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)>, (which means if you
61 use this option, that C<s> can't have embedded C<NUL> characters and has to
62 have a terminating C<NUL> byte).
64 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
70 Perl_is_ascii_string(const U8 *s, STRLEN len)
72 const U8* const send = s + (len ? len : strlen((const char *)s));
75 PERL_ARGS_ASSERT_IS_ASCII_STRING;
77 for (; x < send; ++x) {
78 if (!UTF8_IS_INVARIANT(*x))
86 =for apidoc uvoffuni_to_utf8_flags
88 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
89 Instead, B<Almost all code should use L</uvchr_to_utf8> or
90 L</uvchr_to_utf8_flags>>.
92 This function is like them, but the input is a strict Unicode
93 (as opposed to native) code point. Only in very rare circumstances should code
94 not be using the native code point.
96 For details, see the description for L</uvchr_to_utf8_flags>>.
102 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
104 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
106 if (UNI_IS_INVARIANT(uv)) {
107 *d++ = (U8) LATIN1_TO_NATIVE(uv);
111 /* The first problematic code point is the first surrogate */
112 if (uv >= UNICODE_SURROGATE_FIRST
113 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
115 if (UNICODE_IS_SURROGATE(uv)) {
116 if (flags & UNICODE_WARN_SURROGATE) {
117 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
118 "UTF-16 surrogate U+%04"UVXf, uv);
120 if (flags & UNICODE_DISALLOW_SURROGATE) {
124 else if (UNICODE_IS_SUPER(uv)) {
125 if (flags & UNICODE_WARN_SUPER
126 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
128 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
129 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
131 if (flags & UNICODE_DISALLOW_SUPER
132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
137 else if (UNICODE_IS_NONCHAR(uv)) {
138 if (flags & UNICODE_WARN_NONCHAR) {
139 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
140 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
143 if (flags & UNICODE_DISALLOW_NONCHAR) {
151 STRLEN len = OFFUNISKIP(uv);
154 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
155 uv >>= UTF_ACCUMULATION_SHIFT;
157 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
160 #else /* Non loop style */
162 *d++ = (U8)(( uv >> 6) | 0xc0);
163 *d++ = (U8)(( uv & 0x3f) | 0x80);
167 *d++ = (U8)(( uv >> 12) | 0xe0);
168 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
169 *d++ = (U8)(( uv & 0x3f) | 0x80);
173 *d++ = (U8)(( uv >> 18) | 0xf0);
174 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
175 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
176 *d++ = (U8)(( uv & 0x3f) | 0x80);
179 if (uv < 0x4000000) {
180 *d++ = (U8)(( uv >> 24) | 0xf8);
181 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
182 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
184 *d++ = (U8)(( uv & 0x3f) | 0x80);
187 if (uv < 0x80000000) {
188 *d++ = (U8)(( uv >> 30) | 0xfc);
189 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
191 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
192 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
193 *d++ = (U8)(( uv & 0x3f) | 0x80);
197 if (uv < UTF8_QUAD_MAX)
200 *d++ = 0xfe; /* Can't match U+FEFF! */
201 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
205 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
206 *d++ = (U8)(( uv & 0x3f) | 0x80);
211 *d++ = 0xff; /* Can't match U+FFFE! */
212 *d++ = 0x80; /* 6 Reserved bits */
213 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
214 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
221 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
222 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
223 *d++ = (U8)(( uv & 0x3f) | 0x80);
227 #endif /* Non loop style */
230 =for apidoc uvchr_to_utf8
232 Adds the UTF-8 representation of the native code point C<uv> to the end
233 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
234 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
235 the byte after the end of the new character. In other words,
237 d = uvchr_to_utf8(d, uv);
239 is the recommended wide native character-aware way of saying
243 This function accepts any UV as input. To forbid or warn on non-Unicode code
244 points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
249 /* This is also a macro */
250 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
253 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
255 return uvchr_to_utf8(d, uv);
259 =for apidoc uvchr_to_utf8_flags
261 Adds the UTF-8 representation of the native code point C<uv> to the end
262 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
263 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
264 the byte after the end of the new character. In other words,
266 d = uvchr_to_utf8_flags(d, uv, flags);
270 d = uvchr_to_utf8_flags(d, uv, 0);
272 This is the Unicode-aware way of saying
276 This function will convert to UTF-8 (and not warn) even code points that aren't
277 legal Unicode or are problematic, unless C<flags> contains one or more of the
280 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
281 the function will raise a warning, provided UTF8 warnings are enabled. If instead
282 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
283 If both flags are set, the function will both warn and return NULL.
285 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
286 affect how the function handles a Unicode non-character. And likewise, the
287 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
289 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
290 even less portable) can be warned and/or disallowed even if other above-Unicode
291 code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
294 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
295 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
301 /* This is also a macro */
302 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
305 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
307 return uvchr_to_utf8_flags(d, uv, flags);
312 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
313 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC) character is a
314 valid UTF-8 character. The number of bytes in the UTF-8 character
315 will be returned if it is valid, otherwise 0.
317 This is the "slow" version as opposed to the "fast" version which is
318 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
319 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
320 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
321 you should use the _slow(). In practice this means that the _slow()
322 will be used very rarely, since the maximum Unicode code point (as of
323 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
324 the "Perl extended UTF-8" (e.g, the infamous 'v-strings') will encode into
328 PERL_STATIC_INLINE STRLEN
329 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
331 dTHX; /* The function called below requires thread context */
335 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
337 utf8n_to_uvchr(s, len, &actual_len, UTF8_CHECK_ONLY);
339 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
343 =for apidoc is_utf8_char_buf
345 Returns the number of bytes that comprise the first UTF-8 encoded character in
346 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
347 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
350 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
351 machines) is a valid UTF-8 character.
356 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
361 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
363 if (buf_end <= buf) {
368 if (len > UTF8SKIP(buf)) {
372 if (IS_UTF8_CHAR_FAST(len))
373 return IS_UTF8_CHAR(buf, len) ? len : 0;
374 return is_utf8_char_slow(buf, len);
378 =for apidoc is_utf8_char
380 Tests if some arbitrary number of bytes begins in a valid UTF-8
381 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
382 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
383 character will be returned if it is valid, otherwise 0.
385 This function is deprecated due to the possibility that malformed input could
386 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
392 Perl_is_utf8_char(const U8 *s)
394 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
396 /* Assumes we have enough space, which is why this is deprecated */
397 return is_utf8_char_buf(s, s + UTF8SKIP(s));
402 =for apidoc is_utf8_string
404 Returns true if the first C<len> bytes of string C<s> form a valid
405 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
406 using C<strlen(s)> (which means if you use this option, that C<s> can't have
407 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
408 that all characters being ASCII constitute 'a valid UTF-8 string'.
410 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
416 Perl_is_utf8_string(const U8 *s, STRLEN len)
418 const U8* const send = s + (len ? len : strlen((const char *)s));
421 PERL_ARGS_ASSERT_IS_UTF8_STRING;
424 /* Inline the easy bits of is_utf8_char() here for speed... */
425 if (UTF8_IS_INVARIANT(*x)) {
429 /* ... and call is_utf8_char() only if really needed. */
430 const STRLEN c = UTF8SKIP(x);
431 const U8* const next_char_ptr = x + c;
433 if (next_char_ptr > send) {
437 if (IS_UTF8_CHAR_FAST(c)) {
438 if (!IS_UTF8_CHAR(x, c))
441 else if (! is_utf8_char_slow(x, c)) {
452 Implemented as a macro in utf8.h
454 =for apidoc is_utf8_string_loc
456 Like L</is_utf8_string> but stores the location of the failure (in the
457 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
458 "utf8ness success") in the C<ep>.
460 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
462 =for apidoc is_utf8_string_loclen
464 Like L</is_utf8_string>() but stores the location of the failure (in the
465 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
466 "utf8ness success") in the C<ep>, and the number of UTF-8
467 encoded characters in the C<el>.
469 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
475 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
477 const U8* const send = s + (len ? len : strlen((const char *)s));
482 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
485 const U8* next_char_ptr;
487 /* Inline the easy bits of is_utf8_char() here for speed... */
488 if (UTF8_IS_INVARIANT(*x))
489 next_char_ptr = x + 1;
491 /* ... and call is_utf8_char() only if really needed. */
493 next_char_ptr = c + x;
494 if (next_char_ptr > send) {
497 if (IS_UTF8_CHAR_FAST(c)) {
498 if (!IS_UTF8_CHAR(x, c))
501 c = is_utf8_char_slow(x, c);
520 =for apidoc utf8n_to_uvchr
522 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
523 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
525 Bottom level UTF-8 decode routine.
526 Returns the native code point value of the first character in the string C<s>,
527 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
528 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
529 the length, in bytes, of that character.
531 The value of C<flags> determines the behavior when C<s> does not point to a
532 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
533 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
534 next possible position in C<s> that could begin a non-malformed character.
535 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
537 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
538 individual types of malformations, such as the sequence being overlong (that
539 is, when there is a shorter sequence that can express the same code point;
540 overlong sequences are expressly forbidden in the UTF-8 standard due to
541 potential security issues). Another malformation example is the first byte of
542 a character not being a legal first byte. See F<utf8.h> for the list of such
543 flags. For allowed 0 length strings, this function returns 0; for allowed
544 overlong sequences, the computed code point is returned; for all other allowed
545 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
546 determinable reasonable value.
548 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
549 flags) malformation is found. If this flag is set, the routine assumes that
550 the caller will raise a warning, and this function will silently just set
551 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
553 Note that this API requires disambiguation between successful decoding a C<NUL>
554 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
555 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
556 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
559 Certain code points are considered problematic. These are Unicode surrogates,
560 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
561 By default these are considered regular code points, but certain situations
562 warrant special handling for them. If C<flags> contains
563 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
564 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
565 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
566 maximum) can be set to disallow these categories individually.
568 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
569 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
570 for their respective categories, but otherwise the code points are considered
571 valid (not malformations). To get a category to both be treated as a
572 malformation and raise a warning, specify both the WARN and DISALLOW flags.
573 (But note that warnings are not raised if lexically disabled nor if
574 UTF8_CHECK_ONLY is also specified.)
576 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
577 the others that are above the Unicode legal maximum. There are several
578 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
579 not representable at all on EBCDIC platforms, and the original UTF-8
580 specification never went above this number (the current 0x10FFFF limit was
581 imposed later). (The smaller ones, those that fit into 32 bits, are
582 representable by a UV on ASCII platforms, but not by an IV, which means that
583 the number of operations that can be performed on them is quite restricted.)
584 The UTF-8 encoding on ASCII platforms for these large code points begins with a
585 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
586 be treated as malformations, while allowing smaller above-Unicode code points.
587 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
588 including these, as malformations.)
589 Similarly, UTF8_WARN_FE_FF acts just like
590 the other WARN flags, but applies just to these code points.
592 All other code points corresponding to Unicode characters, including private
593 use and those yet to be assigned, are never considered malformed and never
600 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
603 const U8 * const s0 = s;
604 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
609 UV outlier_ret = 0; /* return value when input is in error or problematic
611 UV pack_warn = 0; /* Save result of packWARN() for later */
612 bool unexpected_non_continuation = FALSE;
613 bool overflowed = FALSE;
614 bool do_overlong_test = TRUE; /* May have to skip this test */
616 const char* const malformed_text = "Malformed UTF-8 character";
618 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
620 /* The order of malformation tests here is important. We should consume as
621 * few bytes as possible in order to not skip any valid character. This is
622 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
623 * http://unicode.org/reports/tr36 for more discussion as to why. For
624 * example, once we've done a UTF8SKIP, we can tell the expected number of
625 * bytes, and could fail right off the bat if the input parameters indicate
626 * that there are too few available. But it could be that just that first
627 * byte is garbled, and the intended character occupies fewer bytes. If we
628 * blindly assumed that the first byte is correct, and skipped based on
629 * that number, we could skip over a valid input character. So instead, we
630 * always examine the sequence byte-by-byte.
632 * We also should not consume too few bytes, otherwise someone could inject
633 * things. For example, an input could be deliberately designed to
634 * overflow, and if this code bailed out immediately upon discovering that,
635 * returning to the caller C<*retlen> pointing to the very next byte (one
636 * which is actually part of of the overflowing sequence), that could look
637 * legitimate to the caller, which could discard the initial partial
638 * sequence and process the rest, inappropriately */
640 /* Zero length strings, if allowed, of necessity are zero */
641 if (UNLIKELY(curlen == 0)) {
646 if (flags & UTF8_ALLOW_EMPTY) {
649 if (! (flags & UTF8_CHECK_ONLY)) {
650 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
655 expectlen = UTF8SKIP(s);
657 /* A well-formed UTF-8 character, as the vast majority of calls to this
658 * function will be for, has this expected length. For efficiency, set
659 * things up here to return it. It will be overriden only in those rare
660 * cases where a malformation is found */
665 /* An invariant is trivially well-formed */
666 if (UTF8_IS_INVARIANT(uv)) {
670 /* A continuation character can't start a valid sequence */
671 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
672 if (flags & UTF8_ALLOW_CONTINUATION) {
676 return UNICODE_REPLACEMENT;
679 if (! (flags & UTF8_CHECK_ONLY)) {
680 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
686 /* Here is not a continuation byte, nor an invariant. The only thing left
687 * is a start byte (possibly for an overlong) */
690 uv = NATIVE_UTF8_TO_I8(uv);
693 /* Remove the leading bits that indicate the number of bytes in the
694 * character's whole UTF-8 sequence, leaving just the bits that are part of
696 uv &= UTF_START_MASK(expectlen);
698 /* Now, loop through the remaining bytes in the character's sequence,
699 * accumulating each into the working value as we go. Be sure to not look
700 * past the end of the input string */
701 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
703 for (s = s0 + 1; s < send; s++) {
704 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
705 #ifndef EBCDIC /* Can't overflow in EBCDIC */
706 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
708 /* The original implementors viewed this malformation as more
709 * serious than the others (though I, khw, don't understand
710 * why, since other malformations also give very very wrong
711 * results), so there is no way to turn off checking for it.
712 * Set a flag, but keep going in the loop, so that we absorb
713 * the rest of the bytes that comprise the character. */
715 overflow_byte = *s; /* Save for warning message's use */
718 uv = UTF8_ACCUMULATE(uv, *s);
721 /* Here, found a non-continuation before processing all expected
722 * bytes. This byte begins a new character, so quit, even if
723 * allowing this malformation. */
724 unexpected_non_continuation = TRUE;
727 } /* End of loop through the character's bytes */
729 /* Save how many bytes were actually in the character */
732 /* The loop above finds two types of malformations: non-continuation and/or
733 * overflow. The non-continuation malformation is really a too-short
734 * malformation, as it means that the current character ended before it was
735 * expected to (being terminated prematurely by the beginning of the next
736 * character, whereas in the too-short malformation there just are too few
737 * bytes available to hold the character. In both cases, the check below
738 * that we have found the expected number of bytes would fail if executed.)
739 * Thus the non-continuation malformation is really unnecessary, being a
740 * subset of the too-short malformation. But there may be existing
741 * applications that are expecting the non-continuation type, so we retain
742 * it, and return it in preference to the too-short malformation. (If this
743 * code were being written from scratch, the two types might be collapsed
744 * into one.) I, khw, am also giving priority to returning the
745 * non-continuation and too-short malformations over overflow when multiple
746 * ones are present. I don't know of any real reason to prefer one over
747 * the other, except that it seems to me that multiple-byte errors trumps
748 * errors from a single byte */
749 if (UNLIKELY(unexpected_non_continuation)) {
750 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
751 if (! (flags & UTF8_CHECK_ONLY)) {
753 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
756 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));
761 uv = UNICODE_REPLACEMENT;
763 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
764 * as what the original expectations were. */
765 do_overlong_test = FALSE;
770 else if (UNLIKELY(curlen < expectlen)) {
771 if (! (flags & UTF8_ALLOW_SHORT)) {
772 if (! (flags & UTF8_CHECK_ONLY)) {
773 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));
777 uv = UNICODE_REPLACEMENT;
778 do_overlong_test = FALSE;
784 #ifndef EBCDIC /* EBCDIC can't overflow */
785 if (UNLIKELY(overflowed)) {
786 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
792 && expectlen > (STRLEN) OFFUNISKIP(uv)
793 && ! (flags & UTF8_ALLOW_LONG))
795 /* The overlong malformation has lower precedence than the others.
796 * Note that if this malformation is allowed, we return the actual
797 * value, instead of the replacement character. This is because this
798 * value is actually well-defined. */
799 if (! (flags & UTF8_CHECK_ONLY)) {
800 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));
805 /* Here, the input is considered to be well-formed, but it still could be a
806 * problematic code point that is not allowed by the input parameters. */
807 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
808 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
809 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
811 if (UNICODE_IS_SURROGATE(uv)) {
813 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
814 * generation of the sv, since no warnings are raised under CHECK */
815 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
816 && ckWARN_d(WARN_SURROGATE))
818 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
819 pack_warn = packWARN(WARN_SURROGATE);
821 if (flags & UTF8_DISALLOW_SURROGATE) {
825 else if ((uv > PERL_UNICODE_MAX)) {
826 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
827 && ckWARN_d(WARN_NON_UNICODE))
829 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
830 pack_warn = packWARN(WARN_NON_UNICODE);
832 #ifndef EBCDIC /* EBCDIC always allows FE, FF */
834 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
835 * points. We test for these after the regular SUPER ones, and
836 * before possibly bailing out, so that the more dire warning
837 * overrides the regular one, if applicable */
838 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
839 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
841 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
843 && ckWARN_d(WARN_UTF8))
845 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
846 pack_warn = packWARN(WARN_UTF8);
848 if (flags & UTF8_DISALLOW_FE_FF) {
853 if (flags & UTF8_DISALLOW_SUPER) {
857 else if (UNICODE_IS_NONCHAR(uv)) {
858 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
859 && ckWARN_d(WARN_NONCHAR))
861 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
862 pack_warn = packWARN(WARN_NONCHAR);
864 if (flags & UTF8_DISALLOW_NONCHAR) {
870 outlier_ret = uv; /* Note we don't bother to convert to native,
871 as all the outlier code points are the same
872 in both ASCII and EBCDIC */
876 /* Here, this is not considered a malformed character, so drop through
880 return UNI_TO_NATIVE(uv);
882 /* There are three cases which get to beyond this point. In all 3 cases:
883 * <sv> if not null points to a string to print as a warning.
884 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
886 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
887 * This is done by initializing it to 0, and changing it only
890 * 1) The input is valid but problematic, and to be warned about. The
891 * return value is the resultant code point; <*retlen> is set to
892 * <curlen>, the number of bytes that comprise the code point.
893 * <pack_warn> contains the result of packWARN() for the warning
894 * types. The entry point for this case is the label <do_warn>;
895 * 2) The input is a valid code point but disallowed by the parameters to
896 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
897 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
898 * comprise the code point. <pack_warn> contains the result of
899 * packWARN() for the warning types. The entry point for this case is
900 * the label <disallowed>.
901 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
902 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
903 * bytes that comprise the malformation. All such malformations are
904 * assumed to be warning type <utf8>. The entry point for this case
905 * is the label <malformed>.
910 if (sv && ckWARN_d(WARN_UTF8)) {
911 pack_warn = packWARN(WARN_UTF8);
916 if (flags & UTF8_CHECK_ONLY) {
918 *retlen = ((STRLEN) -1);
924 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
925 if warnings are to be raised. */
926 const char * const string = SvPVX_const(sv);
929 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
931 Perl_warner(aTHX_ pack_warn, "%s", string);
942 =for apidoc utf8_to_uvchr_buf
944 Returns the native code point of the first character in the string C<s> which
945 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
946 C<*retlen> will be set to the length, in bytes, of that character.
948 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
949 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
950 NULL) to -1. If those warnings are off, the computed value, if well-defined
951 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
952 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
953 the next possible position in C<s> that could begin a non-malformed character.
954 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
962 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
966 return utf8n_to_uvchr(s, send - s, retlen,
967 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
970 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
971 * there are no malformations in the input UTF-8 string C<s>. surrogates,
972 * non-character code points, and non-Unicode code points are allowed. */
975 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
977 UV expectlen = UTF8SKIP(s);
978 const U8* send = s + expectlen;
981 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
987 /* An invariant is trivially returned */
988 if (expectlen == 1) {
993 uv = NATIVE_UTF8_TO_I8(uv);
996 /* Remove the leading bits that indicate the number of bytes, leaving just
997 * the bits that are part of the value */
998 uv &= UTF_START_MASK(expectlen);
1000 /* Now, loop through the remaining bytes, accumulating each into the
1001 * working total as we go. (I khw tried unrolling the loop for up to 4
1002 * bytes, but there was no performance improvement) */
1003 for (++s; s < send; s++) {
1004 uv = UTF8_ACCUMULATE(uv, *s);
1007 return UNI_TO_NATIVE(uv);
1012 =for apidoc utf8_to_uvchr
1014 Returns the native code point of the first character in the string C<s>
1015 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1016 length, in bytes, of that character.
1018 Some, but not all, UTF-8 malformations are detected, and in fact, some
1019 malformed input could cause reading beyond the end of the input buffer, which
1020 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
1022 If C<s> points to one of the detected malformations, and UTF8 warnings are
1023 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1024 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1025 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1026 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1027 next possible position in C<s> that could begin a non-malformed character.
1028 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1034 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
1036 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
1038 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
1042 =for apidoc utf8_to_uvuni_buf
1044 Only in very rare circumstances should code need to be dealing in Unicode
1045 (as opposed to native) code points. In those few cases, use
1046 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1048 Returns the Unicode (not-native) code point of the first character in the
1050 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1051 C<retlen> will be set to the length, in bytes, of that character.
1053 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1054 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1055 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1056 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1057 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1058 next possible position in C<s> that could begin a non-malformed character.
1059 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1065 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1067 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1071 /* Call the low level routine asking for checks */
1072 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1073 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
1077 * Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1078 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1079 * non-character code points, and non-Unicode code points are allowed */
1082 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1084 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1086 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
1090 =for apidoc utf8_to_uvuni
1092 Returns the Unicode code point of the first character in the string C<s>
1093 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1094 length, in bytes, of that character.
1096 Some, but not all, UTF-8 malformations are detected, and in fact, some
1097 malformed input could cause reading beyond the end of the input buffer, which
1098 is one reason why this function is deprecated. The other is that only in
1099 extremely limited circumstances should the Unicode versus native code point be
1100 of any interest to you. See L</utf8_to_uvuni_buf> for alternatives.
1102 If C<s> points to one of the detected malformations, and UTF8 warnings are
1103 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1104 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1105 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1106 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1107 next possible position in C<s> that could begin a non-malformed character.
1108 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1114 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1116 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1118 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
1122 =for apidoc utf8_length
1124 Return the length of the UTF-8 char encoded string C<s> in characters.
1125 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1126 up past C<e>, croaks.
1132 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1137 PERL_ARGS_ASSERT_UTF8_LENGTH;
1139 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1140 * the bitops (especially ~) can create illegal UTF-8.
1141 * In other words: in Perl UTF-8 is not just for Unicode. */
1144 goto warn_and_return;
1154 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1155 "%s in %s", unees, OP_DESC(PL_op));
1157 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1164 =for apidoc utf8_distance
1166 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1169 WARNING: use only if you *know* that the pointers point inside the
1176 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1178 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1180 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1184 =for apidoc utf8_hop
1186 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1187 forward or backward.
1189 WARNING: do not use the following unless you *know* C<off> is within
1190 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1191 on the first byte of character or just after the last byte of a character.
1197 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1199 PERL_ARGS_ASSERT_UTF8_HOP;
1201 PERL_UNUSED_CONTEXT;
1202 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1203 * the bitops (especially ~) can create illegal UTF-8.
1204 * In other words: in Perl UTF-8 is not just for Unicode. */
1213 while (UTF8_IS_CONTINUATION(*s))
1221 =for apidoc bytes_cmp_utf8
1223 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1224 sequence of characters (stored as UTF-8)
1225 in C<u>, C<ulen>. Returns 0 if they are
1226 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1227 if the first string is greater than the second string.
1229 -1 or +1 is returned if the shorter string was identical to the start of the
1230 longer string. -2 or +2 is returned if
1231 there was a difference between characters
1238 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1240 const U8 *const bend = b + blen;
1241 const U8 *const uend = u + ulen;
1243 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1245 PERL_UNUSED_CONTEXT;
1247 while (b < bend && u < uend) {
1249 if (!UTF8_IS_INVARIANT(c)) {
1250 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1253 if (UTF8_IS_CONTINUATION(c1)) {
1254 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1256 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1257 "Malformed UTF-8 character "
1258 "(unexpected non-continuation byte 0x%02x"
1259 ", immediately after start byte 0x%02x)"
1260 /* Dear diag.t, it's in the pod. */
1262 PL_op ? " in " : "",
1263 PL_op ? OP_DESC(PL_op) : "");
1268 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1269 "%s in %s", unees, OP_DESC(PL_op));
1271 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1272 return -2; /* Really want to return undef :-) */
1279 return *b < c ? -2 : +2;
1284 if (b == bend && u == uend)
1287 return b < bend ? +1 : -1;
1291 =for apidoc utf8_to_bytes
1293 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1294 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1295 updates C<len> to contain the new length.
1296 Returns zero on failure, setting C<len> to -1.
1298 If you need a copy of the string, see L</bytes_from_utf8>.
1304 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1306 U8 * const save = s;
1307 U8 * const send = s + *len;
1310 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1312 /* ensure valid UTF-8 and chars < 256 before updating string */
1314 if (! UTF8_IS_INVARIANT(*s)) {
1315 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1316 *len = ((STRLEN) -1);
1327 if (! UTF8_IS_INVARIANT(c)) {
1328 /* Then it is two-byte encoded */
1329 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1340 =for apidoc bytes_from_utf8
1342 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1343 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1344 the newly-created string, and updates C<len> to contain the new
1345 length. Returns the original string if no conversion occurs, C<len>
1346 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1347 0 if C<s> is converted or consisted entirely of characters that are invariant
1348 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1354 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1357 const U8 *start = s;
1361 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1363 PERL_UNUSED_CONTEXT;
1367 /* ensure valid UTF-8 and chars < 256 before converting string */
1368 for (send = s + *len; s < send;) {
1369 if (! UTF8_IS_INVARIANT(*s)) {
1370 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1381 Newx(d, (*len) - count + 1, U8);
1382 s = start; start = d;
1385 if (! UTF8_IS_INVARIANT(c)) {
1386 /* Then it is two-byte encoded */
1387 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1398 =for apidoc bytes_to_utf8
1400 Converts a string C<s> of length C<len> bytes from the native encoding into
1402 Returns a pointer to the newly-created string, and sets C<len> to
1403 reflect the new length in bytes.
1405 A C<NUL> character will be written after the end of the string.
1407 If you want to convert to UTF-8 from encodings other than
1408 the native (Latin1 or EBCDIC),
1409 see L</sv_recode_to_utf8>().
1414 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1415 likewise need duplication. */
1418 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1420 const U8 * const send = s + (*len);
1424 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1425 PERL_UNUSED_CONTEXT;
1427 Newx(d, (*len) * 2 + 1, U8);
1431 append_utf8_from_native_byte(*s, &d);
1440 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1442 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1443 * We optimize for native, for obvious reasons. */
1446 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1451 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1454 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1459 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1461 if (UNI_IS_INVARIANT(uv)) {
1462 *d++ = LATIN1_TO_NATIVE((U8) uv);
1465 if (uv <= MAX_UTF8_TWO_BYTE) {
1466 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1467 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1470 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1471 #define LAST_HIGH_SURROGATE 0xDBFF
1472 #define FIRST_LOW_SURROGATE 0xDC00
1473 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1474 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1476 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1478 UV low = (p[0] << 8) + p[1];
1480 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1481 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1482 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1483 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1485 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1486 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1489 d = uvoffuni_to_utf8_flags(d, uv, 0);
1492 *d++ = (U8)(( uv >> 12) | 0xe0);
1493 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1494 *d++ = (U8)(( uv & 0x3f) | 0x80);
1498 *d++ = (U8)(( uv >> 18) | 0xf0);
1499 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1500 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1501 *d++ = (U8)(( uv & 0x3f) | 0x80);
1506 *newlen = d - dstart;
1510 /* Note: this one is slightly destructive of the source. */
1513 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1516 U8* const send = s + bytelen;
1518 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1521 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1525 const U8 tmp = s[0];
1530 return utf16_to_utf8(p, d, bytelen, newlen);
1534 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1536 U8 tmpbuf[UTF8_MAXBYTES+1];
1537 uvchr_to_utf8(tmpbuf, c);
1538 return _is_utf8_FOO(classnum, tmpbuf);
1541 /* Internal function so we can deprecate the external one, and call
1542 this one from other deprecated functions in this file */
1544 PERL_STATIC_INLINE bool
1545 S_is_utf8_idfirst(pTHX_ const U8 *p)
1551 /* is_utf8_idstart would be more logical. */
1552 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1556 Perl_is_uni_idfirst(pTHX_ UV c)
1558 U8 tmpbuf[UTF8_MAXBYTES+1];
1559 uvchr_to_utf8(tmpbuf, c);
1560 return S_is_utf8_idfirst(aTHX_ tmpbuf);
1564 Perl__is_uni_perl_idcont(pTHX_ UV c)
1566 U8 tmpbuf[UTF8_MAXBYTES+1];
1567 uvchr_to_utf8(tmpbuf, c);
1568 return _is_utf8_perl_idcont(tmpbuf);
1572 Perl__is_uni_perl_idstart(pTHX_ UV c)
1574 U8 tmpbuf[UTF8_MAXBYTES+1];
1575 uvchr_to_utf8(tmpbuf, c);
1576 return _is_utf8_perl_idstart(tmpbuf);
1580 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1582 /* We have the latin1-range values compiled into the core, so just use
1583 * those, converting the result to utf8. The only difference between upper
1584 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1585 * either "SS" or "Ss". Which one to use is passed into the routine in
1586 * 'S_or_s' to avoid a test */
1588 UV converted = toUPPER_LATIN1_MOD(c);
1590 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1592 assert(S_or_s == 'S' || S_or_s == 's');
1594 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1595 characters in this range */
1596 *p = (U8) converted;
1601 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1602 * which it maps to one of them, so as to only have to have one check for
1603 * it in the main case */
1604 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1606 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1607 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1610 converted = GREEK_CAPITAL_LETTER_MU;
1612 case LATIN_SMALL_LETTER_SHARP_S:
1618 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1619 assert(0); /* NOTREACHED */
1623 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1624 *p = UTF8_TWO_BYTE_LO(converted);
1630 /* Call the function to convert a UTF-8 encoded character to the specified case.
1631 * Note that there may be more than one character in the result.
1632 * INP is a pointer to the first byte of the input character
1633 * OUTP will be set to the first byte of the string of changed characters. It
1634 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1635 * LENP will be set to the length in bytes of the string of changed characters
1637 * The functions return the ordinal of the first character in the string of OUTP */
1638 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1639 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1640 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1642 /* This additionally has the input parameter SPECIALS, which if non-zero will
1643 * cause this to use the SPECIALS hash for folding (meaning get full case
1644 * folding); otherwise, when zero, this implies a simple case fold */
1645 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1648 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1652 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1653 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1654 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1655 * the changed version may be longer than the original character.
1657 * The ordinal of the first character of the changed version is returned
1658 * (but note, as explained above, that there may be more.) */
1660 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1663 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1666 uvchr_to_utf8(p, c);
1667 return CALL_UPPER_CASE(p, p, lenp);
1671 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1675 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1678 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1681 uvchr_to_utf8(p, c);
1682 return CALL_TITLE_CASE(p, p, lenp);
1686 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1688 /* We have the latin1-range values compiled into the core, so just use
1689 * those, converting the result to utf8. Since the result is always just
1690 * one character, we allow <p> to be NULL */
1692 U8 converted = toLOWER_LATIN1(c);
1695 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1700 *p = UTF8_TWO_BYTE_HI(converted);
1701 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1709 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1713 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1716 return to_lower_latin1((U8) c, p, lenp);
1719 uvchr_to_utf8(p, c);
1720 return CALL_LOWER_CASE(p, p, lenp);
1724 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1726 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1727 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1728 * FOLD_FLAGS_FULL iff full folding is to be used;
1730 * Not to be used for locale folds
1735 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1737 assert (! (flags & FOLD_FLAGS_LOCALE));
1739 if (c == MICRO_SIGN) {
1740 converted = GREEK_SMALL_LETTER_MU;
1742 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1744 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1745 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1746 * under those circumstances. */
1747 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1748 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1749 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1751 return LATIN_SMALL_LETTER_LONG_S;
1760 else { /* In this range the fold of all other characters is their lower
1762 converted = toLOWER_LATIN1(c);
1765 if (UVCHR_IS_INVARIANT(converted)) {
1766 *p = (U8) converted;
1770 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1771 *p = UTF8_TWO_BYTE_LO(converted);
1779 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1782 /* Not currently externally documented, and subject to change
1783 * <flags> bits meanings:
1784 * FOLD_FLAGS_FULL iff full folding is to be used;
1785 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1786 * locale are to be used.
1787 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1790 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1792 /* Tread a UTF-8 locale as not being in locale at all */
1793 if (IN_UTF8_CTYPE_LOCALE) {
1794 flags &= ~FOLD_FLAGS_LOCALE;
1798 UV result = _to_fold_latin1((U8) c, p, lenp,
1799 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1800 /* It is illegal for the fold to cross the 255/256 boundary under
1801 * locale; in this case return the original */
1802 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1807 /* If no special needs, just use the macro */
1808 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1809 uvchr_to_utf8(p, c);
1810 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1812 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1813 the special flags. */
1814 U8 utf8_c[UTF8_MAXBYTES + 1];
1815 uvchr_to_utf8(utf8_c, c);
1816 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1820 PERL_STATIC_INLINE bool
1821 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1822 const char *const swashname, SV* const invlist)
1824 /* returns a boolean giving whether or not the UTF8-encoded character that
1825 * starts at <p> is in the swash indicated by <swashname>. <swash>
1826 * contains a pointer to where the swash indicated by <swashname>
1827 * is to be stored; which this routine will do, so that future calls will
1828 * look at <*swash> and only generate a swash if it is not null. <invlist>
1829 * is NULL or an inversion list that defines the swash. If not null, it
1830 * saves time during initialization of the swash.
1832 * Note that it is assumed that the buffer length of <p> is enough to
1833 * contain all the bytes that comprise the character. Thus, <*p> should
1834 * have been checked before this call for mal-formedness enough to assure
1839 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1841 /* The API should have included a length for the UTF-8 character in <p>,
1842 * but it doesn't. We therefore assume that p has been validated at least
1843 * as far as there being enough bytes available in it to accommodate the
1844 * character without reading beyond the end, and pass that number on to the
1845 * validating routine */
1846 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
1847 if (ckWARN_d(WARN_UTF8)) {
1848 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1849 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1850 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1851 what the malformation is */
1852 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1858 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1859 *swash = _core_swash_init("utf8",
1861 /* Only use the name if there is no inversion
1862 * list; otherwise will go out to disk */
1863 (invlist) ? "" : swashname,
1865 &PL_sv_undef, 1, 0, invlist, &flags);
1868 return swash_fetch(*swash, p, TRUE) != 0;
1872 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1876 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1878 assert(classnum < _FIRST_NON_SWASH_CC);
1880 return is_utf8_common(p,
1881 &PL_utf8_swash_ptrs[classnum],
1882 swash_property_names[classnum],
1883 PL_XPosix_ptrs[classnum]);
1887 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
1891 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1893 return S_is_utf8_idfirst(aTHX_ p);
1897 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1901 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1905 /* is_utf8_idstart would be more logical. */
1906 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1910 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1915 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1917 if (! PL_utf8_perl_idstart) {
1918 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1920 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
1924 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1929 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1931 if (! PL_utf8_perl_idcont) {
1932 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1934 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
1939 Perl_is_utf8_idcont(pTHX_ const U8 *p)
1943 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
1945 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1949 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
1953 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
1955 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1959 Perl__is_utf8_mark(pTHX_ const U8 *p)
1963 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1965 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1969 =for apidoc to_utf8_case
1971 C<p> contains the pointer to the UTF-8 string encoding
1972 the character that is being converted. This routine assumes that the character
1973 at C<p> is well-formed.
1975 C<ustrp> is a pointer to the character buffer to put the
1976 conversion result to. C<lenp> is a pointer to the length
1979 C<swashp> is a pointer to the swash to use.
1981 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1982 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
1983 but not always, a multicharacter mapping), is tried first.
1985 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1986 any special mappings; C<""> means to use the special mappings. Values other
1987 than these two are treated as the name of the hash containing the special
1988 mappings, like C<"utf8::ToSpecLower">.
1990 C<normal> is a string like "ToLower" which means the swash
1996 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1997 SV **swashp, const char *normal, const char *special)
2001 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
2003 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2005 /* Note that swash_fetch() doesn't output warnings for these because it
2006 * assumes we will */
2007 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2008 if (uv1 <= UNICODE_SURROGATE_LAST) {
2009 if (ckWARN_d(WARN_SURROGATE)) {
2010 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2011 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2012 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2015 else if (UNICODE_IS_SUPER(uv1)) {
2016 if (ckWARN_d(WARN_NON_UNICODE)) {
2017 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2018 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2019 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2023 /* Note that non-characters are perfectly legal, so no warning should
2027 if (!*swashp) /* load on-demand */
2028 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2031 /* It might be "special" (sometimes, but not always,
2032 * a multicharacter mapping) */
2036 /* If passed in the specials name, use that; otherwise use any
2037 * given in the swash */
2038 if (*special != '\0') {
2039 hv = get_hv(special, 0);
2042 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2044 hv = MUTABLE_HV(SvRV(*svp));
2049 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
2054 s = SvPV_const(*svp, len);
2057 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2059 Copy(s, ustrp, len, U8);
2064 if (!len && *swashp) {
2065 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
2068 /* It was "normal" (a single character mapping). */
2069 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2077 return valid_utf8_to_uvchr(ustrp, 0);
2080 /* Here, there was no mapping defined, which means that the code point maps
2081 * to itself. Return the inputs */
2083 if (p != ustrp) { /* Don't copy onto itself */
2084 Copy(p, ustrp, len, U8);
2095 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2097 /* This is called when changing the case of a utf8-encoded character above
2098 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2099 * result contains a character that crosses the 255/256 boundary, disallow
2100 * the change, and return the original code point. See L<perlfunc/lc> for
2103 * p points to the original string whose case was changed; assumed
2104 * by this routine to be well-formed
2105 * result the code point of the first character in the changed-case string
2106 * ustrp points to the changed-case string (<result> represents its first char)
2107 * lenp points to the length of <ustrp> */
2109 UV original; /* To store the first code point of <p> */
2111 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2113 assert(UTF8_IS_ABOVE_LATIN1(*p));
2115 /* We know immediately if the first character in the string crosses the
2116 * boundary, so can skip */
2119 /* Look at every character in the result; if any cross the
2120 * boundary, the whole thing is disallowed */
2121 U8* s = ustrp + UTF8SKIP(ustrp);
2122 U8* e = ustrp + *lenp;
2124 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2130 /* Here, no characters crossed, result is ok as-is */
2136 /* Failed, have to return the original */
2137 original = valid_utf8_to_uvchr(p, lenp);
2138 Copy(p, ustrp, *lenp, char);
2143 =for apidoc to_utf8_upper
2145 Instead use L</toUPPER_utf8>.
2149 /* Not currently externally documented, and subject to change:
2150 * <flags> is set iff iff the rules from the current underlying locale are to
2154 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2160 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2162 if (flags && IN_UTF8_CTYPE_LOCALE) {
2166 if (UTF8_IS_INVARIANT(*p)) {
2168 result = toUPPER_LC(*p);
2171 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2174 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2176 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2177 result = toUPPER_LC(c);
2180 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2184 else { /* utf8, ord above 255 */
2185 result = CALL_UPPER_CASE(p, ustrp, lenp);
2188 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2193 /* Here, used locale rules. Convert back to utf8 */
2194 if (UTF8_IS_INVARIANT(result)) {
2195 *ustrp = (U8) result;
2199 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2200 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2208 =for apidoc to_utf8_title
2210 Instead use L</toTITLE_utf8>.
2214 /* Not currently externally documented, and subject to change:
2215 * <flags> is set iff the rules from the current underlying locale are to be
2216 * used. Since titlecase is not defined in POSIX, for other than a
2217 * UTF-8 locale, uppercase is used instead for code points < 256.
2221 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2227 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2229 if (flags && IN_UTF8_CTYPE_LOCALE) {
2233 if (UTF8_IS_INVARIANT(*p)) {
2235 result = toUPPER_LC(*p);
2238 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2241 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2243 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2244 result = toUPPER_LC(c);
2247 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2251 else { /* utf8, ord above 255 */
2252 result = CALL_TITLE_CASE(p, ustrp, lenp);
2255 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2260 /* Here, used locale rules. Convert back to utf8 */
2261 if (UTF8_IS_INVARIANT(result)) {
2262 *ustrp = (U8) result;
2266 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2267 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2275 =for apidoc to_utf8_lower
2277 Instead use L</toLOWER_utf8>.
2281 /* Not currently externally documented, and subject to change:
2282 * <flags> is set iff iff the rules from the current underlying locale are to
2287 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2293 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2295 if (flags && IN_UTF8_CTYPE_LOCALE) {
2299 if (UTF8_IS_INVARIANT(*p)) {
2301 result = toLOWER_LC(*p);
2304 return to_lower_latin1(*p, ustrp, lenp);
2307 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2309 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2310 result = toLOWER_LC(c);
2313 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2317 else { /* utf8, ord above 255 */
2318 result = CALL_LOWER_CASE(p, ustrp, lenp);
2321 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2327 /* Here, used locale rules. Convert back to utf8 */
2328 if (UTF8_IS_INVARIANT(result)) {
2329 *ustrp = (U8) result;
2333 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2334 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2342 =for apidoc to_utf8_fold
2344 Instead use L</toFOLD_utf8>.
2348 /* Not currently externally documented, and subject to change,
2350 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2351 * locale are to be used.
2352 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2353 * otherwise simple folds
2354 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2359 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2365 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2367 /* These are mutually exclusive */
2368 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2370 assert(p != ustrp); /* Otherwise overwrites */
2372 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2373 flags &= ~FOLD_FLAGS_LOCALE;
2376 if (UTF8_IS_INVARIANT(*p)) {
2377 if (flags & FOLD_FLAGS_LOCALE) {
2378 result = toFOLD_LC(*p);
2381 return _to_fold_latin1(*p, ustrp, lenp,
2382 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2385 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2386 if (flags & FOLD_FLAGS_LOCALE) {
2387 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2388 result = toFOLD_LC(c);
2391 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2393 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2396 else { /* utf8, ord above 255 */
2397 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2399 if (flags & FOLD_FLAGS_LOCALE) {
2401 /* Special case these two characters, as what normally gets
2402 * returned under locale doesn't work */
2403 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2404 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2405 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2409 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2410 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2411 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2413 goto return_ligature_st;
2415 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2417 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2421 /* This is called when changing the case of a utf8-encoded
2422 * character above the ASCII range, and the result should not
2423 * contain an ASCII character. */
2425 UV original; /* To store the first code point of <p> */
2427 /* Look at every character in the result; if any cross the
2428 * boundary, the whole thing is disallowed */
2430 U8* e = ustrp + *lenp;
2433 /* Crossed, have to return the original */
2434 original = valid_utf8_to_uvchr(p, lenp);
2436 /* But in these instances, there is an alternative we can
2437 * return that is valid */
2438 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2439 || original == LATIN_SMALL_LETTER_SHARP_S)
2443 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2444 goto return_ligature_st;
2446 Copy(p, ustrp, *lenp, char);
2452 /* Here, no characters crossed, result is ok as-is */
2457 /* Here, used locale rules. Convert back to utf8 */
2458 if (UTF8_IS_INVARIANT(result)) {
2459 *ustrp = (U8) result;
2463 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2464 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2471 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2472 * folds to a string of two of these characters. By returning this
2473 * instead, then, e.g.,
2474 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2477 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2478 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2480 return LATIN_SMALL_LETTER_LONG_S;
2483 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2484 * have the other one fold to it */
2486 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2487 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2488 return LATIN_SMALL_LIGATURE_ST;
2492 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2493 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2494 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2498 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2500 PERL_ARGS_ASSERT_SWASH_INIT;
2502 /* Returns a copy of a swash initiated by the called function. This is the
2503 * public interface, and returning a copy prevents others from doing
2504 * mischief on the original */
2506 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2510 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2512 /* Initialize and return a swash, creating it if necessary. It does this
2513 * by calling utf8_heavy.pl in the general case. The returned value may be
2514 * the swash's inversion list instead if the input parameters allow it.
2515 * Which is returned should be immaterial to callers, as the only
2516 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2517 * and swash_to_invlist() handle both these transparently.
2519 * This interface should only be used by functions that won't destroy or
2520 * adversely change the swash, as doing so affects all other uses of the
2521 * swash in the program; the general public should use 'Perl_swash_init'
2524 * pkg is the name of the package that <name> should be in.
2525 * name is the name of the swash to find. Typically it is a Unicode
2526 * property name, including user-defined ones
2527 * listsv is a string to initialize the swash with. It must be of the form
2528 * documented as the subroutine return value in
2529 * L<perlunicode/User-Defined Character Properties>
2530 * minbits is the number of bits required to represent each data element.
2531 * It is '1' for binary properties.
2532 * none I (khw) do not understand this one, but it is used only in tr///.
2533 * invlist is an inversion list to initialize the swash with (or NULL)
2534 * flags_p if non-NULL is the address of various input and output flag bits
2535 * to the routine, as follows: ('I' means is input to the routine;
2536 * 'O' means output from the routine. Only flags marked O are
2537 * meaningful on return.)
2538 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2539 * came from a user-defined property. (I O)
2540 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2541 * when the swash cannot be located, to simply return NULL. (I)
2542 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2543 * return of an inversion list instead of a swash hash if this routine
2544 * thinks that would result in faster execution of swash_fetch() later
2547 * Thus there are three possible inputs to find the swash: <name>,
2548 * <listsv>, and <invlist>. At least one must be specified. The result
2549 * will be the union of the specified ones, although <listsv>'s various
2550 * actions can intersect, etc. what <name> gives. To avoid going out to
2551 * disk at all, <invlist> should specify completely what the swash should
2552 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2554 * <invlist> is only valid for binary properties */
2557 SV* retval = &PL_sv_undef;
2558 HV* swash_hv = NULL;
2559 const int invlist_swash_boundary =
2560 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2561 ? 512 /* Based on some benchmarking, but not extensive, see commit
2563 : -1; /* Never return just an inversion list */
2565 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2566 assert(! invlist || minbits == 1);
2568 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2570 if (listsv != &PL_sv_undef || strNE(name, "")) {
2572 const size_t pkg_len = strlen(pkg);
2573 const size_t name_len = strlen(name);
2574 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2578 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2580 PUSHSTACKi(PERLSI_MAGIC);
2584 /* We might get here via a subroutine signature which uses a utf8
2585 * parameter name, at which point PL_subname will have been set
2586 * but not yet used. */
2587 save_item(PL_subname);
2588 if (PL_parser && PL_parser->error_count)
2589 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2590 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2591 if (!method) { /* demand load utf8 */
2593 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2594 GvSV(PL_errgv) = NULL;
2595 /* It is assumed that callers of this routine are not passing in
2596 * any user derived data. */
2597 /* Need to do this after save_re_context() as it will set
2598 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2599 * in Perl_magic_get). Even line to create errsv_save can turn on
2601 #ifndef NO_TAINT_SUPPORT
2602 SAVEBOOL(TAINT_get);
2605 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2608 /* Not ERRSV, as there is no need to vivify a scalar we are
2609 about to discard. */
2610 SV * const errsv = GvSV(PL_errgv);
2611 if (!SvTRUE(errsv)) {
2612 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2613 SvREFCNT_dec(errsv);
2621 mPUSHp(pkg, pkg_len);
2622 mPUSHp(name, name_len);
2627 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2628 GvSV(PL_errgv) = NULL;
2629 /* If we already have a pointer to the method, no need to use
2630 * call_method() to repeat the lookup. */
2632 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2633 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2635 retval = *PL_stack_sp--;
2636 SvREFCNT_inc(retval);
2639 /* Not ERRSV. See above. */
2640 SV * const errsv = GvSV(PL_errgv);
2641 if (!SvTRUE(errsv)) {
2642 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2643 SvREFCNT_dec(errsv);
2648 if (IN_PERL_COMPILETIME) {
2649 CopHINTS_set(PL_curcop, PL_hints);
2651 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2654 /* If caller wants to handle missing properties, let them */
2655 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2659 "Can't find Unicode property definition \"%"SVf"\"",
2661 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2663 } /* End of calling the module to find the swash */
2665 /* If this operation fetched a swash, and we will need it later, get it */
2666 if (retval != &PL_sv_undef
2667 && (minbits == 1 || (flags_p
2669 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2671 swash_hv = MUTABLE_HV(SvRV(retval));
2673 /* If we don't already know that there is a user-defined component to
2674 * this swash, and the user has indicated they wish to know if there is
2675 * one (by passing <flags_p>), find out */
2676 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2677 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2678 if (user_defined && SvUV(*user_defined)) {
2679 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2684 /* Make sure there is an inversion list for binary properties */
2686 SV** swash_invlistsvp = NULL;
2687 SV* swash_invlist = NULL;
2688 bool invlist_in_swash_is_valid = FALSE;
2689 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2690 an unclaimed reference count */
2692 /* If this operation fetched a swash, get its already existing
2693 * inversion list, or create one for it */
2696 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2697 if (swash_invlistsvp) {
2698 swash_invlist = *swash_invlistsvp;
2699 invlist_in_swash_is_valid = TRUE;
2702 swash_invlist = _swash_to_invlist(retval);
2703 swash_invlist_unclaimed = TRUE;
2707 /* If an inversion list was passed in, have to include it */
2710 /* Any fetched swash will by now have an inversion list in it;
2711 * otherwise <swash_invlist> will be NULL, indicating that we
2712 * didn't fetch a swash */
2713 if (swash_invlist) {
2715 /* Add the passed-in inversion list, which invalidates the one
2716 * already stored in the swash */
2717 invlist_in_swash_is_valid = FALSE;
2718 _invlist_union(invlist, swash_invlist, &swash_invlist);
2722 /* Here, there is no swash already. Set up a minimal one, if
2723 * we are going to return a swash */
2724 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2726 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2728 swash_invlist = invlist;
2732 /* Here, we have computed the union of all the passed-in data. It may
2733 * be that there was an inversion list in the swash which didn't get
2734 * touched; otherwise save the computed one */
2735 if (! invlist_in_swash_is_valid
2736 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2738 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2740 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2742 /* We just stole a reference count. */
2743 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2744 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2747 SvREADONLY_on(swash_invlist);
2749 /* Use the inversion list stand-alone if small enough */
2750 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2751 SvREFCNT_dec(retval);
2752 if (!swash_invlist_unclaimed)
2753 SvREFCNT_inc_simple_void_NN(swash_invlist);
2754 retval = newRV_noinc(swash_invlist);
2762 /* This API is wrong for special case conversions since we may need to
2763 * return several Unicode characters for a single Unicode character
2764 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2765 * the lower-level routine, and it is similarly broken for returning
2766 * multiple values. --jhi
2767 * For those, you should use to_utf8_case() instead */
2768 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2771 * Returns the value of property/mapping C<swash> for the first character
2772 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2773 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2774 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2776 * A "swash" is a hash which contains initially the keys/values set up by
2777 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2778 * property for all possible code points. Things are stored in a compact form
2779 * (see utf8_heavy.pl) so that calculation is required to find the actual
2780 * property value for a given code point. As code points are looked up, new
2781 * key/value pairs are added to the hash, so that the calculation doesn't have
2782 * to ever be re-done. Further, each calculation is done, not just for the
2783 * desired one, but for a whole block of code points adjacent to that one.
2784 * For binary properties on ASCII machines, the block is usually for 64 code
2785 * points, starting with a code point evenly divisible by 64. Thus if the
2786 * property value for code point 257 is requested, the code goes out and
2787 * calculates the property values for all 64 code points between 256 and 319,
2788 * and stores these as a single 64-bit long bit vector, called a "swatch",
2789 * under the key for code point 256. The key is the UTF-8 encoding for code
2790 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2791 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2792 * for code point 258 is then requested, this code realizes that it would be
2793 * stored under the key for 256, and would find that value and extract the
2794 * relevant bit, offset from 256.
2796 * Non-binary properties are stored in as many bits as necessary to represent
2797 * their values (32 currently, though the code is more general than that), not
2798 * as single bits, but the principal is the same: the value for each key is a
2799 * vector that encompasses the property values for all code points whose UTF-8
2800 * representations are represented by the key. That is, for all code points
2801 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2805 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2808 HV *const hv = MUTABLE_HV(SvRV(swash));
2813 const U8 *tmps = NULL;
2817 PERL_ARGS_ASSERT_SWASH_FETCH;
2819 /* If it really isn't a hash, it isn't really swash; must be an inversion
2821 if (SvTYPE(hv) != SVt_PVHV) {
2822 return _invlist_contains_cp((SV*)hv,
2824 ? valid_utf8_to_uvchr(ptr, NULL)
2828 /* We store the values in a "swatch" which is a vec() value in a swash
2829 * hash. Code points 0-255 are a single vec() stored with key length
2830 * (klen) 0. All other code points have a UTF-8 representation
2831 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2832 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2833 * length for them is the length of the encoded char - 1. ptr[klen] is the
2834 * final byte in the sequence representing the character */
2835 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2840 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2843 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
2846 klen = UTF8SKIP(ptr) - 1;
2848 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2849 * the vec is the final byte in the sequence. (In EBCDIC this is
2850 * converted to I8 to get consecutive values.) To help you visualize
2852 * Straight 1047 After final byte
2853 * UTF-8 UTF-EBCDIC I8 transform
2854 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2855 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2857 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2858 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2860 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2861 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2863 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2864 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2866 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2867 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2869 * (There are no discontinuities in the elided (...) entries.)
2870 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2871 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2872 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2873 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2874 * index into the vec() swatch (after subtracting 0x80, which we
2875 * actually do with an '&').
2876 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2877 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2878 * dicontinuities which go away by transforming it into I8, and we
2879 * effectively subtract 0xA0 to get the index. */
2880 needents = (1 << UTF_ACCUMULATION_SHIFT);
2881 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2885 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2886 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2887 * it's nothing to sniff at.) Pity we usually come through at least
2888 * two function calls to get here...
2890 * NB: this code assumes that swatches are never modified, once generated!
2893 if (hv == PL_last_swash_hv &&
2894 klen == PL_last_swash_klen &&
2895 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2897 tmps = PL_last_swash_tmps;
2898 slen = PL_last_swash_slen;
2901 /* Try our second-level swatch cache, kept in a hash. */
2902 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2904 /* If not cached, generate it via swatch_get */
2905 if (!svp || !SvPOK(*svp)
2906 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2909 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2910 swatch = swatch_get(swash,
2911 code_point & ~((UV)needents - 1),
2914 else { /* For the first 256 code points, the swatch has a key of
2916 swatch = swatch_get(swash, 0, needents);
2919 if (IN_PERL_COMPILETIME)
2920 CopHINTS_set(PL_curcop, PL_hints);
2922 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2924 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2925 || (slen << 3) < needents)
2926 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2927 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2928 svp, tmps, (UV)slen, (UV)needents);
2931 PL_last_swash_hv = hv;
2932 assert(klen <= sizeof(PL_last_swash_key));
2933 PL_last_swash_klen = (U8)klen;
2934 /* FIXME change interpvar.h? */
2935 PL_last_swash_tmps = (U8 *) tmps;
2936 PL_last_swash_slen = slen;
2938 Copy(ptr, PL_last_swash_key, klen, U8);
2941 switch ((int)((slen << 3) / needents)) {
2943 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2945 return ((UV) tmps[off]);
2949 ((UV) tmps[off ] << 8) +
2950 ((UV) tmps[off + 1]);
2954 ((UV) tmps[off ] << 24) +
2955 ((UV) tmps[off + 1] << 16) +
2956 ((UV) tmps[off + 2] << 8) +
2957 ((UV) tmps[off + 3]);
2959 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2960 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2961 NORETURN_FUNCTION_END;
2964 /* Read a single line of the main body of the swash input text. These are of
2967 * where each number is hex. The first two numbers form the minimum and
2968 * maximum of a range, and the third is the value associated with the range.
2969 * Not all swashes should have a third number
2971 * On input: l points to the beginning of the line to be examined; it points
2972 * to somewhere in the string of the whole input text, and is
2973 * terminated by a \n or the null string terminator.
2974 * lend points to the null terminator of that string
2975 * wants_value is non-zero if the swash expects a third number
2976 * typestr is the name of the swash's mapping, like 'ToLower'
2977 * On output: *min, *max, and *val are set to the values read from the line.
2978 * returns a pointer just beyond the line examined. If there was no
2979 * valid min number on the line, returns lend+1
2983 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2984 const bool wants_value, const U8* const typestr)
2986 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2987 STRLEN numlen; /* Length of the number */
2988 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2989 | PERL_SCAN_DISALLOW_PREFIX
2990 | PERL_SCAN_SILENT_NON_PORTABLE;
2992 /* nl points to the next \n in the scan */
2993 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2995 /* Get the first number on the line: the range minimum */
2997 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2998 if (numlen) /* If found a hex number, position past it */
3000 else if (nl) { /* Else, go handle next line, if any */
3001 return nl + 1; /* 1 is length of "\n" */
3003 else { /* Else, no next line */
3004 return lend + 1; /* to LIST's end at which \n is not found */
3007 /* The max range value follows, separated by a BLANK */
3010 flags = PERL_SCAN_SILENT_ILLDIGIT
3011 | PERL_SCAN_DISALLOW_PREFIX
3012 | PERL_SCAN_SILENT_NON_PORTABLE;
3014 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3017 else /* If no value here, it is a single element range */
3020 /* Non-binary tables have a third entry: what the first element of the
3021 * range maps to. The map for those currently read here is in hex */
3025 flags = PERL_SCAN_SILENT_ILLDIGIT
3026 | PERL_SCAN_DISALLOW_PREFIX
3027 | PERL_SCAN_SILENT_NON_PORTABLE;
3029 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3038 /* diag_listed_as: To%s: illegal mapping '%s' */
3039 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3045 *val = 0; /* bits == 1, then any val should be ignored */
3047 else { /* Nothing following range min, should be single element with no
3053 /* diag_listed_as: To%s: illegal mapping '%s' */
3054 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3058 *val = 0; /* bits == 1, then val should be ignored */
3061 /* Position to next line if any, or EOF */
3071 * Returns a swatch (a bit vector string) for a code point sequence
3072 * that starts from the value C<start> and comprises the number C<span>.
3073 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3074 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3077 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3080 U8 *l, *lend, *x, *xend, *s, *send;
3081 STRLEN lcur, xcur, scur;
3082 HV *const hv = MUTABLE_HV(SvRV(swash));
3083 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3085 SV** listsvp = NULL; /* The string containing the main body of the table */
3086 SV** extssvp = NULL;
3087 SV** invert_it_svp = NULL;
3090 STRLEN octets; /* if bits == 1, then octets == 0 */
3092 UV end = start + span;
3094 if (invlistsvp == NULL) {
3095 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3096 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3097 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3098 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3099 listsvp = hv_fetchs(hv, "LIST", FALSE);
3100 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3102 bits = SvUV(*bitssvp);
3103 none = SvUV(*nonesvp);
3104 typestr = (U8*)SvPV_nolen(*typesvp);
3110 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3112 PERL_ARGS_ASSERT_SWATCH_GET;
3114 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3115 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3119 /* If overflowed, use the max possible */
3125 /* create and initialize $swatch */
3126 scur = octets ? (span * octets) : (span + 7) / 8;
3127 swatch = newSV(scur);
3129 s = (U8*)SvPVX(swatch);
3130 if (octets && none) {
3131 const U8* const e = s + scur;
3134 *s++ = (U8)(none & 0xff);
3135 else if (bits == 16) {
3136 *s++ = (U8)((none >> 8) & 0xff);
3137 *s++ = (U8)( none & 0xff);
3139 else if (bits == 32) {
3140 *s++ = (U8)((none >> 24) & 0xff);
3141 *s++ = (U8)((none >> 16) & 0xff);
3142 *s++ = (U8)((none >> 8) & 0xff);
3143 *s++ = (U8)( none & 0xff);
3149 (void)memzero((U8*)s, scur + 1);
3151 SvCUR_set(swatch, scur);
3152 s = (U8*)SvPVX(swatch);
3154 if (invlistsvp) { /* If has an inversion list set up use that */
3155 _invlist_populate_swatch(*invlistsvp, start, end, s);
3159 /* read $swash->{LIST} */
3160 l = (U8*)SvPV(*listsvp, lcur);
3163 UV min, max, val, upper;
3164 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3165 cBOOL(octets), typestr);
3170 /* If looking for something beyond this range, go try the next one */
3174 /* <end> is generally 1 beyond where we want to set things, but at the
3175 * platform's infinity, where we can't go any higher, we want to
3176 * include the code point at <end> */
3179 : (max != UV_MAX || end != UV_MAX)
3186 if (!none || val < none) {
3191 for (key = min; key <= upper; key++) {
3193 /* offset must be non-negative (start <= min <= key < end) */
3194 offset = octets * (key - start);
3196 s[offset] = (U8)(val & 0xff);
3197 else if (bits == 16) {
3198 s[offset ] = (U8)((val >> 8) & 0xff);
3199 s[offset + 1] = (U8)( val & 0xff);
3201 else if (bits == 32) {
3202 s[offset ] = (U8)((val >> 24) & 0xff);
3203 s[offset + 1] = (U8)((val >> 16) & 0xff);
3204 s[offset + 2] = (U8)((val >> 8) & 0xff);
3205 s[offset + 3] = (U8)( val & 0xff);
3208 if (!none || val < none)
3212 else { /* bits == 1, then val should be ignored */
3217 for (key = min; key <= upper; key++) {
3218 const STRLEN offset = (STRLEN)(key - start);
3219 s[offset >> 3] |= 1 << (offset & 7);
3224 /* Invert if the data says it should be. Assumes that bits == 1 */
3225 if (invert_it_svp && SvUV(*invert_it_svp)) {
3227 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3228 * be 0, and their inversion should also be 0, as we don't succeed any
3229 * Unicode property matches for non-Unicode code points */
3230 if (start <= PERL_UNICODE_MAX) {
3232 /* The code below assumes that we never cross the
3233 * Unicode/above-Unicode boundary in a range, as otherwise we would
3234 * have to figure out where to stop flipping the bits. Since this
3235 * boundary is divisible by a large power of 2, and swatches comes
3236 * in small powers of 2, this should be a valid assumption */
3237 assert(start + span - 1 <= PERL_UNICODE_MAX);
3247 /* read $swash->{EXTRAS}
3248 * This code also copied to swash_to_invlist() below */
3249 x = (U8*)SvPV(*extssvp, xcur);
3257 SV **otherbitssvp, *other;
3261 const U8 opc = *x++;
3265 nl = (U8*)memchr(x, '\n', xend - x);
3267 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3269 x = nl + 1; /* 1 is length of "\n" */
3273 x = xend; /* to EXTRAS' end at which \n is not found */
3280 namelen = nl - namestr;
3284 namelen = xend - namestr;
3288 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3289 otherhv = MUTABLE_HV(SvRV(*othersvp));
3290 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3291 otherbits = (STRLEN)SvUV(*otherbitssvp);
3292 if (bits < otherbits)
3293 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3294 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3296 /* The "other" swatch must be destroyed after. */
3297 other = swatch_get(*othersvp, start, span);
3298 o = (U8*)SvPV(other, olen);
3301 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3303 s = (U8*)SvPV(swatch, slen);
3304 if (bits == 1 && otherbits == 1) {
3306 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3307 "mismatch, slen=%"UVuf", olen=%"UVuf,
3308 (UV)slen, (UV)olen);
3332 STRLEN otheroctets = otherbits >> 3;
3334 U8* const send = s + slen;
3339 if (otherbits == 1) {
3340 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3344 STRLEN vlen = otheroctets;
3352 if (opc == '+' && otherval)
3353 NOOP; /* replace with otherval */
3354 else if (opc == '!' && !otherval)
3356 else if (opc == '-' && otherval)
3358 else if (opc == '&' && !otherval)
3361 s += octets; /* no replacement */
3366 *s++ = (U8)( otherval & 0xff);
3367 else if (bits == 16) {
3368 *s++ = (U8)((otherval >> 8) & 0xff);
3369 *s++ = (U8)( otherval & 0xff);
3371 else if (bits == 32) {
3372 *s++ = (U8)((otherval >> 24) & 0xff);
3373 *s++ = (U8)((otherval >> 16) & 0xff);
3374 *s++ = (U8)((otherval >> 8) & 0xff);
3375 *s++ = (U8)( otherval & 0xff);
3379 sv_free(other); /* through with it! */
3385 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3388 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3389 * Can't be used on a property that is subject to user override, as it
3390 * relies on the value of SPECIALS in the swash which would be set by
3391 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3392 * for overridden properties
3394 * Returns a hash which is the inversion and closure of a swash mapping.
3395 * For example, consider the input lines:
3400 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3401 * 006C. The value for each key is an array. For 006C, the array would
3402 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3403 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3405 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3406 * keys are only code points that are folded-to, so it isn't a full closure.
3408 * Essentially, for any code point, it gives all the code points that map to
3409 * it, or the list of 'froms' for that point.
3411 * Currently it ignores any additions or deletions from other swashes,
3412 * looking at just the main body of the swash, and if there are SPECIALS
3413 * in the swash, at that hash
3415 * The specials hash can be extra code points, and most likely consists of
3416 * maps from single code points to multiple ones (each expressed as a string
3417 * of utf8 characters). This function currently returns only 1-1 mappings.
3418 * However consider this possible input in the specials hash:
3419 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3420 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3422 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3423 * currently handle. But it also means that FB05 and FB06 are equivalent in
3424 * a 1-1 mapping which we should handle, and this relationship may not be in
3425 * the main table. Therefore this function examines all the multi-char
3426 * sequences and adds the 1-1 mappings that come out of that. */
3430 HV *const hv = MUTABLE_HV(SvRV(swash));
3432 /* The string containing the main body of the table. This will have its
3433 * assertion fail if the swash has been converted to its inversion list */
3434 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3436 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3437 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3438 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3439 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3440 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3441 const STRLEN bits = SvUV(*bitssvp);
3442 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3443 const UV none = SvUV(*nonesvp);
3444 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3448 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3450 /* Must have at least 8 bits to get the mappings */
3451 if (bits != 8 && bits != 16 && bits != 32) {
3452 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3456 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3457 mapping to more than one character */
3459 /* Construct an inverse mapping hash for the specials */
3460 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3461 HV * specials_inverse = newHV();
3462 char *char_from; /* the lhs of the map */
3463 I32 from_len; /* its byte length */
3464 char *char_to; /* the rhs of the map */
3465 I32 to_len; /* its byte length */
3466 SV *sv_to; /* and in a sv */
3467 AV* from_list; /* list of things that map to each 'to' */
3469 hv_iterinit(specials_hv);
3471 /* The keys are the characters (in utf8) that map to the corresponding
3472 * utf8 string value. Iterate through the list creating the inverse
3474 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3476 if (! SvPOK(sv_to)) {
3477 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3478 "unexpectedly is not a string, flags=%lu",
3479 (unsigned long)SvFLAGS(sv_to));
3481 /*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)));*/
3483 /* Each key in the inverse list is a mapped-to value, and the key's
3484 * hash value is a list of the strings (each in utf8) that map to
3485 * it. Those strings are all one character long */
3486 if ((listp = hv_fetch(specials_inverse,
3490 from_list = (AV*) *listp;
3492 else { /* No entry yet for it: create one */
3493 from_list = newAV();
3494 if (! hv_store(specials_inverse,
3497 (SV*) from_list, 0))
3499 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3503 /* Here have the list associated with this 'to' (perhaps newly
3504 * created and empty). Just add to it. Note that we ASSUME that
3505 * the input is guaranteed to not have duplications, so we don't
3506 * check for that. Duplications just slow down execution time. */
3507 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3510 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3511 * it looking for cases like the FB05/FB06 examples above. There would
3512 * be an entry in the hash like
3513 * 'st' => [ FB05, FB06 ]
3514 * In this example we will create two lists that get stored in the
3515 * returned hash, 'ret':
3516 * FB05 => [ FB05, FB06 ]
3517 * FB06 => [ FB05, FB06 ]
3519 * Note that there is nothing to do if the array only has one element.
3520 * (In the normal 1-1 case handled below, we don't have to worry about
3521 * two lists, as everything gets tied to the single list that is
3522 * generated for the single character 'to'. But here, we are omitting
3523 * that list, ('st' in the example), so must have multiple lists.) */
3524 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3525 &char_to, &to_len)))
3527 if (av_tindex(from_list) > 0) {
3530 /* We iterate over all combinations of i,j to place each code
3531 * point on each list */
3532 for (i = 0; i <= av_tindex(from_list); i++) {
3534 AV* i_list = newAV();
3535 SV** entryp = av_fetch(from_list, i, FALSE);
3536 if (entryp == NULL) {
3537 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3539 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3540 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3542 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3543 (SV*) i_list, FALSE))
3545 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3548 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3549 for (j = 0; j <= av_tindex(from_list); j++) {
3550 entryp = av_fetch(from_list, j, FALSE);
3551 if (entryp == NULL) {
3552 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3555 /* When i==j this adds itself to the list */
3556 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3557 (U8*) SvPVX(*entryp),
3558 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3560 /*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));*/
3565 SvREFCNT_dec(specials_inverse); /* done with it */
3566 } /* End of specials */
3568 /* read $swash->{LIST} */
3569 l = (U8*)SvPV(*listsvp, lcur);
3572 /* Go through each input line */
3576 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3577 cBOOL(octets), typestr);
3582 /* Each element in the range is to be inverted */
3583 for (inverse = min; inverse <= max; inverse++) {
3587 bool found_key = FALSE;
3588 bool found_inverse = FALSE;
3590 /* The key is the inverse mapping */
3591 char key[UTF8_MAXBYTES+1];
3592 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3593 STRLEN key_len = key_end - key;
3595 /* Get the list for the map */
3596 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3597 list = (AV*) *listp;
3599 else { /* No entry yet for it: create one */
3601 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3602 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3606 /* Look through list to see if this inverse mapping already is
3607 * listed, or if there is a mapping to itself already */
3608 for (i = 0; i <= av_tindex(list); i++) {
3609 SV** entryp = av_fetch(list, i, FALSE);
3611 if (entryp == NULL) {
3612 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3615 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3616 if (SvUV(entry) == val) {
3619 if (SvUV(entry) == inverse) {
3620 found_inverse = TRUE;
3623 /* No need to continue searching if found everything we are
3625 if (found_key && found_inverse) {
3630 /* Make sure there is a mapping to itself on the list */
3632 av_push(list, newSVuv(val));
3633 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3637 /* Simply add the value to the list */
3638 if (! found_inverse) {
3639 av_push(list, newSVuv(inverse));
3640 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3643 /* swatch_get() increments the value of val for each element in the
3644 * range. That makes more compact tables possible. You can
3645 * express the capitalization, for example, of all consecutive
3646 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3647 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3648 * and it's not documented; it appears to be used only in
3649 * implementing tr//; I copied the semantics from swatch_get(), just
3651 if (!none || val < none) {
3661 Perl__swash_to_invlist(pTHX_ SV* const swash)
3664 /* Subject to change or removal. For use only in one place in regcomp.c.
3665 * Ownership is given to one reference count in the returned SV* */
3670 HV *const hv = MUTABLE_HV(SvRV(swash));
3671 UV elements = 0; /* Number of elements in the inversion list */
3681 STRLEN octets; /* if bits == 1, then octets == 0 */
3687 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3689 /* If not a hash, it must be the swash's inversion list instead */
3690 if (SvTYPE(hv) != SVt_PVHV) {
3691 return SvREFCNT_inc_simple_NN((SV*) hv);
3694 /* The string containing the main body of the table */
3695 listsvp = hv_fetchs(hv, "LIST", FALSE);
3696 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3697 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3698 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3699 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3701 typestr = (U8*)SvPV_nolen(*typesvp);
3702 bits = SvUV(*bitssvp);
3703 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3705 /* read $swash->{LIST} */
3706 if (SvPOK(*listsvp)) {
3707 l = (U8*)SvPV(*listsvp, lcur);
3710 /* LIST legitimately doesn't contain a string during compilation phases
3711 * of Perl itself, before the Unicode tables are generated. In this
3712 * case, just fake things up by creating an empty list */
3719 if (*l == 'V') { /* Inversion list format */
3720 char *after_strtol = (char *) lend;
3722 UV* other_elements_ptr;
3724 /* The first number is a count of the rest */
3726 elements = Strtoul((char *)l, &after_strtol, 10);
3727 if (elements == 0) {
3728 invlist = _new_invlist(0);
3731 l = (U8 *) after_strtol;
3733 /* Get the 0th element, which is needed to setup the inversion list */
3734 element0 = (UV) Strtoul((char *)l, &after_strtol, 10);
3735 l = (U8 *) after_strtol;
3736 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3739 /* Then just populate the rest of the input */
3740 while (elements-- > 0) {
3742 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3744 *other_elements_ptr++ = (UV) Strtoul((char *)l, &after_strtol, 10);
3745 l = (U8 *) after_strtol;
3751 /* Scan the input to count the number of lines to preallocate array
3752 * size based on worst possible case, which is each line in the input
3753 * creates 2 elements in the inversion list: 1) the beginning of a
3754 * range in the list; 2) the beginning of a range not in the list. */
3755 while ((loc = (strchr(loc, '\n'))) != NULL) {
3760 /* If the ending is somehow corrupt and isn't a new line, add another
3761 * element for the final range that isn't in the inversion list */
3762 if (! (*lend == '\n'
3763 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3768 invlist = _new_invlist(elements);
3770 /* Now go through the input again, adding each range to the list */
3773 UV val; /* Not used by this function */
3775 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
3776 cBOOL(octets), typestr);
3782 invlist = _add_range_to_invlist(invlist, start, end);
3786 /* Invert if the data says it should be */
3787 if (invert_it_svp && SvUV(*invert_it_svp)) {
3788 _invlist_invert(invlist);
3791 /* This code is copied from swatch_get()
3792 * read $swash->{EXTRAS} */
3793 x = (U8*)SvPV(*extssvp, xcur);
3801 SV **otherbitssvp, *other;
3804 const U8 opc = *x++;
3808 nl = (U8*)memchr(x, '\n', xend - x);
3810 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3812 x = nl + 1; /* 1 is length of "\n" */
3816 x = xend; /* to EXTRAS' end at which \n is not found */
3823 namelen = nl - namestr;
3827 namelen = xend - namestr;
3831 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3832 otherhv = MUTABLE_HV(SvRV(*othersvp));
3833 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3834 otherbits = (STRLEN)SvUV(*otherbitssvp);
3836 if (bits != otherbits || bits != 1) {
3837 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3838 "properties, bits=%"UVuf", otherbits=%"UVuf,
3839 (UV)bits, (UV)otherbits);
3842 /* The "other" swatch must be destroyed after. */
3843 other = _swash_to_invlist((SV *)*othersvp);
3845 /* End of code copied from swatch_get() */
3848 _invlist_union(invlist, other, &invlist);
3851 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3854 _invlist_subtract(invlist, other, &invlist);
3857 _invlist_intersection(invlist, other, &invlist);
3862 sv_free(other); /* through with it! */
3865 SvREADONLY_on(invlist);
3870 Perl__get_swash_invlist(pTHX_ SV* const swash)
3874 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3876 if (! SvROK(swash)) {
3880 /* If it really isn't a hash, it isn't really swash; must be an inversion
3882 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3886 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3895 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3897 /* May change: warns if surrogates, non-character code points, or
3898 * non-Unicode code points are in s which has length len bytes. Returns
3899 * TRUE if none found; FALSE otherwise. The only other validity check is
3900 * to make sure that this won't exceed the string's length */
3902 const U8* const e = s + len;
3905 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3908 if (UTF8SKIP(s) > len) {
3909 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3910 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3913 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3915 if (UTF8_IS_SUPER(s)) {
3916 if (ckWARN_d(WARN_NON_UNICODE)) {
3917 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3918 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3919 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3923 else if (UTF8_IS_SURROGATE(s)) {
3924 if (ckWARN_d(WARN_SURROGATE)) {
3925 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3926 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3927 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3932 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3933 && (ckWARN_d(WARN_NONCHAR)))
3935 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3936 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3937 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3948 =for apidoc pv_uni_display
3950 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3951 length C<len>, the displayable version being at most C<pvlim> bytes long
3952 (if longer, the rest is truncated and "..." will be appended).
3954 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
3955 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3956 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3957 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3958 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3959 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3961 The pointer to the PV of the C<dsv> is returned.
3965 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3970 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3974 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3976 /* This serves double duty as a flag and a character to print after
3977 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3981 if (pvlim && SvCUR(dsv) >= pvlim) {
3985 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3987 const unsigned char c = (unsigned char)u & 0xFF;
3988 if (flags & UNI_DISPLAY_BACKSLASH) {
4005 const char string = ok;
4006 sv_catpvs(dsv, "\\");
4007 sv_catpvn(dsv, &string, 1);
4010 /* isPRINT() is the locale-blind version. */
4011 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4012 const char string = c;
4013 sv_catpvn(dsv, &string, 1);
4018 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4021 sv_catpvs(dsv, "...");
4027 =for apidoc sv_uni_display
4029 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4030 the displayable version being at most C<pvlim> bytes long
4031 (if longer, the rest is truncated and "..." will be appended).
4033 The C<flags> argument is as in L</pv_uni_display>().
4035 The pointer to the PV of the C<dsv> is returned.
4040 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4042 const char * const ptr =
4043 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4045 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4047 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4048 SvCUR(ssv), pvlim, flags);
4052 =for apidoc foldEQ_utf8
4054 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4055 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4056 How far into the strings to compare is determined by other input parameters.
4058 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4059 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4060 with respect to C<s2>.
4062 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4063 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4064 scan will not be considered to be a match unless the goal is reached, and
4065 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4068 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4069 considered an end pointer to the position 1 byte past the maximum point
4070 in C<s1> beyond which scanning will not continue under any circumstances.
4071 (This routine assumes that UTF-8 encoded input strings are not malformed;
4072 malformed input can cause it to read past C<pe1>).
4073 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4074 is less than C<s1>+C<l1>, the match will never be successful because it can
4076 get as far as its goal (and in fact is asserted against). Correspondingly for
4077 C<pe2> with respect to C<s2>.
4079 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4080 C<l2> must be non-zero), and if both do, both have to be
4081 reached for a successful match. Also, if the fold of a character is multiple
4082 characters, all of them must be matched (see tr21 reference below for
4085 Upon a successful match, if C<pe1> is non-NULL,
4086 it will be set to point to the beginning of the I<next> character of C<s1>
4087 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4089 For case-insensitiveness, the "casefolding" of Unicode is used
4090 instead of upper/lowercasing both the characters, see
4091 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4095 /* A flags parameter has been added which may change, and hence isn't
4096 * externally documented. Currently it is:
4097 * 0 for as-documented above
4098 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4099 ASCII one, to not match
4100 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4101 * locale are to be used.
4102 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4103 * routine. This allows that step to be skipped.
4104 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4107 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)
4110 const U8 *p1 = (const U8*)s1; /* Point to current char */
4111 const U8 *p2 = (const U8*)s2;
4112 const U8 *g1 = NULL; /* goal for s1 */
4113 const U8 *g2 = NULL;
4114 const U8 *e1 = NULL; /* Don't scan s1 past this */
4115 U8 *f1 = NULL; /* Point to current folded */
4116 const U8 *e2 = NULL;
4118 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4119 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4120 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4122 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4124 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4125 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4126 /* The algorithm is to trial the folds without regard to the flags on
4127 * the first line of the above assert(), and then see if the result
4128 * violates them. This means that the inputs can't be pre-folded to a
4129 * violating result, hence the assert. This could be changed, with the
4130 * addition of extra tests here for the already-folded case, which would
4131 * slow it down. That cost is more than any possible gain for when these
4132 * flags are specified, as the flags indicate /il or /iaa matching which
4133 * is less common than /iu, and I (khw) also believe that real-world /il
4134 * and /iaa matches are most likely to involve code points 0-255, and this
4135 * function only under rare conditions gets called for 0-255. */
4137 if (IN_UTF8_CTYPE_LOCALE) {
4138 flags &= ~FOLDEQ_LOCALE;
4146 g1 = (const U8*)s1 + l1;
4154 g2 = (const U8*)s2 + l2;
4157 /* Must have at least one goal */
4162 /* Will never match if goal is out-of-bounds */
4163 assert(! e1 || e1 >= g1);
4165 /* Here, there isn't an end pointer, or it is beyond the goal. We
4166 * only go as far as the goal */
4170 assert(e1); /* Must have an end for looking at s1 */
4173 /* Same for goal for s2 */
4175 assert(! e2 || e2 >= g2);
4182 /* If both operands are already folded, we could just do a memEQ on the
4183 * whole strings at once, but it would be better if the caller realized
4184 * this and didn't even call us */
4186 /* Look through both strings, a character at a time */
4187 while (p1 < e1 && p2 < e2) {
4189 /* If at the beginning of a new character in s1, get its fold to use
4190 * and the length of the fold. (exception: locale rules just get the
4191 * character to a single byte) */
4193 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4198 /* If in locale matching, we use two sets of rules, depending
4199 * on if the code point is above or below 255. Here, we test
4200 * for and handle locale rules */
4201 if ((flags & FOLDEQ_LOCALE)
4202 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
4204 /* There is no mixing of code points above and below 255. */
4205 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
4209 /* We handle locale rules by converting, if necessary, the
4210 * code point to a single byte. */
4211 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4215 *foldbuf1 = TWO_BYTE_UTF8_TO_NATIVE(*p1, *(p1 + 1));
4219 else if (isASCII(*p1)) { /* Note, that here won't be both
4220 ASCII and using locale rules */
4222 /* If trying to mix non- with ASCII, and not supposed to,
4224 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4228 *foldbuf1 = toFOLD(*p1);
4231 to_utf8_fold(p1, foldbuf1, &n1);
4233 else { /* Not utf8, get utf8 fold */
4234 to_uni_fold(*p1, foldbuf1, &n1);
4240 if (n2 == 0) { /* Same for s2 */
4241 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4246 if ((flags & FOLDEQ_LOCALE)
4247 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4249 /* Here, the next char in s2 is < 256. We've already
4250 * worked on s1, and if it isn't also < 256, can't match */
4251 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4254 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4258 *foldbuf2 = TWO_BYTE_UTF8_TO_NATIVE(*p2, *(p2 + 1));
4261 /* Use another function to handle locale rules. We've made
4262 * sure that both characters to compare are single bytes */
4263 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4268 else if (isASCII(*p2)) {
4269 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4273 *foldbuf2 = toFOLD(*p2);
4276 to_utf8_fold(p2, foldbuf2, &n2);
4279 to_uni_fold(*p2, foldbuf2, &n2);
4285 /* Here f1 and f2 point to the beginning of the strings to compare.
4286 * These strings are the folds of the next character from each input
4287 * string, stored in utf8. */
4289 /* While there is more to look for in both folds, see if they
4290 * continue to match */
4292 U8 fold_length = UTF8SKIP(f1);
4293 if (fold_length != UTF8SKIP(f2)
4294 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4295 function call for single
4297 || memNE((char*)f1, (char*)f2, fold_length))
4299 return 0; /* mismatch */
4302 /* Here, they matched, advance past them */
4309 /* When reach the end of any fold, advance the input past it */
4311 p1 += u1 ? UTF8SKIP(p1) : 1;
4314 p2 += u2 ? UTF8SKIP(p2) : 1;
4316 } /* End of loop through both strings */
4318 /* A match is defined by each scan that specified an explicit length
4319 * reaching its final goal, and the other not having matched a partial
4320 * character (which can happen when the fold of a character is more than one
4322 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4326 /* Successful match. Set output pointers */
4336 /* XXX The next four functions should likely be moved to mathoms.c once all
4337 * occurrences of them are removed from the core; some cpan-upstream modules
4341 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4343 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4345 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4349 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4351 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4353 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4357 =for apidoc uvuni_to_utf8_flags
4359 Instead you almost certainly want to use L</uvchr_to_utf8> or
4360 L</uvchr_to_utf8_flags>>.
4362 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4363 which itself, while not deprecated, should be used only in isolated
4364 circumstances. These functions were useful for code that wanted to handle
4365 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4366 v5.20, the distinctions between the platforms have mostly been made invisible
4367 to most code, so this function is quite unlikely to be what you want.
4373 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4375 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4377 return uvoffuni_to_utf8_flags(d, uv, flags);
4381 =for apidoc utf8n_to_uvuni
4383 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4385 This function was useful for code that wanted to handle both EBCDIC and
4386 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4387 distinctions between the platforms have mostly been made invisible to most
4388 code, so this function is quite unlikely to be what you want. If you do need
4389 this precise functionality, use instead
4390 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4391 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4398 * c-indentation-style: bsd
4400 * indent-tabs-mode: nil
4403 * ex: set ts=8 sts=4 sw=4 et: