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"
35 #include "charclass_invlists.h"
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
41 =head1 Unicode Support
43 This file contains various utility functions for manipulating UTF8-encoded
44 strings. For the uninitiated, this is a method of representing arbitrary
45 Unicode characters as a variable number of bytes, in such a way that
46 characters in the ASCII range are unmodified, and a zero byte never appears
47 within non-zero characters.
53 =for apidoc is_ascii_string
55 Returns true if the first C<len> bytes of the string C<s> are the same whether
56 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
57 is, if they are invariant. On ASCII-ish machines, only ASCII characters
58 fit this definition, hence the function's name.
60 If C<len> is 0, it will be calculated using C<strlen(s)>.
62 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
68 Perl_is_ascii_string(const U8 *s, STRLEN len)
70 const U8* const send = s + (len ? len : strlen((const char *)s));
73 PERL_ARGS_ASSERT_IS_ASCII_STRING;
75 for (; x < send; ++x) {
76 if (!UTF8_IS_INVARIANT(*x))
84 =for apidoc uvoffuni_to_utf8_flags
86 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
87 Instead, B<Almost all code should use L</uvchr_to_utf8> or
88 L</uvchr_to_utf8_flags>>.
90 This function is like them, but the input is a strict Unicode
91 (as opposed to native) code point. Only in very rare circumstances should code
92 not be using the native code point.
94 For details, see the description for L</uvchr_to_utf8_flags>>.
100 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
102 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
104 if (UNI_IS_INVARIANT(uv)) {
105 *d++ = (U8) LATIN1_TO_NATIVE(uv);
109 /* The first problematic code point is the first surrogate */
110 if (uv >= UNICODE_SURROGATE_FIRST
111 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
113 if (UNICODE_IS_SURROGATE(uv)) {
114 if (flags & UNICODE_WARN_SURROGATE) {
115 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
116 "UTF-16 surrogate U+%04"UVXf, uv);
118 if (flags & UNICODE_DISALLOW_SURROGATE) {
122 else if (UNICODE_IS_SUPER(uv)) {
123 if (flags & UNICODE_WARN_SUPER
124 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
126 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
127 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
129 if (flags & UNICODE_DISALLOW_SUPER
130 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
135 else if (UNICODE_IS_NONCHAR(uv)) {
136 if (flags & UNICODE_WARN_NONCHAR) {
137 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
138 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
141 if (flags & UNICODE_DISALLOW_NONCHAR) {
149 STRLEN len = OFFUNISKIP(uv);
152 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
153 uv >>= UTF_ACCUMULATION_SHIFT;
155 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
158 #else /* Non loop style */
160 *d++ = (U8)(( uv >> 6) | 0xc0);
161 *d++ = (U8)(( uv & 0x3f) | 0x80);
165 *d++ = (U8)(( uv >> 12) | 0xe0);
166 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
167 *d++ = (U8)(( uv & 0x3f) | 0x80);
171 *d++ = (U8)(( uv >> 18) | 0xf0);
172 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
173 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
174 *d++ = (U8)(( uv & 0x3f) | 0x80);
177 if (uv < 0x4000000) {
178 *d++ = (U8)(( uv >> 24) | 0xf8);
179 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
180 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
181 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
182 *d++ = (U8)(( uv & 0x3f) | 0x80);
185 if (uv < 0x80000000) {
186 *d++ = (U8)(( uv >> 30) | 0xfc);
187 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
188 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
189 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
191 *d++ = (U8)(( uv & 0x3f) | 0x80);
195 if (uv < UTF8_QUAD_MAX)
198 *d++ = 0xfe; /* Can't match U+FEFF! */
199 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
204 *d++ = (U8)(( uv & 0x3f) | 0x80);
209 *d++ = 0xff; /* Can't match U+FFFE! */
210 *d++ = 0x80; /* 6 Reserved bits */
211 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
212 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
221 *d++ = (U8)(( uv & 0x3f) | 0x80);
225 #endif /* Non loop style */
228 =for apidoc uvchr_to_utf8
230 Adds the UTF-8 representation of the native code point C<uv> to the end
231 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
232 bytes available. The return value is the pointer to the byte after the
233 end of the new character. In other words,
235 d = uvchr_to_utf8(d, uv);
237 is the recommended wide native character-aware way of saying
241 This function accepts any UV as input. To forbid or warn on non-Unicode code
242 points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
247 /* This is also a macro */
248 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
251 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
253 return uvchr_to_utf8(d, uv);
257 =for apidoc uvchr_to_utf8_flags
259 Adds the UTF-8 representation of the native code point C<uv> to the end
260 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
261 bytes available. The return value is the pointer to the byte after the
262 end of the new character. In other words,
264 d = uvchr_to_utf8_flags(d, uv, flags);
268 d = uvchr_to_utf8_flags(d, uv, 0);
270 This is the Unicode-aware way of saying
274 This function will convert to UTF-8 (and not warn) even code points that aren't
275 legal Unicode or are problematic, unless C<flags> contains one or more of the
278 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
279 the function will raise a warning, provided UTF8 warnings are enabled. If instead
280 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
281 If both flags are set, the function will both warn and return NULL.
283 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
284 affect how the function handles a Unicode non-character. And likewise, the
285 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
287 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
288 even less portable) can be warned and/or disallowed even if other above-Unicode
289 code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
292 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
293 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
299 /* This is also a macro */
300 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
303 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
305 return uvchr_to_utf8_flags(d, uv, flags);
310 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
311 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC) character is a
312 valid UTF-8 character. The number of bytes in the UTF-8 character
313 will be returned if it is valid, otherwise 0.
315 This is the "slow" version as opposed to the "fast" version which is
316 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
317 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
318 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
319 you should use the _slow(). In practice this means that the _slow()
320 will be used very rarely, since the maximum Unicode code point (as of
321 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
322 the "Perl extended UTF-8" (e.g, the infamous 'v-strings') will encode into
326 PERL_STATIC_INLINE STRLEN
327 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
329 dTHX; /* The function called below requires thread context */
333 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
335 utf8n_to_uvchr(s, len, &actual_len, UTF8_CHECK_ONLY);
337 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
341 =for apidoc is_utf8_char_buf
343 Returns the number of bytes that comprise the first UTF-8 encoded character in
344 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
345 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
348 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
349 machines) is a valid UTF-8 character.
354 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
359 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
361 if (buf_end <= buf) {
366 if (len > UTF8SKIP(buf)) {
370 if (IS_UTF8_CHAR_FAST(len))
371 return IS_UTF8_CHAR(buf, len) ? len : 0;
372 return is_utf8_char_slow(buf, len);
376 =for apidoc is_utf8_char
378 Tests if some arbitrary number of bytes begins in a valid UTF-8
379 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
380 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
381 character will be returned if it is valid, otherwise 0.
383 This function is deprecated due to the possibility that malformed input could
384 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
390 Perl_is_utf8_char(const U8 *s)
392 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
394 /* Assumes we have enough space, which is why this is deprecated */
395 return is_utf8_char_buf(s, s + UTF8SKIP(s));
400 =for apidoc is_utf8_string
402 Returns true if the first C<len> bytes of string C<s> form a valid
403 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
404 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
405 terminating NUL byte). Note that all characters being ASCII constitute 'a
408 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
414 Perl_is_utf8_string(const U8 *s, STRLEN len)
416 const U8* const send = s + (len ? len : strlen((const char *)s));
419 PERL_ARGS_ASSERT_IS_UTF8_STRING;
422 /* Inline the easy bits of is_utf8_char() here for speed... */
423 if (UTF8_IS_INVARIANT(*x)) {
427 /* ... and call is_utf8_char() only if really needed. */
428 const STRLEN c = UTF8SKIP(x);
429 const U8* const next_char_ptr = x + c;
431 if (next_char_ptr > send) {
435 if (IS_UTF8_CHAR_FAST(c)) {
436 if (!IS_UTF8_CHAR(x, c))
439 else if (! is_utf8_char_slow(x, c)) {
450 Implemented as a macro in utf8.h
452 =for apidoc is_utf8_string_loc
454 Like L</is_utf8_string> but stores the location of the failure (in the
455 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
456 "utf8ness success") in the C<ep>.
458 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
460 =for apidoc is_utf8_string_loclen
462 Like L</is_utf8_string>() but stores the location of the failure (in the
463 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
464 "utf8ness success") in the C<ep>, and the number of UTF-8
465 encoded characters in the C<el>.
467 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
473 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
475 const U8* const send = s + (len ? len : strlen((const char *)s));
480 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
483 const U8* next_char_ptr;
485 /* Inline the easy bits of is_utf8_char() here for speed... */
486 if (UTF8_IS_INVARIANT(*x))
487 next_char_ptr = x + 1;
489 /* ... and call is_utf8_char() only if really needed. */
491 next_char_ptr = c + x;
492 if (next_char_ptr > send) {
495 if (IS_UTF8_CHAR_FAST(c)) {
496 if (!IS_UTF8_CHAR(x, c))
499 c = is_utf8_char_slow(x, c);
518 =for apidoc utf8n_to_uvchr
520 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
521 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
523 Bottom level UTF-8 decode routine.
524 Returns the native code point value of the first character in the string C<s>,
525 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
526 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
527 the length, in bytes, of that character.
529 The value of C<flags> determines the behavior when C<s> does not point to a
530 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
531 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
532 next possible position in C<s> that could begin a non-malformed character.
533 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
535 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
536 individual types of malformations, such as the sequence being overlong (that
537 is, when there is a shorter sequence that can express the same code point;
538 overlong sequences are expressly forbidden in the UTF-8 standard due to
539 potential security issues). Another malformation example is the first byte of
540 a character not being a legal first byte. See F<utf8.h> for the list of such
541 flags. For allowed 0 length strings, this function returns 0; for allowed
542 overlong sequences, the computed code point is returned; for all other allowed
543 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
544 determinable reasonable value.
546 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
547 flags) malformation is found. If this flag is set, the routine assumes that
548 the caller will raise a warning, and this function will silently just set
549 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
551 Note that this API requires disambiguation between successful decoding a NUL
552 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
553 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
554 first byte of C<s> is 0 as well. If so, the input was a NUL; if not, the input
557 Certain code points are considered problematic. These are Unicode surrogates,
558 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
559 By default these are considered regular code points, but certain situations
560 warrant special handling for them. If C<flags> contains
561 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
562 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
563 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
564 maximum) can be set to disallow these categories individually.
566 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
567 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
568 for their respective categories, but otherwise the code points are considered
569 valid (not malformations). To get a category to both be treated as a
570 malformation and raise a warning, specify both the WARN and DISALLOW flags.
571 (But note that warnings are not raised if lexically disabled nor if
572 UTF8_CHECK_ONLY is also specified.)
574 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
575 the others that are above the Unicode legal maximum. There are several
576 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
577 not representable at all on EBCDIC platforms, and the original UTF-8
578 specification never went above this number (the current 0x10FFFF limit was
579 imposed later). (The smaller ones, those that fit into 32 bits, are
580 representable by a UV on ASCII platforms, but not by an IV, which means that
581 the number of operations that can be performed on them is quite restricted.)
582 The UTF-8 encoding on ASCII platforms for these large code points begins with a
583 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
584 be treated as malformations, while allowing smaller above-Unicode code points.
585 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
586 including these, as malformations.)
587 Similarly, UTF8_WARN_FE_FF acts just like
588 the other WARN flags, but applies just to these code points.
590 All other code points corresponding to Unicode characters, including private
591 use and those yet to be assigned, are never considered malformed and never
598 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
601 const U8 * const s0 = s;
602 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
607 UV outlier_ret = 0; /* return value when input is in error or problematic
609 UV pack_warn = 0; /* Save result of packWARN() for later */
610 bool unexpected_non_continuation = FALSE;
611 bool overflowed = FALSE;
612 bool do_overlong_test = TRUE; /* May have to skip this test */
614 const char* const malformed_text = "Malformed UTF-8 character";
616 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
618 /* The order of malformation tests here is important. We should consume as
619 * few bytes as possible in order to not skip any valid character. This is
620 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
621 * http://unicode.org/reports/tr36 for more discussion as to why. For
622 * example, once we've done a UTF8SKIP, we can tell the expected number of
623 * bytes, and could fail right off the bat if the input parameters indicate
624 * that there are too few available. But it could be that just that first
625 * byte is garbled, and the intended character occupies fewer bytes. If we
626 * blindly assumed that the first byte is correct, and skipped based on
627 * that number, we could skip over a valid input character. So instead, we
628 * always examine the sequence byte-by-byte.
630 * We also should not consume too few bytes, otherwise someone could inject
631 * things. For example, an input could be deliberately designed to
632 * overflow, and if this code bailed out immediately upon discovering that,
633 * returning to the caller C<*retlen> pointing to the very next byte (one
634 * which is actually part of of the overflowing sequence), that could look
635 * legitimate to the caller, which could discard the initial partial
636 * sequence and process the rest, inappropriately */
638 /* Zero length strings, if allowed, of necessity are zero */
639 if (UNLIKELY(curlen == 0)) {
644 if (flags & UTF8_ALLOW_EMPTY) {
647 if (! (flags & UTF8_CHECK_ONLY)) {
648 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
653 expectlen = UTF8SKIP(s);
655 /* A well-formed UTF-8 character, as the vast majority of calls to this
656 * function will be for, has this expected length. For efficiency, set
657 * things up here to return it. It will be overriden only in those rare
658 * cases where a malformation is found */
663 /* An invariant is trivially well-formed */
664 if (UTF8_IS_INVARIANT(uv)) {
668 /* A continuation character can't start a valid sequence */
669 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
670 if (flags & UTF8_ALLOW_CONTINUATION) {
674 return UNICODE_REPLACEMENT;
677 if (! (flags & UTF8_CHECK_ONLY)) {
678 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
684 /* Here is not a continuation byte, nor an invariant. The only thing left
685 * is a start byte (possibly for an overlong) */
688 uv = NATIVE_UTF8_TO_I8(uv);
691 /* Remove the leading bits that indicate the number of bytes in the
692 * character's whole UTF-8 sequence, leaving just the bits that are part of
694 uv &= UTF_START_MASK(expectlen);
696 /* Now, loop through the remaining bytes in the character's sequence,
697 * accumulating each into the working value as we go. Be sure to not look
698 * past the end of the input string */
699 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
701 for (s = s0 + 1; s < send; s++) {
702 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
703 #ifndef EBCDIC /* Can't overflow in EBCDIC */
704 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
706 /* The original implementors viewed this malformation as more
707 * serious than the others (though I, khw, don't understand
708 * why, since other malformations also give very very wrong
709 * results), so there is no way to turn off checking for it.
710 * Set a flag, but keep going in the loop, so that we absorb
711 * the rest of the bytes that comprise the character. */
713 overflow_byte = *s; /* Save for warning message's use */
716 uv = UTF8_ACCUMULATE(uv, *s);
719 /* Here, found a non-continuation before processing all expected
720 * bytes. This byte begins a new character, so quit, even if
721 * allowing this malformation. */
722 unexpected_non_continuation = TRUE;
725 } /* End of loop through the character's bytes */
727 /* Save how many bytes were actually in the character */
730 /* The loop above finds two types of malformations: non-continuation and/or
731 * overflow. The non-continuation malformation is really a too-short
732 * malformation, as it means that the current character ended before it was
733 * expected to (being terminated prematurely by the beginning of the next
734 * character, whereas in the too-short malformation there just are too few
735 * bytes available to hold the character. In both cases, the check below
736 * that we have found the expected number of bytes would fail if executed.)
737 * Thus the non-continuation malformation is really unnecessary, being a
738 * subset of the too-short malformation. But there may be existing
739 * applications that are expecting the non-continuation type, so we retain
740 * it, and return it in preference to the too-short malformation. (If this
741 * code were being written from scratch, the two types might be collapsed
742 * into one.) I, khw, am also giving priority to returning the
743 * non-continuation and too-short malformations over overflow when multiple
744 * ones are present. I don't know of any real reason to prefer one over
745 * the other, except that it seems to me that multiple-byte errors trumps
746 * errors from a single byte */
747 if (UNLIKELY(unexpected_non_continuation)) {
748 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
749 if (! (flags & UTF8_CHECK_ONLY)) {
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
754 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));
759 uv = UNICODE_REPLACEMENT;
761 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
762 * as what the original expectations were. */
763 do_overlong_test = FALSE;
768 else if (UNLIKELY(curlen < expectlen)) {
769 if (! (flags & UTF8_ALLOW_SHORT)) {
770 if (! (flags & UTF8_CHECK_ONLY)) {
771 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));
775 uv = UNICODE_REPLACEMENT;
776 do_overlong_test = FALSE;
782 #ifndef EBCDIC /* EBCDIC can't overflow */
783 if (UNLIKELY(overflowed)) {
784 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
790 && expectlen > (STRLEN) OFFUNISKIP(uv)
791 && ! (flags & UTF8_ALLOW_LONG))
793 /* The overlong malformation has lower precedence than the others.
794 * Note that if this malformation is allowed, we return the actual
795 * value, instead of the replacement character. This is because this
796 * value is actually well-defined. */
797 if (! (flags & UTF8_CHECK_ONLY)) {
798 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));
803 /* Here, the input is considered to be well-formed, but it still could be a
804 * problematic code point that is not allowed by the input parameters. */
805 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
806 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
807 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
809 if (UNICODE_IS_SURROGATE(uv)) {
811 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
812 * generation of the sv, since no warnings are raised under CHECK */
813 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
814 && ckWARN_d(WARN_SURROGATE))
816 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
817 pack_warn = packWARN(WARN_SURROGATE);
819 if (flags & UTF8_DISALLOW_SURROGATE) {
823 else if ((uv > PERL_UNICODE_MAX)) {
824 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
825 && ckWARN_d(WARN_NON_UNICODE))
827 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
828 pack_warn = packWARN(WARN_NON_UNICODE);
830 #ifndef EBCDIC /* EBCDIC always allows FE, FF */
832 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
833 * points. We test for these after the regular SUPER ones, and
834 * before possibly bailing out, so that the more dire warning
835 * overrides the regular one, if applicable */
836 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
837 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
839 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
841 && ckWARN_d(WARN_UTF8))
843 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
844 pack_warn = packWARN(WARN_UTF8);
846 if (flags & UTF8_DISALLOW_FE_FF) {
851 if (flags & UTF8_DISALLOW_SUPER) {
855 else if (UNICODE_IS_NONCHAR(uv)) {
856 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
857 && ckWARN_d(WARN_NONCHAR))
859 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
860 pack_warn = packWARN(WARN_NONCHAR);
862 if (flags & UTF8_DISALLOW_NONCHAR) {
868 outlier_ret = uv; /* Note we don't bother to convert to native,
869 as all the outlier code points are the same
870 in both ASCII and EBCDIC */
874 /* Here, this is not considered a malformed character, so drop through
878 return UNI_TO_NATIVE(uv);
880 /* There are three cases which get to beyond this point. In all 3 cases:
881 * <sv> if not null points to a string to print as a warning.
882 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
884 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
885 * This is done by initializing it to 0, and changing it only
888 * 1) The input is valid but problematic, and to be warned about. The
889 * return value is the resultant code point; <*retlen> is set to
890 * <curlen>, the number of bytes that comprise the code point.
891 * <pack_warn> contains the result of packWARN() for the warning
892 * types. The entry point for this case is the label <do_warn>;
893 * 2) The input is a valid code point but disallowed by the parameters to
894 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
895 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
896 * comprise the code point. <pack_warn> contains the result of
897 * packWARN() for the warning types. The entry point for this case is
898 * the label <disallowed>.
899 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
900 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
901 * bytes that comprise the malformation. All such malformations are
902 * assumed to be warning type <utf8>. The entry point for this case
903 * is the label <malformed>.
908 if (sv && ckWARN_d(WARN_UTF8)) {
909 pack_warn = packWARN(WARN_UTF8);
914 if (flags & UTF8_CHECK_ONLY) {
916 *retlen = ((STRLEN) -1);
922 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
923 if warnings are to be raised. */
924 const char * const string = SvPVX_const(sv);
927 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
929 Perl_warner(aTHX_ pack_warn, "%s", string);
940 =for apidoc utf8_to_uvchr_buf
942 Returns the native code point of the first character in the string C<s> which
943 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
944 C<*retlen> will be set to the length, in bytes, of that character.
946 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
947 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
948 NULL) to -1. If those warnings are off, the computed value, if well-defined
949 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
950 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
951 the next possible position in C<s> that could begin a non-malformed character.
952 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
960 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
964 return utf8n_to_uvchr(s, send - s, retlen,
965 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
968 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
969 * there are no malformations in the input UTF-8 string C<s>. surrogates,
970 * non-character code points, and non-Unicode code points are allowed. */
973 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
975 UV expectlen = UTF8SKIP(s);
976 const U8* send = s + expectlen;
979 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
985 /* An invariant is trivially returned */
986 if (expectlen == 1) {
991 uv = NATIVE_UTF8_TO_I8(uv);
994 /* Remove the leading bits that indicate the number of bytes, leaving just
995 * the bits that are part of the value */
996 uv &= UTF_START_MASK(expectlen);
998 /* Now, loop through the remaining bytes, accumulating each into the
999 * working total as we go. (I khw tried unrolling the loop for up to 4
1000 * bytes, but there was no performance improvement) */
1001 for (++s; s < send; s++) {
1002 uv = UTF8_ACCUMULATE(uv, *s);
1005 return UNI_TO_NATIVE(uv);
1010 =for apidoc utf8_to_uvchr
1012 Returns the native code point of the first character in the string C<s>
1013 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1014 length, in bytes, of that character.
1016 Some, but not all, UTF-8 malformations are detected, and in fact, some
1017 malformed input could cause reading beyond the end of the input buffer, which
1018 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
1020 If C<s> points to one of the detected malformations, and UTF8 warnings are
1021 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1022 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1023 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1024 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1025 next possible position in C<s> that could begin a non-malformed character.
1026 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1032 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
1034 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
1036 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
1040 =for apidoc utf8_to_uvuni_buf
1042 Only in very rare circumstances should code need to be dealing in Unicode
1043 (as opposed to native) code points. In those few cases, use
1044 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1046 Returns the Unicode (not-native) code point of the first character in the
1048 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1049 C<retlen> will be set to the length, in bytes, of that character.
1051 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1052 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1053 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1054 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1055 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1056 next possible position in C<s> that could begin a non-malformed character.
1057 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1063 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1065 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1069 /* Call the low level routine asking for checks */
1070 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1071 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
1075 * Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1076 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1077 * non-character code points, and non-Unicode code points are allowed */
1080 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1082 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1084 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
1088 =for apidoc utf8_to_uvuni
1090 Returns the Unicode code point of the first character in the string C<s>
1091 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1092 length, in bytes, of that character.
1094 Some, but not all, UTF-8 malformations are detected, and in fact, some
1095 malformed input could cause reading beyond the end of the input buffer, which
1096 is one reason why this function is deprecated. The other is that only in
1097 extremely limited circumstances should the Unicode versus native code point be
1098 of any interest to you. See L</utf8_to_uvuni_buf> for alternatives.
1100 If C<s> points to one of the detected malformations, and UTF8 warnings are
1101 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1102 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1103 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1104 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1105 next possible position in C<s> that could begin a non-malformed character.
1106 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1112 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1114 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1116 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
1120 =for apidoc utf8_length
1122 Return the length of the UTF-8 char encoded string C<s> in characters.
1123 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1124 up past C<e>, croaks.
1130 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1135 PERL_ARGS_ASSERT_UTF8_LENGTH;
1137 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1138 * the bitops (especially ~) can create illegal UTF-8.
1139 * In other words: in Perl UTF-8 is not just for Unicode. */
1142 goto warn_and_return;
1152 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1153 "%s in %s", unees, OP_DESC(PL_op));
1155 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1162 =for apidoc utf8_distance
1164 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1167 WARNING: use only if you *know* that the pointers point inside the
1174 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1176 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1178 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1182 =for apidoc utf8_hop
1184 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1185 forward or backward.
1187 WARNING: do not use the following unless you *know* C<off> is within
1188 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1189 on the first byte of character or just after the last byte of a character.
1195 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1197 PERL_ARGS_ASSERT_UTF8_HOP;
1199 PERL_UNUSED_CONTEXT;
1200 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1201 * the bitops (especially ~) can create illegal UTF-8.
1202 * In other words: in Perl UTF-8 is not just for Unicode. */
1211 while (UTF8_IS_CONTINUATION(*s))
1219 =for apidoc bytes_cmp_utf8
1221 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1222 sequence of characters (stored as UTF-8)
1223 in C<u>, C<ulen>. Returns 0 if they are
1224 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1225 if the first string is greater than the second string.
1227 -1 or +1 is returned if the shorter string was identical to the start of the
1228 longer string. -2 or +2 is returned if
1229 there was a difference between characters
1236 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1238 const U8 *const bend = b + blen;
1239 const U8 *const uend = u + ulen;
1241 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1243 PERL_UNUSED_CONTEXT;
1245 while (b < bend && u < uend) {
1247 if (!UTF8_IS_INVARIANT(c)) {
1248 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1251 if (UTF8_IS_CONTINUATION(c1)) {
1252 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1254 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1255 "Malformed UTF-8 character "
1256 "(unexpected non-continuation byte 0x%02x"
1257 ", immediately after start byte 0x%02x)"
1258 /* Dear diag.t, it's in the pod. */
1260 PL_op ? " in " : "",
1261 PL_op ? OP_DESC(PL_op) : "");
1266 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1267 "%s in %s", unees, OP_DESC(PL_op));
1269 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1270 return -2; /* Really want to return undef :-) */
1277 return *b < c ? -2 : +2;
1282 if (b == bend && u == uend)
1285 return b < bend ? +1 : -1;
1289 =for apidoc utf8_to_bytes
1291 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1292 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1293 updates C<len> to contain the new length.
1294 Returns zero on failure, setting C<len> to -1.
1296 If you need a copy of the string, see L</bytes_from_utf8>.
1302 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1304 U8 * const save = s;
1305 U8 * const send = s + *len;
1308 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1310 /* ensure valid UTF-8 and chars < 256 before updating string */
1312 if (! UTF8_IS_INVARIANT(*s)) {
1313 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1314 *len = ((STRLEN) -1);
1325 if (! UTF8_IS_INVARIANT(c)) {
1326 /* Then it is two-byte encoded */
1327 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1338 =for apidoc bytes_from_utf8
1340 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1341 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1342 the newly-created string, and updates C<len> to contain the new
1343 length. Returns the original string if no conversion occurs, C<len>
1344 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1345 0 if C<s> is converted or consisted entirely of characters that are invariant
1346 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1352 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1355 const U8 *start = s;
1359 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1361 PERL_UNUSED_CONTEXT;
1365 /* ensure valid UTF-8 and chars < 256 before converting string */
1366 for (send = s + *len; s < send;) {
1367 if (! UTF8_IS_INVARIANT(*s)) {
1368 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1379 Newx(d, (*len) - count + 1, U8);
1380 s = start; start = d;
1383 if (! UTF8_IS_INVARIANT(c)) {
1384 /* Then it is two-byte encoded */
1385 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1396 =for apidoc bytes_to_utf8
1398 Converts a string C<s> of length C<len> bytes from the native encoding into
1400 Returns a pointer to the newly-created string, and sets C<len> to
1401 reflect the new length in bytes.
1403 A NUL character will be written after the end of the string.
1405 If you want to convert to UTF-8 from encodings other than
1406 the native (Latin1 or EBCDIC),
1407 see L</sv_recode_to_utf8>().
1412 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1413 likewise need duplication. */
1416 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1418 const U8 * const send = s + (*len);
1422 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1423 PERL_UNUSED_CONTEXT;
1425 Newx(d, (*len) * 2 + 1, U8);
1429 append_utf8_from_native_byte(*s, &d);
1438 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1440 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1441 * We optimize for native, for obvious reasons. */
1444 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1449 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1452 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1457 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1459 if (UNI_IS_INVARIANT(uv)) {
1460 *d++ = LATIN1_TO_NATIVE((U8) uv);
1463 if (uv <= MAX_UTF8_TWO_BYTE) {
1464 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1465 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1468 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1469 #define LAST_HIGH_SURROGATE 0xDBFF
1470 #define FIRST_LOW_SURROGATE 0xDC00
1471 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1472 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1474 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1476 UV low = (p[0] << 8) + p[1];
1478 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1479 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1480 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1481 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1483 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1484 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1487 d = uvoffuni_to_utf8_flags(d, uv, 0);
1490 *d++ = (U8)(( uv >> 12) | 0xe0);
1491 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1492 *d++ = (U8)(( uv & 0x3f) | 0x80);
1496 *d++ = (U8)(( uv >> 18) | 0xf0);
1497 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1498 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1499 *d++ = (U8)(( uv & 0x3f) | 0x80);
1504 *newlen = d - dstart;
1508 /* Note: this one is slightly destructive of the source. */
1511 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1514 U8* const send = s + bytelen;
1516 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1519 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1523 const U8 tmp = s[0];
1528 return utf16_to_utf8(p, d, bytelen, newlen);
1532 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1534 U8 tmpbuf[UTF8_MAXBYTES+1];
1535 uvchr_to_utf8(tmpbuf, c);
1536 return _is_utf8_FOO(classnum, tmpbuf);
1539 /* Internal function so we can deprecate the external one, and call
1540 this one from other deprecated functions in this file */
1542 PERL_STATIC_INLINE bool
1543 S_is_utf8_idfirst(pTHX_ const U8 *p)
1549 /* is_utf8_idstart would be more logical. */
1550 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1554 Perl_is_uni_idfirst(pTHX_ UV c)
1556 U8 tmpbuf[UTF8_MAXBYTES+1];
1557 uvchr_to_utf8(tmpbuf, c);
1558 return S_is_utf8_idfirst(aTHX_ tmpbuf);
1562 Perl__is_uni_perl_idcont(pTHX_ UV c)
1564 U8 tmpbuf[UTF8_MAXBYTES+1];
1565 uvchr_to_utf8(tmpbuf, c);
1566 return _is_utf8_perl_idcont(tmpbuf);
1570 Perl__is_uni_perl_idstart(pTHX_ UV c)
1572 U8 tmpbuf[UTF8_MAXBYTES+1];
1573 uvchr_to_utf8(tmpbuf, c);
1574 return _is_utf8_perl_idstart(tmpbuf);
1578 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1580 /* We have the latin1-range values compiled into the core, so just use
1581 * those, converting the result to utf8. The only difference between upper
1582 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1583 * either "SS" or "Ss". Which one to use is passed into the routine in
1584 * 'S_or_s' to avoid a test */
1586 UV converted = toUPPER_LATIN1_MOD(c);
1588 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1590 assert(S_or_s == 'S' || S_or_s == 's');
1592 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1593 characters in this range */
1594 *p = (U8) converted;
1599 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1600 * which it maps to one of them, so as to only have to have one check for
1601 * it in the main case */
1602 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1604 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1605 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1608 converted = GREEK_CAPITAL_LETTER_MU;
1610 case LATIN_SMALL_LETTER_SHARP_S:
1616 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1617 assert(0); /* NOTREACHED */
1621 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1622 *p = UTF8_TWO_BYTE_LO(converted);
1628 /* Call the function to convert a UTF-8 encoded character to the specified case.
1629 * Note that there may be more than one character in the result.
1630 * INP is a pointer to the first byte of the input character
1631 * OUTP will be set to the first byte of the string of changed characters. It
1632 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1633 * LENP will be set to the length in bytes of the string of changed characters
1635 * The functions return the ordinal of the first character in the string of OUTP */
1636 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1637 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1638 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1640 /* This additionally has the input parameter SPECIALS, which if non-zero will
1641 * cause this to use the SPECIALS hash for folding (meaning get full case
1642 * folding); otherwise, when zero, this implies a simple case fold */
1643 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1646 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1650 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1651 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1652 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1653 * the changed version may be longer than the original character.
1655 * The ordinal of the first character of the changed version is returned
1656 * (but note, as explained above, that there may be more.) */
1658 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1661 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1664 uvchr_to_utf8(p, c);
1665 return CALL_UPPER_CASE(p, p, lenp);
1669 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1673 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1676 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1679 uvchr_to_utf8(p, c);
1680 return CALL_TITLE_CASE(p, p, lenp);
1684 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1686 /* We have the latin1-range values compiled into the core, so just use
1687 * those, converting the result to utf8. Since the result is always just
1688 * one character, we allow <p> to be NULL */
1690 U8 converted = toLOWER_LATIN1(c);
1693 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1698 *p = UTF8_TWO_BYTE_HI(converted);
1699 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1707 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1711 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1714 return to_lower_latin1((U8) c, p, lenp);
1717 uvchr_to_utf8(p, c);
1718 return CALL_LOWER_CASE(p, p, lenp);
1722 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1724 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1725 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1726 * FOLD_FLAGS_FULL iff full folding is to be used;
1728 * Not to be used for locale folds
1733 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1735 assert (! (flags & FOLD_FLAGS_LOCALE));
1737 if (c == MICRO_SIGN) {
1738 converted = GREEK_SMALL_LETTER_MU;
1740 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1742 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1743 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1744 * under those circumstances. */
1745 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1746 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1747 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1749 return LATIN_SMALL_LETTER_LONG_S;
1758 else { /* In this range the fold of all other characters is their lower
1760 converted = toLOWER_LATIN1(c);
1763 if (UVCHR_IS_INVARIANT(converted)) {
1764 *p = (U8) converted;
1768 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1769 *p = UTF8_TWO_BYTE_LO(converted);
1777 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
1780 /* Not currently externally documented, and subject to change
1781 * <flags> bits meanings:
1782 * FOLD_FLAGS_FULL iff full folding is to be used;
1783 * FOLD_FLAGS_LOCALE iff in locale
1784 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1787 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1790 UV result = _to_fold_latin1((U8) c, p, lenp,
1791 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1792 /* It is illegal for the fold to cross the 255/256 boundary under
1793 * locale; in this case return the original */
1794 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1799 /* If no special needs, just use the macro */
1800 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1801 uvchr_to_utf8(p, c);
1802 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1804 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1805 the special flags. */
1806 U8 utf8_c[UTF8_MAXBYTES + 1];
1807 uvchr_to_utf8(utf8_c, c);
1808 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1812 PERL_STATIC_INLINE bool
1813 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1814 const char *const swashname, SV* const invlist)
1816 /* returns a boolean giving whether or not the UTF8-encoded character that
1817 * starts at <p> is in the swash indicated by <swashname>. <swash>
1818 * contains a pointer to where the swash indicated by <swashname>
1819 * is to be stored; which this routine will do, so that future calls will
1820 * look at <*swash> and only generate a swash if it is not null. <invlist>
1821 * is NULL or an inversion list that defines the swash. If not null, it
1822 * saves time during initialization of the swash.
1824 * Note that it is assumed that the buffer length of <p> is enough to
1825 * contain all the bytes that comprise the character. Thus, <*p> should
1826 * have been checked before this call for mal-formedness enough to assure
1831 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1833 /* The API should have included a length for the UTF-8 character in <p>,
1834 * but it doesn't. We therefore assume that p has been validated at least
1835 * as far as there being enough bytes available in it to accommodate the
1836 * character without reading beyond the end, and pass that number on to the
1837 * validating routine */
1838 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
1839 if (ckWARN_d(WARN_UTF8)) {
1840 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1841 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1842 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1843 what the malformation is */
1844 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1850 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1851 *swash = _core_swash_init("utf8",
1853 /* Only use the name if there is no inversion
1854 * list; otherwise will go out to disk */
1855 (invlist) ? "" : swashname,
1857 &PL_sv_undef, 1, 0, invlist, &flags);
1860 return swash_fetch(*swash, p, TRUE) != 0;
1864 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1868 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1870 assert(classnum < _FIRST_NON_SWASH_CC);
1872 return is_utf8_common(p,
1873 &PL_utf8_swash_ptrs[classnum],
1874 swash_property_names[classnum],
1875 PL_XPosix_ptrs[classnum]);
1879 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
1883 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1885 return S_is_utf8_idfirst(aTHX_ p);
1889 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1893 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1897 /* is_utf8_idstart would be more logical. */
1898 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1902 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1907 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1909 if (! PL_utf8_perl_idstart) {
1910 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1912 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
1916 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1921 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1923 if (! PL_utf8_perl_idcont) {
1924 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1926 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
1931 Perl_is_utf8_idcont(pTHX_ const U8 *p)
1935 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
1937 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1941 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
1945 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
1947 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1951 Perl__is_utf8_mark(pTHX_ const U8 *p)
1955 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1957 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1961 =for apidoc to_utf8_case
1963 C<p> contains the pointer to the UTF-8 string encoding
1964 the character that is being converted. This routine assumes that the character
1965 at C<p> is well-formed.
1967 C<ustrp> is a pointer to the character buffer to put the
1968 conversion result to. C<lenp> is a pointer to the length
1971 C<swashp> is a pointer to the swash to use.
1973 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1974 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
1975 but not always, a multicharacter mapping), is tried first.
1977 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1978 any special mappings; C<""> means to use the special mappings. Values other
1979 than these two are treated as the name of the hash containing the special
1980 mappings, like C<"utf8::ToSpecLower">.
1982 C<normal> is a string like "ToLower" which means the swash
1988 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1989 SV **swashp, const char *normal, const char *special)
1993 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1995 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1997 /* Note that swash_fetch() doesn't output warnings for these because it
1998 * assumes we will */
1999 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2000 if (uv1 <= UNICODE_SURROGATE_LAST) {
2001 if (ckWARN_d(WARN_SURROGATE)) {
2002 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2003 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2004 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2007 else if (UNICODE_IS_SUPER(uv1)) {
2008 if (ckWARN_d(WARN_NON_UNICODE)) {
2009 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2010 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2011 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2015 /* Note that non-characters are perfectly legal, so no warning should
2019 if (!*swashp) /* load on-demand */
2020 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2023 /* It might be "special" (sometimes, but not always,
2024 * a multicharacter mapping) */
2028 /* If passed in the specials name, use that; otherwise use any
2029 * given in the swash */
2030 if (*special != '\0') {
2031 hv = get_hv(special, 0);
2034 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2036 hv = MUTABLE_HV(SvRV(*svp));
2041 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
2046 s = SvPV_const(*svp, len);
2049 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2051 Copy(s, ustrp, len, U8);
2056 if (!len && *swashp) {
2057 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
2060 /* It was "normal" (a single character mapping). */
2061 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2069 return valid_utf8_to_uvchr(ustrp, 0);
2072 /* Here, there was no mapping defined, which means that the code point maps
2073 * to itself. Return the inputs */
2075 if (p != ustrp) { /* Don't copy onto itself */
2076 Copy(p, ustrp, len, U8);
2087 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2089 /* This is called when changing the case of a utf8-encoded character above
2090 * the Latin1 range, and the operation is in locale. If the result
2091 * contains a character that crosses the 255/256 boundary, disallow the
2092 * change, and return the original code point. See L<perlfunc/lc> for why;
2094 * p points to the original string whose case was changed; assumed
2095 * by this routine to be well-formed
2096 * result the code point of the first character in the changed-case string
2097 * ustrp points to the changed-case string (<result> represents its first char)
2098 * lenp points to the length of <ustrp> */
2100 UV original; /* To store the first code point of <p> */
2102 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2104 assert(UTF8_IS_ABOVE_LATIN1(*p));
2106 /* We know immediately if the first character in the string crosses the
2107 * boundary, so can skip */
2110 /* Look at every character in the result; if any cross the
2111 * boundary, the whole thing is disallowed */
2112 U8* s = ustrp + UTF8SKIP(ustrp);
2113 U8* e = ustrp + *lenp;
2115 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2121 /* Here, no characters crossed, result is ok as-is */
2127 /* Failed, have to return the original */
2128 original = valid_utf8_to_uvchr(p, lenp);
2129 Copy(p, ustrp, *lenp, char);
2134 =for apidoc to_utf8_upper
2136 Instead use L</toUPPER_utf8>.
2140 /* Not currently externally documented, and subject to change:
2141 * <flags> is set iff locale semantics are to be used for code points < 256
2142 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2143 * were used in the calculation; otherwise unchanged. */
2146 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2152 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2154 if (UTF8_IS_INVARIANT(*p)) {
2156 result = toUPPER_LC(*p);
2159 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2162 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2164 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2165 result = toUPPER_LC(c);
2168 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2172 else { /* utf8, ord above 255 */
2173 result = CALL_UPPER_CASE(p, ustrp, lenp);
2176 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2181 /* Here, used locale rules. Convert back to utf8 */
2182 if (UTF8_IS_INVARIANT(result)) {
2183 *ustrp = (U8) result;
2187 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2188 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2193 *tainted_ptr = TRUE;
2199 =for apidoc to_utf8_title
2201 Instead use L</toTITLE_utf8>.
2205 /* Not currently externally documented, and subject to change:
2206 * <flags> is set iff locale semantics are to be used for code points < 256
2207 * Since titlecase is not defined in POSIX, uppercase is used instead
2209 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2210 * were used in the calculation; otherwise unchanged. */
2213 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2219 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2221 if (UTF8_IS_INVARIANT(*p)) {
2223 result = toUPPER_LC(*p);
2226 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2229 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2231 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2232 result = toUPPER_LC(c);
2235 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2239 else { /* utf8, ord above 255 */
2240 result = CALL_TITLE_CASE(p, ustrp, lenp);
2243 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2248 /* Here, used locale rules. Convert back to utf8 */
2249 if (UTF8_IS_INVARIANT(result)) {
2250 *ustrp = (U8) result;
2254 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2255 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2260 *tainted_ptr = TRUE;
2266 =for apidoc to_utf8_lower
2268 Instead use L</toLOWER_utf8>.
2272 /* Not currently externally documented, and subject to change:
2273 * <flags> is set iff locale semantics are to be used for code points < 256
2274 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2275 * were used in the calculation; otherwise unchanged. */
2278 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2284 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2286 if (UTF8_IS_INVARIANT(*p)) {
2288 result = toLOWER_LC(*p);
2291 return to_lower_latin1(*p, ustrp, lenp);
2294 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2296 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2297 result = toLOWER_LC(c);
2300 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2304 else { /* utf8, ord above 255 */
2305 result = CALL_LOWER_CASE(p, ustrp, lenp);
2308 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2314 /* Here, used locale rules. Convert back to utf8 */
2315 if (UTF8_IS_INVARIANT(result)) {
2316 *ustrp = (U8) result;
2320 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2321 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2326 *tainted_ptr = TRUE;
2332 =for apidoc to_utf8_fold
2334 Instead use L</toFOLD_utf8>.
2338 /* Not currently externally documented, and subject to change,
2340 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2341 * points < 256. Since foldcase is not defined in
2342 * POSIX, lowercase is used instead
2343 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2344 * otherwise simple folds
2345 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2347 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2348 * were used in the calculation; otherwise unchanged. */
2351 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2357 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2359 /* These are mutually exclusive */
2360 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2362 assert(p != ustrp); /* Otherwise overwrites */
2364 if (UTF8_IS_INVARIANT(*p)) {
2365 if (flags & FOLD_FLAGS_LOCALE) {
2366 result = toFOLD_LC(*p);
2369 return _to_fold_latin1(*p, ustrp, lenp,
2370 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2373 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2374 if (flags & FOLD_FLAGS_LOCALE) {
2375 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2376 result = toFOLD_LC(c);
2379 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2381 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2384 else { /* utf8, ord above 255 */
2385 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2387 if (flags & FOLD_FLAGS_LOCALE) {
2389 /* Special case these two characters, as what normally gets
2390 * returned under locale doesn't work */
2391 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2392 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2393 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2397 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2398 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2399 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2401 goto return_ligature_st;
2403 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2405 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2409 /* This is called when changing the case of a utf8-encoded
2410 * character above the ASCII range, and the result should not
2411 * contain an ASCII character. */
2413 UV original; /* To store the first code point of <p> */
2415 /* Look at every character in the result; if any cross the
2416 * boundary, the whole thing is disallowed */
2418 U8* e = ustrp + *lenp;
2421 /* Crossed, have to return the original */
2422 original = valid_utf8_to_uvchr(p, lenp);
2424 /* But in these instances, there is an alternative we can
2425 * return that is valid */
2426 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2427 || original == LATIN_SMALL_LETTER_SHARP_S)
2431 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2432 goto return_ligature_st;
2434 Copy(p, ustrp, *lenp, char);
2440 /* Here, no characters crossed, result is ok as-is */
2445 /* Here, used locale rules. Convert back to utf8 */
2446 if (UTF8_IS_INVARIANT(result)) {
2447 *ustrp = (U8) result;
2451 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2452 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2457 *tainted_ptr = TRUE;
2462 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2463 * folds to a string of two of these characters. By returning this
2464 * instead, then, e.g.,
2465 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2468 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2469 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2471 return LATIN_SMALL_LETTER_LONG_S;
2474 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2475 * have the other one fold to it */
2477 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2478 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2479 return LATIN_SMALL_LIGATURE_ST;
2483 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2484 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2485 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2489 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2491 PERL_ARGS_ASSERT_SWASH_INIT;
2493 /* Returns a copy of a swash initiated by the called function. This is the
2494 * public interface, and returning a copy prevents others from doing
2495 * mischief on the original */
2497 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2501 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2503 /* Initialize and return a swash, creating it if necessary. It does this
2504 * by calling utf8_heavy.pl in the general case. The returned value may be
2505 * the swash's inversion list instead if the input parameters allow it.
2506 * Which is returned should be immaterial to callers, as the only
2507 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2508 * and swash_to_invlist() handle both these transparently.
2510 * This interface should only be used by functions that won't destroy or
2511 * adversely change the swash, as doing so affects all other uses of the
2512 * swash in the program; the general public should use 'Perl_swash_init'
2515 * pkg is the name of the package that <name> should be in.
2516 * name is the name of the swash to find. Typically it is a Unicode
2517 * property name, including user-defined ones
2518 * listsv is a string to initialize the swash with. It must be of the form
2519 * documented as the subroutine return value in
2520 * L<perlunicode/User-Defined Character Properties>
2521 * minbits is the number of bits required to represent each data element.
2522 * It is '1' for binary properties.
2523 * none I (khw) do not understand this one, but it is used only in tr///.
2524 * invlist is an inversion list to initialize the swash with (or NULL)
2525 * flags_p if non-NULL is the address of various input and output flag bits
2526 * to the routine, as follows: ('I' means is input to the routine;
2527 * 'O' means output from the routine. Only flags marked O are
2528 * meaningful on return.)
2529 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2530 * came from a user-defined property. (I O)
2531 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2532 * when the swash cannot be located, to simply return NULL. (I)
2533 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2534 * return of an inversion list instead of a swash hash if this routine
2535 * thinks that would result in faster execution of swash_fetch() later
2538 * Thus there are three possible inputs to find the swash: <name>,
2539 * <listsv>, and <invlist>. At least one must be specified. The result
2540 * will be the union of the specified ones, although <listsv>'s various
2541 * actions can intersect, etc. what <name> gives. To avoid going out to
2542 * disk at all, <invlist> should specify completely what the swash should
2543 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2545 * <invlist> is only valid for binary properties */
2548 SV* retval = &PL_sv_undef;
2549 HV* swash_hv = NULL;
2550 const int invlist_swash_boundary =
2551 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2552 ? 512 /* Based on some benchmarking, but not extensive, see commit
2554 : -1; /* Never return just an inversion list */
2556 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2557 assert(! invlist || minbits == 1);
2559 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2561 if (listsv != &PL_sv_undef || strNE(name, "")) {
2563 const size_t pkg_len = strlen(pkg);
2564 const size_t name_len = strlen(name);
2565 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2569 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2571 PUSHSTACKi(PERLSI_MAGIC);
2575 /* We might get here via a subroutine signature which uses a utf8
2576 * parameter name, at which point PL_subname will have been set
2577 * but not yet used. */
2578 save_item(PL_subname);
2579 if (PL_parser && PL_parser->error_count)
2580 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2581 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2582 if (!method) { /* demand load utf8 */
2584 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2585 GvSV(PL_errgv) = NULL;
2586 /* It is assumed that callers of this routine are not passing in
2587 * any user derived data. */
2588 /* Need to do this after save_re_context() as it will set
2589 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2590 * in Perl_magic_get). Even line to create errsv_save can turn on
2592 #ifndef NO_TAINT_SUPPORT
2593 SAVEBOOL(TAINT_get);
2596 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2599 /* Not ERRSV, as there is no need to vivify a scalar we are
2600 about to discard. */
2601 SV * const errsv = GvSV(PL_errgv);
2602 if (!SvTRUE(errsv)) {
2603 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2604 SvREFCNT_dec(errsv);
2612 mPUSHp(pkg, pkg_len);
2613 mPUSHp(name, name_len);
2618 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2619 GvSV(PL_errgv) = NULL;
2620 /* If we already have a pointer to the method, no need to use
2621 * call_method() to repeat the lookup. */
2623 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2624 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2626 retval = *PL_stack_sp--;
2627 SvREFCNT_inc(retval);
2630 /* Not ERRSV. See above. */
2631 SV * const errsv = GvSV(PL_errgv);
2632 if (!SvTRUE(errsv)) {
2633 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2634 SvREFCNT_dec(errsv);
2639 if (IN_PERL_COMPILETIME) {
2640 CopHINTS_set(PL_curcop, PL_hints);
2642 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2645 /* If caller wants to handle missing properties, let them */
2646 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2650 "Can't find Unicode property definition \"%"SVf"\"",
2652 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2654 } /* End of calling the module to find the swash */
2656 /* If this operation fetched a swash, and we will need it later, get it */
2657 if (retval != &PL_sv_undef
2658 && (minbits == 1 || (flags_p
2660 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2662 swash_hv = MUTABLE_HV(SvRV(retval));
2664 /* If we don't already know that there is a user-defined component to
2665 * this swash, and the user has indicated they wish to know if there is
2666 * one (by passing <flags_p>), find out */
2667 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2668 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2669 if (user_defined && SvUV(*user_defined)) {
2670 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2675 /* Make sure there is an inversion list for binary properties */
2677 SV** swash_invlistsvp = NULL;
2678 SV* swash_invlist = NULL;
2679 bool invlist_in_swash_is_valid = FALSE;
2680 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2681 an unclaimed reference count */
2683 /* If this operation fetched a swash, get its already existing
2684 * inversion list, or create one for it */
2687 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2688 if (swash_invlistsvp) {
2689 swash_invlist = *swash_invlistsvp;
2690 invlist_in_swash_is_valid = TRUE;
2693 swash_invlist = _swash_to_invlist(retval);
2694 swash_invlist_unclaimed = TRUE;
2698 /* If an inversion list was passed in, have to include it */
2701 /* Any fetched swash will by now have an inversion list in it;
2702 * otherwise <swash_invlist> will be NULL, indicating that we
2703 * didn't fetch a swash */
2704 if (swash_invlist) {
2706 /* Add the passed-in inversion list, which invalidates the one
2707 * already stored in the swash */
2708 invlist_in_swash_is_valid = FALSE;
2709 _invlist_union(invlist, swash_invlist, &swash_invlist);
2713 /* Here, there is no swash already. Set up a minimal one, if
2714 * we are going to return a swash */
2715 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2717 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2719 swash_invlist = invlist;
2723 /* Here, we have computed the union of all the passed-in data. It may
2724 * be that there was an inversion list in the swash which didn't get
2725 * touched; otherwise save the computed one */
2726 if (! invlist_in_swash_is_valid
2727 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2729 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2731 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2733 /* We just stole a reference count. */
2734 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2735 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2738 SvREADONLY_on(swash_invlist);
2740 /* Use the inversion list stand-alone if small enough */
2741 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2742 SvREFCNT_dec(retval);
2743 if (!swash_invlist_unclaimed)
2744 SvREFCNT_inc_simple_void_NN(swash_invlist);
2745 retval = newRV_noinc(swash_invlist);
2753 /* This API is wrong for special case conversions since we may need to
2754 * return several Unicode characters for a single Unicode character
2755 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2756 * the lower-level routine, and it is similarly broken for returning
2757 * multiple values. --jhi
2758 * For those, you should use to_utf8_case() instead */
2759 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2762 * Returns the value of property/mapping C<swash> for the first character
2763 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2764 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2765 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2767 * A "swash" is a hash which contains initially the keys/values set up by
2768 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2769 * property for all possible code points. Things are stored in a compact form
2770 * (see utf8_heavy.pl) so that calculation is required to find the actual
2771 * property value for a given code point. As code points are looked up, new
2772 * key/value pairs are added to the hash, so that the calculation doesn't have
2773 * to ever be re-done. Further, each calculation is done, not just for the
2774 * desired one, but for a whole block of code points adjacent to that one.
2775 * For binary properties on ASCII machines, the block is usually for 64 code
2776 * points, starting with a code point evenly divisible by 64. Thus if the
2777 * property value for code point 257 is requested, the code goes out and
2778 * calculates the property values for all 64 code points between 256 and 319,
2779 * and stores these as a single 64-bit long bit vector, called a "swatch",
2780 * under the key for code point 256. The key is the UTF-8 encoding for code
2781 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2782 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2783 * for code point 258 is then requested, this code realizes that it would be
2784 * stored under the key for 256, and would find that value and extract the
2785 * relevant bit, offset from 256.
2787 * Non-binary properties are stored in as many bits as necessary to represent
2788 * their values (32 currently, though the code is more general than that), not
2789 * as single bits, but the principal is the same: the value for each key is a
2790 * vector that encompasses the property values for all code points whose UTF-8
2791 * representations are represented by the key. That is, for all code points
2792 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2796 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2799 HV *const hv = MUTABLE_HV(SvRV(swash));
2804 const U8 *tmps = NULL;
2809 PERL_ARGS_ASSERT_SWASH_FETCH;
2811 /* If it really isn't a hash, it isn't really swash; must be an inversion
2813 if (SvTYPE(hv) != SVt_PVHV) {
2814 return _invlist_contains_cp((SV*)hv,
2816 ? valid_utf8_to_uvchr(ptr, NULL)
2820 /* We store the values in a "swatch" which is a vec() value in a swash
2821 * hash. Code points 0-255 are a single vec() stored with key length
2822 * (klen) 0. All other code points have a UTF-8 representation
2823 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2824 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2825 * length for them is the length of the encoded char - 1. ptr[klen] is the
2826 * final byte in the sequence representing the character */
2827 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2832 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2835 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
2838 klen = UTF8SKIP(ptr) - 1;
2840 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2841 * the vec is the final byte in the sequence. (In EBCDIC this is
2842 * converted to I8 to get consecutive values.) To help you visualize
2844 * Straight 1047 After final byte
2845 * UTF-8 UTF-EBCDIC I8 transform
2846 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2847 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2849 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2850 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2852 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2853 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2855 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2856 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2858 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2859 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2861 * (There are no discontinuities in the elided (...) entries.)
2862 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2863 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2864 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2865 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2866 * index into the vec() swatch (after subtracting 0x80, which we
2867 * actually do with an '&').
2868 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2869 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2870 * dicontinuities which go away by transforming it into I8, and we
2871 * effectively subtract 0xA0 to get the index. */
2872 needents = (1 << UTF_ACCUMULATION_SHIFT);
2873 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2877 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2878 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2879 * it's nothing to sniff at.) Pity we usually come through at least
2880 * two function calls to get here...
2882 * NB: this code assumes that swatches are never modified, once generated!
2885 if (hv == PL_last_swash_hv &&
2886 klen == PL_last_swash_klen &&
2887 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2889 tmps = PL_last_swash_tmps;
2890 slen = PL_last_swash_slen;
2893 /* Try our second-level swatch cache, kept in a hash. */
2894 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2896 /* If not cached, generate it via swatch_get */
2897 if (!svp || !SvPOK(*svp)
2898 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2901 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2902 swatch = swatch_get(swash,
2903 code_point & ~((UV)needents - 1),
2906 else { /* For the first 256 code points, the swatch has a key of
2908 swatch = swatch_get(swash, 0, needents);
2911 if (IN_PERL_COMPILETIME)
2912 CopHINTS_set(PL_curcop, PL_hints);
2914 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2916 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2917 || (slen << 3) < needents)
2918 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2919 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2920 svp, tmps, (UV)slen, (UV)needents);
2923 PL_last_swash_hv = hv;
2924 assert(klen <= sizeof(PL_last_swash_key));
2925 PL_last_swash_klen = (U8)klen;
2926 /* FIXME change interpvar.h? */
2927 PL_last_swash_tmps = (U8 *) tmps;
2928 PL_last_swash_slen = slen;
2930 Copy(ptr, PL_last_swash_key, klen, U8);
2933 switch ((int)((slen << 3) / needents)) {
2935 bit = 1 << (off & 7);
2937 return (tmps[off] & bit) != 0;
2942 return (tmps[off] << 8) + tmps[off + 1] ;
2945 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
2947 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2948 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2949 NORETURN_FUNCTION_END;
2952 /* Read a single line of the main body of the swash input text. These are of
2955 * where each number is hex. The first two numbers form the minimum and
2956 * maximum of a range, and the third is the value associated with the range.
2957 * Not all swashes should have a third number
2959 * On input: l points to the beginning of the line to be examined; it points
2960 * to somewhere in the string of the whole input text, and is
2961 * terminated by a \n or the null string terminator.
2962 * lend points to the null terminator of that string
2963 * wants_value is non-zero if the swash expects a third number
2964 * typestr is the name of the swash's mapping, like 'ToLower'
2965 * On output: *min, *max, and *val are set to the values read from the line.
2966 * returns a pointer just beyond the line examined. If there was no
2967 * valid min number on the line, returns lend+1
2971 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2972 const bool wants_value, const U8* const typestr)
2974 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2975 STRLEN numlen; /* Length of the number */
2976 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2977 | PERL_SCAN_DISALLOW_PREFIX
2978 | PERL_SCAN_SILENT_NON_PORTABLE;
2980 /* nl points to the next \n in the scan */
2981 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2983 /* Get the first number on the line: the range minimum */
2985 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2986 if (numlen) /* If found a hex number, position past it */
2988 else if (nl) { /* Else, go handle next line, if any */
2989 return nl + 1; /* 1 is length of "\n" */
2991 else { /* Else, no next line */
2992 return lend + 1; /* to LIST's end at which \n is not found */
2995 /* The max range value follows, separated by a BLANK */
2998 flags = PERL_SCAN_SILENT_ILLDIGIT
2999 | PERL_SCAN_DISALLOW_PREFIX
3000 | PERL_SCAN_SILENT_NON_PORTABLE;
3002 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3005 else /* If no value here, it is a single element range */
3008 /* Non-binary tables have a third entry: what the first element of the
3009 * range maps to. The map for those currently read here is in hex */
3013 flags = PERL_SCAN_SILENT_ILLDIGIT
3014 | PERL_SCAN_DISALLOW_PREFIX
3015 | PERL_SCAN_SILENT_NON_PORTABLE;
3017 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3026 /* diag_listed_as: To%s: illegal mapping '%s' */
3027 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3033 *val = 0; /* bits == 1, then any val should be ignored */
3035 else { /* Nothing following range min, should be single element with no
3041 /* diag_listed_as: To%s: illegal mapping '%s' */
3042 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3046 *val = 0; /* bits == 1, then val should be ignored */
3049 /* Position to next line if any, or EOF */
3059 * Returns a swatch (a bit vector string) for a code point sequence
3060 * that starts from the value C<start> and comprises the number C<span>.
3061 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3062 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3065 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3068 U8 *l, *lend, *x, *xend, *s, *send;
3069 STRLEN lcur, xcur, scur;
3070 HV *const hv = MUTABLE_HV(SvRV(swash));
3071 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3073 SV** listsvp = NULL; /* The string containing the main body of the table */
3074 SV** extssvp = NULL;
3075 SV** invert_it_svp = NULL;
3078 STRLEN octets; /* if bits == 1, then octets == 0 */
3080 UV end = start + span;
3082 if (invlistsvp == NULL) {
3083 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3084 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3085 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3086 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3087 listsvp = hv_fetchs(hv, "LIST", FALSE);
3088 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3090 bits = SvUV(*bitssvp);
3091 none = SvUV(*nonesvp);
3092 typestr = (U8*)SvPV_nolen(*typesvp);
3098 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3100 PERL_ARGS_ASSERT_SWATCH_GET;
3102 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3103 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3107 /* If overflowed, use the max possible */
3113 /* create and initialize $swatch */
3114 scur = octets ? (span * octets) : (span + 7) / 8;
3115 swatch = newSV(scur);
3117 s = (U8*)SvPVX(swatch);
3118 if (octets && none) {
3119 const U8* const e = s + scur;
3122 *s++ = (U8)(none & 0xff);
3123 else if (bits == 16) {
3124 *s++ = (U8)((none >> 8) & 0xff);
3125 *s++ = (U8)( none & 0xff);
3127 else if (bits == 32) {
3128 *s++ = (U8)((none >> 24) & 0xff);
3129 *s++ = (U8)((none >> 16) & 0xff);
3130 *s++ = (U8)((none >> 8) & 0xff);
3131 *s++ = (U8)( none & 0xff);
3137 (void)memzero((U8*)s, scur + 1);
3139 SvCUR_set(swatch, scur);
3140 s = (U8*)SvPVX(swatch);
3142 if (invlistsvp) { /* If has an inversion list set up use that */
3143 _invlist_populate_swatch(*invlistsvp, start, end, s);
3147 /* read $swash->{LIST} */
3148 l = (U8*)SvPV(*listsvp, lcur);
3151 UV min, max, val, upper;
3152 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3153 cBOOL(octets), typestr);
3158 /* If looking for something beyond this range, go try the next one */
3162 /* <end> is generally 1 beyond where we want to set things, but at the
3163 * platform's infinity, where we can't go any higher, we want to
3164 * include the code point at <end> */
3167 : (max != UV_MAX || end != UV_MAX)
3174 if (!none || val < none) {
3179 for (key = min; key <= upper; key++) {
3181 /* offset must be non-negative (start <= min <= key < end) */
3182 offset = octets * (key - start);
3184 s[offset] = (U8)(val & 0xff);
3185 else if (bits == 16) {
3186 s[offset ] = (U8)((val >> 8) & 0xff);
3187 s[offset + 1] = (U8)( val & 0xff);
3189 else if (bits == 32) {
3190 s[offset ] = (U8)((val >> 24) & 0xff);
3191 s[offset + 1] = (U8)((val >> 16) & 0xff);
3192 s[offset + 2] = (U8)((val >> 8) & 0xff);
3193 s[offset + 3] = (U8)( val & 0xff);
3196 if (!none || val < none)
3200 else { /* bits == 1, then val should be ignored */
3205 for (key = min; key <= upper; key++) {
3206 const STRLEN offset = (STRLEN)(key - start);
3207 s[offset >> 3] |= 1 << (offset & 7);
3212 /* Invert if the data says it should be. Assumes that bits == 1 */
3213 if (invert_it_svp && SvUV(*invert_it_svp)) {
3215 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3216 * be 0, and their inversion should also be 0, as we don't succeed any
3217 * Unicode property matches for non-Unicode code points */
3218 if (start <= PERL_UNICODE_MAX) {
3220 /* The code below assumes that we never cross the
3221 * Unicode/above-Unicode boundary in a range, as otherwise we would
3222 * have to figure out where to stop flipping the bits. Since this
3223 * boundary is divisible by a large power of 2, and swatches comes
3224 * in small powers of 2, this should be a valid assumption */
3225 assert(start + span - 1 <= PERL_UNICODE_MAX);
3235 /* read $swash->{EXTRAS}
3236 * This code also copied to swash_to_invlist() below */
3237 x = (U8*)SvPV(*extssvp, xcur);
3245 SV **otherbitssvp, *other;
3249 const U8 opc = *x++;
3253 nl = (U8*)memchr(x, '\n', xend - x);
3255 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3257 x = nl + 1; /* 1 is length of "\n" */
3261 x = xend; /* to EXTRAS' end at which \n is not found */
3268 namelen = nl - namestr;
3272 namelen = xend - namestr;
3276 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3277 otherhv = MUTABLE_HV(SvRV(*othersvp));
3278 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3279 otherbits = (STRLEN)SvUV(*otherbitssvp);
3280 if (bits < otherbits)
3281 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3282 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3284 /* The "other" swatch must be destroyed after. */
3285 other = swatch_get(*othersvp, start, span);
3286 o = (U8*)SvPV(other, olen);
3289 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3291 s = (U8*)SvPV(swatch, slen);
3292 if (bits == 1 && otherbits == 1) {
3294 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3295 "mismatch, slen=%"UVuf", olen=%"UVuf,
3296 (UV)slen, (UV)olen);
3320 STRLEN otheroctets = otherbits >> 3;
3322 U8* const send = s + slen;
3327 if (otherbits == 1) {
3328 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3332 STRLEN vlen = otheroctets;
3340 if (opc == '+' && otherval)
3341 NOOP; /* replace with otherval */
3342 else if (opc == '!' && !otherval)
3344 else if (opc == '-' && otherval)
3346 else if (opc == '&' && !otherval)
3349 s += octets; /* no replacement */
3354 *s++ = (U8)( otherval & 0xff);
3355 else if (bits == 16) {
3356 *s++ = (U8)((otherval >> 8) & 0xff);
3357 *s++ = (U8)( otherval & 0xff);
3359 else if (bits == 32) {
3360 *s++ = (U8)((otherval >> 24) & 0xff);
3361 *s++ = (U8)((otherval >> 16) & 0xff);
3362 *s++ = (U8)((otherval >> 8) & 0xff);
3363 *s++ = (U8)( otherval & 0xff);
3367 sv_free(other); /* through with it! */
3373 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3376 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3377 * Can't be used on a property that is subject to user override, as it
3378 * relies on the value of SPECIALS in the swash which would be set by
3379 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3380 * for overridden properties
3382 * Returns a hash which is the inversion and closure of a swash mapping.
3383 * For example, consider the input lines:
3388 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3389 * 006C. The value for each key is an array. For 006C, the array would
3390 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3391 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3393 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3394 * keys are only code points that are folded-to, so it isn't a full closure.
3396 * Essentially, for any code point, it gives all the code points that map to
3397 * it, or the list of 'froms' for that point.
3399 * Currently it ignores any additions or deletions from other swashes,
3400 * looking at just the main body of the swash, and if there are SPECIALS
3401 * in the swash, at that hash
3403 * The specials hash can be extra code points, and most likely consists of
3404 * maps from single code points to multiple ones (each expressed as a string
3405 * of utf8 characters). This function currently returns only 1-1 mappings.
3406 * However consider this possible input in the specials hash:
3407 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3408 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3410 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3411 * currently handle. But it also means that FB05 and FB06 are equivalent in
3412 * a 1-1 mapping which we should handle, and this relationship may not be in
3413 * the main table. Therefore this function examines all the multi-char
3414 * sequences and adds the 1-1 mappings that come out of that. */
3418 HV *const hv = MUTABLE_HV(SvRV(swash));
3420 /* The string containing the main body of the table. This will have its
3421 * assertion fail if the swash has been converted to its inversion list */
3422 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3424 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3425 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3426 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3427 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3428 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3429 const STRLEN bits = SvUV(*bitssvp);
3430 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3431 const UV none = SvUV(*nonesvp);
3432 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3436 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3438 /* Must have at least 8 bits to get the mappings */
3439 if (bits != 8 && bits != 16 && bits != 32) {
3440 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3444 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3445 mapping to more than one character */
3447 /* Construct an inverse mapping hash for the specials */
3448 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3449 HV * specials_inverse = newHV();
3450 char *char_from; /* the lhs of the map */
3451 I32 from_len; /* its byte length */
3452 char *char_to; /* the rhs of the map */
3453 I32 to_len; /* its byte length */
3454 SV *sv_to; /* and in a sv */
3455 AV* from_list; /* list of things that map to each 'to' */
3457 hv_iterinit(specials_hv);
3459 /* The keys are the characters (in utf8) that map to the corresponding
3460 * utf8 string value. Iterate through the list creating the inverse
3462 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3464 if (! SvPOK(sv_to)) {
3465 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3466 "unexpectedly is not a string, flags=%lu",
3467 (unsigned long)SvFLAGS(sv_to));
3469 /*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)));*/
3471 /* Each key in the inverse list is a mapped-to value, and the key's
3472 * hash value is a list of the strings (each in utf8) that map to
3473 * it. Those strings are all one character long */
3474 if ((listp = hv_fetch(specials_inverse,
3478 from_list = (AV*) *listp;
3480 else { /* No entry yet for it: create one */
3481 from_list = newAV();
3482 if (! hv_store(specials_inverse,
3485 (SV*) from_list, 0))
3487 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3491 /* Here have the list associated with this 'to' (perhaps newly
3492 * created and empty). Just add to it. Note that we ASSUME that
3493 * the input is guaranteed to not have duplications, so we don't
3494 * check for that. Duplications just slow down execution time. */
3495 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3498 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3499 * it looking for cases like the FB05/FB06 examples above. There would
3500 * be an entry in the hash like
3501 * 'st' => [ FB05, FB06 ]
3502 * In this example we will create two lists that get stored in the
3503 * returned hash, 'ret':
3504 * FB05 => [ FB05, FB06 ]
3505 * FB06 => [ FB05, FB06 ]
3507 * Note that there is nothing to do if the array only has one element.
3508 * (In the normal 1-1 case handled below, we don't have to worry about
3509 * two lists, as everything gets tied to the single list that is
3510 * generated for the single character 'to'. But here, we are omitting
3511 * that list, ('st' in the example), so must have multiple lists.) */
3512 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3513 &char_to, &to_len)))
3515 if (av_len(from_list) > 0) {
3518 /* We iterate over all combinations of i,j to place each code
3519 * point on each list */
3520 for (i = 0; i <= av_len(from_list); i++) {
3522 AV* i_list = newAV();
3523 SV** entryp = av_fetch(from_list, i, FALSE);
3524 if (entryp == NULL) {
3525 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3527 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3528 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3530 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3531 (SV*) i_list, FALSE))
3533 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3536 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3537 for (j = 0; j <= av_len(from_list); j++) {
3538 entryp = av_fetch(from_list, j, FALSE);
3539 if (entryp == NULL) {
3540 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3543 /* When i==j this adds itself to the list */
3544 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3545 (U8*) SvPVX(*entryp),
3546 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3548 /*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));*/
3553 SvREFCNT_dec(specials_inverse); /* done with it */
3554 } /* End of specials */
3556 /* read $swash->{LIST} */
3557 l = (U8*)SvPV(*listsvp, lcur);
3560 /* Go through each input line */
3564 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3565 cBOOL(octets), typestr);
3570 /* Each element in the range is to be inverted */
3571 for (inverse = min; inverse <= max; inverse++) {
3575 bool found_key = FALSE;
3576 bool found_inverse = FALSE;
3578 /* The key is the inverse mapping */
3579 char key[UTF8_MAXBYTES+1];
3580 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3581 STRLEN key_len = key_end - key;
3583 /* Get the list for the map */
3584 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3585 list = (AV*) *listp;
3587 else { /* No entry yet for it: create one */
3589 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3590 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3594 /* Look through list to see if this inverse mapping already is
3595 * listed, or if there is a mapping to itself already */
3596 for (i = 0; i <= av_len(list); i++) {
3597 SV** entryp = av_fetch(list, i, FALSE);
3599 if (entryp == NULL) {
3600 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3603 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3604 if (SvUV(entry) == val) {
3607 if (SvUV(entry) == inverse) {
3608 found_inverse = TRUE;
3611 /* No need to continue searching if found everything we are
3613 if (found_key && found_inverse) {
3618 /* Make sure there is a mapping to itself on the list */
3620 av_push(list, newSVuv(val));
3621 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3625 /* Simply add the value to the list */
3626 if (! found_inverse) {
3627 av_push(list, newSVuv(inverse));
3628 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3631 /* swatch_get() increments the value of val for each element in the
3632 * range. That makes more compact tables possible. You can
3633 * express the capitalization, for example, of all consecutive
3634 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3635 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3636 * and it's not documented; it appears to be used only in
3637 * implementing tr//; I copied the semantics from swatch_get(), just
3639 if (!none || val < none) {
3649 Perl__swash_to_invlist(pTHX_ SV* const swash)
3652 /* Subject to change or removal. For use only in one place in regcomp.c.
3653 * Ownership is given to one reference count in the returned SV* */
3658 HV *const hv = MUTABLE_HV(SvRV(swash));
3659 UV elements = 0; /* Number of elements in the inversion list */
3669 STRLEN octets; /* if bits == 1, then octets == 0 */
3675 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3677 /* If not a hash, it must be the swash's inversion list instead */
3678 if (SvTYPE(hv) != SVt_PVHV) {
3679 return SvREFCNT_inc_simple_NN((SV*) hv);
3682 /* The string containing the main body of the table */
3683 listsvp = hv_fetchs(hv, "LIST", FALSE);
3684 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3685 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3686 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3687 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3689 typestr = (U8*)SvPV_nolen(*typesvp);
3690 bits = SvUV(*bitssvp);
3691 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3693 /* read $swash->{LIST} */
3694 if (SvPOK(*listsvp)) {
3695 l = (U8*)SvPV(*listsvp, lcur);
3698 /* LIST legitimately doesn't contain a string during compilation phases
3699 * of Perl itself, before the Unicode tables are generated. In this
3700 * case, just fake things up by creating an empty list */
3707 if (*l == 'V') { /* Inversion list format */
3708 char *after_strtol = (char *) lend;
3710 UV* other_elements_ptr;
3712 /* The first number is a count of the rest */
3714 elements = Strtoul((char *)l, &after_strtol, 10);
3715 l = (U8 *) after_strtol;
3717 /* Get the 0th element, which is needed to setup the inversion list */
3718 element0 = (UV) Strtoul((char *)l, &after_strtol, 10);
3719 l = (U8 *) after_strtol;
3720 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3723 /* Then just populate the rest of the input */
3724 while (elements-- > 0) {
3726 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3728 *other_elements_ptr++ = (UV) Strtoul((char *)l, &after_strtol, 10);
3729 l = (U8 *) after_strtol;
3734 /* Scan the input to count the number of lines to preallocate array
3735 * size based on worst possible case, which is each line in the input
3736 * creates 2 elements in the inversion list: 1) the beginning of a
3737 * range in the list; 2) the beginning of a range not in the list. */
3738 while ((loc = (strchr(loc, '\n'))) != NULL) {
3743 /* If the ending is somehow corrupt and isn't a new line, add another
3744 * element for the final range that isn't in the inversion list */
3745 if (! (*lend == '\n'
3746 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3751 invlist = _new_invlist(elements);
3753 /* Now go through the input again, adding each range to the list */
3756 UV val; /* Not used by this function */
3758 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
3759 cBOOL(octets), typestr);
3765 invlist = _add_range_to_invlist(invlist, start, end);
3769 /* Invert if the data says it should be */
3770 if (invert_it_svp && SvUV(*invert_it_svp)) {
3771 _invlist_invert(invlist);
3774 /* This code is copied from swatch_get()
3775 * read $swash->{EXTRAS} */
3776 x = (U8*)SvPV(*extssvp, xcur);
3784 SV **otherbitssvp, *other;
3787 const U8 opc = *x++;
3791 nl = (U8*)memchr(x, '\n', xend - x);
3793 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3795 x = nl + 1; /* 1 is length of "\n" */
3799 x = xend; /* to EXTRAS' end at which \n is not found */
3806 namelen = nl - namestr;
3810 namelen = xend - namestr;
3814 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3815 otherhv = MUTABLE_HV(SvRV(*othersvp));
3816 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3817 otherbits = (STRLEN)SvUV(*otherbitssvp);
3819 if (bits != otherbits || bits != 1) {
3820 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3821 "properties, bits=%"UVuf", otherbits=%"UVuf,
3822 (UV)bits, (UV)otherbits);
3825 /* The "other" swatch must be destroyed after. */
3826 other = _swash_to_invlist((SV *)*othersvp);
3828 /* End of code copied from swatch_get() */
3831 _invlist_union(invlist, other, &invlist);
3834 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3837 _invlist_subtract(invlist, other, &invlist);
3840 _invlist_intersection(invlist, other, &invlist);
3845 sv_free(other); /* through with it! */
3848 SvREADONLY_on(invlist);
3853 Perl__get_swash_invlist(pTHX_ SV* const swash)
3857 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3859 if (! SvROK(swash)) {
3863 /* If it really isn't a hash, it isn't really swash; must be an inversion
3865 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3869 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3878 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3880 /* May change: warns if surrogates, non-character code points, or
3881 * non-Unicode code points are in s which has length len bytes. Returns
3882 * TRUE if none found; FALSE otherwise. The only other validity check is
3883 * to make sure that this won't exceed the string's length */
3885 const U8* const e = s + len;
3888 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3891 if (UTF8SKIP(s) > len) {
3892 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3893 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3896 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3898 if (UTF8_IS_SUPER(s)) {
3899 if (ckWARN_d(WARN_NON_UNICODE)) {
3900 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3901 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3902 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3906 else if (UTF8_IS_SURROGATE(s)) {
3907 if (ckWARN_d(WARN_SURROGATE)) {
3908 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3909 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3910 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3915 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3916 && (ckWARN_d(WARN_NONCHAR)))
3918 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3919 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3920 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3931 =for apidoc pv_uni_display
3933 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3934 length C<len>, the displayable version being at most C<pvlim> bytes long
3935 (if longer, the rest is truncated and "..." will be appended).
3937 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
3938 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3939 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3940 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3941 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3942 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3944 The pointer to the PV of the C<dsv> is returned.
3948 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3953 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3957 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3959 /* This serves double duty as a flag and a character to print after
3960 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3964 if (pvlim && SvCUR(dsv) >= pvlim) {
3968 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3970 const unsigned char c = (unsigned char)u & 0xFF;
3971 if (flags & UNI_DISPLAY_BACKSLASH) {
3988 const char string = ok;
3989 sv_catpvs(dsv, "\\");
3990 sv_catpvn(dsv, &string, 1);
3993 /* isPRINT() is the locale-blind version. */
3994 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3995 const char string = c;
3996 sv_catpvn(dsv, &string, 1);
4001 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4004 sv_catpvs(dsv, "...");
4010 =for apidoc sv_uni_display
4012 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4013 the displayable version being at most C<pvlim> bytes long
4014 (if longer, the rest is truncated and "..." will be appended).
4016 The C<flags> argument is as in L</pv_uni_display>().
4018 The pointer to the PV of the C<dsv> is returned.
4023 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4025 const char * const ptr =
4026 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4028 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4030 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4031 SvCUR(ssv), pvlim, flags);
4035 =for apidoc foldEQ_utf8
4037 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4038 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4039 How far into the strings to compare is determined by other input parameters.
4041 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4042 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4043 with respect to C<s2>.
4045 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4046 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4047 scan will not be considered to be a match unless the goal is reached, and
4048 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4051 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4052 considered an end pointer to the position 1 byte past the maximum point
4053 in C<s1> beyond which scanning will not continue under any circumstances.
4054 (This routine assumes that UTF-8 encoded input strings are not malformed;
4055 malformed input can cause it to read past C<pe1>).
4056 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4057 is less than C<s1>+C<l1>, the match will never be successful because it can
4059 get as far as its goal (and in fact is asserted against). Correspondingly for
4060 C<pe2> with respect to C<s2>.
4062 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4063 C<l2> must be non-zero), and if both do, both have to be
4064 reached for a successful match. Also, if the fold of a character is multiple
4065 characters, all of them must be matched (see tr21 reference below for
4068 Upon a successful match, if C<pe1> is non-NULL,
4069 it will be set to point to the beginning of the I<next> character of C<s1>
4070 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4072 For case-insensitiveness, the "casefolding" of Unicode is used
4073 instead of upper/lowercasing both the characters, see
4074 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4078 /* A flags parameter has been added which may change, and hence isn't
4079 * externally documented. Currently it is:
4080 * 0 for as-documented above
4081 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4082 ASCII one, to not match
4083 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
4084 * points below 256; unicode rules for above 255; and
4085 * folds that cross those boundaries are disallowed,
4086 * like the NOMIX_ASCII option
4087 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4088 * routine. This allows that step to be skipped.
4089 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4092 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)
4095 const U8 *p1 = (const U8*)s1; /* Point to current char */
4096 const U8 *p2 = (const U8*)s2;
4097 const U8 *g1 = NULL; /* goal for s1 */
4098 const U8 *g2 = NULL;
4099 const U8 *e1 = NULL; /* Don't scan s1 past this */
4100 U8 *f1 = NULL; /* Point to current folded */
4101 const U8 *e2 = NULL;
4103 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4104 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4105 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4107 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4109 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
4110 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4111 /* The algorithm is to trial the folds without regard to the flags on
4112 * the first line of the above assert(), and then see if the result
4113 * violates them. This means that the inputs can't be pre-folded to a
4114 * violating result, hence the assert. This could be changed, with the
4115 * addition of extra tests here for the already-folded case, which would
4116 * slow it down. That cost is more than any possible gain for when these
4117 * flags are specified, as the flags indicate /il or /iaa matching which
4118 * is less common than /iu, and I (khw) also believe that real-world /il
4119 * and /iaa matches are most likely to involve code points 0-255, and this
4120 * function only under rare conditions gets called for 0-255. */
4127 g1 = (const U8*)s1 + l1;
4135 g2 = (const U8*)s2 + l2;
4138 /* Must have at least one goal */
4143 /* Will never match if goal is out-of-bounds */
4144 assert(! e1 || e1 >= g1);
4146 /* Here, there isn't an end pointer, or it is beyond the goal. We
4147 * only go as far as the goal */
4151 assert(e1); /* Must have an end for looking at s1 */
4154 /* Same for goal for s2 */
4156 assert(! e2 || e2 >= g2);
4163 /* If both operands are already folded, we could just do a memEQ on the
4164 * whole strings at once, but it would be better if the caller realized
4165 * this and didn't even call us */
4167 /* Look through both strings, a character at a time */
4168 while (p1 < e1 && p2 < e2) {
4170 /* If at the beginning of a new character in s1, get its fold to use
4171 * and the length of the fold. (exception: locale rules just get the
4172 * character to a single byte) */
4174 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4179 /* If in locale matching, we use two sets of rules, depending
4180 * on if the code point is above or below 255. Here, we test
4181 * for and handle locale rules */
4182 if ((flags & FOLDEQ_UTF8_LOCALE)
4183 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
4185 /* There is no mixing of code points above and below 255. */
4186 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
4190 /* We handle locale rules by converting, if necessary, the
4191 * code point to a single byte. */
4192 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4196 *foldbuf1 = TWO_BYTE_UTF8_TO_NATIVE(*p1, *(p1 + 1));
4200 else if (isASCII(*p1)) { /* Note, that here won't be both
4201 ASCII and using locale rules */
4203 /* If trying to mix non- with ASCII, and not supposed to,
4205 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4209 *foldbuf1 = toFOLD(*p1);
4212 to_utf8_fold(p1, foldbuf1, &n1);
4214 else { /* Not utf8, get utf8 fold */
4215 to_uni_fold(*p1, foldbuf1, &n1);
4221 if (n2 == 0) { /* Same for s2 */
4222 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4227 if ((flags & FOLDEQ_UTF8_LOCALE)
4228 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4230 /* Here, the next char in s2 is < 256. We've already
4231 * worked on s1, and if it isn't also < 256, can't match */
4232 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4235 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4239 *foldbuf2 = TWO_BYTE_UTF8_TO_NATIVE(*p2, *(p2 + 1));
4242 /* Use another function to handle locale rules. We've made
4243 * sure that both characters to compare are single bytes */
4244 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4249 else if (isASCII(*p2)) {
4250 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4254 *foldbuf2 = toFOLD(*p2);
4257 to_utf8_fold(p2, foldbuf2, &n2);
4260 to_uni_fold(*p2, foldbuf2, &n2);
4266 /* Here f1 and f2 point to the beginning of the strings to compare.
4267 * These strings are the folds of the next character from each input
4268 * string, stored in utf8. */
4270 /* While there is more to look for in both folds, see if they
4271 * continue to match */
4273 U8 fold_length = UTF8SKIP(f1);
4274 if (fold_length != UTF8SKIP(f2)
4275 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4276 function call for single
4278 || memNE((char*)f1, (char*)f2, fold_length))
4280 return 0; /* mismatch */
4283 /* Here, they matched, advance past them */
4290 /* When reach the end of any fold, advance the input past it */
4292 p1 += u1 ? UTF8SKIP(p1) : 1;
4295 p2 += u2 ? UTF8SKIP(p2) : 1;
4297 } /* End of loop through both strings */
4299 /* A match is defined by each scan that specified an explicit length
4300 * reaching its final goal, and the other not having matched a partial
4301 * character (which can happen when the fold of a character is more than one
4303 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4307 /* Successful match. Set output pointers */
4317 /* XXX The next four functions should likely be moved to mathoms.c once all
4318 * occurrences of them are removed from the core; some cpan-upstream modules
4322 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4324 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4326 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4330 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4332 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4334 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4338 =for apidoc uvuni_to_utf8_flags
4340 Instead you almost certainly want to use L</uvchr_to_utf8> or
4341 L</uvchr_to_utf8_flags>>.
4343 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4344 which itself, while not deprecated, should be used only in isolated
4345 circumstances. These functions were useful for code that wanted to handle
4346 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4347 v5.20, the distinctions between the platforms have mostly been made invisible
4348 to most code, so this function is quite unlikely to be what you want.
4354 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4356 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4358 return uvoffuni_to_utf8_flags(d, uv, flags);
4362 =for apidoc utf8n_to_uvuni
4364 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4366 This function was useful for code that wanted to handle both EBCDIC and
4367 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4368 distinctions between the platforms have mostly been made invisible to most
4369 code, so this function is quite unlikely to be what you want. If you do need
4370 this precise functionality, use instead
4371 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4372 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4379 * c-indentation-style: bsd
4381 * indent-tabs-mode: nil
4384 * ex: set ts=8 sts=4 sw=4 et: