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 "invlist_inline.h"
36 static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
38 static const char cp_above_legal_max[] =
39 "It is deprecated to use code point 0x%"UVXf"; the permissible max is 0x%"UVXf"";
41 #define MAX_NON_DEPRECATED_CP (IV_MAX)
44 =head1 Unicode Support
45 These are various utility functions for manipulating UTF8-encoded
46 strings. For the uninitiated, this is a method of representing arbitrary
47 Unicode characters as a variable number of bytes, in such a way that
48 characters in the ASCII range are unmodified, and a zero byte never appears
49 within non-zero characters.
55 =for apidoc is_invariant_string
57 Returns true iff the first C<len> bytes of the string C<s> are the same
58 regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
59 EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
60 machines, all the ASCII characters and only the ASCII characters fit this
61 definition. On EBCDIC machines, the ASCII-range characters are invariant, but
62 so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
65 If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
66 use this option, that C<s> can't have embedded C<NUL> characters and has to
67 have a terminating C<NUL> byte).
69 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
75 Perl_is_invariant_string(const U8 *s, STRLEN len)
77 const U8* const send = s + (len ? len : strlen((const char *)s));
80 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
91 =for apidoc uvoffuni_to_utf8_flags
93 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
94 Instead, B<Almost all code should use L</uvchr_to_utf8> or
95 L</uvchr_to_utf8_flags>>.
97 This function is like them, but the input is a strict Unicode
98 (as opposed to native) code point. Only in very rare circumstances should code
99 not be using the native code point.
101 For details, see the description for L</uvchr_to_utf8_flags>.
107 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
109 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
111 if (OFFUNI_IS_INVARIANT(uv)) {
112 *d++ = LATIN1_TO_NATIVE(uv);
116 /* The first problematic code point is the first surrogate */
117 if (uv >= UNICODE_SURROGATE_FIRST) {
118 if (UNICODE_IS_SURROGATE(uv)) {
119 if (flags & UNICODE_WARN_SURROGATE) {
120 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
121 "UTF-16 surrogate U+%04"UVXf, uv);
123 if (flags & UNICODE_DISALLOW_SURROGATE) {
127 else if (UNICODE_IS_SUPER(uv)) {
128 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
129 && ckWARN_d(WARN_DEPRECATED))
131 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
132 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
134 if ( (flags & UNICODE_WARN_SUPER)
135 || (UNICODE_IS_ABOVE_31_BIT(uv) && (flags & UNICODE_WARN_ABOVE_31_BIT)))
137 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
139 /* Choose the more dire applicable warning */
140 (UNICODE_IS_ABOVE_31_BIT(uv))
141 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
142 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
145 if (flags & UNICODE_DISALLOW_SUPER
146 || (UNICODE_IS_ABOVE_31_BIT(uv) && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
151 else if (UNICODE_IS_NONCHAR(uv)) {
152 if (flags & UNICODE_WARN_NONCHAR) {
153 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
154 "Unicode non-character U+%04"UVXf" is not recommended for open interchange",
157 if (flags & UNICODE_DISALLOW_NONCHAR) {
165 STRLEN len = OFFUNISKIP(uv);
168 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
169 uv >>= UTF_ACCUMULATION_SHIFT;
171 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
174 #else /* Non loop style */
176 *d++ = (U8)(( uv >> 6) | 0xc0);
177 *d++ = (U8)(( uv & 0x3f) | 0x80);
181 *d++ = (U8)(( uv >> 12) | 0xe0);
182 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
183 *d++ = (U8)(( uv & 0x3f) | 0x80);
187 *d++ = (U8)(( uv >> 18) | 0xf0);
188 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
189 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
190 *d++ = (U8)(( uv & 0x3f) | 0x80);
193 if (uv < 0x4000000) {
194 *d++ = (U8)(( uv >> 24) | 0xf8);
195 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
196 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
197 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
198 *d++ = (U8)(( uv & 0x3f) | 0x80);
201 if (uv < 0x80000000) {
202 *d++ = (U8)(( uv >> 30) | 0xfc);
203 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
205 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
206 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
207 *d++ = (U8)(( uv & 0x3f) | 0x80);
211 if (uv < UTF8_QUAD_MAX)
214 *d++ = 0xfe; /* Can't match U+FEFF! */
215 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
220 *d++ = (U8)(( uv & 0x3f) | 0x80);
225 *d++ = 0xff; /* Can't match U+FFFE! */
226 *d++ = 0x80; /* 6 Reserved bits */
227 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
228 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
233 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
234 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
235 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
237 *d++ = (U8)(( uv & 0x3f) | 0x80);
241 #endif /* Non loop style */
244 =for apidoc uvchr_to_utf8
246 Adds the UTF-8 representation of the native code point C<uv> to the end
247 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
248 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
249 the byte after the end of the new character. In other words,
251 d = uvchr_to_utf8(d, uv);
253 is the recommended wide native character-aware way of saying
257 This function accepts any UV as input, but very high code points (above
258 C<IV_MAX> on the platform) will raise a deprecation warning. This is
259 typically 0x7FFF_FFFF in a 32-bit word.
261 It is possible to forbid or warn on non-Unicode code points, or those that may
262 be problematic by using L</uvchr_to_utf8_flags>.
267 /* This is also a macro */
268 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
271 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
273 return uvchr_to_utf8(d, uv);
277 =for apidoc uvchr_to_utf8_flags
279 Adds the UTF-8 representation of the native code point C<uv> to the end
280 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
281 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
282 the byte after the end of the new character. In other words,
284 d = uvchr_to_utf8_flags(d, uv, flags);
288 d = uvchr_to_utf8_flags(d, uv, 0);
290 This is the Unicode-aware way of saying
294 If C<flags> is 0, this function accepts any UV as input, but very high code
295 points (above C<IV_MAX> for the platform) will raise a deprecation warning.
296 This is typically 0x7FFF_FFFF in a 32-bit word.
298 Specifying C<flags> can further restrict what is allowed and not warned on, as
301 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
302 the function will raise a warning, provided UTF8 warnings are enabled. If
303 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
304 NULL. If both flags are set, the function will both warn and return NULL.
306 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
307 affect how the function handles a Unicode non-character.
309 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
310 affect the handling of code points that are above the Unicode maximum of
311 0x10FFFF. Languages other than Perl may not be able to accept files that
314 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
315 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
316 three DISALLOW flags.
318 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
319 so using them is more problematic than other above-Unicode code points. Perl
320 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
321 likely that non-Perl languages will not be able to read files that contain
322 these that written by the perl interpreter; nor would Perl understand files
323 written by something that uses a different extension. For these reasons, there
324 is a separate set of flags that can warn and/or disallow these extremely high
325 code points, even if other above-Unicode ones are accepted. These are the
326 C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
327 are entirely independent from the deprecation warning for code points above
328 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
329 code point that needs more than 31 bits to represent. When that happens,
330 effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
331 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
332 above-Unicode code points, including these, as malformations; and
333 C<UNICODE_WARN_SUPER> warns on these.)
335 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
336 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
337 than on ASCII. Prior to that, code points 2**31 and higher were simply
338 unrepresentable, and a different, incompatible method was used to represent
339 code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
340 and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
341 platforms, warning and disallowing 2**31 and higher.
346 /* This is also a macro */
347 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
350 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
352 return uvchr_to_utf8_flags(d, uv, flags);
356 =for apidoc is_utf8_string
358 Returns true if the first C<len> bytes of string C<s> form a valid
359 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
360 using C<strlen(s)> (which means if you use this option, that C<s> can't have
361 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
362 that all characters being ASCII constitute 'a valid UTF-8 string'.
364 See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
370 Perl_is_utf8_string(const U8 *s, STRLEN len)
372 const U8* const send = s + (len ? len : strlen((const char *)s));
375 PERL_ARGS_ASSERT_IS_UTF8_STRING;
378 STRLEN len = isUTF8_CHAR(x, send);
379 if (UNLIKELY(! len)) {
389 Implemented as a macro in utf8.h
391 =for apidoc is_utf8_string_loc
393 Like L</is_utf8_string> but stores the location of the failure (in the
394 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
395 "utf8ness success") in the C<ep>.
397 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
399 =for apidoc is_utf8_string_loclen
401 Like L</is_utf8_string>() but stores the location of the failure (in the
402 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
403 "utf8ness success") in the C<ep>, and the number of UTF-8
404 encoded characters in the C<el>.
406 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
412 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
414 const U8* const send = s + (len ? len : strlen((const char *)s));
418 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
421 STRLEN len = isUTF8_CHAR(x, send);
422 if (UNLIKELY(! len)) {
440 =for apidoc utf8n_to_uvchr
442 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
443 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
445 Bottom level UTF-8 decode routine.
446 Returns the native code point value of the first character in the string C<s>,
447 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
448 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
449 the length, in bytes, of that character.
451 The value of C<flags> determines the behavior when C<s> does not point to a
452 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
453 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
454 next possible position in C<s> that could begin a non-malformed character.
455 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
457 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
458 individual types of malformations, such as the sequence being overlong (that
459 is, when there is a shorter sequence that can express the same code point;
460 overlong sequences are expressly forbidden in the UTF-8 standard due to
461 potential security issues). Another malformation example is the first byte of
462 a character not being a legal first byte. See F<utf8.h> for the list of such
463 flags. For allowed 0 length strings, this function returns 0; for allowed
464 overlong sequences, the computed code point is returned; for all other allowed
465 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
466 determinable reasonable value.
468 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
469 flags) malformation is found. If this flag is set, the routine assumes that
470 the caller will raise a warning, and this function will silently just set
471 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
473 Note that this API requires disambiguation between successful decoding a C<NUL>
474 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
475 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
476 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
479 Certain code points are considered problematic. These are Unicode surrogates,
480 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
481 By default these are considered regular code points, but certain situations
482 warrant special handling for them. If C<flags> contains
483 C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
484 malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
485 C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
486 Unicode maximum) can be set to disallow these categories individually.
488 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
489 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
490 raised for their respective categories, but otherwise the code points are
491 considered valid (not malformations). To get a category to both be treated as
492 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
493 (But note that warnings are not raised if lexically disabled nor if
494 C<UTF8_CHECK_ONLY> is also specified.)
496 It is now deprecated to have very high code points (above C<IV_MAX> on the
497 platforms) and this function will raise a deprecation warning for these (unless
498 such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
501 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
502 so using them is more problematic than other above-Unicode code points. Perl
503 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
504 likely that non-Perl languages will not be able to read files that contain
505 these that written by the perl interpreter; nor would Perl understand files
506 written by something that uses a different extension. For these reasons, there
507 is a separate set of flags that can warn and/or disallow these extremely high
508 code points, even if other above-Unicode ones are accepted. These are the
509 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
510 are entirely independent from the deprecation warning for code points above
511 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
512 code point that needs more than 31 bits to represent. When that happens,
513 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
514 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
515 above-Unicode code points, including these, as malformations; and
516 C<UTF8_WARN_SUPER> warns on these.)
518 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
519 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
520 than on ASCII. Prior to that, code points 2**31 and higher were simply
521 unrepresentable, and a different, incompatible method was used to represent
522 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
523 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
524 platforms, warning and disallowing 2**31 and higher.
526 All other code points corresponding to Unicode characters, including private
527 use and those yet to be assigned, are never considered malformed and never
534 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
536 const U8 * const s0 = s;
537 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
542 UV outlier_ret = 0; /* return value when input is in error or problematic
544 UV pack_warn = 0; /* Save result of packWARN() for later */
545 bool unexpected_non_continuation = FALSE;
546 bool overflowed = FALSE;
547 bool do_overlong_test = TRUE; /* May have to skip this test */
549 const char* const malformed_text = "Malformed UTF-8 character";
551 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
553 /* The order of malformation tests here is important. We should consume as
554 * few bytes as possible in order to not skip any valid character. This is
555 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
556 * http://unicode.org/reports/tr36 for more discussion as to why. For
557 * example, once we've done a UTF8SKIP, we can tell the expected number of
558 * bytes, and could fail right off the bat if the input parameters indicate
559 * that there are too few available. But it could be that just that first
560 * byte is garbled, and the intended character occupies fewer bytes. If we
561 * blindly assumed that the first byte is correct, and skipped based on
562 * that number, we could skip over a valid input character. So instead, we
563 * always examine the sequence byte-by-byte.
565 * We also should not consume too few bytes, otherwise someone could inject
566 * things. For example, an input could be deliberately designed to
567 * overflow, and if this code bailed out immediately upon discovering that,
568 * returning to the caller C<*retlen> pointing to the very next byte (one
569 * which is actually part of of the overflowing sequence), that could look
570 * legitimate to the caller, which could discard the initial partial
571 * sequence and process the rest, inappropriately */
573 /* Zero length strings, if allowed, of necessity are zero */
574 if (UNLIKELY(curlen == 0)) {
579 if (flags & UTF8_ALLOW_EMPTY) {
582 if (! (flags & UTF8_CHECK_ONLY)) {
583 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
588 expectlen = UTF8SKIP(s);
590 /* A well-formed UTF-8 character, as the vast majority of calls to this
591 * function will be for, has this expected length. For efficiency, set
592 * things up here to return it. It will be overriden only in those rare
593 * cases where a malformation is found */
598 /* An invariant is trivially well-formed */
599 if (UTF8_IS_INVARIANT(uv)) {
603 /* A continuation character can't start a valid sequence */
604 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
605 if (flags & UTF8_ALLOW_CONTINUATION) {
609 return UNICODE_REPLACEMENT;
612 if (! (flags & UTF8_CHECK_ONLY)) {
613 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
619 /* Here is not a continuation byte, nor an invariant. The only thing left
620 * is a start byte (possibly for an overlong) */
623 uv = NATIVE_UTF8_TO_I8(uv);
626 /* Remove the leading bits that indicate the number of bytes in the
627 * character's whole UTF-8 sequence, leaving just the bits that are part of
629 uv &= UTF_START_MASK(expectlen);
631 /* Now, loop through the remaining bytes in the character's sequence,
632 * accumulating each into the working value as we go. Be sure to not look
633 * past the end of the input string */
634 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
636 for (s = s0 + 1; s < send; s++) {
637 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
638 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
640 /* The original implementors viewed this malformation as more
641 * serious than the others (though I, khw, don't understand
642 * why, since other malformations also give very very wrong
643 * results), so there is no way to turn off checking for it.
644 * Set a flag, but keep going in the loop, so that we absorb
645 * the rest of the bytes that comprise the character. */
647 overflow_byte = *s; /* Save for warning message's use */
649 uv = UTF8_ACCUMULATE(uv, *s);
652 /* Here, found a non-continuation before processing all expected
653 * bytes. This byte begins a new character, so quit, even if
654 * allowing this malformation. */
655 unexpected_non_continuation = TRUE;
658 } /* End of loop through the character's bytes */
660 /* Save how many bytes were actually in the character */
663 /* The loop above finds two types of malformations: non-continuation and/or
664 * overflow. The non-continuation malformation is really a too-short
665 * malformation, as it means that the current character ended before it was
666 * expected to (being terminated prematurely by the beginning of the next
667 * character, whereas in the too-short malformation there just are too few
668 * bytes available to hold the character. In both cases, the check below
669 * that we have found the expected number of bytes would fail if executed.)
670 * Thus the non-continuation malformation is really unnecessary, being a
671 * subset of the too-short malformation. But there may be existing
672 * applications that are expecting the non-continuation type, so we retain
673 * it, and return it in preference to the too-short malformation. (If this
674 * code were being written from scratch, the two types might be collapsed
675 * into one.) I, khw, am also giving priority to returning the
676 * non-continuation and too-short malformations over overflow when multiple
677 * ones are present. I don't know of any real reason to prefer one over
678 * the other, except that it seems to me that multiple-byte errors trumps
679 * errors from a single byte */
680 if (UNLIKELY(unexpected_non_continuation)) {
681 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
682 if (! (flags & UTF8_CHECK_ONLY)) {
684 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
687 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));
692 uv = UNICODE_REPLACEMENT;
694 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
695 * as what the original expectations were. */
696 do_overlong_test = FALSE;
701 else if (UNLIKELY(curlen < expectlen)) {
702 if (! (flags & UTF8_ALLOW_SHORT)) {
703 if (! (flags & UTF8_CHECK_ONLY)) {
704 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));
708 uv = UNICODE_REPLACEMENT;
709 do_overlong_test = FALSE;
715 if (UNLIKELY(overflowed)) {
716 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
721 && expectlen > (STRLEN) OFFUNISKIP(uv)
722 && ! (flags & UTF8_ALLOW_LONG))
724 /* The overlong malformation has lower precedence than the others.
725 * Note that if this malformation is allowed, we return the actual
726 * value, instead of the replacement character. This is because this
727 * value is actually well-defined. */
728 if (! (flags & UTF8_CHECK_ONLY)) {
729 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", OFFUNISKIP(uv), *s0));
734 /* Here, the input is considered to be well-formed, but it still could be a
735 * problematic code point that is not allowed by the input parameters. */
736 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
737 && ((flags & ( UTF8_DISALLOW_NONCHAR
738 |UTF8_DISALLOW_SURROGATE
740 |UTF8_DISALLOW_ABOVE_31_BIT
744 |UTF8_WARN_ABOVE_31_BIT))
745 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
746 && ckWARN_d(WARN_DEPRECATED))))
748 if (UNICODE_IS_SURROGATE(uv)) {
750 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
751 * generation of the sv, since no warnings are raised under CHECK */
752 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
753 && ckWARN_d(WARN_SURROGATE))
755 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
756 pack_warn = packWARN(WARN_SURROGATE);
758 if (flags & UTF8_DISALLOW_SURROGATE) {
762 else if ((uv > PERL_UNICODE_MAX)) {
763 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
764 && ckWARN_d(WARN_NON_UNICODE))
766 sv = sv_2mortal(Perl_newSVpvf(aTHX_
767 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
769 pack_warn = packWARN(WARN_NON_UNICODE);
772 /* The maximum code point ever specified by a standard was
773 * 2**31 - 1. Anything larger than that is a Perl extension that
774 * very well may not be understood by other applications (including
775 * earlier perl versions on EBCDIC platforms). On ASCII platforms,
776 * these code points are indicated by the first UTF-8 byte being
777 * 0xFE or 0xFF. We test for these after the regular SUPER ones,
778 * and before possibly bailing out, so that the slightly more dire
779 * warning will override the regular one. */
782 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */
784 /* The I8 for 2**31 (U+80000000) is
785 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
786 * and it turns out that on all EBCDIC pages recognized that
787 * the UTF-EBCDIC for that code point is
788 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
789 * For the next lower code point, the 1047 UTF-EBCDIC is
790 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
791 * The other code pages differ only in the bytes following
792 * \x42. Thus the following works (the minimum continuation
794 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41
802 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER
803 |UTF8_DISALLOW_ABOVE_31_BIT)))
805 if ( ! (flags & UTF8_CHECK_ONLY)
806 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
807 && ckWARN_d(WARN_UTF8))
809 sv = sv_2mortal(Perl_newSVpvf(aTHX_
810 "Code point 0x%"UVXf" is not Unicode, and not portable",
812 pack_warn = packWARN(WARN_UTF8);
814 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
819 if (flags & UTF8_DISALLOW_SUPER) {
823 /* The deprecated warning overrides any non-deprecated one */
824 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
826 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
827 uv, MAX_NON_DEPRECATED_CP));
828 pack_warn = packWARN(WARN_DEPRECATED);
831 else if (UNICODE_IS_NONCHAR(uv)) {
832 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
833 && ckWARN_d(WARN_NONCHAR))
835 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
836 pack_warn = packWARN(WARN_NONCHAR);
838 if (flags & UTF8_DISALLOW_NONCHAR) {
844 outlier_ret = uv; /* Note we don't bother to convert to native,
845 as all the outlier code points are the same
846 in both ASCII and EBCDIC */
850 /* Here, this is not considered a malformed character, so drop through
854 return UNI_TO_NATIVE(uv);
856 /* There are three cases which get to beyond this point. In all 3 cases:
857 * <sv> if not null points to a string to print as a warning.
858 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
860 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
861 * This is done by initializing it to 0, and changing it only
864 * 1) The input is valid but problematic, and to be warned about. The
865 * return value is the resultant code point; <*retlen> is set to
866 * <curlen>, the number of bytes that comprise the code point.
867 * <pack_warn> contains the result of packWARN() for the warning
868 * types. The entry point for this case is the label <do_warn>;
869 * 2) The input is a valid code point but disallowed by the parameters to
870 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
871 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
872 * comprise the code point. <pack_warn> contains the result of
873 * packWARN() for the warning types. The entry point for this case is
874 * the label <disallowed>.
875 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
876 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
877 * bytes that comprise the malformation. All such malformations are
878 * assumed to be warning type <utf8>. The entry point for this case
879 * is the label <malformed>.
884 if (sv && ckWARN_d(WARN_UTF8)) {
885 pack_warn = packWARN(WARN_UTF8);
890 if (flags & UTF8_CHECK_ONLY) {
892 *retlen = ((STRLEN) -1);
898 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
899 if warnings are to be raised. */
900 const char * const string = SvPVX_const(sv);
903 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
905 Perl_warner(aTHX_ pack_warn, "%s", string);
916 =for apidoc utf8_to_uvchr_buf
918 Returns the native code point of the first character in the string C<s> which
919 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
920 C<*retlen> will be set to the length, in bytes, of that character.
922 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
923 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
924 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
925 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
926 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
927 the next possible position in C<s> that could begin a non-malformed character.
928 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
931 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
932 unless those are turned off.
939 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
943 return utf8n_to_uvchr(s, send - s, retlen,
944 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
947 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
948 * there are no malformations in the input UTF-8 string C<s>. surrogates,
949 * non-character code points, and non-Unicode code points are allowed. */
952 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
954 UV expectlen = UTF8SKIP(s);
955 const U8* send = s + expectlen;
958 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
965 /* An invariant is trivially returned */
966 if (expectlen == 1) {
971 uv = NATIVE_UTF8_TO_I8(uv);
974 /* Remove the leading bits that indicate the number of bytes, leaving just
975 * the bits that are part of the value */
976 uv &= UTF_START_MASK(expectlen);
978 /* Now, loop through the remaining bytes, accumulating each into the
979 * working total as we go. (I khw tried unrolling the loop for up to 4
980 * bytes, but there was no performance improvement) */
981 for (++s; s < send; s++) {
982 uv = UTF8_ACCUMULATE(uv, *s);
985 return UNI_TO_NATIVE(uv);
990 =for apidoc utf8_to_uvuni_buf
992 Only in very rare circumstances should code need to be dealing in Unicode
993 (as opposed to native) code points. In those few cases, use
994 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
996 Returns the Unicode (not-native) code point of the first character in the
998 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
999 C<retlen> will be set to the length, in bytes, of that character.
1001 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1002 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1003 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1004 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1005 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1006 next possible position in C<s> that could begin a non-malformed character.
1007 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1009 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1010 unless those are turned off.
1016 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1018 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1022 /* Call the low level routine asking for checks */
1023 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1024 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
1028 =for apidoc utf8_length
1030 Return the length of the UTF-8 char encoded string C<s> in characters.
1031 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1032 up past C<e>, croaks.
1038 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1042 PERL_ARGS_ASSERT_UTF8_LENGTH;
1044 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1045 * the bitops (especially ~) can create illegal UTF-8.
1046 * In other words: in Perl UTF-8 is not just for Unicode. */
1049 goto warn_and_return;
1059 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1060 "%s in %s", unees, OP_DESC(PL_op));
1062 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1069 =for apidoc utf8_distance
1071 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1074 WARNING: use only if you *know* that the pointers point inside the
1081 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1083 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1085 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1089 =for apidoc utf8_hop
1091 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1092 forward or backward.
1094 WARNING: do not use the following unless you *know* C<off> is within
1095 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1096 on the first byte of character or just after the last byte of a character.
1102 Perl_utf8_hop(const U8 *s, I32 off)
1104 PERL_ARGS_ASSERT_UTF8_HOP;
1106 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1107 * the bitops (especially ~) can create illegal UTF-8.
1108 * In other words: in Perl UTF-8 is not just for Unicode. */
1117 while (UTF8_IS_CONTINUATION(*s))
1125 =for apidoc bytes_cmp_utf8
1127 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1128 sequence of characters (stored as UTF-8)
1129 in C<u>, C<ulen>. Returns 0 if they are
1130 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1131 if the first string is greater than the second string.
1133 -1 or +1 is returned if the shorter string was identical to the start of the
1134 longer string. -2 or +2 is returned if
1135 there was a difference between characters
1142 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1144 const U8 *const bend = b + blen;
1145 const U8 *const uend = u + ulen;
1147 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1149 while (b < bend && u < uend) {
1151 if (!UTF8_IS_INVARIANT(c)) {
1152 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1155 if (UTF8_IS_CONTINUATION(c1)) {
1156 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1158 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1159 "Malformed UTF-8 character "
1160 "(unexpected non-continuation byte 0x%02x"
1161 ", immediately after start byte 0x%02x)"
1162 /* Dear diag.t, it's in the pod. */
1164 PL_op ? " in " : "",
1165 PL_op ? OP_DESC(PL_op) : "");
1170 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1171 "%s in %s", unees, OP_DESC(PL_op));
1173 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1174 return -2; /* Really want to return undef :-) */
1181 return *b < c ? -2 : +2;
1186 if (b == bend && u == uend)
1189 return b < bend ? +1 : -1;
1193 =for apidoc utf8_to_bytes
1195 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1196 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1197 updates C<len> to contain the new length.
1198 Returns zero on failure, setting C<len> to -1.
1200 If you need a copy of the string, see L</bytes_from_utf8>.
1206 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1208 U8 * const save = s;
1209 U8 * const send = s + *len;
1212 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1213 PERL_UNUSED_CONTEXT;
1215 /* ensure valid UTF-8 and chars < 256 before updating string */
1217 if (! UTF8_IS_INVARIANT(*s)) {
1218 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1219 *len = ((STRLEN) -1);
1230 if (! UTF8_IS_INVARIANT(c)) {
1231 /* Then it is two-byte encoded */
1232 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1243 =for apidoc bytes_from_utf8
1245 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1246 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1247 the newly-created string, and updates C<len> to contain the new
1248 length. Returns the original string if no conversion occurs, C<len>
1249 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1250 0 if C<s> is converted or consisted entirely of characters that are invariant
1251 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
1257 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1260 const U8 *start = s;
1264 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1265 PERL_UNUSED_CONTEXT;
1269 /* ensure valid UTF-8 and chars < 256 before converting string */
1270 for (send = s + *len; s < send;) {
1271 if (! UTF8_IS_INVARIANT(*s)) {
1272 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1283 Newx(d, (*len) - count + 1, U8);
1284 s = start; start = d;
1287 if (! UTF8_IS_INVARIANT(c)) {
1288 /* Then it is two-byte encoded */
1289 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1300 =for apidoc bytes_to_utf8
1302 Converts a string C<s> of length C<len> bytes from the native encoding into
1304 Returns a pointer to the newly-created string, and sets C<len> to
1305 reflect the new length in bytes.
1307 A C<NUL> character will be written after the end of the string.
1309 If you want to convert to UTF-8 from encodings other than
1310 the native (Latin1 or EBCDIC),
1311 see L</sv_recode_to_utf8>().
1316 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1317 likewise need duplication. */
1320 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1322 const U8 * const send = s + (*len);
1326 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1327 PERL_UNUSED_CONTEXT;
1329 Newx(d, (*len) * 2 + 1, U8);
1333 append_utf8_from_native_byte(*s, &d);
1342 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1344 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1345 * We optimize for native, for obvious reasons. */
1348 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1353 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1356 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1361 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1363 if (OFFUNI_IS_INVARIANT(uv)) {
1364 *d++ = LATIN1_TO_NATIVE((U8) uv);
1367 if (uv <= MAX_UTF8_TWO_BYTE) {
1368 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1369 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1372 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1373 #define LAST_HIGH_SURROGATE 0xDBFF
1374 #define FIRST_LOW_SURROGATE 0xDC00
1375 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1377 /* This assumes that most uses will be in the first Unicode plane, not
1378 * needing surrogates */
1379 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1380 && uv <= UNICODE_SURROGATE_LAST))
1382 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1383 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1386 UV low = (p[0] << 8) + p[1];
1387 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1388 || UNLIKELY(low > LAST_LOW_SURROGATE))
1390 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1393 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1394 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1398 d = uvoffuni_to_utf8_flags(d, uv, 0);
1401 *d++ = (U8)(( uv >> 12) | 0xe0);
1402 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1403 *d++ = (U8)(( uv & 0x3f) | 0x80);
1407 *d++ = (U8)(( uv >> 18) | 0xf0);
1408 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1409 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1410 *d++ = (U8)(( uv & 0x3f) | 0x80);
1415 *newlen = d - dstart;
1419 /* Note: this one is slightly destructive of the source. */
1422 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1425 U8* const send = s + bytelen;
1427 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1430 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1434 const U8 tmp = s[0];
1439 return utf16_to_utf8(p, d, bytelen, newlen);
1443 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1445 U8 tmpbuf[UTF8_MAXBYTES+1];
1446 uvchr_to_utf8(tmpbuf, c);
1447 return _is_utf8_FOO(classnum, tmpbuf);
1450 /* Internal function so we can deprecate the external one, and call
1451 this one from other deprecated functions in this file */
1454 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1456 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1460 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1464 Perl__is_uni_perl_idcont(pTHX_ UV c)
1466 U8 tmpbuf[UTF8_MAXBYTES+1];
1467 uvchr_to_utf8(tmpbuf, c);
1468 return _is_utf8_perl_idcont(tmpbuf);
1472 Perl__is_uni_perl_idstart(pTHX_ UV c)
1474 U8 tmpbuf[UTF8_MAXBYTES+1];
1475 uvchr_to_utf8(tmpbuf, c);
1476 return _is_utf8_perl_idstart(tmpbuf);
1480 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1482 /* We have the latin1-range values compiled into the core, so just use
1483 * those, converting the result to UTF-8. The only difference between upper
1484 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1485 * either "SS" or "Ss". Which one to use is passed into the routine in
1486 * 'S_or_s' to avoid a test */
1488 UV converted = toUPPER_LATIN1_MOD(c);
1490 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1492 assert(S_or_s == 'S' || S_or_s == 's');
1494 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1495 characters in this range */
1496 *p = (U8) converted;
1501 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1502 * which it maps to one of them, so as to only have to have one check for
1503 * it in the main case */
1504 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1506 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1507 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1510 converted = GREEK_CAPITAL_LETTER_MU;
1512 #if UNICODE_MAJOR_VERSION > 2 \
1513 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1514 && UNICODE_DOT_DOT_VERSION >= 8)
1515 case LATIN_SMALL_LETTER_SHARP_S:
1522 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1523 NOT_REACHED; /* NOTREACHED */
1527 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1528 *p = UTF8_TWO_BYTE_LO(converted);
1534 /* Call the function to convert a UTF-8 encoded character to the specified case.
1535 * Note that there may be more than one character in the result.
1536 * INP is a pointer to the first byte of the input character
1537 * OUTP will be set to the first byte of the string of changed characters. It
1538 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1539 * LENP will be set to the length in bytes of the string of changed characters
1541 * The functions return the ordinal of the first character in the string of OUTP */
1542 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1543 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1544 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1546 /* This additionally has the input parameter SPECIALS, which if non-zero will
1547 * cause this to use the SPECIALS hash for folding (meaning get full case
1548 * folding); otherwise, when zero, this implies a simple case fold */
1549 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1552 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1554 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1555 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1556 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1557 * the changed version may be longer than the original character.
1559 * The ordinal of the first character of the changed version is returned
1560 * (but note, as explained above, that there may be more.) */
1562 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1565 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1568 uvchr_to_utf8(p, c);
1569 return CALL_UPPER_CASE(p, p, lenp);
1573 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1575 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1578 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1581 uvchr_to_utf8(p, c);
1582 return CALL_TITLE_CASE(p, p, lenp);
1586 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
1588 /* We have the latin1-range values compiled into the core, so just use
1589 * those, converting the result to UTF-8. Since the result is always just
1590 * one character, we allow <p> to be NULL */
1592 U8 converted = toLOWER_LATIN1(c);
1595 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1600 /* Result is known to always be < 256, so can use the EIGHT_BIT
1602 *p = UTF8_EIGHT_BIT_HI(converted);
1603 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
1611 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1613 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1616 return to_lower_latin1((U8) c, p, lenp);
1619 uvchr_to_utf8(p, c);
1620 return CALL_LOWER_CASE(p, p, lenp);
1624 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1626 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1627 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1628 * FOLD_FLAGS_FULL iff full folding is to be used;
1630 * Not to be used for locale folds
1635 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1636 PERL_UNUSED_CONTEXT;
1638 assert (! (flags & FOLD_FLAGS_LOCALE));
1640 if (c == MICRO_SIGN) {
1641 converted = GREEK_SMALL_LETTER_MU;
1643 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1644 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1645 || UNICODE_DOT_DOT_VERSION > 0)
1646 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1648 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1649 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1650 * under those circumstances. */
1651 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1652 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1653 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1655 return LATIN_SMALL_LETTER_LONG_S;
1665 else { /* In this range the fold of all other characters is their lower
1667 converted = toLOWER_LATIN1(c);
1670 if (UVCHR_IS_INVARIANT(converted)) {
1671 *p = (U8) converted;
1675 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1676 *p = UTF8_TWO_BYTE_LO(converted);
1684 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1687 /* Not currently externally documented, and subject to change
1688 * <flags> bits meanings:
1689 * FOLD_FLAGS_FULL iff full folding is to be used;
1690 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1691 * locale are to be used.
1692 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1695 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1697 if (flags & FOLD_FLAGS_LOCALE) {
1698 /* Treat a UTF-8 locale as not being in locale at all */
1699 if (IN_UTF8_CTYPE_LOCALE) {
1700 flags &= ~FOLD_FLAGS_LOCALE;
1703 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1704 goto needs_full_generality;
1709 return _to_fold_latin1((U8) c, p, lenp,
1710 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1713 /* Here, above 255. If no special needs, just use the macro */
1714 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1715 uvchr_to_utf8(p, c);
1716 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1718 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1719 the special flags. */
1720 U8 utf8_c[UTF8_MAXBYTES + 1];
1722 needs_full_generality:
1723 uvchr_to_utf8(utf8_c, c);
1724 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1728 PERL_STATIC_INLINE bool
1729 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1730 const char *const swashname, SV* const invlist)
1732 /* returns a boolean giving whether or not the UTF8-encoded character that
1733 * starts at <p> is in the swash indicated by <swashname>. <swash>
1734 * contains a pointer to where the swash indicated by <swashname>
1735 * is to be stored; which this routine will do, so that future calls will
1736 * look at <*swash> and only generate a swash if it is not null. <invlist>
1737 * is NULL or an inversion list that defines the swash. If not null, it
1738 * saves time during initialization of the swash.
1740 * Note that it is assumed that the buffer length of <p> is enough to
1741 * contain all the bytes that comprise the character. Thus, <*p> should
1742 * have been checked before this call for mal-formedness enough to assure
1745 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1747 /* The API should have included a length for the UTF-8 character in <p>,
1748 * but it doesn't. We therefore assume that p has been validated at least
1749 * as far as there being enough bytes available in it to accommodate the
1750 * character without reading beyond the end, and pass that number on to the
1751 * validating routine */
1752 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1753 if (ckWARN_d(WARN_UTF8)) {
1754 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1755 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1756 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1757 what the malformation is */
1758 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1764 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1765 *swash = _core_swash_init("utf8",
1767 /* Only use the name if there is no inversion
1768 * list; otherwise will go out to disk */
1769 (invlist) ? "" : swashname,
1771 &PL_sv_undef, 1, 0, invlist, &flags);
1774 return swash_fetch(*swash, p, TRUE) != 0;
1778 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1780 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1782 assert(classnum < _FIRST_NON_SWASH_CC);
1784 return is_utf8_common(p,
1785 &PL_utf8_swash_ptrs[classnum],
1786 swash_property_names[classnum],
1787 PL_XPosix_ptrs[classnum]);
1791 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1795 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1797 if (! PL_utf8_perl_idstart) {
1798 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1800 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
1804 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1806 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1810 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1814 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1818 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1820 if (! PL_utf8_perl_idcont) {
1821 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1823 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
1827 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1829 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1831 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1835 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1837 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1839 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1843 Perl__is_utf8_mark(pTHX_ const U8 *p)
1845 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1847 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1851 =for apidoc to_utf8_case
1853 C<p> contains the pointer to the UTF-8 string encoding
1854 the character that is being converted. This routine assumes that the character
1855 at C<p> is well-formed.
1857 C<ustrp> is a pointer to the character buffer to put the
1858 conversion result to. C<lenp> is a pointer to the length
1861 C<swashp> is a pointer to the swash to use.
1863 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1864 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
1865 but not always, a multicharacter mapping), is tried first.
1867 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1868 any special mappings; C<""> means to use the special mappings. Values other
1869 than these two are treated as the name of the hash containing the special
1870 mappings, like C<"utf8::ToSpecLower">.
1872 C<normal> is a string like C<"ToLower"> which means the swash
1875 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1876 unless those are turned off.
1881 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1882 SV **swashp, const char *normal, const char *special)
1885 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1887 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1889 /* Note that swash_fetch() doesn't output warnings for these because it
1890 * assumes we will */
1891 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1892 if (uv1 <= UNICODE_SURROGATE_LAST) {
1893 if (ckWARN_d(WARN_SURROGATE)) {
1894 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1895 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1896 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1899 else if (UNICODE_IS_SUPER(uv1)) {
1900 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1901 && ckWARN_d(WARN_DEPRECATED))
1903 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1904 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1906 if (ckWARN_d(WARN_NON_UNICODE)) {
1907 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1908 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1909 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1913 /* Note that non-characters are perfectly legal, so no warning should
1917 if (!*swashp) /* load on-demand */
1918 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1921 /* It might be "special" (sometimes, but not always,
1922 * a multicharacter mapping) */
1926 /* If passed in the specials name, use that; otherwise use any
1927 * given in the swash */
1928 if (*special != '\0') {
1929 hv = get_hv(special, 0);
1932 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1934 hv = MUTABLE_HV(SvRV(*svp));
1939 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
1944 s = SvPV_const(*svp, len);
1947 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1949 Copy(s, ustrp, len, U8);
1954 if (!len && *swashp) {
1955 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
1958 /* It was "normal" (a single character mapping). */
1959 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1967 return valid_utf8_to_uvchr(ustrp, 0);
1970 /* Here, there was no mapping defined, which means that the code point maps
1971 * to itself. Return the inputs */
1973 if (p != ustrp) { /* Don't copy onto itself */
1974 Copy(p, ustrp, len, U8);
1985 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1987 /* This is called when changing the case of a UTF-8-encoded character above
1988 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1989 * result contains a character that crosses the 255/256 boundary, disallow
1990 * the change, and return the original code point. See L<perlfunc/lc> for
1993 * p points to the original string whose case was changed; assumed
1994 * by this routine to be well-formed
1995 * result the code point of the first character in the changed-case string
1996 * ustrp points to the changed-case string (<result> represents its first char)
1997 * lenp points to the length of <ustrp> */
1999 UV original; /* To store the first code point of <p> */
2001 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2003 assert(UTF8_IS_ABOVE_LATIN1(*p));
2005 /* We know immediately if the first character in the string crosses the
2006 * boundary, so can skip */
2009 /* Look at every character in the result; if any cross the
2010 * boundary, the whole thing is disallowed */
2011 U8* s = ustrp + UTF8SKIP(ustrp);
2012 U8* e = ustrp + *lenp;
2014 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2020 /* Here, no characters crossed, result is ok as-is, but we warn. */
2021 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
2027 /* Failed, have to return the original */
2028 original = valid_utf8_to_uvchr(p, lenp);
2030 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2031 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2032 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2033 "resolved to \"\\x{%"UVXf"}\".",
2037 Copy(p, ustrp, *lenp, char);
2042 =for apidoc to_utf8_upper
2044 Instead use L</toUPPER_utf8>.
2048 /* Not currently externally documented, and subject to change:
2049 * <flags> is set iff iff the rules from the current underlying locale are to
2053 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2057 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2060 /* Treat a UTF-8 locale as not being in locale at all */
2061 if (IN_UTF8_CTYPE_LOCALE) {
2065 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2069 if (UTF8_IS_INVARIANT(*p)) {
2071 result = toUPPER_LC(*p);
2074 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2077 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2079 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2080 result = toUPPER_LC(c);
2083 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2087 else { /* UTF-8, ord above 255 */
2088 result = CALL_UPPER_CASE(p, ustrp, lenp);
2091 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2096 /* Here, used locale rules. Convert back to UTF-8 */
2097 if (UTF8_IS_INVARIANT(result)) {
2098 *ustrp = (U8) result;
2102 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2103 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2111 =for apidoc to_utf8_title
2113 Instead use L</toTITLE_utf8>.
2117 /* Not currently externally documented, and subject to change:
2118 * <flags> is set iff the rules from the current underlying locale are to be
2119 * used. Since titlecase is not defined in POSIX, for other than a
2120 * UTF-8 locale, uppercase is used instead for code points < 256.
2124 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2128 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2131 /* Treat a UTF-8 locale as not being in locale at all */
2132 if (IN_UTF8_CTYPE_LOCALE) {
2136 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2140 if (UTF8_IS_INVARIANT(*p)) {
2142 result = toUPPER_LC(*p);
2145 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2148 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2150 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2151 result = toUPPER_LC(c);
2154 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2158 else { /* UTF-8, ord above 255 */
2159 result = CALL_TITLE_CASE(p, ustrp, lenp);
2162 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2167 /* Here, used locale rules. Convert back to UTF-8 */
2168 if (UTF8_IS_INVARIANT(result)) {
2169 *ustrp = (U8) result;
2173 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2174 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2182 =for apidoc to_utf8_lower
2184 Instead use L</toLOWER_utf8>.
2188 /* Not currently externally documented, and subject to change:
2189 * <flags> is set iff iff the rules from the current underlying locale are to
2194 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2198 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2201 /* Treat a UTF-8 locale as not being in locale at all */
2202 if (IN_UTF8_CTYPE_LOCALE) {
2206 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2210 if (UTF8_IS_INVARIANT(*p)) {
2212 result = toLOWER_LC(*p);
2215 return to_lower_latin1(*p, ustrp, lenp);
2218 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2220 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2221 result = toLOWER_LC(c);
2224 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2228 else { /* UTF-8, ord above 255 */
2229 result = CALL_LOWER_CASE(p, ustrp, lenp);
2232 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2238 /* Here, used locale rules. Convert back to UTF-8 */
2239 if (UTF8_IS_INVARIANT(result)) {
2240 *ustrp = (U8) result;
2244 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2245 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2253 =for apidoc to_utf8_fold
2255 Instead use L</toFOLD_utf8>.
2259 /* Not currently externally documented, and subject to change,
2261 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2262 * locale are to be used.
2263 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2264 * otherwise simple folds
2265 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2270 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2274 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2276 /* These are mutually exclusive */
2277 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2279 assert(p != ustrp); /* Otherwise overwrites */
2281 if (flags & FOLD_FLAGS_LOCALE) {
2282 /* Treat a UTF-8 locale as not being in locale at all */
2283 if (IN_UTF8_CTYPE_LOCALE) {
2284 flags &= ~FOLD_FLAGS_LOCALE;
2287 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2291 if (UTF8_IS_INVARIANT(*p)) {
2292 if (flags & FOLD_FLAGS_LOCALE) {
2293 result = toFOLD_LC(*p);
2296 return _to_fold_latin1(*p, ustrp, lenp,
2297 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2300 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2301 if (flags & FOLD_FLAGS_LOCALE) {
2302 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2303 result = toFOLD_LC(c);
2306 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2308 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2311 else { /* UTF-8, ord above 255 */
2312 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2314 if (flags & FOLD_FLAGS_LOCALE) {
2316 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2317 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2319 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2320 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2322 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2324 /* Special case these two characters, as what normally gets
2325 * returned under locale doesn't work */
2326 if (UTF8SKIP(p) == cap_sharp_s_len
2327 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
2329 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2330 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2331 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2332 "resolved to \"\\x{17F}\\x{17F}\".");
2337 if (UTF8SKIP(p) == long_s_t_len
2338 && memEQ((char *) p, LONG_S_T, long_s_t_len))
2340 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2341 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2342 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2343 "resolved to \"\\x{FB06}\".");
2344 goto return_ligature_st;
2347 #if UNICODE_MAJOR_VERSION == 3 \
2348 && UNICODE_DOT_VERSION == 0 \
2349 && UNICODE_DOT_DOT_VERSION == 1
2350 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2352 /* And special case this on this Unicode version only, for the same
2353 * reaons the other two are special cased. They would cross the
2354 * 255/256 boundary which is forbidden under /l, and so the code
2355 * wouldn't catch that they are equivalent (which they are only in
2357 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2358 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2360 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2361 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2362 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2363 "resolved to \"\\x{0131}\".");
2364 goto return_dotless_i;
2368 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2370 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2374 /* This is called when changing the case of a UTF-8-encoded
2375 * character above the ASCII range, and the result should not
2376 * contain an ASCII character. */
2378 UV original; /* To store the first code point of <p> */
2380 /* Look at every character in the result; if any cross the
2381 * boundary, the whole thing is disallowed */
2383 U8* e = ustrp + *lenp;
2386 /* Crossed, have to return the original */
2387 original = valid_utf8_to_uvchr(p, lenp);
2389 /* But in these instances, there is an alternative we can
2390 * return that is valid */
2391 if (original == LATIN_SMALL_LETTER_SHARP_S
2392 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2393 || original == LATIN_CAPITAL_LETTER_SHARP_S
2398 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2399 goto return_ligature_st;
2401 #if UNICODE_MAJOR_VERSION == 3 \
2402 && UNICODE_DOT_VERSION == 0 \
2403 && UNICODE_DOT_DOT_VERSION == 1
2405 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2406 goto return_dotless_i;
2409 Copy(p, ustrp, *lenp, char);
2415 /* Here, no characters crossed, result is ok as-is */
2420 /* Here, used locale rules. Convert back to UTF-8 */
2421 if (UTF8_IS_INVARIANT(result)) {
2422 *ustrp = (U8) result;
2426 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2427 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2434 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2435 * folds to a string of two of these characters. By returning this
2436 * instead, then, e.g.,
2437 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2440 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2441 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2443 return LATIN_SMALL_LETTER_LONG_S;
2446 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2447 * have the other one fold to it */
2449 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2450 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2451 return LATIN_SMALL_LIGATURE_ST;
2453 #if UNICODE_MAJOR_VERSION == 3 \
2454 && UNICODE_DOT_VERSION == 0 \
2455 && UNICODE_DOT_DOT_VERSION == 1
2458 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2459 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2460 return LATIN_SMALL_LETTER_DOTLESS_I;
2467 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2468 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2469 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2473 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2475 PERL_ARGS_ASSERT_SWASH_INIT;
2477 /* Returns a copy of a swash initiated by the called function. This is the
2478 * public interface, and returning a copy prevents others from doing
2479 * mischief on the original */
2481 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2485 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2488 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2489 * use the following define */
2491 #define CORE_SWASH_INIT_RETURN(x) \
2492 PL_curpm= old_PL_curpm; \
2495 /* Initialize and return a swash, creating it if necessary. It does this
2496 * by calling utf8_heavy.pl in the general case. The returned value may be
2497 * the swash's inversion list instead if the input parameters allow it.
2498 * Which is returned should be immaterial to callers, as the only
2499 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2500 * and swash_to_invlist() handle both these transparently.
2502 * This interface should only be used by functions that won't destroy or
2503 * adversely change the swash, as doing so affects all other uses of the
2504 * swash in the program; the general public should use 'Perl_swash_init'
2507 * pkg is the name of the package that <name> should be in.
2508 * name is the name of the swash to find. Typically it is a Unicode
2509 * property name, including user-defined ones
2510 * listsv is a string to initialize the swash with. It must be of the form
2511 * documented as the subroutine return value in
2512 * L<perlunicode/User-Defined Character Properties>
2513 * minbits is the number of bits required to represent each data element.
2514 * It is '1' for binary properties.
2515 * none I (khw) do not understand this one, but it is used only in tr///.
2516 * invlist is an inversion list to initialize the swash with (or NULL)
2517 * flags_p if non-NULL is the address of various input and output flag bits
2518 * to the routine, as follows: ('I' means is input to the routine;
2519 * 'O' means output from the routine. Only flags marked O are
2520 * meaningful on return.)
2521 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2522 * came from a user-defined property. (I O)
2523 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2524 * when the swash cannot be located, to simply return NULL. (I)
2525 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2526 * return of an inversion list instead of a swash hash if this routine
2527 * thinks that would result in faster execution of swash_fetch() later
2530 * Thus there are three possible inputs to find the swash: <name>,
2531 * <listsv>, and <invlist>. At least one must be specified. The result
2532 * will be the union of the specified ones, although <listsv>'s various
2533 * actions can intersect, etc. what <name> gives. To avoid going out to
2534 * disk at all, <invlist> should specify completely what the swash should
2535 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2537 * <invlist> is only valid for binary properties */
2539 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2541 SV* retval = &PL_sv_undef;
2542 HV* swash_hv = NULL;
2543 const int invlist_swash_boundary =
2544 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2545 ? 512 /* Based on some benchmarking, but not extensive, see commit
2547 : -1; /* Never return just an inversion list */
2549 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2550 assert(! invlist || minbits == 1);
2552 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2553 that triggered the swash init and the swash init perl logic itself.
2556 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2558 if (listsv != &PL_sv_undef || strNE(name, "")) {
2560 const size_t pkg_len = strlen(pkg);
2561 const size_t name_len = strlen(name);
2562 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2566 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2568 PUSHSTACKi(PERLSI_MAGIC);
2572 /* We might get here via a subroutine signature which uses a utf8
2573 * parameter name, at which point PL_subname will have been set
2574 * but not yet used. */
2575 save_item(PL_subname);
2576 if (PL_parser && PL_parser->error_count)
2577 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2578 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2579 if (!method) { /* demand load UTF-8 */
2581 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2582 GvSV(PL_errgv) = NULL;
2583 #ifndef NO_TAINT_SUPPORT
2584 /* It is assumed that callers of this routine are not passing in
2585 * any user derived data. */
2586 /* Need to do this after save_re_context() as it will set
2587 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2588 * in Perl_magic_get). Even line to create errsv_save can turn on
2590 SAVEBOOL(TAINT_get);
2593 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2596 /* Not ERRSV, as there is no need to vivify a scalar we are
2597 about to discard. */
2598 SV * const errsv = GvSV(PL_errgv);
2599 if (!SvTRUE(errsv)) {
2600 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2601 SvREFCNT_dec(errsv);
2609 mPUSHp(pkg, pkg_len);
2610 mPUSHp(name, name_len);
2615 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2616 GvSV(PL_errgv) = NULL;
2617 /* If we already have a pointer to the method, no need to use
2618 * call_method() to repeat the lookup. */
2620 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2621 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2623 retval = *PL_stack_sp--;
2624 SvREFCNT_inc(retval);
2627 /* Not ERRSV. See above. */
2628 SV * const errsv = GvSV(PL_errgv);
2629 if (!SvTRUE(errsv)) {
2630 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2631 SvREFCNT_dec(errsv);
2636 if (IN_PERL_COMPILETIME) {
2637 CopHINTS_set(PL_curcop, PL_hints);
2639 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2642 /* If caller wants to handle missing properties, let them */
2643 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2644 CORE_SWASH_INIT_RETURN(NULL);
2647 "Can't find Unicode property definition \"%"SVf"\"",
2649 NOT_REACHED; /* NOTREACHED */
2651 } /* End of calling the module to find the swash */
2653 /* If this operation fetched a swash, and we will need it later, get it */
2654 if (retval != &PL_sv_undef
2655 && (minbits == 1 || (flags_p
2657 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2659 swash_hv = MUTABLE_HV(SvRV(retval));
2661 /* If we don't already know that there is a user-defined component to
2662 * this swash, and the user has indicated they wish to know if there is
2663 * one (by passing <flags_p>), find out */
2664 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2665 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2666 if (user_defined && SvUV(*user_defined)) {
2667 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2672 /* Make sure there is an inversion list for binary properties */
2674 SV** swash_invlistsvp = NULL;
2675 SV* swash_invlist = NULL;
2676 bool invlist_in_swash_is_valid = FALSE;
2677 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2678 an unclaimed reference count */
2680 /* If this operation fetched a swash, get its already existing
2681 * inversion list, or create one for it */
2684 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2685 if (swash_invlistsvp) {
2686 swash_invlist = *swash_invlistsvp;
2687 invlist_in_swash_is_valid = TRUE;
2690 swash_invlist = _swash_to_invlist(retval);
2691 swash_invlist_unclaimed = TRUE;
2695 /* If an inversion list was passed in, have to include it */
2698 /* Any fetched swash will by now have an inversion list in it;
2699 * otherwise <swash_invlist> will be NULL, indicating that we
2700 * didn't fetch a swash */
2701 if (swash_invlist) {
2703 /* Add the passed-in inversion list, which invalidates the one
2704 * already stored in the swash */
2705 invlist_in_swash_is_valid = FALSE;
2706 _invlist_union(invlist, swash_invlist, &swash_invlist);
2710 /* Here, there is no swash already. Set up a minimal one, if
2711 * we are going to return a swash */
2712 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2714 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2716 swash_invlist = invlist;
2720 /* Here, we have computed the union of all the passed-in data. It may
2721 * be that there was an inversion list in the swash which didn't get
2722 * touched; otherwise save the computed one */
2723 if (! invlist_in_swash_is_valid
2724 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2726 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2728 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2730 /* We just stole a reference count. */
2731 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2732 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2735 SvREADONLY_on(swash_invlist);
2737 /* Use the inversion list stand-alone if small enough */
2738 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2739 SvREFCNT_dec(retval);
2740 if (!swash_invlist_unclaimed)
2741 SvREFCNT_inc_simple_void_NN(swash_invlist);
2742 retval = newRV_noinc(swash_invlist);
2746 CORE_SWASH_INIT_RETURN(retval);
2747 #undef CORE_SWASH_INIT_RETURN
2751 /* This API is wrong for special case conversions since we may need to
2752 * return several Unicode characters for a single Unicode character
2753 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2754 * the lower-level routine, and it is similarly broken for returning
2755 * multiple values. --jhi
2756 * For those, you should use to_utf8_case() instead */
2757 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2760 * Returns the value of property/mapping C<swash> for the first character
2761 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2762 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
2763 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2765 * A "swash" is a hash which contains initially the keys/values set up by
2766 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2767 * property for all possible code points. Things are stored in a compact form
2768 * (see utf8_heavy.pl) so that calculation is required to find the actual
2769 * property value for a given code point. As code points are looked up, new
2770 * key/value pairs are added to the hash, so that the calculation doesn't have
2771 * to ever be re-done. Further, each calculation is done, not just for the
2772 * desired one, but for a whole block of code points adjacent to that one.
2773 * For binary properties on ASCII machines, the block is usually for 64 code
2774 * points, starting with a code point evenly divisible by 64. Thus if the
2775 * property value for code point 257 is requested, the code goes out and
2776 * calculates the property values for all 64 code points between 256 and 319,
2777 * and stores these as a single 64-bit long bit vector, called a "swatch",
2778 * under the key for code point 256. The key is the UTF-8 encoding for code
2779 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2780 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2781 * for code point 258 is then requested, this code realizes that it would be
2782 * stored under the key for 256, and would find that value and extract the
2783 * relevant bit, offset from 256.
2785 * Non-binary properties are stored in as many bits as necessary to represent
2786 * their values (32 currently, though the code is more general than that), not
2787 * as single bits, but the principal is the same: the value for each key is a
2788 * vector that encompasses the property values for all code points whose UTF-8
2789 * representations are represented by the key. That is, for all code points
2790 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2794 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2796 HV *const hv = MUTABLE_HV(SvRV(swash));
2801 const U8 *tmps = NULL;
2805 PERL_ARGS_ASSERT_SWASH_FETCH;
2807 /* If it really isn't a hash, it isn't really swash; must be an inversion
2809 if (SvTYPE(hv) != SVt_PVHV) {
2810 return _invlist_contains_cp((SV*)hv,
2812 ? valid_utf8_to_uvchr(ptr, NULL)
2816 /* We store the values in a "swatch" which is a vec() value in a swash
2817 * hash. Code points 0-255 are a single vec() stored with key length
2818 * (klen) 0. All other code points have a UTF-8 representation
2819 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2820 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2821 * length for them is the length of the encoded char - 1. ptr[klen] is the
2822 * final byte in the sequence representing the character */
2823 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2828 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2831 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
2834 klen = UTF8SKIP(ptr) - 1;
2836 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2837 * the vec is the final byte in the sequence. (In EBCDIC this is
2838 * converted to I8 to get consecutive values.) To help you visualize
2840 * Straight 1047 After final byte
2841 * UTF-8 UTF-EBCDIC I8 transform
2842 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2843 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2845 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2846 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2848 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2849 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2851 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2852 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2854 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2855 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2857 * (There are no discontinuities in the elided (...) entries.)
2858 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2859 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2860 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2861 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2862 * index into the vec() swatch (after subtracting 0x80, which we
2863 * actually do with an '&').
2864 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2865 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2866 * dicontinuities which go away by transforming it into I8, and we
2867 * effectively subtract 0xA0 to get the index. */
2868 needents = (1 << UTF_ACCUMULATION_SHIFT);
2869 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2873 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
2874 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2875 * it's nothing to sniff at.) Pity we usually come through at least
2876 * two function calls to get here...
2878 * NB: this code assumes that swatches are never modified, once generated!
2881 if (hv == PL_last_swash_hv &&
2882 klen == PL_last_swash_klen &&
2883 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2885 tmps = PL_last_swash_tmps;
2886 slen = PL_last_swash_slen;
2889 /* Try our second-level swatch cache, kept in a hash. */
2890 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2892 /* If not cached, generate it via swatch_get */
2893 if (!svp || !SvPOK(*svp)
2894 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2897 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2898 swatch = swatch_get(swash,
2899 code_point & ~((UV)needents - 1),
2902 else { /* For the first 256 code points, the swatch has a key of
2904 swatch = swatch_get(swash, 0, needents);
2907 if (IN_PERL_COMPILETIME)
2908 CopHINTS_set(PL_curcop, PL_hints);
2910 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2912 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2913 || (slen << 3) < needents)
2914 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2915 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2916 svp, tmps, (UV)slen, (UV)needents);
2919 PL_last_swash_hv = hv;
2920 assert(klen <= sizeof(PL_last_swash_key));
2921 PL_last_swash_klen = (U8)klen;
2922 /* FIXME change interpvar.h? */
2923 PL_last_swash_tmps = (U8 *) tmps;
2924 PL_last_swash_slen = slen;
2926 Copy(ptr, PL_last_swash_key, klen, U8);
2929 switch ((int)((slen << 3) / needents)) {
2931 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2933 return ((UV) tmps[off]);
2937 ((UV) tmps[off ] << 8) +
2938 ((UV) tmps[off + 1]);
2942 ((UV) tmps[off ] << 24) +
2943 ((UV) tmps[off + 1] << 16) +
2944 ((UV) tmps[off + 2] << 8) +
2945 ((UV) tmps[off + 3]);
2947 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2948 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2949 NORETURN_FUNCTION_END;
2952 /* Read a single line of the main body of the swash input text. These are of
2955 * where each number is hex. The first two numbers form the minimum and
2956 * maximum of a range, and the third is the value associated with the range.
2957 * Not all swashes should have a third number
2959 * On input: l points to the beginning of the line to be examined; it points
2960 * to somewhere in the string of the whole input text, and is
2961 * terminated by a \n or the null string terminator.
2962 * lend points to the null terminator of that string
2963 * wants_value is non-zero if the swash expects a third number
2964 * typestr is the name of the swash's mapping, like 'ToLower'
2965 * On output: *min, *max, and *val are set to the values read from the line.
2966 * returns a pointer just beyond the line examined. If there was no
2967 * valid min number on the line, returns lend+1
2971 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2972 const bool wants_value, const U8* const typestr)
2974 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2975 STRLEN numlen; /* Length of the number */
2976 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2977 | PERL_SCAN_DISALLOW_PREFIX
2978 | PERL_SCAN_SILENT_NON_PORTABLE;
2980 /* nl points to the next \n in the scan */
2981 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2983 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2985 /* Get the first number on the line: the range minimum */
2987 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2988 *max = *min; /* So can never return without setting max */
2989 if (numlen) /* If found a hex number, position past it */
2991 else if (nl) { /* Else, go handle next line, if any */
2992 return nl + 1; /* 1 is length of "\n" */
2994 else { /* Else, no next line */
2995 return lend + 1; /* to LIST's end at which \n is not found */
2998 /* The max range value follows, separated by a BLANK */
3001 flags = PERL_SCAN_SILENT_ILLDIGIT
3002 | PERL_SCAN_DISALLOW_PREFIX
3003 | PERL_SCAN_SILENT_NON_PORTABLE;
3005 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3008 else /* If no value here, it is a single element range */
3011 /* Non-binary tables have a third entry: what the first element of the
3012 * range maps to. The map for those currently read here is in hex */
3016 flags = PERL_SCAN_SILENT_ILLDIGIT
3017 | PERL_SCAN_DISALLOW_PREFIX
3018 | PERL_SCAN_SILENT_NON_PORTABLE;
3020 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3029 /* diag_listed_as: To%s: illegal mapping '%s' */
3030 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3036 *val = 0; /* bits == 1, then any val should be ignored */
3038 else { /* Nothing following range min, should be single element with no
3043 /* diag_listed_as: To%s: illegal mapping '%s' */
3044 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3048 *val = 0; /* bits == 1, then val should be ignored */
3051 /* Position to next line if any, or EOF */
3061 * Returns a swatch (a bit vector string) for a code point sequence
3062 * that starts from the value C<start> and comprises the number C<span>.
3063 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3064 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3067 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3070 U8 *l, *lend, *x, *xend, *s, *send;
3071 STRLEN lcur, xcur, scur;
3072 HV *const hv = MUTABLE_HV(SvRV(swash));
3073 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3075 SV** listsvp = NULL; /* The string containing the main body of the table */
3076 SV** extssvp = NULL;
3077 SV** invert_it_svp = NULL;
3080 STRLEN octets; /* if bits == 1, then octets == 0 */
3082 UV end = start + span;
3084 if (invlistsvp == NULL) {
3085 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3086 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3087 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3088 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3089 listsvp = hv_fetchs(hv, "LIST", FALSE);
3090 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3092 bits = SvUV(*bitssvp);
3093 none = SvUV(*nonesvp);
3094 typestr = (U8*)SvPV_nolen(*typesvp);
3100 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3102 PERL_ARGS_ASSERT_SWATCH_GET;
3104 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3105 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3109 /* If overflowed, use the max possible */
3115 /* create and initialize $swatch */
3116 scur = octets ? (span * octets) : (span + 7) / 8;
3117 swatch = newSV(scur);
3119 s = (U8*)SvPVX(swatch);
3120 if (octets && none) {
3121 const U8* const e = s + scur;
3124 *s++ = (U8)(none & 0xff);
3125 else if (bits == 16) {
3126 *s++ = (U8)((none >> 8) & 0xff);
3127 *s++ = (U8)( none & 0xff);
3129 else if (bits == 32) {
3130 *s++ = (U8)((none >> 24) & 0xff);
3131 *s++ = (U8)((none >> 16) & 0xff);
3132 *s++ = (U8)((none >> 8) & 0xff);
3133 *s++ = (U8)( none & 0xff);
3139 (void)memzero((U8*)s, scur + 1);
3141 SvCUR_set(swatch, scur);
3142 s = (U8*)SvPVX(swatch);
3144 if (invlistsvp) { /* If has an inversion list set up use that */
3145 _invlist_populate_swatch(*invlistsvp, start, end, s);
3149 /* read $swash->{LIST} */
3150 l = (U8*)SvPV(*listsvp, lcur);
3153 UV min, max, val, upper;
3154 l = swash_scan_list_line(l, lend, &min, &max, &val,
3155 cBOOL(octets), typestr);
3160 /* If looking for something beyond this range, go try the next one */
3164 /* <end> is generally 1 beyond where we want to set things, but at the
3165 * platform's infinity, where we can't go any higher, we want to
3166 * include the code point at <end> */
3169 : (max != UV_MAX || end != UV_MAX)
3176 if (!none || val < none) {
3181 for (key = min; key <= upper; key++) {
3183 /* offset must be non-negative (start <= min <= key < end) */
3184 offset = octets * (key - start);
3186 s[offset] = (U8)(val & 0xff);
3187 else if (bits == 16) {
3188 s[offset ] = (U8)((val >> 8) & 0xff);
3189 s[offset + 1] = (U8)( val & 0xff);
3191 else if (bits == 32) {
3192 s[offset ] = (U8)((val >> 24) & 0xff);
3193 s[offset + 1] = (U8)((val >> 16) & 0xff);
3194 s[offset + 2] = (U8)((val >> 8) & 0xff);
3195 s[offset + 3] = (U8)( val & 0xff);
3198 if (!none || val < none)
3202 else { /* bits == 1, then val should be ignored */
3207 for (key = min; key <= upper; key++) {
3208 const STRLEN offset = (STRLEN)(key - start);
3209 s[offset >> 3] |= 1 << (offset & 7);
3214 /* Invert if the data says it should be. Assumes that bits == 1 */
3215 if (invert_it_svp && SvUV(*invert_it_svp)) {
3217 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3218 * be 0, and their inversion should also be 0, as we don't succeed any
3219 * Unicode property matches for non-Unicode code points */
3220 if (start <= PERL_UNICODE_MAX) {
3222 /* The code below assumes that we never cross the
3223 * Unicode/above-Unicode boundary in a range, as otherwise we would
3224 * have to figure out where to stop flipping the bits. Since this
3225 * boundary is divisible by a large power of 2, and swatches comes
3226 * in small powers of 2, this should be a valid assumption */
3227 assert(start + span - 1 <= PERL_UNICODE_MAX);
3237 /* read $swash->{EXTRAS}
3238 * This code also copied to swash_to_invlist() below */
3239 x = (U8*)SvPV(*extssvp, xcur);
3247 SV **otherbitssvp, *other;
3251 const U8 opc = *x++;
3255 nl = (U8*)memchr(x, '\n', xend - x);
3257 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3259 x = nl + 1; /* 1 is length of "\n" */
3263 x = xend; /* to EXTRAS' end at which \n is not found */
3270 namelen = nl - namestr;
3274 namelen = xend - namestr;
3278 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3279 otherhv = MUTABLE_HV(SvRV(*othersvp));
3280 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3281 otherbits = (STRLEN)SvUV(*otherbitssvp);
3282 if (bits < otherbits)
3283 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3284 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3286 /* The "other" swatch must be destroyed after. */
3287 other = swatch_get(*othersvp, start, span);
3288 o = (U8*)SvPV(other, olen);
3291 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3293 s = (U8*)SvPV(swatch, slen);
3294 if (bits == 1 && otherbits == 1) {
3296 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3297 "mismatch, slen=%"UVuf", olen=%"UVuf,
3298 (UV)slen, (UV)olen);
3322 STRLEN otheroctets = otherbits >> 3;
3324 U8* const send = s + slen;
3329 if (otherbits == 1) {
3330 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3334 STRLEN vlen = otheroctets;
3342 if (opc == '+' && otherval)
3343 NOOP; /* replace with otherval */
3344 else if (opc == '!' && !otherval)
3346 else if (opc == '-' && otherval)
3348 else if (opc == '&' && !otherval)
3351 s += octets; /* no replacement */
3356 *s++ = (U8)( otherval & 0xff);
3357 else if (bits == 16) {
3358 *s++ = (U8)((otherval >> 8) & 0xff);
3359 *s++ = (U8)( otherval & 0xff);
3361 else if (bits == 32) {
3362 *s++ = (U8)((otherval >> 24) & 0xff);
3363 *s++ = (U8)((otherval >> 16) & 0xff);
3364 *s++ = (U8)((otherval >> 8) & 0xff);
3365 *s++ = (U8)( otherval & 0xff);
3369 sv_free(other); /* through with it! */
3375 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3378 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3379 * Can't be used on a property that is subject to user override, as it
3380 * relies on the value of SPECIALS in the swash which would be set by
3381 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3382 * for overridden properties
3384 * Returns a hash which is the inversion and closure of a swash mapping.
3385 * For example, consider the input lines:
3390 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
3391 * 006C. The value for each key is an array. For 006C, the array would
3392 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3393 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
3395 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3396 * keys are only code points that are folded-to, so it isn't a full closure.
3398 * Essentially, for any code point, it gives all the code points that map to
3399 * it, or the list of 'froms' for that point.
3401 * Currently it ignores any additions or deletions from other swashes,
3402 * looking at just the main body of the swash, and if there are SPECIALS
3403 * in the swash, at that hash
3405 * The specials hash can be extra code points, and most likely consists of
3406 * maps from single code points to multiple ones (each expressed as a string
3407 * of UTF-8 characters). This function currently returns only 1-1 mappings.
3408 * However consider this possible input in the specials hash:
3409 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3410 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3412 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3413 * currently handle. But it also means that FB05 and FB06 are equivalent in
3414 * a 1-1 mapping which we should handle, and this relationship may not be in
3415 * the main table. Therefore this function examines all the multi-char
3416 * sequences and adds the 1-1 mappings that come out of that.
3418 * XXX This function was originally intended to be multipurpose, but its
3419 * only use is quite likely to remain for constructing the inversion of
3420 * the CaseFolding (//i) property. If it were more general purpose for
3421 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3422 * because certain folds are prohibited under /iaa and /il. As an example,
3423 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3424 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3425 * prohibited, so we would not figure out that they fold to each other.
3426 * Code could be written to automatically figure this out, similar to the
3427 * code that does this for multi-character folds, but this is the only case
3428 * where something like this is ever likely to happen, as all the single
3429 * char folds to the 0-255 range are now quite settled. Instead there is a
3430 * little special code that is compiled only for this Unicode version. This
3431 * is smaller and didn't require much coding time to do. But this makes
3432 * this routine strongly tied to being used just for CaseFolding. If ever
3433 * it should be generalized, this would have to be fixed */
3437 HV *const hv = MUTABLE_HV(SvRV(swash));
3439 /* The string containing the main body of the table. This will have its
3440 * assertion fail if the swash has been converted to its inversion list */
3441 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3443 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3444 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3445 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3446 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3447 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3448 const STRLEN bits = SvUV(*bitssvp);
3449 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3450 const UV none = SvUV(*nonesvp);
3451 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3455 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3457 /* Must have at least 8 bits to get the mappings */
3458 if (bits != 8 && bits != 16 && bits != 32) {
3459 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3463 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3464 mapping to more than one character */
3466 /* Construct an inverse mapping hash for the specials */
3467 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3468 HV * specials_inverse = newHV();
3469 char *char_from; /* the lhs of the map */
3470 I32 from_len; /* its byte length */
3471 char *char_to; /* the rhs of the map */
3472 I32 to_len; /* its byte length */
3473 SV *sv_to; /* and in a sv */
3474 AV* from_list; /* list of things that map to each 'to' */
3476 hv_iterinit(specials_hv);
3478 /* The keys are the characters (in UTF-8) that map to the corresponding
3479 * UTF-8 string value. Iterate through the list creating the inverse
3481 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3483 if (! SvPOK(sv_to)) {
3484 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3485 "unexpectedly is not a string, flags=%lu",
3486 (unsigned long)SvFLAGS(sv_to));
3488 /*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)));*/
3490 /* Each key in the inverse list is a mapped-to value, and the key's
3491 * hash value is a list of the strings (each in UTF-8) that map to
3492 * it. Those strings are all one character long */
3493 if ((listp = hv_fetch(specials_inverse,
3497 from_list = (AV*) *listp;
3499 else { /* No entry yet for it: create one */
3500 from_list = newAV();
3501 if (! hv_store(specials_inverse,
3504 (SV*) from_list, 0))
3506 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3510 /* Here have the list associated with this 'to' (perhaps newly
3511 * created and empty). Just add to it. Note that we ASSUME that
3512 * the input is guaranteed to not have duplications, so we don't
3513 * check for that. Duplications just slow down execution time. */
3514 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3517 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3518 * it looking for cases like the FB05/FB06 examples above. There would
3519 * be an entry in the hash like
3520 * 'st' => [ FB05, FB06 ]
3521 * In this example we will create two lists that get stored in the
3522 * returned hash, 'ret':
3523 * FB05 => [ FB05, FB06 ]
3524 * FB06 => [ FB05, FB06 ]
3526 * Note that there is nothing to do if the array only has one element.
3527 * (In the normal 1-1 case handled below, we don't have to worry about
3528 * two lists, as everything gets tied to the single list that is
3529 * generated for the single character 'to'. But here, we are omitting
3530 * that list, ('st' in the example), so must have multiple lists.) */
3531 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3532 &char_to, &to_len)))
3534 if (av_tindex(from_list) > 0) {
3537 /* We iterate over all combinations of i,j to place each code
3538 * point on each list */
3539 for (i = 0; i <= av_tindex(from_list); i++) {
3541 AV* i_list = newAV();
3542 SV** entryp = av_fetch(from_list, i, FALSE);
3543 if (entryp == NULL) {
3544 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3546 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3547 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3549 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3550 (SV*) i_list, FALSE))
3552 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3555 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3556 for (j = 0; j <= av_tindex(from_list); j++) {
3557 entryp = av_fetch(from_list, j, FALSE);
3558 if (entryp == NULL) {
3559 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3562 /* When i==j this adds itself to the list */
3563 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3564 (U8*) SvPVX(*entryp),
3565 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3567 /*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));*/
3572 SvREFCNT_dec(specials_inverse); /* done with it */
3573 } /* End of specials */
3575 /* read $swash->{LIST} */
3577 #if UNICODE_MAJOR_VERSION == 3 \
3578 && UNICODE_DOT_VERSION == 0 \
3579 && UNICODE_DOT_DOT_VERSION == 1
3581 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3582 * rule so that things work under /iaa and /il */
3584 SV * mod_listsv = sv_mortalcopy(*listsvp);
3585 sv_catpv(mod_listsv, "130\t130\t131\n");
3586 l = (U8*)SvPV(mod_listsv, lcur);
3590 l = (U8*)SvPV(*listsvp, lcur);
3596 /* Go through each input line */
3600 l = swash_scan_list_line(l, lend, &min, &max, &val,
3601 cBOOL(octets), typestr);
3606 /* Each element in the range is to be inverted */
3607 for (inverse = min; inverse <= max; inverse++) {
3611 bool found_key = FALSE;
3612 bool found_inverse = FALSE;
3614 /* The key is the inverse mapping */
3615 char key[UTF8_MAXBYTES+1];
3616 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3617 STRLEN key_len = key_end - key;
3619 /* Get the list for the map */
3620 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3621 list = (AV*) *listp;
3623 else { /* No entry yet for it: create one */
3625 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3626 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3630 /* Look through list to see if this inverse mapping already is
3631 * listed, or if there is a mapping to itself already */
3632 for (i = 0; i <= av_tindex(list); i++) {
3633 SV** entryp = av_fetch(list, i, FALSE);
3636 if (entryp == NULL) {
3637 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3641 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3645 if (uv == inverse) {
3646 found_inverse = TRUE;
3649 /* No need to continue searching if found everything we are
3651 if (found_key && found_inverse) {
3656 /* Make sure there is a mapping to itself on the list */
3658 av_push(list, newSVuv(val));
3659 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3663 /* Simply add the value to the list */
3664 if (! found_inverse) {
3665 av_push(list, newSVuv(inverse));
3666 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3669 /* swatch_get() increments the value of val for each element in the
3670 * range. That makes more compact tables possible. You can
3671 * express the capitalization, for example, of all consecutive
3672 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3673 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3674 * and it's not documented; it appears to be used only in
3675 * implementing tr//; I copied the semantics from swatch_get(), just
3677 if (!none || val < none) {
3687 Perl__swash_to_invlist(pTHX_ SV* const swash)
3690 /* Subject to change or removal. For use only in one place in regcomp.c.
3691 * Ownership is given to one reference count in the returned SV* */
3696 HV *const hv = MUTABLE_HV(SvRV(swash));
3697 UV elements = 0; /* Number of elements in the inversion list */
3707 STRLEN octets; /* if bits == 1, then octets == 0 */
3713 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3715 /* If not a hash, it must be the swash's inversion list instead */
3716 if (SvTYPE(hv) != SVt_PVHV) {
3717 return SvREFCNT_inc_simple_NN((SV*) hv);
3720 /* The string containing the main body of the table */
3721 listsvp = hv_fetchs(hv, "LIST", FALSE);
3722 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3723 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3724 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3725 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3727 typestr = (U8*)SvPV_nolen(*typesvp);
3728 bits = SvUV(*bitssvp);
3729 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3731 /* read $swash->{LIST} */
3732 if (SvPOK(*listsvp)) {
3733 l = (U8*)SvPV(*listsvp, lcur);
3736 /* LIST legitimately doesn't contain a string during compilation phases
3737 * of Perl itself, before the Unicode tables are generated. In this
3738 * case, just fake things up by creating an empty list */
3745 if (*l == 'V') { /* Inversion list format */
3746 const char *after_atou = (char *) lend;
3748 UV* other_elements_ptr;
3750 /* The first number is a count of the rest */
3752 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
3753 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
3755 if (elements == 0) {
3756 invlist = _new_invlist(0);
3759 while (isSPACE(*l)) l++;
3760 l = (U8 *) after_atou;
3762 /* Get the 0th element, which is needed to setup the inversion list */
3763 while (isSPACE(*l)) l++;
3764 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
3765 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
3767 l = (U8 *) after_atou;
3768 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3771 /* Then just populate the rest of the input */
3772 while (elements-- > 0) {
3774 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3776 while (isSPACE(*l)) l++;
3777 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
3778 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
3780 l = (U8 *) after_atou;
3786 /* Scan the input to count the number of lines to preallocate array
3787 * size based on worst possible case, which is each line in the input
3788 * creates 2 elements in the inversion list: 1) the beginning of a
3789 * range in the list; 2) the beginning of a range not in the list. */
3790 while ((loc = (strchr(loc, '\n'))) != NULL) {
3795 /* If the ending is somehow corrupt and isn't a new line, add another
3796 * element for the final range that isn't in the inversion list */
3797 if (! (*lend == '\n'
3798 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3803 invlist = _new_invlist(elements);
3805 /* Now go through the input again, adding each range to the list */
3808 UV val; /* Not used by this function */
3810 l = swash_scan_list_line(l, lend, &start, &end, &val,
3811 cBOOL(octets), typestr);
3817 invlist = _add_range_to_invlist(invlist, start, end);
3821 /* Invert if the data says it should be */
3822 if (invert_it_svp && SvUV(*invert_it_svp)) {
3823 _invlist_invert(invlist);
3826 /* This code is copied from swatch_get()
3827 * read $swash->{EXTRAS} */
3828 x = (U8*)SvPV(*extssvp, xcur);
3836 SV **otherbitssvp, *other;
3839 const U8 opc = *x++;
3843 nl = (U8*)memchr(x, '\n', xend - x);
3845 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3847 x = nl + 1; /* 1 is length of "\n" */
3851 x = xend; /* to EXTRAS' end at which \n is not found */
3858 namelen = nl - namestr;
3862 namelen = xend - namestr;
3866 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3867 otherhv = MUTABLE_HV(SvRV(*othersvp));
3868 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3869 otherbits = (STRLEN)SvUV(*otherbitssvp);
3871 if (bits != otherbits || bits != 1) {
3872 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3873 "properties, bits=%"UVuf", otherbits=%"UVuf,
3874 (UV)bits, (UV)otherbits);
3877 /* The "other" swatch must be destroyed after. */
3878 other = _swash_to_invlist((SV *)*othersvp);
3880 /* End of code copied from swatch_get() */
3883 _invlist_union(invlist, other, &invlist);
3886 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3889 _invlist_subtract(invlist, other, &invlist);
3892 _invlist_intersection(invlist, other, &invlist);
3897 sv_free(other); /* through with it! */
3900 SvREADONLY_on(invlist);
3905 Perl__get_swash_invlist(pTHX_ SV* const swash)
3909 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3911 if (! SvROK(swash)) {
3915 /* If it really isn't a hash, it isn't really swash; must be an inversion
3917 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3921 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3930 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3932 /* May change: warns if surrogates, non-character code points, or
3933 * non-Unicode code points are in s which has length len bytes. Returns
3934 * TRUE if none found; FALSE otherwise. The only other validity check is
3935 * to make sure that this won't exceed the string's length.
3937 * Code points above the platform's C<IV_MAX> will raise a deprecation
3938 * warning, unless those are turned off. */
3940 const U8* const e = s + len;
3943 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3946 if (UTF8SKIP(s) > len) {
3947 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3948 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3951 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
3953 if (UTF8_IS_SUPER(s, e)) {
3954 if ( ckWARN_d(WARN_NON_UNICODE)
3955 || ( ckWARN_d(WARN_DEPRECATED)
3956 #if defined(UV_IS_QUAD)
3957 /* 2**63 and up meet these conditions provided we have
3960 && *s == 0xFE && e - s >= UTF8_MAXBYTES
3963 && *s == 0xFF && e -s >= UTF8_MAXBYTES
3966 #else /* Below is 32-bit words */
3967 /* 2**31 and above meet these conditions on all EBCDIC
3968 * pages recognized for 32-bit platforms */
3970 && *s == 0xFE && e - s >= UTF8_MAXBYTES
3977 /* A side effect of this function will be to warn */
3978 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_SUPER);
3982 else if (UTF8_IS_SURROGATE(s, e)) {
3983 if (ckWARN_d(WARN_SURROGATE)) {
3984 /* This has a different warning than the one the called
3985 * function would output, so can't just call it, unlike we
3986 * do for the non-chars and above-unicodes */
3987 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3988 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3989 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3993 else if ((UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
3994 /* A side effect of this function will be to warn */
3995 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_NONCHAR);
4006 =for apidoc pv_uni_display
4008 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4009 length C<len>, the displayable version being at most C<pvlim> bytes long
4010 (if longer, the rest is truncated and C<"..."> will be appended).
4012 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4013 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4014 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4015 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4016 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4017 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4019 The pointer to the PV of the C<dsv> is returned.
4021 See also L</sv_uni_display>.
4025 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4030 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4034 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4036 /* This serves double duty as a flag and a character to print after
4037 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4041 if (pvlim && SvCUR(dsv) >= pvlim) {
4045 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4047 const unsigned char c = (unsigned char)u & 0xFF;
4048 if (flags & UNI_DISPLAY_BACKSLASH) {
4065 const char string = ok;
4066 sv_catpvs(dsv, "\\");
4067 sv_catpvn(dsv, &string, 1);
4070 /* isPRINT() is the locale-blind version. */
4071 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4072 const char string = c;
4073 sv_catpvn(dsv, &string, 1);
4078 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4081 sv_catpvs(dsv, "...");
4087 =for apidoc sv_uni_display
4089 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4090 the displayable version being at most C<pvlim> bytes long
4091 (if longer, the rest is truncated and "..." will be appended).
4093 The C<flags> argument is as in L</pv_uni_display>().
4095 The pointer to the PV of the C<dsv> is returned.
4100 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4102 const char * const ptr =
4103 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4105 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4107 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4108 SvCUR(ssv), pvlim, flags);
4112 =for apidoc foldEQ_utf8
4114 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4115 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4116 How far into the strings to compare is determined by other input parameters.
4118 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4119 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4120 with respect to C<s2>.
4122 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4123 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4124 scan will not be considered to be a match unless the goal is reached, and
4125 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4128 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
4129 considered an end pointer to the position 1 byte past the maximum point
4130 in C<s1> beyond which scanning will not continue under any circumstances.
4131 (This routine assumes that UTF-8 encoded input strings are not malformed;
4132 malformed input can cause it to read past C<pe1>).
4133 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4134 is less than C<s1>+C<l1>, the match will never be successful because it can
4136 get as far as its goal (and in fact is asserted against). Correspondingly for
4137 C<pe2> with respect to C<s2>.
4139 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4140 C<l2> must be non-zero), and if both do, both have to be
4141 reached for a successful match. Also, if the fold of a character is multiple
4142 characters, all of them must be matched (see tr21 reference below for
4145 Upon a successful match, if C<pe1> is non-C<NULL>,
4146 it will be set to point to the beginning of the I<next> character of C<s1>
4147 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4149 For case-insensitiveness, the "casefolding" of Unicode is used
4150 instead of upper/lowercasing both the characters, see
4151 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4155 /* A flags parameter has been added which may change, and hence isn't
4156 * externally documented. Currently it is:
4157 * 0 for as-documented above
4158 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4159 ASCII one, to not match
4160 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4161 * locale are to be used.
4162 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4163 * routine. This allows that step to be skipped.
4164 * Currently, this requires s1 to be encoded as UTF-8
4165 * (u1 must be true), which is asserted for.
4166 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4167 * cross certain boundaries. Hence, the caller should
4168 * let this function do the folding instead of
4169 * pre-folding. This code contains an assertion to
4170 * that effect. However, if the caller knows what
4171 * it's doing, it can pass this flag to indicate that,
4172 * and the assertion is skipped.
4173 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4174 * FOLDEQ_S2_FOLDS_SANE
4177 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)
4179 const U8 *p1 = (const U8*)s1; /* Point to current char */
4180 const U8 *p2 = (const U8*)s2;
4181 const U8 *g1 = NULL; /* goal for s1 */
4182 const U8 *g2 = NULL;
4183 const U8 *e1 = NULL; /* Don't scan s1 past this */
4184 U8 *f1 = NULL; /* Point to current folded */
4185 const U8 *e2 = NULL;
4187 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4188 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4189 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4190 U8 flags_for_folder = FOLD_FLAGS_FULL;
4192 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4194 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4195 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
4196 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4197 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
4198 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4199 /* The algorithm is to trial the folds without regard to the flags on
4200 * the first line of the above assert(), and then see if the result
4201 * violates them. This means that the inputs can't be pre-folded to a
4202 * violating result, hence the assert. This could be changed, with the
4203 * addition of extra tests here for the already-folded case, which would
4204 * slow it down. That cost is more than any possible gain for when these
4205 * flags are specified, as the flags indicate /il or /iaa matching which
4206 * is less common than /iu, and I (khw) also believe that real-world /il
4207 * and /iaa matches are most likely to involve code points 0-255, and this
4208 * function only under rare conditions gets called for 0-255. */
4210 if (flags & FOLDEQ_LOCALE) {
4211 if (IN_UTF8_CTYPE_LOCALE) {
4212 flags &= ~FOLDEQ_LOCALE;
4215 flags_for_folder |= FOLD_FLAGS_LOCALE;
4224 g1 = (const U8*)s1 + l1;
4232 g2 = (const U8*)s2 + l2;
4235 /* Must have at least one goal */
4240 /* Will never match if goal is out-of-bounds */
4241 assert(! e1 || e1 >= g1);
4243 /* Here, there isn't an end pointer, or it is beyond the goal. We
4244 * only go as far as the goal */
4248 assert(e1); /* Must have an end for looking at s1 */
4251 /* Same for goal for s2 */
4253 assert(! e2 || e2 >= g2);
4260 /* If both operands are already folded, we could just do a memEQ on the
4261 * whole strings at once, but it would be better if the caller realized
4262 * this and didn't even call us */
4264 /* Look through both strings, a character at a time */
4265 while (p1 < e1 && p2 < e2) {
4267 /* If at the beginning of a new character in s1, get its fold to use
4268 * and the length of the fold. */
4270 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4276 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4278 /* We have to forbid mixing ASCII with non-ASCII if the
4279 * flags so indicate. And, we can short circuit having to
4280 * call the general functions for this common ASCII case,
4281 * all of whose non-locale folds are also ASCII, and hence
4282 * UTF-8 invariants, so the UTF8ness of the strings is not
4284 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4288 *foldbuf1 = toFOLD(*p1);
4291 _to_utf8_fold_flags(p1, foldbuf1, &n1, flags_for_folder);
4293 else { /* Not UTF-8, get UTF-8 fold */
4294 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4300 if (n2 == 0) { /* Same for s2 */
4301 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4307 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4308 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4312 *foldbuf2 = toFOLD(*p2);
4315 _to_utf8_fold_flags(p2, foldbuf2, &n2, flags_for_folder);
4318 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4324 /* Here f1 and f2 point to the beginning of the strings to compare.
4325 * These strings are the folds of the next character from each input
4326 * string, stored in UTF-8. */
4328 /* While there is more to look for in both folds, see if they
4329 * continue to match */
4331 U8 fold_length = UTF8SKIP(f1);
4332 if (fold_length != UTF8SKIP(f2)
4333 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4334 function call for single
4336 || memNE((char*)f1, (char*)f2, fold_length))
4338 return 0; /* mismatch */
4341 /* Here, they matched, advance past them */
4348 /* When reach the end of any fold, advance the input past it */
4350 p1 += u1 ? UTF8SKIP(p1) : 1;
4353 p2 += u2 ? UTF8SKIP(p2) : 1;
4355 } /* End of loop through both strings */
4357 /* A match is defined by each scan that specified an explicit length
4358 * reaching its final goal, and the other not having matched a partial
4359 * character (which can happen when the fold of a character is more than one
4361 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4365 /* Successful match. Set output pointers */
4375 /* XXX The next two functions should likely be moved to mathoms.c once all
4376 * occurrences of them are removed from the core; some cpan-upstream modules
4380 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4382 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4384 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4388 =for apidoc utf8n_to_uvuni
4390 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4392 This function was useful for code that wanted to handle both EBCDIC and
4393 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4394 distinctions between the platforms have mostly been made invisible to most
4395 code, so this function is quite unlikely to be what you want. If you do need
4396 this precise functionality, use instead
4397 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4398 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4404 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4406 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4408 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4412 =for apidoc uvuni_to_utf8_flags
4414 Instead you almost certainly want to use L</uvchr_to_utf8> or
4415 L</uvchr_to_utf8_flags>.
4417 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4418 which itself, while not deprecated, should be used only in isolated
4419 circumstances. These functions were useful for code that wanted to handle
4420 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4421 v5.20, the distinctions between the platforms have mostly been made invisible
4422 to most code, so this function is quite unlikely to be what you want.
4428 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4430 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4432 return uvoffuni_to_utf8_flags(d, uv, flags);
4436 * ex: set ts=8 sts=4 sw=4 et: