4724 =for apidoc uvuni_to_utf8_flags
4726 Instead you almost certainly want to use L</uvchr_to_utf8> or
4727 L</uvchr_to_utf8_flags>>.
4729 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4730 which itself, while not deprecated, should be used only in isolated
4731 circumstances. These functions were useful for code that wanted to handle
4732 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4733 v5.20, the distinctions between the platforms have mostly been made invisible
4734 to most code, so this function is quite unlikely to be what you want.
4740 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4742 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4744 return uvoffuni_to_utf8_flags(d, uv, flags);
4748 =for apidoc utf8n_to_uvuni
4750 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4752 This function was useful for code that wanted to handle both EBCDIC and
4753 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4754 distinctions between the platforms have mostly been made invisible to most
4755 code, so this function is quite unlikely to be what you want. If you do need
4756 this precise functionality, use instead
4757 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4758 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4765 * c-indentation-style: bsd
4767 * indent-tabs-mode: nil
4770 * ex: set ts=8 sts=4 sw=4 et: