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>.
106 #define HANDLE_UNICODE_SURROGATE(uv, flags) \
108 if (flags & UNICODE_WARN_SURROGATE) { \
109 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
110 "UTF-16 surrogate U+%04"UVXf, uv); \
112 if (flags & UNICODE_DISALLOW_SURROGATE) { \
117 #define HANDLE_UNICODE_NONCHAR(uv, flags) \
119 if (flags & UNICODE_WARN_NONCHAR) { \
120 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
121 "Unicode non-character U+%04"UVXf" is not " \
122 "recommended for open interchange", uv); \
124 if (flags & UNICODE_DISALLOW_NONCHAR) { \
129 /* Use shorter names internally in this file */
130 #define SHIFT UTF_ACCUMULATION_SHIFT
132 #define MARK UTF_CONTINUATION_MARK
133 #define MASK UTF_CONTINUATION_MASK
136 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
138 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
140 if (OFFUNI_IS_INVARIANT(uv)) {
141 *d++ = LATIN1_TO_NATIVE(uv);
144 if (uv <= MAX_UTF8_TWO_BYTE) {
145 *d++ = UTF8_TWO_BYTE_HI(uv);
146 *d++ = UTF8_TWO_BYTE_LO(uv);
150 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
151 * below, the 16 is for start bytes E0-EF (which are all the possible ones
152 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
153 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
154 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
155 * 0x800-0xFFFF on ASCII */
156 if (uv < (16 * (1U << (2 * SHIFT)))) {
157 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
158 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
159 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
161 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
162 aren't tested here */
163 /* The most likely code points in this range are below the surrogates.
164 * Do an extra test to quickly exclude those. */
165 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
166 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
167 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
169 HANDLE_UNICODE_NONCHAR(uv, flags);
171 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
172 HANDLE_UNICODE_SURROGATE(uv, flags);
179 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
180 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
181 * happen starting with 4-byte characters on ASCII platforms. We unify the
182 * code for these with EBCDIC, even though some of them require 5-bytes on
183 * those, because khw believes the code saving is worth the very slight
184 * performance hit on these high EBCDIC code points. */
186 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
187 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
188 && ckWARN_d(WARN_DEPRECATED))
190 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
191 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
193 if ( (flags & UNICODE_WARN_SUPER)
194 || ( UNICODE_IS_ABOVE_31_BIT(uv)
195 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
197 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
199 /* Choose the more dire applicable warning */
200 (UNICODE_IS_ABOVE_31_BIT(uv))
201 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
202 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
205 if (flags & UNICODE_DISALLOW_SUPER
206 || ( UNICODE_IS_ABOVE_31_BIT(uv)
207 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
212 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
213 HANDLE_UNICODE_NONCHAR(uv, flags);
216 /* Test for and handle 4-byte result. In the test immediately below, the
217 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
218 * characters). The 3 is for 3 continuation bytes; these each contribute
219 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
220 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
221 * 0x1_0000-0x1F_FFFF on ASCII */
222 if (uv < (8 * (1U << (3 * SHIFT)))) {
223 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
224 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
225 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
226 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
228 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
229 characters. The end-plane non-characters for EBCDIC were
230 handled just above */
231 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
232 HANDLE_UNICODE_NONCHAR(uv, flags);
234 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
235 HANDLE_UNICODE_SURROGATE(uv, flags);
242 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
243 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
244 * format. The unrolled version above turns out to not save all that much
245 * time, and at these high code points (well above the legal Unicode range
246 * on ASCII platforms, and well above anything in common use in EBCDIC),
247 * khw believes that less code outweighs slight performance gains. */
250 STRLEN len = OFFUNISKIP(uv);
253 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
254 uv >>= UTF_ACCUMULATION_SHIFT;
256 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
262 =for apidoc uvchr_to_utf8
264 Adds the UTF-8 representation of the native code point C<uv> to the end
265 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
266 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
267 the byte after the end of the new character. In other words,
269 d = uvchr_to_utf8(d, uv);
271 is the recommended wide native character-aware way of saying
275 This function accepts any UV as input, but very high code points (above
276 C<IV_MAX> on the platform) will raise a deprecation warning. This is
277 typically 0x7FFF_FFFF in a 32-bit word.
279 It is possible to forbid or warn on non-Unicode code points, or those that may
280 be problematic by using L</uvchr_to_utf8_flags>.
285 /* This is also a macro */
286 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
289 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
291 return uvchr_to_utf8(d, uv);
295 =for apidoc uvchr_to_utf8_flags
297 Adds the UTF-8 representation of the native code point C<uv> to the end
298 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
299 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
300 the byte after the end of the new character. In other words,
302 d = uvchr_to_utf8_flags(d, uv, flags);
306 d = uvchr_to_utf8_flags(d, uv, 0);
308 This is the Unicode-aware way of saying
312 If C<flags> is 0, this function accepts any UV as input, but very high code
313 points (above C<IV_MAX> for the platform) will raise a deprecation warning.
314 This is typically 0x7FFF_FFFF in a 32-bit word.
316 Specifying C<flags> can further restrict what is allowed and not warned on, as
319 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
320 the function will raise a warning, provided UTF8 warnings are enabled. If
321 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
322 NULL. If both flags are set, the function will both warn and return NULL.
324 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
325 affect how the function handles a Unicode non-character.
327 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
328 affect the handling of code points that are above the Unicode maximum of
329 0x10FFFF. Languages other than Perl may not be able to accept files that
332 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
333 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
334 three DISALLOW flags.
336 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
337 so using them is more problematic than other above-Unicode code points. Perl
338 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
339 likely that non-Perl languages will not be able to read files that contain
340 these that written by the perl interpreter; nor would Perl understand files
341 written by something that uses a different extension. For these reasons, there
342 is a separate set of flags that can warn and/or disallow these extremely high
343 code points, even if other above-Unicode ones are accepted. These are the
344 C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
345 are entirely independent from the deprecation warning for code points above
346 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
347 code point that needs more than 31 bits to represent. When that happens,
348 effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
349 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
350 above-Unicode code points, including these, as malformations; and
351 C<UNICODE_WARN_SUPER> warns on these.)
353 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
354 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
355 than on ASCII. Prior to that, code points 2**31 and higher were simply
356 unrepresentable, and a different, incompatible method was used to represent
357 code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
358 and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
359 platforms, warning and disallowing 2**31 and higher.
364 /* This is also a macro */
365 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
368 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
370 return uvchr_to_utf8_flags(d, uv, flags);
374 =for apidoc is_utf8_string
376 Returns true if the first C<len> bytes of string C<s> form a valid
377 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
378 using C<strlen(s)> (which means if you use this option, that C<s> can't have
379 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
380 that all characters being ASCII constitute 'a valid UTF-8 string'.
382 See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
388 Perl_is_utf8_string(const U8 *s, STRLEN len)
390 const U8* const send = s + (len ? len : strlen((const char *)s));
393 PERL_ARGS_ASSERT_IS_UTF8_STRING;
396 STRLEN len = isUTF8_CHAR(x, send);
397 if (UNLIKELY(! len)) {
407 Implemented as a macro in utf8.h
409 =for apidoc is_utf8_string_loc
411 Like L</is_utf8_string> but stores the location of the failure (in the
412 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
413 "utf8ness success") in the C<ep>.
415 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
417 =for apidoc is_utf8_string_loclen
419 Like L</is_utf8_string>() but stores the location of the failure (in the
420 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
421 "utf8ness success") in the C<ep>, and the number of UTF-8
422 encoded characters in the C<el>.
424 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
430 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
432 const U8* const send = s + (len ? len : strlen((const char *)s));
436 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
439 STRLEN len = isUTF8_CHAR(x, send);
440 if (UNLIKELY(! len)) {
458 =for apidoc utf8n_to_uvchr
460 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
461 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
463 Bottom level UTF-8 decode routine.
464 Returns the native code point value of the first character in the string C<s>,
465 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
466 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
467 the length, in bytes, of that character.
469 The value of C<flags> determines the behavior when C<s> does not point to a
470 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
471 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
472 next possible position in C<s> that could begin a non-malformed character.
473 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
475 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
476 individual types of malformations, such as the sequence being overlong (that
477 is, when there is a shorter sequence that can express the same code point;
478 overlong sequences are expressly forbidden in the UTF-8 standard due to
479 potential security issues). Another malformation example is the first byte of
480 a character not being a legal first byte. See F<utf8.h> for the list of such
481 flags. For allowed 0 length strings, this function returns 0; for allowed
482 overlong sequences, the computed code point is returned; for all other allowed
483 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
484 determinable reasonable value.
486 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
487 flags) malformation is found. If this flag is set, the routine assumes that
488 the caller will raise a warning, and this function will silently just set
489 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
491 Note that this API requires disambiguation between successful decoding a C<NUL>
492 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
493 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
494 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
497 Certain code points are considered problematic. These are Unicode surrogates,
498 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
499 By default these are considered regular code points, but certain situations
500 warrant special handling for them. If C<flags> contains
501 C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
502 malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
503 C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
504 Unicode maximum) can be set to disallow these categories individually.
506 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
507 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
508 raised for their respective categories, but otherwise the code points are
509 considered valid (not malformations). To get a category to both be treated as
510 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
511 (But note that warnings are not raised if lexically disabled nor if
512 C<UTF8_CHECK_ONLY> is also specified.)
514 It is now deprecated to have very high code points (above C<IV_MAX> on the
515 platforms) and this function will raise a deprecation warning for these (unless
516 such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
519 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
520 so using them is more problematic than other above-Unicode code points. Perl
521 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
522 likely that non-Perl languages will not be able to read files that contain
523 these that written by the perl interpreter; nor would Perl understand files
524 written by something that uses a different extension. For these reasons, there
525 is a separate set of flags that can warn and/or disallow these extremely high
526 code points, even if other above-Unicode ones are accepted. These are the
527 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
528 are entirely independent from the deprecation warning for code points above
529 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
530 code point that needs more than 31 bits to represent. When that happens,
531 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
532 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
533 above-Unicode code points, including these, as malformations; and
534 C<UTF8_WARN_SUPER> warns on these.)
536 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
537 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
538 than on ASCII. Prior to that, code points 2**31 and higher were simply
539 unrepresentable, and a different, incompatible method was used to represent
540 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
541 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
542 platforms, warning and disallowing 2**31 and higher.
544 All other code points corresponding to Unicode characters, including private
545 use and those yet to be assigned, are never considered malformed and never
552 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
554 const U8 * const s0 = s;
555 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
560 UV outlier_ret = 0; /* return value when input is in error or problematic
562 UV pack_warn = 0; /* Save result of packWARN() for later */
563 bool unexpected_non_continuation = FALSE;
564 bool overflowed = FALSE;
565 bool do_overlong_test = TRUE; /* May have to skip this test */
567 const char* const malformed_text = "Malformed UTF-8 character";
569 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
571 /* The order of malformation tests here is important. We should consume as
572 * few bytes as possible in order to not skip any valid character. This is
573 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
574 * http://unicode.org/reports/tr36 for more discussion as to why. For
575 * example, once we've done a UTF8SKIP, we can tell the expected number of
576 * bytes, and could fail right off the bat if the input parameters indicate
577 * that there are too few available. But it could be that just that first
578 * byte is garbled, and the intended character occupies fewer bytes. If we
579 * blindly assumed that the first byte is correct, and skipped based on
580 * that number, we could skip over a valid input character. So instead, we
581 * always examine the sequence byte-by-byte.
583 * We also should not consume too few bytes, otherwise someone could inject
584 * things. For example, an input could be deliberately designed to
585 * overflow, and if this code bailed out immediately upon discovering that,
586 * returning to the caller C<*retlen> pointing to the very next byte (one
587 * which is actually part of of the overflowing sequence), that could look
588 * legitimate to the caller, which could discard the initial partial
589 * sequence and process the rest, inappropriately */
591 /* Zero length strings, if allowed, of necessity are zero */
592 if (UNLIKELY(curlen == 0)) {
597 if (flags & UTF8_ALLOW_EMPTY) {
600 if (! (flags & UTF8_CHECK_ONLY)) {
601 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
606 expectlen = UTF8SKIP(s);
608 /* A well-formed UTF-8 character, as the vast majority of calls to this
609 * function will be for, has this expected length. For efficiency, set
610 * things up here to return it. It will be overriden only in those rare
611 * cases where a malformation is found */
616 /* An invariant is trivially well-formed */
617 if (UTF8_IS_INVARIANT(uv)) {
621 /* A continuation character can't start a valid sequence */
622 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
623 if (flags & UTF8_ALLOW_CONTINUATION) {
627 return UNICODE_REPLACEMENT;
630 if (! (flags & UTF8_CHECK_ONLY)) {
631 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
637 /* Here is not a continuation byte, nor an invariant. The only thing left
638 * is a start byte (possibly for an overlong) */
641 uv = NATIVE_UTF8_TO_I8(uv);
644 /* Remove the leading bits that indicate the number of bytes in the
645 * character's whole UTF-8 sequence, leaving just the bits that are part of
647 uv &= UTF_START_MASK(expectlen);
649 /* Now, loop through the remaining bytes in the character's sequence,
650 * accumulating each into the working value as we go. Be sure to not look
651 * past the end of the input string */
652 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
654 for (s = s0 + 1; s < send; s++) {
655 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
656 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
658 /* The original implementors viewed this malformation as more
659 * serious than the others (though I, khw, don't understand
660 * why, since other malformations also give very very wrong
661 * results), so there is no way to turn off checking for it.
662 * Set a flag, but keep going in the loop, so that we absorb
663 * the rest of the bytes that comprise the character. */
665 overflow_byte = *s; /* Save for warning message's use */
667 uv = UTF8_ACCUMULATE(uv, *s);
670 /* Here, found a non-continuation before processing all expected
671 * bytes. This byte begins a new character, so quit, even if
672 * allowing this malformation. */
673 unexpected_non_continuation = TRUE;
676 } /* End of loop through the character's bytes */
678 /* Save how many bytes were actually in the character */
681 /* The loop above finds two types of malformations: non-continuation and/or
682 * overflow. The non-continuation malformation is really a too-short
683 * malformation, as it means that the current character ended before it was
684 * expected to (being terminated prematurely by the beginning of the next
685 * character, whereas in the too-short malformation there just are too few
686 * bytes available to hold the character. In both cases, the check below
687 * that we have found the expected number of bytes would fail if executed.)
688 * Thus the non-continuation malformation is really unnecessary, being a
689 * subset of the too-short malformation. But there may be existing
690 * applications that are expecting the non-continuation type, so we retain
691 * it, and return it in preference to the too-short malformation. (If this
692 * code were being written from scratch, the two types might be collapsed
693 * into one.) I, khw, am also giving priority to returning the
694 * non-continuation and too-short malformations over overflow when multiple
695 * ones are present. I don't know of any real reason to prefer one over
696 * the other, except that it seems to me that multiple-byte errors trumps
697 * errors from a single byte */
698 if (UNLIKELY(unexpected_non_continuation)) {
699 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
700 if (! (flags & UTF8_CHECK_ONLY)) {
702 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
705 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));
710 uv = UNICODE_REPLACEMENT;
712 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
713 * as what the original expectations were. */
714 do_overlong_test = FALSE;
719 else if (UNLIKELY(curlen < expectlen)) {
720 if (! (flags & UTF8_ALLOW_SHORT)) {
721 if (! (flags & UTF8_CHECK_ONLY)) {
722 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));
726 uv = UNICODE_REPLACEMENT;
727 do_overlong_test = FALSE;
733 if (UNLIKELY(overflowed)) {
734 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
739 && expectlen > (STRLEN) OFFUNISKIP(uv)
740 && ! (flags & UTF8_ALLOW_LONG))
742 /* The overlong malformation has lower precedence than the others.
743 * Note that if this malformation is allowed, we return the actual
744 * value, instead of the replacement character. This is because this
745 * value is actually well-defined. */
746 if (! (flags & UTF8_CHECK_ONLY)) {
747 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));
752 /* Here, the input is considered to be well-formed, but it still could be a
753 * problematic code point that is not allowed by the input parameters. */
754 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
755 && ((flags & ( UTF8_DISALLOW_NONCHAR
756 |UTF8_DISALLOW_SURROGATE
758 |UTF8_DISALLOW_ABOVE_31_BIT
762 |UTF8_WARN_ABOVE_31_BIT))
763 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
764 && ckWARN_d(WARN_DEPRECATED))))
766 if (UNICODE_IS_SURROGATE(uv)) {
768 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
769 * generation of the sv, since no warnings are raised under CHECK */
770 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
771 && ckWARN_d(WARN_SURROGATE))
773 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
774 pack_warn = packWARN(WARN_SURROGATE);
776 if (flags & UTF8_DISALLOW_SURROGATE) {
780 else if ((uv > PERL_UNICODE_MAX)) {
781 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
782 && ckWARN_d(WARN_NON_UNICODE))
784 sv = sv_2mortal(Perl_newSVpvf(aTHX_
785 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
787 pack_warn = packWARN(WARN_NON_UNICODE);
790 /* The maximum code point ever specified by a standard was
791 * 2**31 - 1. Anything larger than that is a Perl extension that
792 * very well may not be understood by other applications (including
793 * earlier perl versions on EBCDIC platforms). On ASCII platforms,
794 * these code points are indicated by the first UTF-8 byte being
795 * 0xFE or 0xFF. We test for these after the regular SUPER ones,
796 * and before possibly bailing out, so that the slightly more dire
797 * warning will override the regular one. */
800 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */
802 /* The I8 for 2**31 (U+80000000) is
803 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
804 * and it turns out that on all EBCDIC pages recognized that
805 * the UTF-EBCDIC for that code point is
806 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
807 * For the next lower code point, the 1047 UTF-EBCDIC is
808 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
809 * The other code pages differ only in the bytes following
810 * \x42. Thus the following works (the minimum continuation
812 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41
820 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER
821 |UTF8_DISALLOW_ABOVE_31_BIT)))
823 if ( ! (flags & UTF8_CHECK_ONLY)
824 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
825 && ckWARN_d(WARN_UTF8))
827 sv = sv_2mortal(Perl_newSVpvf(aTHX_
828 "Code point 0x%"UVXf" is not Unicode, and not portable",
830 pack_warn = packWARN(WARN_UTF8);
832 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
837 if (flags & UTF8_DISALLOW_SUPER) {
841 /* The deprecated warning overrides any non-deprecated one */
842 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
844 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
845 uv, MAX_NON_DEPRECATED_CP));
846 pack_warn = packWARN(WARN_DEPRECATED);
849 else if (UNICODE_IS_NONCHAR(uv)) {
850 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
851 && ckWARN_d(WARN_NONCHAR))
853 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
854 pack_warn = packWARN(WARN_NONCHAR);
856 if (flags & UTF8_DISALLOW_NONCHAR) {
862 outlier_ret = uv; /* Note we don't bother to convert to native,
863 as all the outlier code points are the same
864 in both ASCII and EBCDIC */
868 /* Here, this is not considered a malformed character, so drop through
872 return UNI_TO_NATIVE(uv);
874 /* There are three cases which get to beyond this point. In all 3 cases:
875 * <sv> if not null points to a string to print as a warning.
876 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
878 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
879 * This is done by initializing it to 0, and changing it only
882 * 1) The input is valid but problematic, and to be warned about. The
883 * return value is the resultant code point; <*retlen> is set to
884 * <curlen>, the number of bytes that comprise the code point.
885 * <pack_warn> contains the result of packWARN() for the warning
886 * types. The entry point for this case is the label <do_warn>;
887 * 2) The input is a valid code point but disallowed by the parameters to
888 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
889 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
890 * comprise the code point. <pack_warn> contains the result of
891 * packWARN() for the warning types. The entry point for this case is
892 * the label <disallowed>.
893 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
894 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
895 * bytes that comprise the malformation. All such malformations are
896 * assumed to be warning type <utf8>. The entry point for this case
897 * is the label <malformed>.
902 if (sv && ckWARN_d(WARN_UTF8)) {
903 pack_warn = packWARN(WARN_UTF8);
908 if (flags & UTF8_CHECK_ONLY) {
910 *retlen = ((STRLEN) -1);
916 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
917 if warnings are to be raised. */
918 const char * const string = SvPVX_const(sv);
921 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
923 Perl_warner(aTHX_ pack_warn, "%s", string);
934 =for apidoc utf8_to_uvchr_buf
936 Returns the native code point of the first character in the string C<s> which
937 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
938 C<*retlen> will be set to the length, in bytes, of that character.
940 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
941 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
942 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
943 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
944 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
945 the next possible position in C<s> that could begin a non-malformed character.
946 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
949 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
950 unless those are turned off.
957 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
961 return utf8n_to_uvchr(s, send - s, retlen,
962 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
965 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
966 * there are no malformations in the input UTF-8 string C<s>. surrogates,
967 * non-character code points, and non-Unicode code points are allowed. */
970 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
972 UV expectlen = UTF8SKIP(s);
973 const U8* send = s + expectlen;
976 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
983 /* An invariant is trivially returned */
984 if (expectlen == 1) {
989 uv = NATIVE_UTF8_TO_I8(uv);
992 /* Remove the leading bits that indicate the number of bytes, leaving just
993 * the bits that are part of the value */
994 uv &= UTF_START_MASK(expectlen);
996 /* Now, loop through the remaining bytes, accumulating each into the
997 * working total as we go. (I khw tried unrolling the loop for up to 4
998 * bytes, but there was no performance improvement) */
999 for (++s; s < send; s++) {
1000 uv = UTF8_ACCUMULATE(uv, *s);
1003 return UNI_TO_NATIVE(uv);
1008 =for apidoc utf8_to_uvuni_buf
1010 Only in very rare circumstances should code need to be dealing in Unicode
1011 (as opposed to native) code points. In those few cases, use
1012 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1014 Returns the Unicode (not-native) code point of the first character in the
1016 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1017 C<retlen> will be set to the length, in bytes, of that character.
1019 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1020 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1021 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1022 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1023 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1024 next possible position in C<s> that could begin a non-malformed character.
1025 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1027 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1028 unless those are turned off.
1034 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1036 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1040 /* Call the low level routine asking for checks */
1041 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1042 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
1046 =for apidoc utf8_length
1048 Return the length of the UTF-8 char encoded string C<s> in characters.
1049 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1050 up past C<e>, croaks.
1056 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1060 PERL_ARGS_ASSERT_UTF8_LENGTH;
1062 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1063 * the bitops (especially ~) can create illegal UTF-8.
1064 * In other words: in Perl UTF-8 is not just for Unicode. */
1067 goto warn_and_return;
1077 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1078 "%s in %s", unees, OP_DESC(PL_op));
1080 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1087 =for apidoc utf8_distance
1089 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1092 WARNING: use only if you *know* that the pointers point inside the
1099 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1101 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1103 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1107 =for apidoc utf8_hop
1109 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1110 forward or backward.
1112 WARNING: do not use the following unless you *know* C<off> is within
1113 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1114 on the first byte of character or just after the last byte of a character.
1120 Perl_utf8_hop(const U8 *s, I32 off)
1122 PERL_ARGS_ASSERT_UTF8_HOP;
1124 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1125 * the bitops (especially ~) can create illegal UTF-8.
1126 * In other words: in Perl UTF-8 is not just for Unicode. */
1135 while (UTF8_IS_CONTINUATION(*s))
1143 =for apidoc bytes_cmp_utf8
1145 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1146 sequence of characters (stored as UTF-8)
1147 in C<u>, C<ulen>. Returns 0 if they are
1148 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1149 if the first string is greater than the second string.
1151 -1 or +1 is returned if the shorter string was identical to the start of the
1152 longer string. -2 or +2 is returned if
1153 there was a difference between characters
1160 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1162 const U8 *const bend = b + blen;
1163 const U8 *const uend = u + ulen;
1165 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1167 while (b < bend && u < uend) {
1169 if (!UTF8_IS_INVARIANT(c)) {
1170 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1173 if (UTF8_IS_CONTINUATION(c1)) {
1174 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1176 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1177 "Malformed UTF-8 character "
1178 "(unexpected non-continuation byte 0x%02x"
1179 ", immediately after start byte 0x%02x)"
1180 /* Dear diag.t, it's in the pod. */
1182 PL_op ? " in " : "",
1183 PL_op ? OP_DESC(PL_op) : "");
1188 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1189 "%s in %s", unees, OP_DESC(PL_op));
1191 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1192 return -2; /* Really want to return undef :-) */
1199 return *b < c ? -2 : +2;
1204 if (b == bend && u == uend)
1207 return b < bend ? +1 : -1;
1211 =for apidoc utf8_to_bytes
1213 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1214 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1215 updates C<len> to contain the new length.
1216 Returns zero on failure, setting C<len> to -1.
1218 If you need a copy of the string, see L</bytes_from_utf8>.
1224 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1226 U8 * const save = s;
1227 U8 * const send = s + *len;
1230 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1231 PERL_UNUSED_CONTEXT;
1233 /* ensure valid UTF-8 and chars < 256 before updating string */
1235 if (! UTF8_IS_INVARIANT(*s)) {
1236 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1237 *len = ((STRLEN) -1);
1248 if (! UTF8_IS_INVARIANT(c)) {
1249 /* Then it is two-byte encoded */
1250 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1261 =for apidoc bytes_from_utf8
1263 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1264 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1265 the newly-created string, and updates C<len> to contain the new
1266 length. Returns the original string if no conversion occurs, C<len>
1267 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1268 0 if C<s> is converted or consisted entirely of characters that are invariant
1269 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
1275 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1278 const U8 *start = s;
1282 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1283 PERL_UNUSED_CONTEXT;
1287 /* ensure valid UTF-8 and chars < 256 before converting string */
1288 for (send = s + *len; s < send;) {
1289 if (! UTF8_IS_INVARIANT(*s)) {
1290 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1301 Newx(d, (*len) - count + 1, U8);
1302 s = start; start = d;
1305 if (! UTF8_IS_INVARIANT(c)) {
1306 /* Then it is two-byte encoded */
1307 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1318 =for apidoc bytes_to_utf8
1320 Converts a string C<s> of length C<len> bytes from the native encoding into
1322 Returns a pointer to the newly-created string, and sets C<len> to
1323 reflect the new length in bytes.
1325 A C<NUL> character will be written after the end of the string.
1327 If you want to convert to UTF-8 from encodings other than
1328 the native (Latin1 or EBCDIC),
1329 see L</sv_recode_to_utf8>().
1334 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1335 likewise need duplication. */
1338 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1340 const U8 * const send = s + (*len);
1344 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1345 PERL_UNUSED_CONTEXT;
1347 Newx(d, (*len) * 2 + 1, U8);
1351 append_utf8_from_native_byte(*s, &d);
1360 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1362 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1363 * We optimize for native, for obvious reasons. */
1366 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1371 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1374 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1379 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1381 if (OFFUNI_IS_INVARIANT(uv)) {
1382 *d++ = LATIN1_TO_NATIVE((U8) uv);
1385 if (uv <= MAX_UTF8_TWO_BYTE) {
1386 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1387 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1390 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1391 #define LAST_HIGH_SURROGATE 0xDBFF
1392 #define FIRST_LOW_SURROGATE 0xDC00
1393 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1395 /* This assumes that most uses will be in the first Unicode plane, not
1396 * needing surrogates */
1397 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1398 && uv <= UNICODE_SURROGATE_LAST))
1400 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1401 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1404 UV low = (p[0] << 8) + p[1];
1405 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1406 || UNLIKELY(low > LAST_LOW_SURROGATE))
1408 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1411 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1412 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1416 d = uvoffuni_to_utf8_flags(d, uv, 0);
1419 *d++ = (U8)(( uv >> 12) | 0xe0);
1420 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1421 *d++ = (U8)(( uv & 0x3f) | 0x80);
1425 *d++ = (U8)(( uv >> 18) | 0xf0);
1426 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1427 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1428 *d++ = (U8)(( uv & 0x3f) | 0x80);
1433 *newlen = d - dstart;
1437 /* Note: this one is slightly destructive of the source. */
1440 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1443 U8* const send = s + bytelen;
1445 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1448 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1452 const U8 tmp = s[0];
1457 return utf16_to_utf8(p, d, bytelen, newlen);
1461 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1463 U8 tmpbuf[UTF8_MAXBYTES+1];
1464 uvchr_to_utf8(tmpbuf, c);
1465 return _is_utf8_FOO(classnum, tmpbuf);
1468 /* Internal function so we can deprecate the external one, and call
1469 this one from other deprecated functions in this file */
1472 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1474 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1478 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1482 Perl__is_uni_perl_idcont(pTHX_ UV c)
1484 U8 tmpbuf[UTF8_MAXBYTES+1];
1485 uvchr_to_utf8(tmpbuf, c);
1486 return _is_utf8_perl_idcont(tmpbuf);
1490 Perl__is_uni_perl_idstart(pTHX_ UV c)
1492 U8 tmpbuf[UTF8_MAXBYTES+1];
1493 uvchr_to_utf8(tmpbuf, c);
1494 return _is_utf8_perl_idstart(tmpbuf);
1498 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1500 /* We have the latin1-range values compiled into the core, so just use
1501 * those, converting the result to UTF-8. The only difference between upper
1502 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1503 * either "SS" or "Ss". Which one to use is passed into the routine in
1504 * 'S_or_s' to avoid a test */
1506 UV converted = toUPPER_LATIN1_MOD(c);
1508 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1510 assert(S_or_s == 'S' || S_or_s == 's');
1512 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1513 characters in this range */
1514 *p = (U8) converted;
1519 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1520 * which it maps to one of them, so as to only have to have one check for
1521 * it in the main case */
1522 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1524 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1525 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1528 converted = GREEK_CAPITAL_LETTER_MU;
1530 #if UNICODE_MAJOR_VERSION > 2 \
1531 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1532 && UNICODE_DOT_DOT_VERSION >= 8)
1533 case LATIN_SMALL_LETTER_SHARP_S:
1540 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1541 NOT_REACHED; /* NOTREACHED */
1545 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1546 *p = UTF8_TWO_BYTE_LO(converted);
1552 /* Call the function to convert a UTF-8 encoded character to the specified case.
1553 * Note that there may be more than one character in the result.
1554 * INP is a pointer to the first byte of the input character
1555 * OUTP will be set to the first byte of the string of changed characters. It
1556 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1557 * LENP will be set to the length in bytes of the string of changed characters
1559 * The functions return the ordinal of the first character in the string of OUTP */
1560 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1561 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1562 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1564 /* This additionally has the input parameter SPECIALS, which if non-zero will
1565 * cause this to use the SPECIALS hash for folding (meaning get full case
1566 * folding); otherwise, when zero, this implies a simple case fold */
1567 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1570 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1572 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1573 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1574 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1575 * the changed version may be longer than the original character.
1577 * The ordinal of the first character of the changed version is returned
1578 * (but note, as explained above, that there may be more.) */
1580 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1583 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1586 uvchr_to_utf8(p, c);
1587 return CALL_UPPER_CASE(p, p, lenp);
1591 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1593 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1596 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1599 uvchr_to_utf8(p, c);
1600 return CALL_TITLE_CASE(p, p, lenp);
1604 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
1606 /* We have the latin1-range values compiled into the core, so just use
1607 * those, converting the result to UTF-8. Since the result is always just
1608 * one character, we allow <p> to be NULL */
1610 U8 converted = toLOWER_LATIN1(c);
1613 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1618 /* Result is known to always be < 256, so can use the EIGHT_BIT
1620 *p = UTF8_EIGHT_BIT_HI(converted);
1621 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
1629 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1631 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1634 return to_lower_latin1((U8) c, p, lenp);
1637 uvchr_to_utf8(p, c);
1638 return CALL_LOWER_CASE(p, p, lenp);
1642 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1644 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1645 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1646 * FOLD_FLAGS_FULL iff full folding is to be used;
1648 * Not to be used for locale folds
1653 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1654 PERL_UNUSED_CONTEXT;
1656 assert (! (flags & FOLD_FLAGS_LOCALE));
1658 if (c == MICRO_SIGN) {
1659 converted = GREEK_SMALL_LETTER_MU;
1661 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1662 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1663 || UNICODE_DOT_DOT_VERSION > 0)
1664 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1666 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1667 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1668 * under those circumstances. */
1669 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1670 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1671 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1673 return LATIN_SMALL_LETTER_LONG_S;
1683 else { /* In this range the fold of all other characters is their lower
1685 converted = toLOWER_LATIN1(c);
1688 if (UVCHR_IS_INVARIANT(converted)) {
1689 *p = (U8) converted;
1693 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1694 *p = UTF8_TWO_BYTE_LO(converted);
1702 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1705 /* Not currently externally documented, and subject to change
1706 * <flags> bits meanings:
1707 * FOLD_FLAGS_FULL iff full folding is to be used;
1708 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1709 * locale are to be used.
1710 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1713 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1715 if (flags & FOLD_FLAGS_LOCALE) {
1716 /* Treat a UTF-8 locale as not being in locale at all */
1717 if (IN_UTF8_CTYPE_LOCALE) {
1718 flags &= ~FOLD_FLAGS_LOCALE;
1721 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1722 goto needs_full_generality;
1727 return _to_fold_latin1((U8) c, p, lenp,
1728 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1731 /* Here, above 255. If no special needs, just use the macro */
1732 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1733 uvchr_to_utf8(p, c);
1734 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1736 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1737 the special flags. */
1738 U8 utf8_c[UTF8_MAXBYTES + 1];
1740 needs_full_generality:
1741 uvchr_to_utf8(utf8_c, c);
1742 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1746 PERL_STATIC_INLINE bool
1747 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1748 const char *const swashname, SV* const invlist)
1750 /* returns a boolean giving whether or not the UTF8-encoded character that
1751 * starts at <p> is in the swash indicated by <swashname>. <swash>
1752 * contains a pointer to where the swash indicated by <swashname>
1753 * is to be stored; which this routine will do, so that future calls will
1754 * look at <*swash> and only generate a swash if it is not null. <invlist>
1755 * is NULL or an inversion list that defines the swash. If not null, it
1756 * saves time during initialization of the swash.
1758 * Note that it is assumed that the buffer length of <p> is enough to
1759 * contain all the bytes that comprise the character. Thus, <*p> should
1760 * have been checked before this call for mal-formedness enough to assure
1763 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1765 /* The API should have included a length for the UTF-8 character in <p>,
1766 * but it doesn't. We therefore assume that p has been validated at least
1767 * as far as there being enough bytes available in it to accommodate the
1768 * character without reading beyond the end, and pass that number on to the
1769 * validating routine */
1770 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1771 if (ckWARN_d(WARN_UTF8)) {
1772 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1773 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1774 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1775 what the malformation is */
1776 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1782 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1783 *swash = _core_swash_init("utf8",
1785 /* Only use the name if there is no inversion
1786 * list; otherwise will go out to disk */
1787 (invlist) ? "" : swashname,
1789 &PL_sv_undef, 1, 0, invlist, &flags);
1792 return swash_fetch(*swash, p, TRUE) != 0;
1796 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1798 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1800 assert(classnum < _FIRST_NON_SWASH_CC);
1802 return is_utf8_common(p,
1803 &PL_utf8_swash_ptrs[classnum],
1804 swash_property_names[classnum],
1805 PL_XPosix_ptrs[classnum]);
1809 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1813 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1815 if (! PL_utf8_perl_idstart) {
1816 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1818 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
1822 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1824 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1828 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1832 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1836 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1838 if (! PL_utf8_perl_idcont) {
1839 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1841 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
1845 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1847 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1849 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1853 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1855 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1857 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1861 Perl__is_utf8_mark(pTHX_ const U8 *p)
1863 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1865 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1869 =for apidoc to_utf8_case
1871 C<p> contains the pointer to the UTF-8 string encoding
1872 the character that is being converted. This routine assumes that the character
1873 at C<p> is well-formed.
1875 C<ustrp> is a pointer to the character buffer to put the
1876 conversion result to. C<lenp> is a pointer to the length
1879 C<swashp> is a pointer to the swash to use.
1881 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1882 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
1883 but not always, a multicharacter mapping), is tried first.
1885 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1886 any special mappings; C<""> means to use the special mappings. Values other
1887 than these two are treated as the name of the hash containing the special
1888 mappings, like C<"utf8::ToSpecLower">.
1890 C<normal> is a string like C<"ToLower"> which means the swash
1893 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1894 unless those are turned off.
1899 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1900 SV **swashp, const char *normal, const char *special)
1903 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1905 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1907 /* Note that swash_fetch() doesn't output warnings for these because it
1908 * assumes we will */
1909 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1910 if (uv1 <= UNICODE_SURROGATE_LAST) {
1911 if (ckWARN_d(WARN_SURROGATE)) {
1912 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1913 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1914 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1917 else if (UNICODE_IS_SUPER(uv1)) {
1918 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1919 && ckWARN_d(WARN_DEPRECATED))
1921 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1922 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1924 if (ckWARN_d(WARN_NON_UNICODE)) {
1925 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1926 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1927 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1931 /* Note that non-characters are perfectly legal, so no warning should
1935 if (!*swashp) /* load on-demand */
1936 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1939 /* It might be "special" (sometimes, but not always,
1940 * a multicharacter mapping) */
1944 /* If passed in the specials name, use that; otherwise use any
1945 * given in the swash */
1946 if (*special != '\0') {
1947 hv = get_hv(special, 0);
1950 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1952 hv = MUTABLE_HV(SvRV(*svp));
1957 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
1962 s = SvPV_const(*svp, len);
1965 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1967 Copy(s, ustrp, len, U8);
1972 if (!len && *swashp) {
1973 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
1976 /* It was "normal" (a single character mapping). */
1977 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1985 return valid_utf8_to_uvchr(ustrp, 0);
1988 /* Here, there was no mapping defined, which means that the code point maps
1989 * to itself. Return the inputs */
1991 if (p != ustrp) { /* Don't copy onto itself */
1992 Copy(p, ustrp, len, U8);
2003 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2005 /* This is called when changing the case of a UTF-8-encoded character above
2006 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2007 * result contains a character that crosses the 255/256 boundary, disallow
2008 * the change, and return the original code point. See L<perlfunc/lc> for
2011 * p points to the original string whose case was changed; assumed
2012 * by this routine to be well-formed
2013 * result the code point of the first character in the changed-case string
2014 * ustrp points to the changed-case string (<result> represents its first char)
2015 * lenp points to the length of <ustrp> */
2017 UV original; /* To store the first code point of <p> */
2019 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2021 assert(UTF8_IS_ABOVE_LATIN1(*p));
2023 /* We know immediately if the first character in the string crosses the
2024 * boundary, so can skip */
2027 /* Look at every character in the result; if any cross the
2028 * boundary, the whole thing is disallowed */
2029 U8* s = ustrp + UTF8SKIP(ustrp);
2030 U8* e = ustrp + *lenp;
2032 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2038 /* Here, no characters crossed, result is ok as-is, but we warn. */
2039 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
2045 /* Failed, have to return the original */
2046 original = valid_utf8_to_uvchr(p, lenp);
2048 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2049 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2050 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2051 "resolved to \"\\x{%"UVXf"}\".",
2055 Copy(p, ustrp, *lenp, char);
2060 =for apidoc to_utf8_upper
2062 Instead use L</toUPPER_utf8>.
2066 /* Not currently externally documented, and subject to change:
2067 * <flags> is set iff iff the rules from the current underlying locale are to
2071 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2075 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2078 /* Treat a UTF-8 locale as not being in locale at all */
2079 if (IN_UTF8_CTYPE_LOCALE) {
2083 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2087 if (UTF8_IS_INVARIANT(*p)) {
2089 result = toUPPER_LC(*p);
2092 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2095 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2097 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2098 result = toUPPER_LC(c);
2101 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2105 else { /* UTF-8, ord above 255 */
2106 result = CALL_UPPER_CASE(p, ustrp, lenp);
2109 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2114 /* Here, used locale rules. Convert back to UTF-8 */
2115 if (UTF8_IS_INVARIANT(result)) {
2116 *ustrp = (U8) result;
2120 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2121 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2129 =for apidoc to_utf8_title
2131 Instead use L</toTITLE_utf8>.
2135 /* Not currently externally documented, and subject to change:
2136 * <flags> is set iff the rules from the current underlying locale are to be
2137 * used. Since titlecase is not defined in POSIX, for other than a
2138 * UTF-8 locale, uppercase is used instead for code points < 256.
2142 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2146 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2149 /* Treat a UTF-8 locale as not being in locale at all */
2150 if (IN_UTF8_CTYPE_LOCALE) {
2154 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2158 if (UTF8_IS_INVARIANT(*p)) {
2160 result = toUPPER_LC(*p);
2163 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2166 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2168 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2169 result = toUPPER_LC(c);
2172 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2176 else { /* UTF-8, ord above 255 */
2177 result = CALL_TITLE_CASE(p, ustrp, lenp);
2180 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2185 /* Here, used locale rules. Convert back to UTF-8 */
2186 if (UTF8_IS_INVARIANT(result)) {
2187 *ustrp = (U8) result;
2191 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2192 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2200 =for apidoc to_utf8_lower
2202 Instead use L</toLOWER_utf8>.
2206 /* Not currently externally documented, and subject to change:
2207 * <flags> is set iff iff the rules from the current underlying locale are to
2212 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2216 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2219 /* Treat a UTF-8 locale as not being in locale at all */
2220 if (IN_UTF8_CTYPE_LOCALE) {
2224 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2228 if (UTF8_IS_INVARIANT(*p)) {
2230 result = toLOWER_LC(*p);
2233 return to_lower_latin1(*p, ustrp, lenp);
2236 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2238 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2239 result = toLOWER_LC(c);
2242 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2246 else { /* UTF-8, ord above 255 */
2247 result = CALL_LOWER_CASE(p, ustrp, lenp);
2250 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2256 /* Here, used locale rules. Convert back to UTF-8 */
2257 if (UTF8_IS_INVARIANT(result)) {
2258 *ustrp = (U8) result;
2262 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2263 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2271 =for apidoc to_utf8_fold
2273 Instead use L</toFOLD_utf8>.
2277 /* Not currently externally documented, and subject to change,
2279 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2280 * locale are to be used.
2281 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2282 * otherwise simple folds
2283 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2288 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2292 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2294 /* These are mutually exclusive */
2295 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2297 assert(p != ustrp); /* Otherwise overwrites */
2299 if (flags & FOLD_FLAGS_LOCALE) {
2300 /* Treat a UTF-8 locale as not being in locale at all */
2301 if (IN_UTF8_CTYPE_LOCALE) {
2302 flags &= ~FOLD_FLAGS_LOCALE;
2305 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2309 if (UTF8_IS_INVARIANT(*p)) {
2310 if (flags & FOLD_FLAGS_LOCALE) {
2311 result = toFOLD_LC(*p);
2314 return _to_fold_latin1(*p, ustrp, lenp,
2315 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2318 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2319 if (flags & FOLD_FLAGS_LOCALE) {
2320 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2321 result = toFOLD_LC(c);
2324 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2326 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2329 else { /* UTF-8, ord above 255 */
2330 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2332 if (flags & FOLD_FLAGS_LOCALE) {
2334 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2335 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2337 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2338 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2340 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2342 /* Special case these two characters, as what normally gets
2343 * returned under locale doesn't work */
2344 if (UTF8SKIP(p) == cap_sharp_s_len
2345 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
2347 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2348 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2349 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2350 "resolved to \"\\x{17F}\\x{17F}\".");
2355 if (UTF8SKIP(p) == long_s_t_len
2356 && memEQ((char *) p, LONG_S_T, long_s_t_len))
2358 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2359 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2360 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2361 "resolved to \"\\x{FB06}\".");
2362 goto return_ligature_st;
2365 #if UNICODE_MAJOR_VERSION == 3 \
2366 && UNICODE_DOT_VERSION == 0 \
2367 && UNICODE_DOT_DOT_VERSION == 1
2368 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2370 /* And special case this on this Unicode version only, for the same
2371 * reaons the other two are special cased. They would cross the
2372 * 255/256 boundary which is forbidden under /l, and so the code
2373 * wouldn't catch that they are equivalent (which they are only in
2375 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2376 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2378 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2379 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2380 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2381 "resolved to \"\\x{0131}\".");
2382 goto return_dotless_i;
2386 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2388 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2392 /* This is called when changing the case of a UTF-8-encoded
2393 * character above the ASCII range, and the result should not
2394 * contain an ASCII character. */
2396 UV original; /* To store the first code point of <p> */
2398 /* Look at every character in the result; if any cross the
2399 * boundary, the whole thing is disallowed */
2401 U8* e = ustrp + *lenp;
2404 /* Crossed, have to return the original */
2405 original = valid_utf8_to_uvchr(p, lenp);
2407 /* But in these instances, there is an alternative we can
2408 * return that is valid */
2409 if (original == LATIN_SMALL_LETTER_SHARP_S
2410 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2411 || original == LATIN_CAPITAL_LETTER_SHARP_S
2416 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2417 goto return_ligature_st;
2419 #if UNICODE_MAJOR_VERSION == 3 \
2420 && UNICODE_DOT_VERSION == 0 \
2421 && UNICODE_DOT_DOT_VERSION == 1
2423 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2424 goto return_dotless_i;
2427 Copy(p, ustrp, *lenp, char);
2433 /* Here, no characters crossed, result is ok as-is */
2438 /* Here, used locale rules. Convert back to UTF-8 */
2439 if (UTF8_IS_INVARIANT(result)) {
2440 *ustrp = (U8) result;
2444 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2445 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2452 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2453 * folds to a string of two of these characters. By returning this
2454 * instead, then, e.g.,
2455 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2458 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2459 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2461 return LATIN_SMALL_LETTER_LONG_S;
2464 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2465 * have the other one fold to it */
2467 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2468 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2469 return LATIN_SMALL_LIGATURE_ST;
2471 #if UNICODE_MAJOR_VERSION == 3 \
2472 && UNICODE_DOT_VERSION == 0 \
2473 && UNICODE_DOT_DOT_VERSION == 1
2476 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2477 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2478 return LATIN_SMALL_LETTER_DOTLESS_I;
2485 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2486 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2487 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2491 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2493 PERL_ARGS_ASSERT_SWASH_INIT;
2495 /* Returns a copy of a swash initiated by the called function. This is the
2496 * public interface, and returning a copy prevents others from doing
2497 * mischief on the original */
2499 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2503 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2506 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2507 * use the following define */
2509 #define CORE_SWASH_INIT_RETURN(x) \
2510 PL_curpm= old_PL_curpm; \
2513 /* Initialize and return a swash, creating it if necessary. It does this
2514 * by calling utf8_heavy.pl in the general case. The returned value may be
2515 * the swash's inversion list instead if the input parameters allow it.
2516 * Which is returned should be immaterial to callers, as the only
2517 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2518 * and swash_to_invlist() handle both these transparently.
2520 * This interface should only be used by functions that won't destroy or
2521 * adversely change the swash, as doing so affects all other uses of the
2522 * swash in the program; the general public should use 'Perl_swash_init'
2525 * pkg is the name of the package that <name> should be in.
2526 * name is the name of the swash to find. Typically it is a Unicode
2527 * property name, including user-defined ones
2528 * listsv is a string to initialize the swash with. It must be of the form
2529 * documented as the subroutine return value in
2530 * L<perlunicode/User-Defined Character Properties>
2531 * minbits is the number of bits required to represent each data element.
2532 * It is '1' for binary properties.
2533 * none I (khw) do not understand this one, but it is used only in tr///.
2534 * invlist is an inversion list to initialize the swash with (or NULL)
2535 * flags_p if non-NULL is the address of various input and output flag bits
2536 * to the routine, as follows: ('I' means is input to the routine;
2537 * 'O' means output from the routine. Only flags marked O are
2538 * meaningful on return.)
2539 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2540 * came from a user-defined property. (I O)
2541 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2542 * when the swash cannot be located, to simply return NULL. (I)
2543 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2544 * return of an inversion list instead of a swash hash if this routine
2545 * thinks that would result in faster execution of swash_fetch() later
2548 * Thus there are three possible inputs to find the swash: <name>,
2549 * <listsv>, and <invlist>. At least one must be specified. The result
2550 * will be the union of the specified ones, although <listsv>'s various
2551 * actions can intersect, etc. what <name> gives. To avoid going out to
2552 * disk at all, <invlist> should specify completely what the swash should
2553 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2555 * <invlist> is only valid for binary properties */
2557 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2559 SV* retval = &PL_sv_undef;
2560 HV* swash_hv = NULL;
2561 const int invlist_swash_boundary =
2562 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2563 ? 512 /* Based on some benchmarking, but not extensive, see commit
2565 : -1; /* Never return just an inversion list */
2567 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2568 assert(! invlist || minbits == 1);
2570 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2571 that triggered the swash init and the swash init perl logic itself.
2574 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2576 if (listsv != &PL_sv_undef || strNE(name, "")) {
2578 const size_t pkg_len = strlen(pkg);
2579 const size_t name_len = strlen(name);
2580 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2584 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2586 PUSHSTACKi(PERLSI_MAGIC);
2590 /* We might get here via a subroutine signature which uses a utf8
2591 * parameter name, at which point PL_subname will have been set
2592 * but not yet used. */
2593 save_item(PL_subname);
2594 if (PL_parser && PL_parser->error_count)
2595 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2596 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2597 if (!method) { /* demand load UTF-8 */
2599 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2600 GvSV(PL_errgv) = NULL;
2601 #ifndef NO_TAINT_SUPPORT
2602 /* It is assumed that callers of this routine are not passing in
2603 * any user derived data. */
2604 /* Need to do this after save_re_context() as it will set
2605 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2606 * in Perl_magic_get). Even line to create errsv_save can turn on
2608 SAVEBOOL(TAINT_get);
2611 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2614 /* Not ERRSV, as there is no need to vivify a scalar we are
2615 about to discard. */
2616 SV * const errsv = GvSV(PL_errgv);
2617 if (!SvTRUE(errsv)) {
2618 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2619 SvREFCNT_dec(errsv);
2627 mPUSHp(pkg, pkg_len);
2628 mPUSHp(name, name_len);
2633 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2634 GvSV(PL_errgv) = NULL;
2635 /* If we already have a pointer to the method, no need to use
2636 * call_method() to repeat the lookup. */
2638 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2639 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2641 retval = *PL_stack_sp--;
2642 SvREFCNT_inc(retval);
2645 /* Not ERRSV. See above. */
2646 SV * const errsv = GvSV(PL_errgv);
2647 if (!SvTRUE(errsv)) {
2648 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2649 SvREFCNT_dec(errsv);
2654 if (IN_PERL_COMPILETIME) {
2655 CopHINTS_set(PL_curcop, PL_hints);
2657 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2660 /* If caller wants to handle missing properties, let them */
2661 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2662 CORE_SWASH_INIT_RETURN(NULL);
2665 "Can't find Unicode property definition \"%"SVf"\"",
2667 NOT_REACHED; /* NOTREACHED */
2669 } /* End of calling the module to find the swash */
2671 /* If this operation fetched a swash, and we will need it later, get it */
2672 if (retval != &PL_sv_undef
2673 && (minbits == 1 || (flags_p
2675 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2677 swash_hv = MUTABLE_HV(SvRV(retval));
2679 /* If we don't already know that there is a user-defined component to
2680 * this swash, and the user has indicated they wish to know if there is
2681 * one (by passing <flags_p>), find out */
2682 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2683 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2684 if (user_defined && SvUV(*user_defined)) {
2685 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2690 /* Make sure there is an inversion list for binary properties */
2692 SV** swash_invlistsvp = NULL;
2693 SV* swash_invlist = NULL;
2694 bool invlist_in_swash_is_valid = FALSE;
2695 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2696 an unclaimed reference count */
2698 /* If this operation fetched a swash, get its already existing
2699 * inversion list, or create one for it */
2702 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2703 if (swash_invlistsvp) {
2704 swash_invlist = *swash_invlistsvp;
2705 invlist_in_swash_is_valid = TRUE;
2708 swash_invlist = _swash_to_invlist(retval);
2709 swash_invlist_unclaimed = TRUE;
2713 /* If an inversion list was passed in, have to include it */
2716 /* Any fetched swash will by now have an inversion list in it;
2717 * otherwise <swash_invlist> will be NULL, indicating that we
2718 * didn't fetch a swash */
2719 if (swash_invlist) {
2721 /* Add the passed-in inversion list, which invalidates the one
2722 * already stored in the swash */
2723 invlist_in_swash_is_valid = FALSE;
2724 _invlist_union(invlist, swash_invlist, &swash_invlist);
2728 /* Here, there is no swash already. Set up a minimal one, if
2729 * we are going to return a swash */
2730 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2732 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2734 swash_invlist = invlist;
2738 /* Here, we have computed the union of all the passed-in data. It may
2739 * be that there was an inversion list in the swash which didn't get
2740 * touched; otherwise save the computed one */
2741 if (! invlist_in_swash_is_valid
2742 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2744 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2746 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2748 /* We just stole a reference count. */
2749 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2750 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2753 SvREADONLY_on(swash_invlist);
2755 /* Use the inversion list stand-alone if small enough */
2756 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2757 SvREFCNT_dec(retval);
2758 if (!swash_invlist_unclaimed)
2759 SvREFCNT_inc_simple_void_NN(swash_invlist);
2760 retval = newRV_noinc(swash_invlist);
2764 CORE_SWASH_INIT_RETURN(retval);
2765 #undef CORE_SWASH_INIT_RETURN
2769 /* This API is wrong for special case conversions since we may need to
2770 * return several Unicode characters for a single Unicode character
2771 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2772 * the lower-level routine, and it is similarly broken for returning
2773 * multiple values. --jhi
2774 * For those, you should use to_utf8_case() instead */
2775 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2778 * Returns the value of property/mapping C<swash> for the first character
2779 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2780 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
2781 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2783 * A "swash" is a hash which contains initially the keys/values set up by
2784 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2785 * property for all possible code points. Things are stored in a compact form
2786 * (see utf8_heavy.pl) so that calculation is required to find the actual
2787 * property value for a given code point. As code points are looked up, new
2788 * key/value pairs are added to the hash, so that the calculation doesn't have
2789 * to ever be re-done. Further, each calculation is done, not just for the
2790 * desired one, but for a whole block of code points adjacent to that one.
2791 * For binary properties on ASCII machines, the block is usually for 64 code
2792 * points, starting with a code point evenly divisible by 64. Thus if the
2793 * property value for code point 257 is requested, the code goes out and
2794 * calculates the property values for all 64 code points between 256 and 319,
2795 * and stores these as a single 64-bit long bit vector, called a "swatch",
2796 * under the key for code point 256. The key is the UTF-8 encoding for code
2797 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2798 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2799 * for code point 258 is then requested, this code realizes that it would be
2800 * stored under the key for 256, and would find that value and extract the
2801 * relevant bit, offset from 256.
2803 * Non-binary properties are stored in as many bits as necessary to represent
2804 * their values (32 currently, though the code is more general than that), not
2805 * as single bits, but the principal is the same: the value for each key is a
2806 * vector that encompasses the property values for all code points whose UTF-8
2807 * representations are represented by the key. That is, for all code points
2808 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2812 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2814 HV *const hv = MUTABLE_HV(SvRV(swash));
2819 const U8 *tmps = NULL;
2823 PERL_ARGS_ASSERT_SWASH_FETCH;
2825 /* If it really isn't a hash, it isn't really swash; must be an inversion
2827 if (SvTYPE(hv) != SVt_PVHV) {
2828 return _invlist_contains_cp((SV*)hv,
2830 ? valid_utf8_to_uvchr(ptr, NULL)
2834 /* We store the values in a "swatch" which is a vec() value in a swash
2835 * hash. Code points 0-255 are a single vec() stored with key length
2836 * (klen) 0. All other code points have a UTF-8 representation
2837 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2838 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2839 * length for them is the length of the encoded char - 1. ptr[klen] is the
2840 * final byte in the sequence representing the character */
2841 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2846 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2849 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
2852 klen = UTF8SKIP(ptr) - 1;
2854 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2855 * the vec is the final byte in the sequence. (In EBCDIC this is
2856 * converted to I8 to get consecutive values.) To help you visualize
2858 * Straight 1047 After final byte
2859 * UTF-8 UTF-EBCDIC I8 transform
2860 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2861 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2863 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2864 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2866 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2867 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2869 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2870 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2872 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2873 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2875 * (There are no discontinuities in the elided (...) entries.)
2876 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2877 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2878 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2879 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2880 * index into the vec() swatch (after subtracting 0x80, which we
2881 * actually do with an '&').
2882 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2883 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2884 * dicontinuities which go away by transforming it into I8, and we
2885 * effectively subtract 0xA0 to get the index. */
2886 needents = (1 << UTF_ACCUMULATION_SHIFT);
2887 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2891 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
2892 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2893 * it's nothing to sniff at.) Pity we usually come through at least
2894 * two function calls to get here...
2896 * NB: this code assumes that swatches are never modified, once generated!
2899 if (hv == PL_last_swash_hv &&
2900 klen == PL_last_swash_klen &&
2901 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2903 tmps = PL_last_swash_tmps;
2904 slen = PL_last_swash_slen;
2907 /* Try our second-level swatch cache, kept in a hash. */
2908 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2910 /* If not cached, generate it via swatch_get */
2911 if (!svp || !SvPOK(*svp)
2912 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2915 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2916 swatch = swatch_get(swash,
2917 code_point & ~((UV)needents - 1),
2920 else { /* For the first 256 code points, the swatch has a key of
2922 swatch = swatch_get(swash, 0, needents);
2925 if (IN_PERL_COMPILETIME)
2926 CopHINTS_set(PL_curcop, PL_hints);
2928 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2930 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2931 || (slen << 3) < needents)
2932 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2933 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2934 svp, tmps, (UV)slen, (UV)needents);
2937 PL_last_swash_hv = hv;
2938 assert(klen <= sizeof(PL_last_swash_key));
2939 PL_last_swash_klen = (U8)klen;
2940 /* FIXME change interpvar.h? */
2941 PL_last_swash_tmps = (U8 *) tmps;
2942 PL_last_swash_slen = slen;
2944 Copy(ptr, PL_last_swash_key, klen, U8);
2947 switch ((int)((slen << 3) / needents)) {
2949 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2951 return ((UV) tmps[off]);
2955 ((UV) tmps[off ] << 8) +
2956 ((UV) tmps[off + 1]);
2960 ((UV) tmps[off ] << 24) +
2961 ((UV) tmps[off + 1] << 16) +
2962 ((UV) tmps[off + 2] << 8) +
2963 ((UV) tmps[off + 3]);
2965 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2966 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2967 NORETURN_FUNCTION_END;
2970 /* Read a single line of the main body of the swash input text. These are of
2973 * where each number is hex. The first two numbers form the minimum and
2974 * maximum of a range, and the third is the value associated with the range.
2975 * Not all swashes should have a third number
2977 * On input: l points to the beginning of the line to be examined; it points
2978 * to somewhere in the string of the whole input text, and is
2979 * terminated by a \n or the null string terminator.
2980 * lend points to the null terminator of that string
2981 * wants_value is non-zero if the swash expects a third number
2982 * typestr is the name of the swash's mapping, like 'ToLower'
2983 * On output: *min, *max, and *val are set to the values read from the line.
2984 * returns a pointer just beyond the line examined. If there was no
2985 * valid min number on the line, returns lend+1
2989 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2990 const bool wants_value, const U8* const typestr)
2992 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2993 STRLEN numlen; /* Length of the number */
2994 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2995 | PERL_SCAN_DISALLOW_PREFIX
2996 | PERL_SCAN_SILENT_NON_PORTABLE;
2998 /* nl points to the next \n in the scan */
2999 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3001 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3003 /* Get the first number on the line: the range minimum */
3005 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3006 *max = *min; /* So can never return without setting max */
3007 if (numlen) /* If found a hex number, position past it */
3009 else if (nl) { /* Else, go handle next line, if any */
3010 return nl + 1; /* 1 is length of "\n" */
3012 else { /* Else, no next line */
3013 return lend + 1; /* to LIST's end at which \n is not found */
3016 /* The max range value follows, separated by a BLANK */
3019 flags = PERL_SCAN_SILENT_ILLDIGIT
3020 | PERL_SCAN_DISALLOW_PREFIX
3021 | PERL_SCAN_SILENT_NON_PORTABLE;
3023 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3026 else /* If no value here, it is a single element range */
3029 /* Non-binary tables have a third entry: what the first element of the
3030 * range maps to. The map for those currently read here is in hex */
3034 flags = PERL_SCAN_SILENT_ILLDIGIT
3035 | PERL_SCAN_DISALLOW_PREFIX
3036 | PERL_SCAN_SILENT_NON_PORTABLE;
3038 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3047 /* diag_listed_as: To%s: illegal mapping '%s' */
3048 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3054 *val = 0; /* bits == 1, then any val should be ignored */
3056 else { /* Nothing following range min, should be single element with no
3061 /* diag_listed_as: To%s: illegal mapping '%s' */
3062 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3066 *val = 0; /* bits == 1, then val should be ignored */
3069 /* Position to next line if any, or EOF */
3079 * Returns a swatch (a bit vector string) for a code point sequence
3080 * that starts from the value C<start> and comprises the number C<span>.
3081 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3082 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3085 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3088 U8 *l, *lend, *x, *xend, *s, *send;
3089 STRLEN lcur, xcur, scur;
3090 HV *const hv = MUTABLE_HV(SvRV(swash));
3091 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3093 SV** listsvp = NULL; /* The string containing the main body of the table */
3094 SV** extssvp = NULL;
3095 SV** invert_it_svp = NULL;
3098 STRLEN octets; /* if bits == 1, then octets == 0 */
3100 UV end = start + span;
3102 if (invlistsvp == NULL) {
3103 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3104 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3105 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3106 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3107 listsvp = hv_fetchs(hv, "LIST", FALSE);
3108 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3110 bits = SvUV(*bitssvp);
3111 none = SvUV(*nonesvp);
3112 typestr = (U8*)SvPV_nolen(*typesvp);
3118 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3120 PERL_ARGS_ASSERT_SWATCH_GET;
3122 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3123 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3127 /* If overflowed, use the max possible */
3133 /* create and initialize $swatch */
3134 scur = octets ? (span * octets) : (span + 7) / 8;
3135 swatch = newSV(scur);
3137 s = (U8*)SvPVX(swatch);
3138 if (octets && none) {
3139 const U8* const e = s + scur;
3142 *s++ = (U8)(none & 0xff);
3143 else if (bits == 16) {
3144 *s++ = (U8)((none >> 8) & 0xff);
3145 *s++ = (U8)( none & 0xff);
3147 else if (bits == 32) {
3148 *s++ = (U8)((none >> 24) & 0xff);
3149 *s++ = (U8)((none >> 16) & 0xff);
3150 *s++ = (U8)((none >> 8) & 0xff);
3151 *s++ = (U8)( none & 0xff);
3157 (void)memzero((U8*)s, scur + 1);
3159 SvCUR_set(swatch, scur);
3160 s = (U8*)SvPVX(swatch);
3162 if (invlistsvp) { /* If has an inversion list set up use that */
3163 _invlist_populate_swatch(*invlistsvp, start, end, s);
3167 /* read $swash->{LIST} */
3168 l = (U8*)SvPV(*listsvp, lcur);
3171 UV min, max, val, upper;
3172 l = swash_scan_list_line(l, lend, &min, &max, &val,
3173 cBOOL(octets), typestr);
3178 /* If looking for something beyond this range, go try the next one */
3182 /* <end> is generally 1 beyond where we want to set things, but at the
3183 * platform's infinity, where we can't go any higher, we want to
3184 * include the code point at <end> */
3187 : (max != UV_MAX || end != UV_MAX)
3194 if (!none || val < none) {
3199 for (key = min; key <= upper; key++) {
3201 /* offset must be non-negative (start <= min <= key < end) */
3202 offset = octets * (key - start);
3204 s[offset] = (U8)(val & 0xff);
3205 else if (bits == 16) {
3206 s[offset ] = (U8)((val >> 8) & 0xff);
3207 s[offset + 1] = (U8)( val & 0xff);
3209 else if (bits == 32) {
3210 s[offset ] = (U8)((val >> 24) & 0xff);
3211 s[offset + 1] = (U8)((val >> 16) & 0xff);
3212 s[offset + 2] = (U8)((val >> 8) & 0xff);
3213 s[offset + 3] = (U8)( val & 0xff);
3216 if (!none || val < none)
3220 else { /* bits == 1, then val should be ignored */
3225 for (key = min; key <= upper; key++) {
3226 const STRLEN offset = (STRLEN)(key - start);
3227 s[offset >> 3] |= 1 << (offset & 7);
3232 /* Invert if the data says it should be. Assumes that bits == 1 */
3233 if (invert_it_svp && SvUV(*invert_it_svp)) {
3235 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3236 * be 0, and their inversion should also be 0, as we don't succeed any
3237 * Unicode property matches for non-Unicode code points */
3238 if (start <= PERL_UNICODE_MAX) {
3240 /* The code below assumes that we never cross the
3241 * Unicode/above-Unicode boundary in a range, as otherwise we would
3242 * have to figure out where to stop flipping the bits. Since this
3243 * boundary is divisible by a large power of 2, and swatches comes
3244 * in small powers of 2, this should be a valid assumption */
3245 assert(start + span - 1 <= PERL_UNICODE_MAX);
3255 /* read $swash->{EXTRAS}
3256 * This code also copied to swash_to_invlist() below */
3257 x = (U8*)SvPV(*extssvp, xcur);
3265 SV **otherbitssvp, *other;
3269 const U8 opc = *x++;
3273 nl = (U8*)memchr(x, '\n', xend - x);
3275 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3277 x = nl + 1; /* 1 is length of "\n" */
3281 x = xend; /* to EXTRAS' end at which \n is not found */
3288 namelen = nl - namestr;
3292 namelen = xend - namestr;
3296 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3297 otherhv = MUTABLE_HV(SvRV(*othersvp));
3298 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3299 otherbits = (STRLEN)SvUV(*otherbitssvp);
3300 if (bits < otherbits)
3301 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3302 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3304 /* The "other" swatch must be destroyed after. */
3305 other = swatch_get(*othersvp, start, span);
3306 o = (U8*)SvPV(other, olen);
3309 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3311 s = (U8*)SvPV(swatch, slen);
3312 if (bits == 1 && otherbits == 1) {
3314 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3315 "mismatch, slen=%"UVuf", olen=%"UVuf,
3316 (UV)slen, (UV)olen);
3340 STRLEN otheroctets = otherbits >> 3;
3342 U8* const send = s + slen;
3347 if (otherbits == 1) {
3348 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3352 STRLEN vlen = otheroctets;
3360 if (opc == '+' && otherval)
3361 NOOP; /* replace with otherval */
3362 else if (opc == '!' && !otherval)
3364 else if (opc == '-' && otherval)
3366 else if (opc == '&' && !otherval)
3369 s += octets; /* no replacement */
3374 *s++ = (U8)( otherval & 0xff);
3375 else if (bits == 16) {
3376 *s++ = (U8)((otherval >> 8) & 0xff);
3377 *s++ = (U8)( otherval & 0xff);
3379 else if (bits == 32) {
3380 *s++ = (U8)((otherval >> 24) & 0xff);
3381 *s++ = (U8)((otherval >> 16) & 0xff);
3382 *s++ = (U8)((otherval >> 8) & 0xff);
3383 *s++ = (U8)( otherval & 0xff);
3387 sv_free(other); /* through with it! */
3393 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3396 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3397 * Can't be used on a property that is subject to user override, as it
3398 * relies on the value of SPECIALS in the swash which would be set by
3399 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3400 * for overridden properties
3402 * Returns a hash which is the inversion and closure of a swash mapping.
3403 * For example, consider the input lines:
3408 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
3409 * 006C. The value for each key is an array. For 006C, the array would
3410 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3411 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
3413 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3414 * keys are only code points that are folded-to, so it isn't a full closure.
3416 * Essentially, for any code point, it gives all the code points that map to
3417 * it, or the list of 'froms' for that point.
3419 * Currently it ignores any additions or deletions from other swashes,
3420 * looking at just the main body of the swash, and if there are SPECIALS
3421 * in the swash, at that hash
3423 * The specials hash can be extra code points, and most likely consists of
3424 * maps from single code points to multiple ones (each expressed as a string
3425 * of UTF-8 characters). This function currently returns only 1-1 mappings.
3426 * However consider this possible input in the specials hash:
3427 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3428 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3430 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3431 * currently handle. But it also means that FB05 and FB06 are equivalent in
3432 * a 1-1 mapping which we should handle, and this relationship may not be in
3433 * the main table. Therefore this function examines all the multi-char
3434 * sequences and adds the 1-1 mappings that come out of that.
3436 * XXX This function was originally intended to be multipurpose, but its
3437 * only use is quite likely to remain for constructing the inversion of
3438 * the CaseFolding (//i) property. If it were more general purpose for
3439 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3440 * because certain folds are prohibited under /iaa and /il. As an example,
3441 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3442 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3443 * prohibited, so we would not figure out that they fold to each other.
3444 * Code could be written to automatically figure this out, similar to the
3445 * code that does this for multi-character folds, but this is the only case
3446 * where something like this is ever likely to happen, as all the single
3447 * char folds to the 0-255 range are now quite settled. Instead there is a
3448 * little special code that is compiled only for this Unicode version. This
3449 * is smaller and didn't require much coding time to do. But this makes
3450 * this routine strongly tied to being used just for CaseFolding. If ever
3451 * it should be generalized, this would have to be fixed */
3455 HV *const hv = MUTABLE_HV(SvRV(swash));
3457 /* The string containing the main body of the table. This will have its
3458 * assertion fail if the swash has been converted to its inversion list */
3459 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3461 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3462 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3463 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3464 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3465 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3466 const STRLEN bits = SvUV(*bitssvp);
3467 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3468 const UV none = SvUV(*nonesvp);
3469 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3473 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3475 /* Must have at least 8 bits to get the mappings */
3476 if (bits != 8 && bits != 16 && bits != 32) {
3477 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3481 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3482 mapping to more than one character */
3484 /* Construct an inverse mapping hash for the specials */
3485 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3486 HV * specials_inverse = newHV();
3487 char *char_from; /* the lhs of the map */
3488 I32 from_len; /* its byte length */
3489 char *char_to; /* the rhs of the map */
3490 I32 to_len; /* its byte length */
3491 SV *sv_to; /* and in a sv */
3492 AV* from_list; /* list of things that map to each 'to' */
3494 hv_iterinit(specials_hv);
3496 /* The keys are the characters (in UTF-8) that map to the corresponding
3497 * UTF-8 string value. Iterate through the list creating the inverse
3499 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3501 if (! SvPOK(sv_to)) {
3502 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3503 "unexpectedly is not a string, flags=%lu",
3504 (unsigned long)SvFLAGS(sv_to));
3506 /*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)));*/
3508 /* Each key in the inverse list is a mapped-to value, and the key's
3509 * hash value is a list of the strings (each in UTF-8) that map to
3510 * it. Those strings are all one character long */
3511 if ((listp = hv_fetch(specials_inverse,
3515 from_list = (AV*) *listp;
3517 else { /* No entry yet for it: create one */
3518 from_list = newAV();
3519 if (! hv_store(specials_inverse,
3522 (SV*) from_list, 0))
3524 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3528 /* Here have the list associated with this 'to' (perhaps newly
3529 * created and empty). Just add to it. Note that we ASSUME that
3530 * the input is guaranteed to not have duplications, so we don't
3531 * check for that. Duplications just slow down execution time. */
3532 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3535 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3536 * it looking for cases like the FB05/FB06 examples above. There would
3537 * be an entry in the hash like
3538 * 'st' => [ FB05, FB06 ]
3539 * In this example we will create two lists that get stored in the
3540 * returned hash, 'ret':
3541 * FB05 => [ FB05, FB06 ]
3542 * FB06 => [ FB05, FB06 ]
3544 * Note that there is nothing to do if the array only has one element.
3545 * (In the normal 1-1 case handled below, we don't have to worry about
3546 * two lists, as everything gets tied to the single list that is
3547 * generated for the single character 'to'. But here, we are omitting
3548 * that list, ('st' in the example), so must have multiple lists.) */
3549 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3550 &char_to, &to_len)))
3552 if (av_tindex(from_list) > 0) {
3555 /* We iterate over all combinations of i,j to place each code
3556 * point on each list */
3557 for (i = 0; i <= av_tindex(from_list); i++) {
3559 AV* i_list = newAV();
3560 SV** entryp = av_fetch(from_list, i, FALSE);
3561 if (entryp == NULL) {
3562 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3564 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3565 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3567 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3568 (SV*) i_list, FALSE))
3570 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3573 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3574 for (j = 0; j <= av_tindex(from_list); j++) {
3575 entryp = av_fetch(from_list, j, FALSE);
3576 if (entryp == NULL) {
3577 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3580 /* When i==j this adds itself to the list */
3581 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3582 (U8*) SvPVX(*entryp),
3583 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3585 /*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));*/
3590 SvREFCNT_dec(specials_inverse); /* done with it */
3591 } /* End of specials */
3593 /* read $swash->{LIST} */
3595 #if UNICODE_MAJOR_VERSION == 3 \
3596 && UNICODE_DOT_VERSION == 0 \
3597 && UNICODE_DOT_DOT_VERSION == 1
3599 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3600 * rule so that things work under /iaa and /il */
3602 SV * mod_listsv = sv_mortalcopy(*listsvp);
3603 sv_catpv(mod_listsv, "130\t130\t131\n");
3604 l = (U8*)SvPV(mod_listsv, lcur);
3608 l = (U8*)SvPV(*listsvp, lcur);
3614 /* Go through each input line */
3618 l = swash_scan_list_line(l, lend, &min, &max, &val,
3619 cBOOL(octets), typestr);
3624 /* Each element in the range is to be inverted */
3625 for (inverse = min; inverse <= max; inverse++) {
3629 bool found_key = FALSE;
3630 bool found_inverse = FALSE;
3632 /* The key is the inverse mapping */
3633 char key[UTF8_MAXBYTES+1];
3634 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3635 STRLEN key_len = key_end - key;
3637 /* Get the list for the map */
3638 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3639 list = (AV*) *listp;
3641 else { /* No entry yet for it: create one */
3643 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3644 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3648 /* Look through list to see if this inverse mapping already is
3649 * listed, or if there is a mapping to itself already */
3650 for (i = 0; i <= av_tindex(list); i++) {
3651 SV** entryp = av_fetch(list, i, FALSE);
3654 if (entryp == NULL) {
3655 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3659 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3663 if (uv == inverse) {
3664 found_inverse = TRUE;
3667 /* No need to continue searching if found everything we are
3669 if (found_key && found_inverse) {
3674 /* Make sure there is a mapping to itself on the list */
3676 av_push(list, newSVuv(val));
3677 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3681 /* Simply add the value to the list */
3682 if (! found_inverse) {
3683 av_push(list, newSVuv(inverse));
3684 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3687 /* swatch_get() increments the value of val for each element in the
3688 * range. That makes more compact tables possible. You can
3689 * express the capitalization, for example, of all consecutive
3690 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3691 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3692 * and it's not documented; it appears to be used only in
3693 * implementing tr//; I copied the semantics from swatch_get(), just
3695 if (!none || val < none) {
3705 Perl__swash_to_invlist(pTHX_ SV* const swash)
3708 /* Subject to change or removal. For use only in one place in regcomp.c.
3709 * Ownership is given to one reference count in the returned SV* */
3714 HV *const hv = MUTABLE_HV(SvRV(swash));
3715 UV elements = 0; /* Number of elements in the inversion list */
3725 STRLEN octets; /* if bits == 1, then octets == 0 */
3731 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3733 /* If not a hash, it must be the swash's inversion list instead */
3734 if (SvTYPE(hv) != SVt_PVHV) {
3735 return SvREFCNT_inc_simple_NN((SV*) hv);
3738 /* The string containing the main body of the table */
3739 listsvp = hv_fetchs(hv, "LIST", FALSE);
3740 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3741 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3742 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3743 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3745 typestr = (U8*)SvPV_nolen(*typesvp);
3746 bits = SvUV(*bitssvp);
3747 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3749 /* read $swash->{LIST} */
3750 if (SvPOK(*listsvp)) {
3751 l = (U8*)SvPV(*listsvp, lcur);
3754 /* LIST legitimately doesn't contain a string during compilation phases
3755 * of Perl itself, before the Unicode tables are generated. In this
3756 * case, just fake things up by creating an empty list */
3763 if (*l == 'V') { /* Inversion list format */
3764 const char *after_atou = (char *) lend;
3766 UV* other_elements_ptr;
3768 /* The first number is a count of the rest */
3770 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
3771 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
3773 if (elements == 0) {
3774 invlist = _new_invlist(0);
3777 while (isSPACE(*l)) l++;
3778 l = (U8 *) after_atou;
3780 /* Get the 0th element, which is needed to setup the inversion list */
3781 while (isSPACE(*l)) l++;
3782 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
3783 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
3785 l = (U8 *) after_atou;
3786 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3789 /* Then just populate the rest of the input */
3790 while (elements-- > 0) {
3792 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3794 while (isSPACE(*l)) l++;
3795 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
3796 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
3798 l = (U8 *) after_atou;
3804 /* Scan the input to count the number of lines to preallocate array
3805 * size based on worst possible case, which is each line in the input
3806 * creates 2 elements in the inversion list: 1) the beginning of a
3807 * range in the list; 2) the beginning of a range not in the list. */
3808 while ((loc = (strchr(loc, '\n'))) != NULL) {
3813 /* If the ending is somehow corrupt and isn't a new line, add another
3814 * element for the final range that isn't in the inversion list */
3815 if (! (*lend == '\n'
3816 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3821 invlist = _new_invlist(elements);
3823 /* Now go through the input again, adding each range to the list */
3826 UV val; /* Not used by this function */
3828 l = swash_scan_list_line(l, lend, &start, &end, &val,
3829 cBOOL(octets), typestr);
3835 invlist = _add_range_to_invlist(invlist, start, end);
3839 /* Invert if the data says it should be */
3840 if (invert_it_svp && SvUV(*invert_it_svp)) {
3841 _invlist_invert(invlist);
3844 /* This code is copied from swatch_get()
3845 * read $swash->{EXTRAS} */
3846 x = (U8*)SvPV(*extssvp, xcur);
3854 SV **otherbitssvp, *other;
3857 const U8 opc = *x++;
3861 nl = (U8*)memchr(x, '\n', xend - x);
3863 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3865 x = nl + 1; /* 1 is length of "\n" */
3869 x = xend; /* to EXTRAS' end at which \n is not found */
3876 namelen = nl - namestr;
3880 namelen = xend - namestr;
3884 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3885 otherhv = MUTABLE_HV(SvRV(*othersvp));
3886 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3887 otherbits = (STRLEN)SvUV(*otherbitssvp);
3889 if (bits != otherbits || bits != 1) {
3890 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3891 "properties, bits=%"UVuf", otherbits=%"UVuf,
3892 (UV)bits, (UV)otherbits);
3895 /* The "other" swatch must be destroyed after. */
3896 other = _swash_to_invlist((SV *)*othersvp);
3898 /* End of code copied from swatch_get() */
3901 _invlist_union(invlist, other, &invlist);
3904 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3907 _invlist_subtract(invlist, other, &invlist);
3910 _invlist_intersection(invlist, other, &invlist);
3915 sv_free(other); /* through with it! */
3918 SvREADONLY_on(invlist);
3923 Perl__get_swash_invlist(pTHX_ SV* const swash)
3927 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3929 if (! SvROK(swash)) {
3933 /* If it really isn't a hash, it isn't really swash; must be an inversion
3935 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3939 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3948 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3950 /* May change: warns if surrogates, non-character code points, or
3951 * non-Unicode code points are in s which has length len bytes. Returns
3952 * TRUE if none found; FALSE otherwise. The only other validity check is
3953 * to make sure that this won't exceed the string's length.
3955 * Code points above the platform's C<IV_MAX> will raise a deprecation
3956 * warning, unless those are turned off. */
3958 const U8* const e = s + len;
3961 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3964 if (UTF8SKIP(s) > len) {
3965 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3966 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3969 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
3971 if (UTF8_IS_SUPER(s, e)) {
3972 if ( ckWARN_d(WARN_NON_UNICODE)
3973 || ( ckWARN_d(WARN_DEPRECATED)
3974 #if defined(UV_IS_QUAD)
3975 /* 2**63 and up meet these conditions provided we have
3978 && *s == 0xFE && e - s >= UTF8_MAXBYTES
3981 && *s == 0xFF && e -s >= UTF8_MAXBYTES
3984 #else /* Below is 32-bit words */
3985 /* 2**31 and above meet these conditions on all EBCDIC
3986 * pages recognized for 32-bit platforms */
3988 && *s == 0xFE && e - s >= UTF8_MAXBYTES
3995 /* A side effect of this function will be to warn */
3996 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_SUPER);
4000 else if (UTF8_IS_SURROGATE(s, e)) {
4001 if (ckWARN_d(WARN_SURROGATE)) {
4002 /* This has a different warning than the one the called
4003 * function would output, so can't just call it, unlike we
4004 * do for the non-chars and above-unicodes */
4005 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4006 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4007 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
4011 else if ((UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
4012 /* A side effect of this function will be to warn */
4013 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_NONCHAR);
4024 =for apidoc pv_uni_display
4026 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4027 length C<len>, the displayable version being at most C<pvlim> bytes long
4028 (if longer, the rest is truncated and C<"..."> will be appended).
4030 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4031 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4032 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4033 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4034 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4035 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4037 The pointer to the PV of the C<dsv> is returned.
4039 See also L</sv_uni_display>.
4043 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4048 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4052 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4054 /* This serves double duty as a flag and a character to print after
4055 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4059 if (pvlim && SvCUR(dsv) >= pvlim) {
4063 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4065 const unsigned char c = (unsigned char)u & 0xFF;
4066 if (flags & UNI_DISPLAY_BACKSLASH) {
4083 const char string = ok;
4084 sv_catpvs(dsv, "\\");
4085 sv_catpvn(dsv, &string, 1);
4088 /* isPRINT() is the locale-blind version. */
4089 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4090 const char string = c;
4091 sv_catpvn(dsv, &string, 1);
4096 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4099 sv_catpvs(dsv, "...");
4105 =for apidoc sv_uni_display
4107 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4108 the displayable version being at most C<pvlim> bytes long
4109 (if longer, the rest is truncated and "..." will be appended).
4111 The C<flags> argument is as in L</pv_uni_display>().
4113 The pointer to the PV of the C<dsv> is returned.
4118 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4120 const char * const ptr =
4121 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4123 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4125 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4126 SvCUR(ssv), pvlim, flags);
4130 =for apidoc foldEQ_utf8
4132 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4133 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4134 How far into the strings to compare is determined by other input parameters.
4136 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4137 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4138 with respect to C<s2>.
4140 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4141 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4142 scan will not be considered to be a match unless the goal is reached, and
4143 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4146 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
4147 considered an end pointer to the position 1 byte past the maximum point
4148 in C<s1> beyond which scanning will not continue under any circumstances.
4149 (This routine assumes that UTF-8 encoded input strings are not malformed;
4150 malformed input can cause it to read past C<pe1>).
4151 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4152 is less than C<s1>+C<l1>, the match will never be successful because it can
4154 get as far as its goal (and in fact is asserted against). Correspondingly for
4155 C<pe2> with respect to C<s2>.
4157 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4158 C<l2> must be non-zero), and if both do, both have to be
4159 reached for a successful match. Also, if the fold of a character is multiple
4160 characters, all of them must be matched (see tr21 reference below for
4163 Upon a successful match, if C<pe1> is non-C<NULL>,
4164 it will be set to point to the beginning of the I<next> character of C<s1>
4165 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4167 For case-insensitiveness, the "casefolding" of Unicode is used
4168 instead of upper/lowercasing both the characters, see
4169 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4173 /* A flags parameter has been added which may change, and hence isn't
4174 * externally documented. Currently it is:
4175 * 0 for as-documented above
4176 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4177 ASCII one, to not match
4178 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4179 * locale are to be used.
4180 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4181 * routine. This allows that step to be skipped.
4182 * Currently, this requires s1 to be encoded as UTF-8
4183 * (u1 must be true), which is asserted for.
4184 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4185 * cross certain boundaries. Hence, the caller should
4186 * let this function do the folding instead of
4187 * pre-folding. This code contains an assertion to
4188 * that effect. However, if the caller knows what
4189 * it's doing, it can pass this flag to indicate that,
4190 * and the assertion is skipped.
4191 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4192 * FOLDEQ_S2_FOLDS_SANE
4195 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)
4197 const U8 *p1 = (const U8*)s1; /* Point to current char */
4198 const U8 *p2 = (const U8*)s2;
4199 const U8 *g1 = NULL; /* goal for s1 */
4200 const U8 *g2 = NULL;
4201 const U8 *e1 = NULL; /* Don't scan s1 past this */
4202 U8 *f1 = NULL; /* Point to current folded */
4203 const U8 *e2 = NULL;
4205 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4206 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4207 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4208 U8 flags_for_folder = FOLD_FLAGS_FULL;
4210 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4212 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4213 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
4214 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4215 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
4216 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4217 /* The algorithm is to trial the folds without regard to the flags on
4218 * the first line of the above assert(), and then see if the result
4219 * violates them. This means that the inputs can't be pre-folded to a
4220 * violating result, hence the assert. This could be changed, with the
4221 * addition of extra tests here for the already-folded case, which would
4222 * slow it down. That cost is more than any possible gain for when these
4223 * flags are specified, as the flags indicate /il or /iaa matching which
4224 * is less common than /iu, and I (khw) also believe that real-world /il
4225 * and /iaa matches are most likely to involve code points 0-255, and this
4226 * function only under rare conditions gets called for 0-255. */
4228 if (flags & FOLDEQ_LOCALE) {
4229 if (IN_UTF8_CTYPE_LOCALE) {
4230 flags &= ~FOLDEQ_LOCALE;
4233 flags_for_folder |= FOLD_FLAGS_LOCALE;
4242 g1 = (const U8*)s1 + l1;
4250 g2 = (const U8*)s2 + l2;
4253 /* Must have at least one goal */
4258 /* Will never match if goal is out-of-bounds */
4259 assert(! e1 || e1 >= g1);
4261 /* Here, there isn't an end pointer, or it is beyond the goal. We
4262 * only go as far as the goal */
4266 assert(e1); /* Must have an end for looking at s1 */
4269 /* Same for goal for s2 */
4271 assert(! e2 || e2 >= g2);
4278 /* If both operands are already folded, we could just do a memEQ on the
4279 * whole strings at once, but it would be better if the caller realized
4280 * this and didn't even call us */
4282 /* Look through both strings, a character at a time */
4283 while (p1 < e1 && p2 < e2) {
4285 /* If at the beginning of a new character in s1, get its fold to use
4286 * and the length of the fold. */
4288 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4294 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4296 /* We have to forbid mixing ASCII with non-ASCII if the
4297 * flags so indicate. And, we can short circuit having to
4298 * call the general functions for this common ASCII case,
4299 * all of whose non-locale folds are also ASCII, and hence
4300 * UTF-8 invariants, so the UTF8ness of the strings is not
4302 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4306 *foldbuf1 = toFOLD(*p1);
4309 _to_utf8_fold_flags(p1, foldbuf1, &n1, flags_for_folder);
4311 else { /* Not UTF-8, get UTF-8 fold */
4312 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4318 if (n2 == 0) { /* Same for s2 */
4319 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4325 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4326 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4330 *foldbuf2 = toFOLD(*p2);
4333 _to_utf8_fold_flags(p2, foldbuf2, &n2, flags_for_folder);
4336 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4342 /* Here f1 and f2 point to the beginning of the strings to compare.
4343 * These strings are the folds of the next character from each input
4344 * string, stored in UTF-8. */
4346 /* While there is more to look for in both folds, see if they
4347 * continue to match */
4349 U8 fold_length = UTF8SKIP(f1);
4350 if (fold_length != UTF8SKIP(f2)
4351 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4352 function call for single
4354 || memNE((char*)f1, (char*)f2, fold_length))
4356 return 0; /* mismatch */
4359 /* Here, they matched, advance past them */
4366 /* When reach the end of any fold, advance the input past it */
4368 p1 += u1 ? UTF8SKIP(p1) : 1;
4371 p2 += u2 ? UTF8SKIP(p2) : 1;
4373 } /* End of loop through both strings */
4375 /* A match is defined by each scan that specified an explicit length
4376 * reaching its final goal, and the other not having matched a partial
4377 * character (which can happen when the fold of a character is more than one
4379 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4383 /* Successful match. Set output pointers */
4393 /* XXX The next two functions should likely be moved to mathoms.c once all
4394 * occurrences of them are removed from the core; some cpan-upstream modules
4398 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4400 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4402 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4406 =for apidoc utf8n_to_uvuni
4408 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4410 This function was useful for code that wanted to handle both EBCDIC and
4411 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4412 distinctions between the platforms have mostly been made invisible to most
4413 code, so this function is quite unlikely to be what you want. If you do need
4414 this precise functionality, use instead
4415 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4416 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4422 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4424 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4426 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4430 =for apidoc uvuni_to_utf8_flags
4432 Instead you almost certainly want to use L</uvchr_to_utf8> or
4433 L</uvchr_to_utf8_flags>.
4435 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4436 which itself, while not deprecated, should be used only in isolated
4437 circumstances. These functions were useful for code that wanted to handle
4438 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4439 v5.20, the distinctions between the platforms have mostly been made invisible
4440 to most code, so this function is quite unlikely to be what you want.
4446 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4448 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4450 return uvoffuni_to_utf8_flags(d, uv, flags);
4454 * ex: set ts=8 sts=4 sw=4 et: