3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "inline_invlist.c"
37 /* Separate prototypes needed because in ASCII systems these are
38 * usually macros but they still are compiled as code, too. */
39 PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
40 PERL_CALLCONV UV Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen);
41 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
44 static const char unees[] =
45 "Malformed UTF-8 character (unexpected end of string)";
48 =head1 Unicode Support
50 This file contains various utility functions for manipulating UTF8-encoded
51 strings. For the uninitiated, this is a method of representing arbitrary
52 Unicode characters as a variable number of bytes, in such a way that
53 characters in the ASCII range are unmodified, and a zero byte never appears
54 within non-zero characters.
60 =for apidoc is_ascii_string
62 Returns true if the first C<len> bytes of the string C<s> are the same whether
63 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
64 is, if they are invariant. On ASCII-ish machines, only ASCII characters
65 fit this definition, hence the function's name.
67 If C<len> is 0, it will be calculated using C<strlen(s)>.
69 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
75 Perl_is_ascii_string(const U8 *s, STRLEN len)
77 const U8* const send = s + (len ? len : strlen((const char *)s));
80 PERL_ARGS_ASSERT_IS_ASCII_STRING;
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
91 =for apidoc uvuni_to_utf8_flags
93 Adds the UTF-8 representation of the code point C<uv> to the end
94 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
95 bytes available. The return value is the pointer to the byte after the
96 end of the new character. In other words,
98 d = uvuni_to_utf8_flags(d, uv, flags);
102 d = uvuni_to_utf8(d, uv);
104 (which is equivalent to)
106 d = uvuni_to_utf8_flags(d, uv, 0);
108 This is the recommended Unicode-aware way of saying
112 This function will convert to UTF-8 (and not warn) even code points that aren't
113 legal Unicode or are problematic, unless C<flags> contains one or more of the
116 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
117 the function will raise a warning, provided UTF8 warnings are enabled. If instead
118 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
119 If both flags are set, the function will both warn and return NULL.
121 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
122 affect how the function handles a Unicode non-character. And, likewise for the
123 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
124 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
125 even less portable) can be warned and/or disallowed even if other above-Unicode
126 code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
129 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
130 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
138 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
140 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
142 /* The first problematic code point is the first surrogate */
143 if (uv >= UNICODE_SURROGATE_FIRST
144 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
146 if (UNICODE_IS_SURROGATE(uv)) {
147 if (flags & UNICODE_WARN_SURROGATE) {
148 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
149 "UTF-16 surrogate U+%04"UVXf, uv);
151 if (flags & UNICODE_DISALLOW_SURROGATE) {
155 else if (UNICODE_IS_SUPER(uv)) {
156 if (flags & UNICODE_WARN_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
159 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
160 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
162 if (flags & UNICODE_DISALLOW_SUPER
163 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
168 else if (UNICODE_IS_NONCHAR(uv)) {
169 if (flags & UNICODE_WARN_NONCHAR) {
170 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
171 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
174 if (flags & UNICODE_DISALLOW_NONCHAR) {
179 if (UNI_IS_INVARIANT(uv)) {
180 *d++ = (U8)UTF_TO_NATIVE(uv);
185 STRLEN len = UNISKIP(uv);
188 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
189 uv >>= UTF_ACCUMULATION_SHIFT;
191 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
194 #else /* Non loop style */
196 *d++ = (U8)(( uv >> 6) | 0xc0);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
201 *d++ = (U8)(( uv >> 12) | 0xe0);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
207 *d++ = (U8)(( uv >> 18) | 0xf0);
208 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
209 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
210 *d++ = (U8)(( uv & 0x3f) | 0x80);
213 if (uv < 0x4000000) {
214 *d++ = (U8)(( uv >> 24) | 0xf8);
215 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
218 *d++ = (U8)(( uv & 0x3f) | 0x80);
221 if (uv < 0x80000000) {
222 *d++ = (U8)(( uv >> 30) | 0xfc);
223 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
227 *d++ = (U8)(( uv & 0x3f) | 0x80);
231 if (uv < UTF8_QUAD_MAX)
234 *d++ = 0xfe; /* Can't match U+FEFF! */
235 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
239 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
240 *d++ = (U8)(( uv & 0x3f) | 0x80);
245 *d++ = 0xff; /* Can't match U+FFFE! */
246 *d++ = 0x80; /* 6 Reserved bits */
247 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
248 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
257 *d++ = (U8)(( uv & 0x3f) | 0x80);
261 #endif /* Loop style */
266 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
267 character. Note that an INVARIANT (i.e. ASCII) character is a valid
268 UTF-8 character. The number of bytes in the UTF-8 character
269 will be returned if it is valid, otherwise 0.
271 This is the "slow" version as opposed to the "fast" version which is
272 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
273 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
274 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
275 you should use the _slow(). In practice this means that the _slow()
276 will be used very rarely, since the maximum Unicode code point (as of
277 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
278 the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
283 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
285 dTHX; /* The function called below requires thread context */
289 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
291 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
293 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
297 =for apidoc is_utf8_char_buf
299 Returns the number of bytes that comprise the first UTF-8 encoded character in
300 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
301 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
304 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
305 machines) is a valid UTF-8 character.
310 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
315 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
317 if (buf_end <= buf) {
322 if (len > UTF8SKIP(buf)) {
327 if (IS_UTF8_CHAR_FAST(len))
328 return IS_UTF8_CHAR(buf, len) ? len : 0;
329 #endif /* #ifdef IS_UTF8_CHAR */
330 return is_utf8_char_slow(buf, len);
334 =for apidoc is_utf8_char
338 Tests if some arbitrary number of bytes begins in a valid UTF-8
339 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
340 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
341 character will be returned if it is valid, otherwise 0.
343 This function is deprecated due to the possibility that malformed input could
344 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
350 Perl_is_utf8_char(const U8 *s)
352 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
354 /* Assumes we have enough space, which is why this is deprecated */
355 return is_utf8_char_buf(s, s + UTF8SKIP(s));
360 =for apidoc is_utf8_string
362 Returns true if the first C<len> bytes of string C<s> form a valid
363 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
364 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
365 terminating NUL byte). Note that all characters being ASCII constitute 'a
368 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
374 Perl_is_utf8_string(const U8 *s, STRLEN len)
376 const U8* const send = s + (len ? len : strlen((const char *)s));
379 PERL_ARGS_ASSERT_IS_UTF8_STRING;
382 /* Inline the easy bits of is_utf8_char() here for speed... */
383 if (UTF8_IS_INVARIANT(*x)) {
386 else if (!UTF8_IS_START(*x))
389 /* ... and call is_utf8_char() only if really needed. */
390 const STRLEN c = UTF8SKIP(x);
391 const U8* const next_char_ptr = x + c;
393 if (next_char_ptr > send) {
397 if (IS_UTF8_CHAR_FAST(c)) {
398 if (!IS_UTF8_CHAR(x, c))
401 else if (! is_utf8_char_slow(x, c)) {
412 Implemented as a macro in utf8.h
414 =for apidoc is_utf8_string_loc
416 Like L</is_utf8_string> but stores the location of the failure (in the
417 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
418 "utf8ness success") in the C<ep>.
420 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
422 =for apidoc is_utf8_string_loclen
424 Like L</is_utf8_string>() but stores the location of the failure (in the
425 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
426 "utf8ness success") in the C<ep>, and the number of UTF-8
427 encoded characters in the C<el>.
429 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
435 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
437 const U8* const send = s + (len ? len : strlen((const char *)s));
442 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
445 const U8* next_char_ptr;
447 /* Inline the easy bits of is_utf8_char() here for speed... */
448 if (UTF8_IS_INVARIANT(*x))
449 next_char_ptr = x + 1;
450 else if (!UTF8_IS_START(*x))
453 /* ... and call is_utf8_char() only if really needed. */
455 next_char_ptr = c + x;
456 if (next_char_ptr > send) {
459 if (IS_UTF8_CHAR_FAST(c)) {
460 if (!IS_UTF8_CHAR(x, c))
463 c = is_utf8_char_slow(x, c);
482 =for apidoc utf8n_to_uvuni
484 Bottom level UTF-8 decode routine.
485 Returns the code point value of the first character in the string C<s>,
486 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
487 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
488 the length, in bytes, of that character.
490 The value of C<flags> determines the behavior when C<s> does not point to a
491 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
492 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
493 next possible position in C<s> that could begin a non-malformed character.
494 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
496 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
497 individual types of malformations, such as the sequence being overlong (that
498 is, when there is a shorter sequence that can express the same code point;
499 overlong sequences are expressly forbidden in the UTF-8 standard due to
500 potential security issues). Another malformation example is the first byte of
501 a character not being a legal first byte. See F<utf8.h> for the list of such
502 flags. For allowed 0 length strings, this function returns 0; for allowed
503 overlong sequences, the computed code point is returned; for all other allowed
504 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
505 determinable reasonable value.
507 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
508 flags) malformation is found. If this flag is set, the routine assumes that
509 the caller will raise a warning, and this function will silently just set
510 C<retlen> to C<-1> and return zero.
512 Certain code points are considered problematic. These are Unicode surrogates,
513 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
514 By default these are considered regular code points, but certain situations
515 warrant special handling for them. If C<flags> contains
516 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
517 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
518 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
519 maximum) can be set to disallow these categories individually.
521 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
522 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
523 for their respective categories, but otherwise the code points are considered
524 valid (not malformations). To get a category to both be treated as a
525 malformation and raise a warning, specify both the WARN and DISALLOW flags.
526 (But note that warnings are not raised if lexically disabled nor if
527 UTF8_CHECK_ONLY is also specified.)
529 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
530 the others that are above the Unicode legal maximum. There are several
531 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
532 not representable at all on EBCDIC platforms, and the original UTF-8
533 specification never went above this number (the current 0x10FFFF limit was
534 imposed later). (The smaller ones, those that fit into 32 bits, are
535 representable by a UV on ASCII platforms, but not by an IV, which means that
536 the number of operations that can be performed on them is quite restricted.)
537 The UTF-8 encoding on ASCII platforms for these large code points begins with a
538 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
539 be treated as malformations, while allowing smaller above-Unicode code points.
540 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
541 including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
542 the other WARN flags, but applies just to these code points.
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
548 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
554 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
557 const U8 * const s0 = s;
558 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
563 UV outlier_ret = 0; /* return value when input is in error or problematic
565 UV pack_warn = 0; /* Save result of packWARN() for later */
566 bool unexpected_non_continuation = FALSE;
567 bool overflowed = FALSE;
568 bool do_overlong_test = TRUE; /* May have to skip this test */
570 const char* const malformed_text = "Malformed UTF-8 character";
572 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
574 /* The order of malformation tests here is important. We should consume as
575 * few bytes as possible in order to not skip any valid character. This is
576 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
577 * http://unicode.org/reports/tr36 for more discussion as to why. For
578 * example, once we've done a UTF8SKIP, we can tell the expected number of
579 * bytes, and could fail right off the bat if the input parameters indicate
580 * that there are too few available. But it could be that just that first
581 * byte is garbled, and the intended character occupies fewer bytes. If we
582 * blindly assumed that the first byte is correct, and skipped based on
583 * that number, we could skip over a valid input character. So instead, we
584 * always examine the sequence byte-by-byte.
586 * We also should not consume too few bytes, otherwise someone could inject
587 * things. For example, an input could be deliberately designed to
588 * overflow, and if this code bailed out immediately upon discovering that,
589 * returning to the caller *retlen pointing to the very next byte (one
590 * which is actually part of of the overflowing sequence), that could look
591 * legitimate to the caller, which could discard the initial partial
592 * sequence and process the rest, inappropriately */
594 /* Zero length strings, if allowed, of necessity are zero */
595 if (UNLIKELY(curlen == 0)) {
600 if (flags & UTF8_ALLOW_EMPTY) {
603 if (! (flags & UTF8_CHECK_ONLY)) {
604 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
609 expectlen = UTF8SKIP(s);
611 /* A well-formed UTF-8 character, as the vast majority of calls to this
612 * function will be for, has this expected length. For efficiency, set
613 * things up here to return it. It will be overriden only in those rare
614 * cases where a malformation is found */
619 /* An invariant is trivially well-formed */
620 if (UTF8_IS_INVARIANT(uv)) {
621 return (UV) (NATIVE_TO_UTF(*s));
624 /* A continuation character can't start a valid sequence */
625 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
626 if (flags & UTF8_ALLOW_CONTINUATION) {
630 return UNICODE_REPLACEMENT;
633 if (! (flags & UTF8_CHECK_ONLY)) {
634 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
641 uv = NATIVE_TO_UTF(uv);
644 /* Here is not a continuation byte, nor an invariant. The only thing left
645 * is a start byte (possibly for an overlong) */
647 /* Remove the leading bits that indicate the number of bytes in the
648 * character's whole UTF-8 sequence, leaving just the bits that are part of
650 uv &= UTF_START_MASK(expectlen);
652 /* Now, loop through the remaining bytes in the character's sequence,
653 * accumulating each into the working value as we go. Be sure to not look
654 * past the end of the input string */
655 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
657 for (s = s0 + 1; s < send; s++) {
658 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
659 #ifndef EBCDIC /* Can't overflow in EBCDIC */
660 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
662 /* The original implementors viewed this malformation as more
663 * serious than the others (though I, khw, don't understand
664 * why, since other malformations also give very very wrong
665 * results), so there is no way to turn off checking for it.
666 * Set a flag, but keep going in the loop, so that we absorb
667 * the rest of the bytes that comprise the character. */
669 overflow_byte = *s; /* Save for warning message's use */
672 uv = UTF8_ACCUMULATE(uv, *s);
675 /* Here, found a non-continuation before processing all expected
676 * bytes. This byte begins a new character, so quit, even if
677 * allowing this malformation. */
678 unexpected_non_continuation = TRUE;
681 } /* End of loop through the character's bytes */
683 /* Save how many bytes were actually in the character */
686 /* The loop above finds two types of malformations: non-continuation and/or
687 * overflow. The non-continuation malformation is really a too-short
688 * malformation, as it means that the current character ended before it was
689 * expected to (being terminated prematurely by the beginning of the next
690 * character, whereas in the too-short malformation there just are too few
691 * bytes available to hold the character. In both cases, the check below
692 * that we have found the expected number of bytes would fail if executed.)
693 * Thus the non-continuation malformation is really unnecessary, being a
694 * subset of the too-short malformation. But there may be existing
695 * applications that are expecting the non-continuation type, so we retain
696 * it, and return it in preference to the too-short malformation. (If this
697 * code were being written from scratch, the two types might be collapsed
698 * into one.) I, khw, am also giving priority to returning the
699 * non-continuation and too-short malformations over overflow when multiple
700 * ones are present. I don't know of any real reason to prefer one over
701 * the other, except that it seems to me that multiple-byte errors trumps
702 * errors from a single byte */
703 if (UNLIKELY(unexpected_non_continuation)) {
704 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
705 if (! (flags & UTF8_CHECK_ONLY)) {
707 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
710 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));
715 uv = UNICODE_REPLACEMENT;
717 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
718 * as what the original expectations were. */
719 do_overlong_test = FALSE;
724 else if (UNLIKELY(curlen < expectlen)) {
725 if (! (flags & UTF8_ALLOW_SHORT)) {
726 if (! (flags & UTF8_CHECK_ONLY)) {
727 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));
731 uv = UNICODE_REPLACEMENT;
732 do_overlong_test = FALSE;
738 #ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
742 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
743 * generation of the sv, since no warnings are raised under CHECK */
744 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
745 && ckWARN_d(WARN_UTF8))
747 /* This message is deliberately not of the same syntax as the other
748 * messages for malformations, for backwards compatibility in the
749 * unlikely event that code is relying on its precise earlier text
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
752 pack_warn = packWARN(WARN_UTF8);
754 if (flags & UTF8_DISALLOW_FE_FF) {
758 if (UNLIKELY(overflowed)) {
760 /* If the first byte is FF, it will overflow a 32-bit word. If the
761 * first byte is FE, it will overflow a signed 32-bit word. The
762 * above preserves backward compatibility, since its message was used
763 * in earlier versions of this code in preference to overflow */
764 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
770 && expectlen > (STRLEN)UNISKIP(uv)
771 && ! (flags & UTF8_ALLOW_LONG))
773 /* The overlong malformation has lower precedence than the others.
774 * Note that if this malformation is allowed, we return the actual
775 * value, instead of the replacement character. This is because this
776 * value is actually well-defined. */
777 if (! (flags & UTF8_CHECK_ONLY)) {
778 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), *s0));
783 /* Here, the input is considered to be well-formed , but could be a
784 * problematic code point that is not allowed by the input parameters. */
785 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
786 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
787 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
789 if (UNICODE_IS_SURROGATE(uv)) {
790 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
791 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
793 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
794 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
796 if (flags & UTF8_DISALLOW_SURROGATE) {
800 else if ((uv > PERL_UNICODE_MAX)) {
801 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
802 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
804 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
805 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
807 if (flags & UTF8_DISALLOW_SUPER) {
811 else if (UNICODE_IS_NONCHAR(uv)) {
812 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
813 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
815 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
816 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
818 if (flags & UTF8_DISALLOW_NONCHAR) {
828 /* Here, this is not considered a malformed character, so drop through
834 /* There are three cases which get to beyond this point. In all 3 cases:
835 * <sv> if not null points to a string to print as a warning.
836 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
838 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
839 * This is done by initializing it to 0, and changing it only
842 * 1) The input is valid but problematic, and to be warned about. The
843 * return value is the resultant code point; <*retlen> is set to
844 * <curlen>, the number of bytes that comprise the code point.
845 * <pack_warn> contains the result of packWARN() for the warning
846 * types. The entry point for this case is the label <do_warn>;
847 * 2) The input is a valid code point but disallowed by the parameters to
848 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
849 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
850 * comprise the code point. <pack_warn> contains the result of
851 * packWARN() for the warning types. The entry point for this case is
852 * the label <disallowed>.
853 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
854 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
855 * bytes that comprise the malformation. All such malformations are
856 * assumed to be warning type <utf8>. The entry point for this case
857 * is the label <malformed>.
862 if (sv && ckWARN_d(WARN_UTF8)) {
863 pack_warn = packWARN(WARN_UTF8);
868 if (flags & UTF8_CHECK_ONLY) {
870 *retlen = ((STRLEN) -1);
876 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
877 if warnings are to be raised. */
878 const char * const string = SvPVX_const(sv);
881 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
883 Perl_warner(aTHX_ pack_warn, "%s", string);
894 =for apidoc utf8_to_uvchr_buf
896 Returns the native code point of the first character in the string C<s> which
897 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
898 C<*retlen> will be set to the length, in bytes, of that character.
900 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
901 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
902 NULL) to -1. If those warnings are off, the computed value if well-defined (or
903 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
904 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
905 next possible position in C<s> that could begin a non-malformed character.
906 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
913 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
915 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
919 return utf8n_to_uvchr(s, send - s, retlen,
920 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
923 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
924 * there are no malformations in the input UTF-8 string C<s>. surrogates,
925 * non-character code points, and non-Unicode code points are allowed. A macro
926 * in utf8.h is used to normally avoid this function wrapper */
929 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
931 const UV uv = valid_utf8_to_uvuni(s, retlen);
933 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
935 return UNI_TO_NATIVE(uv);
939 =for apidoc utf8_to_uvchr
943 Returns the native code point of the first character in the string C<s>
944 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
945 length, in bytes, of that character.
947 Some, but not all, UTF-8 malformations are detected, and in fact, some
948 malformed input could cause reading beyond the end of the input buffer, which
949 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
951 If C<s> points to one of the detected malformations, and UTF8 warnings are
952 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
953 NULL) to -1. If those warnings are off, the computed value if well-defined (or
954 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
955 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
956 next possible position in C<s> that could begin a non-malformed character.
957 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
963 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
965 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
967 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
971 =for apidoc utf8_to_uvuni_buf
973 Returns the Unicode code point of the first character in the string C<s> which
974 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
975 C<retlen> will be set to the length, in bytes, of that character.
977 This function should only be used when the returned UV is considered
978 an index into the Unicode semantic tables (e.g. swashes).
980 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
981 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
982 NULL) to -1. If those warnings are off, the computed value if well-defined (or
983 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
984 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
985 next possible position in C<s> that could begin a non-malformed character.
986 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
992 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
994 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
998 /* Call the low level routine asking for checks */
999 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1000 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1003 /* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1004 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1005 * non-character code points, and non-Unicode code points are allowed */
1008 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1010 UV expectlen = UTF8SKIP(s);
1011 const U8* send = s + expectlen;
1012 UV uv = NATIVE_TO_UTF(*s);
1014 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1017 *retlen = expectlen;
1020 /* An invariant is trivially returned */
1021 if (expectlen == 1) {
1025 /* Remove the leading bits that indicate the number of bytes, leaving just
1026 * the bits that are part of the value */
1027 uv &= UTF_START_MASK(expectlen);
1029 /* Now, loop through the remaining bytes, accumulating each into the
1030 * working total as we go. (I khw tried unrolling the loop for up to 4
1031 * bytes, but there was no performance improvement) */
1032 for (++s; s < send; s++) {
1033 uv = UTF8_ACCUMULATE(uv, *s);
1040 =for apidoc utf8_to_uvuni
1044 Returns the Unicode code point of the first character in the string C<s>
1045 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1046 length, in bytes, of that character.
1048 This function should only be used when the returned UV is considered
1049 an index into the Unicode semantic tables (e.g. swashes).
1051 Some, but not all, UTF-8 malformations are detected, and in fact, some
1052 malformed input could cause reading beyond the end of the input buffer, which
1053 is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
1055 If C<s> points to one of the detected malformations, and UTF8 warnings are
1056 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1057 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1058 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1059 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1060 next possible position in C<s> that could begin a non-malformed character.
1061 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
1067 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1069 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1071 return valid_utf8_to_uvuni(s, retlen);
1075 =for apidoc utf8_length
1077 Return the length of the UTF-8 char encoded string C<s> in characters.
1078 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1079 up past C<e>, croaks.
1085 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1090 PERL_ARGS_ASSERT_UTF8_LENGTH;
1092 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1093 * the bitops (especially ~) can create illegal UTF-8.
1094 * In other words: in Perl UTF-8 is not just for Unicode. */
1097 goto warn_and_return;
1107 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1108 "%s in %s", unees, OP_DESC(PL_op));
1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1117 =for apidoc utf8_distance
1119 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1122 WARNING: use only if you *know* that the pointers point inside the
1129 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1131 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1133 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1137 =for apidoc utf8_hop
1139 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1140 forward or backward.
1142 WARNING: do not use the following unless you *know* C<off> is within
1143 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1144 on the first byte of character or just after the last byte of a character.
1150 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1152 PERL_ARGS_ASSERT_UTF8_HOP;
1154 PERL_UNUSED_CONTEXT;
1155 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1156 * the bitops (especially ~) can create illegal UTF-8.
1157 * In other words: in Perl UTF-8 is not just for Unicode. */
1166 while (UTF8_IS_CONTINUATION(*s))
1174 =for apidoc bytes_cmp_utf8
1176 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1177 sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
1178 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1179 if the first string is greater than the second string.
1181 -1 or +1 is returned if the shorter string was identical to the start of the
1182 longer string. -2 or +2 is returned if the was a difference between characters
1189 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1191 const U8 *const bend = b + blen;
1192 const U8 *const uend = u + ulen;
1194 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1196 PERL_UNUSED_CONTEXT;
1198 while (b < bend && u < uend) {
1200 if (!UTF8_IS_INVARIANT(c)) {
1201 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1204 if (UTF8_IS_CONTINUATION(c1)) {
1205 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
1207 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1208 "Malformed UTF-8 character "
1209 "(unexpected non-continuation byte 0x%02x"
1210 ", immediately after start byte 0x%02x)"
1211 /* Dear diag.t, it's in the pod. */
1213 PL_op ? " in " : "",
1214 PL_op ? OP_DESC(PL_op) : "");
1219 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1220 "%s in %s", unees, OP_DESC(PL_op));
1222 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1223 return -2; /* Really want to return undef :-) */
1230 return *b < c ? -2 : +2;
1235 if (b == bend && u == uend)
1238 return b < bend ? +1 : -1;
1242 =for apidoc utf8_to_bytes
1244 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1245 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1246 updates C<len> to contain the new length.
1247 Returns zero on failure, setting C<len> to -1.
1249 If you need a copy of the string, see L</bytes_from_utf8>.
1255 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1257 U8 * const save = s;
1258 U8 * const send = s + *len;
1261 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1263 /* ensure valid UTF-8 and chars < 256 before updating string */
1267 if (!UTF8_IS_INVARIANT(c) &&
1268 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1269 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
1270 *len = ((STRLEN) -1);
1278 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
1287 =for apidoc bytes_from_utf8
1289 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1290 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1291 the newly-created string, and updates C<len> to contain the new
1292 length. Returns the original string if no conversion occurs, C<len>
1293 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1294 0 if C<s> is converted or consisted entirely of characters that are invariant
1295 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1301 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1304 const U8 *start = s;
1308 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1310 PERL_UNUSED_CONTEXT;
1314 /* ensure valid UTF-8 and chars < 256 before converting string */
1315 for (send = s + *len; s < send;) {
1317 if (!UTF8_IS_INVARIANT(c)) {
1318 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1319 (c = *s++) && UTF8_IS_CONTINUATION(c))
1328 Newx(d, (*len) - count + 1, U8);
1329 s = start; start = d;
1332 if (!UTF8_IS_INVARIANT(c)) {
1333 /* Then it is two-byte encoded */
1334 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
1344 =for apidoc bytes_to_utf8
1346 Converts a string C<s> of length C<len> bytes from the native encoding into
1348 Returns a pointer to the newly-created string, and sets C<len> to
1349 reflect the new length in bytes.
1351 A NUL character will be written after the end of the string.
1353 If you want to convert to UTF-8 from encodings other than
1354 the native (Latin1 or EBCDIC),
1355 see L</sv_recode_to_utf8>().
1360 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1361 likewise need duplication. */
1364 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1366 const U8 * const send = s + (*len);
1370 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1371 PERL_UNUSED_CONTEXT;
1373 Newx(d, (*len) * 2 + 1, U8);
1377 const UV uv = NATIVE_TO_ASCII(*s++);
1378 if (UNI_IS_INVARIANT(uv))
1379 *d++ = (U8)UTF_TO_NATIVE(uv);
1381 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1382 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
1391 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1393 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1394 * We optimize for native, for obvious reasons. */
1397 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1402 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1405 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1410 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1414 *d++ = UNI_TO_NATIVE(uv);
1421 *d++ = (U8)(( uv >> 6) | 0xc0);
1422 *d++ = (U8)(( uv & 0x3f) | 0x80);
1425 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
1427 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1429 UV low = (p[0] << 8) + p[1];
1431 if (low < 0xdc00 || low > 0xdfff)
1432 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1433 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1435 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1436 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1439 *d++ = (U8)(( uv >> 12) | 0xe0);
1440 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1441 *d++ = (U8)(( uv & 0x3f) | 0x80);
1445 *d++ = (U8)(( uv >> 18) | 0xf0);
1446 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1447 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1448 *d++ = (U8)(( uv & 0x3f) | 0x80);
1452 *newlen = d - dstart;
1456 /* Note: this one is slightly destructive of the source. */
1459 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1462 U8* const send = s + bytelen;
1464 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1467 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1471 const U8 tmp = s[0];
1476 return utf16_to_utf8(p, d, bytelen, newlen);
1479 /* for now these are all defined (inefficiently) in terms of the utf8 versions.
1480 * Note that the macros in handy.h that call these short-circuit calling them
1481 * for Latin-1 range inputs */
1484 Perl_is_uni_alnum(pTHX_ UV c)
1486 U8 tmpbuf[UTF8_MAXBYTES+1];
1487 uvchr_to_utf8(tmpbuf, c);
1488 return is_utf8_alnum(tmpbuf);
1492 Perl_is_uni_idfirst(pTHX_ UV c)
1494 U8 tmpbuf[UTF8_MAXBYTES+1];
1495 uvchr_to_utf8(tmpbuf, c);
1496 return is_utf8_idfirst(tmpbuf);
1500 Perl_is_uni_alpha(pTHX_ UV c)
1502 U8 tmpbuf[UTF8_MAXBYTES+1];
1503 uvchr_to_utf8(tmpbuf, c);
1504 return is_utf8_alpha(tmpbuf);
1508 Perl_is_uni_ascii(pTHX_ UV c)
1514 Perl_is_uni_blank(pTHX_ UV c)
1516 U8 tmpbuf[UTF8_MAXBYTES+1];
1517 uvchr_to_utf8(tmpbuf, c);
1518 return is_utf8_blank(tmpbuf);
1522 Perl_is_uni_space(pTHX_ UV c)
1524 U8 tmpbuf[UTF8_MAXBYTES+1];
1525 uvchr_to_utf8(tmpbuf, c);
1526 return is_utf8_space(tmpbuf);
1530 Perl_is_uni_digit(pTHX_ UV c)
1532 U8 tmpbuf[UTF8_MAXBYTES+1];
1533 uvchr_to_utf8(tmpbuf, c);
1534 return is_utf8_digit(tmpbuf);
1538 Perl_is_uni_upper(pTHX_ UV c)
1540 U8 tmpbuf[UTF8_MAXBYTES+1];
1541 uvchr_to_utf8(tmpbuf, c);
1542 return is_utf8_upper(tmpbuf);
1546 Perl_is_uni_lower(pTHX_ UV c)
1548 U8 tmpbuf[UTF8_MAXBYTES+1];
1549 uvchr_to_utf8(tmpbuf, c);
1550 return is_utf8_lower(tmpbuf);
1554 Perl_is_uni_cntrl(pTHX_ UV c)
1556 return isCNTRL_L1(c);
1560 Perl_is_uni_graph(pTHX_ UV c)
1562 U8 tmpbuf[UTF8_MAXBYTES+1];
1563 uvchr_to_utf8(tmpbuf, c);
1564 return is_utf8_graph(tmpbuf);
1568 Perl_is_uni_print(pTHX_ UV c)
1570 U8 tmpbuf[UTF8_MAXBYTES+1];
1571 uvchr_to_utf8(tmpbuf, c);
1572 return is_utf8_print(tmpbuf);
1576 Perl_is_uni_punct(pTHX_ UV c)
1578 U8 tmpbuf[UTF8_MAXBYTES+1];
1579 uvchr_to_utf8(tmpbuf, c);
1580 return is_utf8_punct(tmpbuf);
1584 Perl_is_uni_xdigit(pTHX_ UV c)
1586 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1587 uvchr_to_utf8(tmpbuf, c);
1588 return is_utf8_xdigit(tmpbuf);
1592 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1594 /* We have the latin1-range values compiled into the core, so just use
1595 * those, converting the result to utf8. The only difference between upper
1596 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1597 * either "SS" or "Ss". Which one to use is passed into the routine in
1598 * 'S_or_s' to avoid a test */
1600 UV converted = toUPPER_LATIN1_MOD(c);
1602 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1604 assert(S_or_s == 'S' || S_or_s == 's');
1606 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1607 characters in this range */
1608 *p = (U8) converted;
1613 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1614 * which it maps to one of them, so as to only have to have one check for
1615 * it in the main case */
1616 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1618 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1619 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1622 converted = GREEK_CAPITAL_LETTER_MU;
1624 case LATIN_SMALL_LETTER_SHARP_S:
1630 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1631 assert(0); /* NOTREACHED */
1635 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1636 *p = UTF8_TWO_BYTE_LO(converted);
1642 /* Call the function to convert a UTF-8 encoded character to the specified case.
1643 * Note that there may be more than one character in the result.
1644 * INP is a pointer to the first byte of the input character
1645 * OUTP will be set to the first byte of the string of changed characters. It
1646 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1647 * LENP will be set to the length in bytes of the string of changed characters
1649 * The functions return the ordinal of the first character in the string of OUTP */
1650 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1651 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1652 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
1654 /* This additionally has the input parameter SPECIALS, which if non-zero will
1655 * cause this to use the SPECIALS hash for folding (meaning get full case
1656 * folding); otherwise, when zero, this implies a simple case fold */
1657 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
1660 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1664 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1665 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1666 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1667 * the changed version may be longer than the original character.
1669 * The ordinal of the first character of the changed version is returned
1670 * (but note, as explained above, that there may be more.) */
1672 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1675 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1678 uvchr_to_utf8(p, c);
1679 return CALL_UPPER_CASE(p, p, lenp);
1683 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1687 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1690 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1693 uvchr_to_utf8(p, c);
1694 return CALL_TITLE_CASE(p, p, lenp);
1698 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1700 /* We have the latin1-range values compiled into the core, so just use
1701 * those, converting the result to utf8. Since the result is always just
1702 * one character, we allow <p> to be NULL */
1704 U8 converted = toLOWER_LATIN1(c);
1707 if (UNI_IS_INVARIANT(converted)) {
1712 *p = UTF8_TWO_BYTE_HI(converted);
1713 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1721 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1725 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1728 return to_lower_latin1((U8) c, p, lenp);
1731 uvchr_to_utf8(p, c);
1732 return CALL_LOWER_CASE(p, p, lenp);
1736 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
1738 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
1743 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1745 if (c == MICRO_SIGN) {
1746 converted = GREEK_SMALL_LETTER_MU;
1748 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1754 else { /* In this range the fold of all other characters is their lower
1756 converted = toLOWER_LATIN1(c);
1759 if (UNI_IS_INVARIANT(converted)) {
1760 *p = (U8) converted;
1764 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1765 *p = UTF8_TWO_BYTE_LO(converted);
1773 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
1776 /* Not currently externally documented, and subject to change
1777 * <flags> bits meanings:
1778 * FOLD_FLAGS_FULL iff full folding is to be used;
1779 * FOLD_FLAGS_LOCALE iff in locale
1780 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1783 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1786 UV result = _to_fold_latin1((U8) c, p, lenp,
1787 cBOOL(((flags & FOLD_FLAGS_FULL)
1788 /* If ASCII-safe, don't allow full folding,
1789 * as that could include SHARP S => ss;
1790 * otherwise there is no crossing of
1791 * ascii/non-ascii in the latin1 range */
1792 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1793 /* It is illegal for the fold to cross the 255/256 boundary under
1794 * locale; in this case return the original */
1795 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1800 /* If no special needs, just use the macro */
1801 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1802 uvchr_to_utf8(p, c);
1803 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1805 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1806 the special flags. */
1807 U8 utf8_c[UTF8_MAXBYTES + 1];
1808 uvchr_to_utf8(utf8_c, c);
1809 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1813 /* for now these all assume no locale info available for Unicode > 255; and
1814 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1815 * called instead, so that these don't get called for < 255 */
1818 Perl_is_uni_alnum_lc(pTHX_ UV c)
1820 return is_uni_alnum(c); /* XXX no locale support yet */
1824 Perl_is_uni_idfirst_lc(pTHX_ UV c)
1826 return is_uni_idfirst(c); /* XXX no locale support yet */
1830 Perl_is_uni_alpha_lc(pTHX_ UV c)
1832 return is_uni_alpha(c); /* XXX no locale support yet */
1836 Perl_is_uni_ascii_lc(pTHX_ UV c)
1838 return is_uni_ascii(c); /* XXX no locale support yet */
1842 Perl_is_uni_blank_lc(pTHX_ UV c)
1844 return is_uni_blank(c); /* XXX no locale support yet */
1848 Perl_is_uni_space_lc(pTHX_ UV c)
1850 return is_uni_space(c); /* XXX no locale support yet */
1854 Perl_is_uni_digit_lc(pTHX_ UV c)
1856 return is_uni_digit(c); /* XXX no locale support yet */
1860 Perl_is_uni_upper_lc(pTHX_ UV c)
1862 return is_uni_upper(c); /* XXX no locale support yet */
1866 Perl_is_uni_lower_lc(pTHX_ UV c)
1868 return is_uni_lower(c); /* XXX no locale support yet */
1872 Perl_is_uni_cntrl_lc(pTHX_ UV c)
1874 return is_uni_cntrl(c); /* XXX no locale support yet */
1878 Perl_is_uni_graph_lc(pTHX_ UV c)
1880 return is_uni_graph(c); /* XXX no locale support yet */
1884 Perl_is_uni_print_lc(pTHX_ UV c)
1886 return is_uni_print(c); /* XXX no locale support yet */
1890 Perl_is_uni_punct_lc(pTHX_ UV c)
1892 return is_uni_punct(c); /* XXX no locale support yet */
1896 Perl_is_uni_xdigit_lc(pTHX_ UV c)
1898 return is_uni_xdigit(c); /* XXX no locale support yet */
1902 Perl_to_uni_upper_lc(pTHX_ U32 c)
1904 /* XXX returns only the first character -- do not use XXX */
1905 /* XXX no locale support yet */
1907 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1908 return (U32)to_uni_upper(c, tmpbuf, &len);
1912 Perl_to_uni_title_lc(pTHX_ U32 c)
1914 /* XXX returns only the first character XXX -- do not use XXX */
1915 /* XXX no locale support yet */
1917 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1918 return (U32)to_uni_title(c, tmpbuf, &len);
1922 Perl_to_uni_lower_lc(pTHX_ U32 c)
1924 /* XXX returns only the first character -- do not use XXX */
1925 /* XXX no locale support yet */
1927 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1928 return (U32)to_uni_lower(c, tmpbuf, &len);
1932 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1933 const char *const swashname)
1935 /* returns a boolean giving whether or not the UTF8-encoded character that
1936 * starts at <p> is in the swash indicated by <swashname>. <swash>
1937 * contains a pointer to where the swash indicated by <swashname>
1938 * is to be stored; which this routine will do, so that future calls will
1939 * look at <*swash> and only generate a swash if it is not null
1941 * Note that it is assumed that the buffer length of <p> is enough to
1942 * contain all the bytes that comprise the character. Thus, <*p> should
1943 * have been checked before this call for mal-formedness enough to assure
1948 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1950 /* The API should have included a length for the UTF-8 character in <p>,
1951 * but it doesn't. We therefor assume that p has been validated at least
1952 * as far as there being enough bytes available in it to accommodate the
1953 * character without reading beyond the end, and pass that number on to the
1954 * validating routine */
1955 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
1958 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1959 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
1961 return swash_fetch(*swash, p, TRUE) != 0;
1965 Perl_is_utf8_alnum(pTHX_ const U8 *p)
1969 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1971 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1972 * descendant of isalnum(3), in other words, it doesn't
1973 * contain the '_'. --jhi */
1974 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
1978 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
1982 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1986 /* is_utf8_idstart would be more logical. */
1987 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1991 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1995 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1999 /* is_utf8_idstart would be more logical. */
2000 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2004 Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2008 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2010 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2014 Perl_is_utf8_idcont(pTHX_ const U8 *p)
2018 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2020 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
2024 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2028 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2030 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2034 Perl_is_utf8_alpha(pTHX_ const U8 *p)
2038 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2040 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
2044 Perl_is_utf8_ascii(pTHX_ const U8 *p)
2048 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2050 /* ASCII characters are the same whether in utf8 or not. So the macro
2051 * works on both utf8 and non-utf8 representations. */
2056 Perl_is_utf8_blank(pTHX_ const U8 *p)
2060 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2062 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2066 Perl_is_utf8_space(pTHX_ const U8 *p)
2070 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2072 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
2076 Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2080 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2082 /* Only true if is an ASCII space-like character, and ASCII is invariant
2083 * under utf8, so can just use the macro */
2084 return isSPACE_A(*p);
2088 Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2092 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2094 /* Only true if is an ASCII word character, and ASCII is invariant
2095 * under utf8, so can just use the macro */
2096 return isWORDCHAR_A(*p);
2100 Perl_is_utf8_digit(pTHX_ const U8 *p)
2104 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2106 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
2110 Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2114 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2116 /* Only true if is an ASCII digit character, and ASCII is invariant
2117 * under utf8, so can just use the macro */
2118 return isDIGIT_A(*p);
2122 Perl_is_utf8_upper(pTHX_ const U8 *p)
2126 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2128 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
2132 Perl_is_utf8_lower(pTHX_ const U8 *p)
2136 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2138 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
2142 Perl_is_utf8_cntrl(pTHX_ const U8 *p)
2146 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2149 return isCNTRL_A(*p);
2152 /* All controls are in Latin1 */
2153 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2156 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2160 Perl_is_utf8_graph(pTHX_ const U8 *p)
2164 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2166 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
2170 Perl_is_utf8_print(pTHX_ const U8 *p)
2174 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2176 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
2180 Perl_is_utf8_punct(pTHX_ const U8 *p)
2184 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2186 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
2190 Perl_is_utf8_xdigit(pTHX_ const U8 *p)
2194 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2196 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
2200 Perl_is_utf8_mark(pTHX_ const U8 *p)
2204 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2206 return is_utf8_common(p, &PL_utf8_mark, "IsM");
2210 Perl_is_utf8_X_regular_begin(pTHX_ const U8 *p)
2214 PERL_ARGS_ASSERT_IS_UTF8_X_REGULAR_BEGIN;
2216 return is_utf8_common(p, &PL_utf8_X_regular_begin, "_X_Regular_Begin");
2220 Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2224 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2226 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2230 =for apidoc to_utf8_case
2232 The C<p> contains the pointer to the UTF-8 string encoding
2233 the character that is being converted. This routine assumes that the character
2234 at C<p> is well-formed.
2236 The C<ustrp> is a pointer to the character buffer to put the
2237 conversion result to. The C<lenp> is a pointer to the length
2240 The C<swashp> is a pointer to the swash to use.
2242 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2243 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
2244 but not always, a multicharacter mapping), is tried first.
2246 The C<special> is a string like "utf8::ToSpecLower", which means the
2247 hash %utf8::ToSpecLower. The access to the hash is through
2248 Perl_to_utf8_case().
2250 The C<normal> is a string like "ToLower" which means the swash
2256 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2257 SV **swashp, const char *normal, const char *special)
2260 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2262 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
2263 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2264 * are necessary in EBCDIC, they are redundant no-ops
2265 * in ASCII-ish platforms, and hopefully optimized away. */
2266 const UV uv1 = NATIVE_TO_UNI(uv0);
2268 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2270 /* Note that swash_fetch() doesn't output warnings for these because it
2271 * assumes we will */
2272 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2273 if (uv1 <= UNICODE_SURROGATE_LAST) {
2274 if (ckWARN_d(WARN_SURROGATE)) {
2275 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2276 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2277 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2280 else if (UNICODE_IS_SUPER(uv1)) {
2281 if (ckWARN_d(WARN_NON_UNICODE)) {
2282 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2283 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2284 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2288 /* Note that non-characters are perfectly legal, so no warning should
2292 uvuni_to_utf8(tmpbuf, uv1);
2294 if (!*swashp) /* load on-demand */
2295 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2298 /* It might be "special" (sometimes, but not always,
2299 * a multicharacter mapping) */
2300 HV * const hv = get_hv(special, 0);
2304 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2308 s = SvPV_const(*svp, len);
2310 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2313 /* If we have EBCDIC we need to remap the characters
2314 * since any characters in the low 256 are Unicode
2315 * code points, not EBCDIC. */
2316 U8 *t = (U8*)s, *tend = t + len, *d;
2323 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2325 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2334 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2339 Copy(tmpbuf, ustrp, len, U8);
2341 Copy(s, ustrp, len, U8);
2347 if (!len && *swashp) {
2348 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
2351 /* It was "normal" (a single character mapping). */
2352 const UV uv3 = UNI_TO_NATIVE(uv2);
2353 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2361 return valid_utf8_to_uvchr(ustrp, 0);
2364 /* Here, there was no mapping defined, which means that the code point maps
2365 * to itself. Return the inputs */
2367 Copy(p, ustrp, len, U8);
2377 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2379 /* This is called when changing the case of a utf8-encoded character above
2380 * the Latin1 range, and the operation is in locale. If the result
2381 * contains a character that crosses the 255/256 boundary, disallow the
2382 * change, and return the original code point. See L<perlfunc/lc> for why;
2384 * p points to the original string whose case was changed; assumed
2385 * by this routine to be well-formed
2386 * result the code point of the first character in the changed-case string
2387 * ustrp points to the changed-case string (<result> represents its first char)
2388 * lenp points to the length of <ustrp> */
2390 UV original; /* To store the first code point of <p> */
2392 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2394 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2396 /* We know immediately if the first character in the string crosses the
2397 * boundary, so can skip */
2400 /* Look at every character in the result; if any cross the
2401 * boundary, the whole thing is disallowed */
2402 U8* s = ustrp + UTF8SKIP(ustrp);
2403 U8* e = ustrp + *lenp;
2405 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2412 /* Here, no characters crossed, result is ok as-is */
2418 /* Failed, have to return the original */
2419 original = valid_utf8_to_uvchr(p, lenp);
2420 Copy(p, ustrp, *lenp, char);
2425 =for apidoc to_utf8_upper
2427 Convert the UTF-8 encoded character at C<p> to its uppercase version and
2428 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2429 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2430 the uppercase version may be longer than the original character.
2432 The first character of the uppercased version is returned
2433 (but note, as explained above, that there may be more.)
2435 The character at C<p> is assumed by this routine to be well-formed.
2439 /* Not currently externally documented, and subject to change:
2440 * <flags> is set iff locale semantics are to be used for code points < 256
2441 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2442 * were used in the calculation; otherwise unchanged. */
2445 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2451 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2453 if (UTF8_IS_INVARIANT(*p)) {
2455 result = toUPPER_LC(*p);
2458 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2461 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2463 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2466 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2470 else { /* utf8, ord above 255 */
2471 result = CALL_UPPER_CASE(p, ustrp, lenp);
2474 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2479 /* Here, used locale rules. Convert back to utf8 */
2480 if (UTF8_IS_INVARIANT(result)) {
2481 *ustrp = (U8) result;
2485 *ustrp = UTF8_EIGHT_BIT_HI(result);
2486 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2491 *tainted_ptr = TRUE;
2497 =for apidoc to_utf8_title
2499 Convert the UTF-8 encoded character at C<p> to its titlecase version and
2500 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2501 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2502 titlecase version may be longer than the original character.
2504 The first character of the titlecased version is returned
2505 (but note, as explained above, that there may be more.)
2507 The character at C<p> is assumed by this routine to be well-formed.
2511 /* Not currently externally documented, and subject to change:
2512 * <flags> is set iff locale semantics are to be used for code points < 256
2513 * Since titlecase is not defined in POSIX, uppercase is used instead
2515 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2516 * were used in the calculation; otherwise unchanged. */
2519 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2525 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2527 if (UTF8_IS_INVARIANT(*p)) {
2529 result = toUPPER_LC(*p);
2532 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2535 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2537 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2540 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2544 else { /* utf8, ord above 255 */
2545 result = CALL_TITLE_CASE(p, ustrp, lenp);
2548 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2553 /* Here, used locale rules. Convert back to utf8 */
2554 if (UTF8_IS_INVARIANT(result)) {
2555 *ustrp = (U8) result;
2559 *ustrp = UTF8_EIGHT_BIT_HI(result);
2560 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2565 *tainted_ptr = TRUE;
2571 =for apidoc to_utf8_lower
2573 Convert the UTF-8 encoded character at C<p> to its lowercase version and
2574 store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2575 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2576 lowercase version may be longer than the original character.
2578 The first character of the lowercased version is returned
2579 (but note, as explained above, that there may be more.)
2581 The character at C<p> is assumed by this routine to be well-formed.
2585 /* Not currently externally documented, and subject to change:
2586 * <flags> is set iff locale semantics are to be used for code points < 256
2587 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2588 * were used in the calculation; otherwise unchanged. */
2591 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2597 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2599 if (UTF8_IS_INVARIANT(*p)) {
2601 result = toLOWER_LC(*p);
2604 return to_lower_latin1(*p, ustrp, lenp);
2607 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2609 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2612 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2616 else { /* utf8, ord above 255 */
2617 result = CALL_LOWER_CASE(p, ustrp, lenp);
2620 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2626 /* Here, used locale rules. Convert back to utf8 */
2627 if (UTF8_IS_INVARIANT(result)) {
2628 *ustrp = (U8) result;
2632 *ustrp = UTF8_EIGHT_BIT_HI(result);
2633 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2638 *tainted_ptr = TRUE;
2644 =for apidoc to_utf8_fold
2646 Convert the UTF-8 encoded character at C<p> to its foldcase version and
2647 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2648 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2649 foldcase version may be longer than the original character (up to
2652 The first character of the foldcased version is returned
2653 (but note, as explained above, that there may be more.)
2655 The character at C<p> is assumed by this routine to be well-formed.
2659 /* Not currently externally documented, and subject to change,
2661 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2662 * points < 256. Since foldcase is not defined in
2663 * POSIX, lowercase is used instead
2664 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2665 * otherwise simple folds
2666 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2668 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2669 * were used in the calculation; otherwise unchanged. */
2672 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2678 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2680 /* These are mutually exclusive */
2681 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2683 assert(p != ustrp); /* Otherwise overwrites */
2685 if (UTF8_IS_INVARIANT(*p)) {
2686 if (flags & FOLD_FLAGS_LOCALE) {
2687 result = toLOWER_LC(*p);
2690 return _to_fold_latin1(*p, ustrp, lenp,
2691 cBOOL(flags & FOLD_FLAGS_FULL));
2694 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2695 if (flags & FOLD_FLAGS_LOCALE) {
2696 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2699 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2701 cBOOL((flags & FOLD_FLAGS_FULL
2702 /* If ASCII safe, don't allow full
2703 * folding, as that could include SHARP
2704 * S => ss; otherwise there is no
2705 * crossing of ascii/non-ascii in the
2707 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
2710 else { /* utf8, ord above 255 */
2711 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2713 if ((flags & FOLD_FLAGS_LOCALE)) {
2714 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2716 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2720 /* This is called when changing the case of a utf8-encoded
2721 * character above the Latin1 range, and the result should not
2722 * contain an ASCII character. */
2724 UV original; /* To store the first code point of <p> */
2726 /* Look at every character in the result; if any cross the
2727 * boundary, the whole thing is disallowed */
2729 U8* e = ustrp + *lenp;
2732 /* Crossed, have to return the original */
2733 original = valid_utf8_to_uvchr(p, lenp);
2734 Copy(p, ustrp, *lenp, char);
2740 /* Here, no characters crossed, result is ok as-is */
2745 /* Here, used locale rules. Convert back to utf8 */
2746 if (UTF8_IS_INVARIANT(result)) {
2747 *ustrp = (U8) result;
2751 *ustrp = UTF8_EIGHT_BIT_HI(result);
2752 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2757 *tainted_ptr = TRUE;
2763 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2764 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2765 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2769 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2771 PERL_ARGS_ASSERT_SWASH_INIT;
2773 /* Returns a copy of a swash initiated by the called function. This is the
2774 * public interface, and returning a copy prevents others from doing
2775 * mischief on the original */
2777 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2781 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2783 /* Initialize and return a swash, creating it if necessary. It does this
2784 * by calling utf8_heavy.pl in the general case. The returned value may be
2785 * the swash's inversion list instead if the input parameters allow it.
2786 * Which is returned should be immaterial to callers, as the only
2787 * operations permitted on a swash, swash_fetch() and
2788 * _get_swash_invlist(), handle both these transparently.
2790 * This interface should only be used by functions that won't destroy or
2791 * adversely change the swash, as doing so affects all other uses of the
2792 * swash in the program; the general public should use 'Perl_swash_init'
2795 * pkg is the name of the package that <name> should be in.
2796 * name is the name of the swash to find. Typically it is a Unicode
2797 * property name, including user-defined ones
2798 * listsv is a string to initialize the swash with. It must be of the form
2799 * documented as the subroutine return value in
2800 * L<perlunicode/User-Defined Character Properties>
2801 * minbits is the number of bits required to represent each data element.
2802 * It is '1' for binary properties.
2803 * none I (khw) do not understand this one, but it is used only in tr///.
2804 * invlist is an inversion list to initialize the swash with (or NULL)
2805 * flags_p if non-NULL is the address of various input and output flag bits
2806 * to the routine, as follows: ('I' means is input to the routine;
2807 * 'O' means output from the routine. Only flags marked O are
2808 * meaningful on return.)
2809 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2810 * came from a user-defined property. (I O)
2811 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2812 * when the swash cannot be located, to simply return NULL. (I)
2813 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2814 * return of an inversion list instead of a swash hash if this routine
2815 * thinks that would result in faster execution of swash_fetch() later
2818 * Thus there are three possible inputs to find the swash: <name>,
2819 * <listsv>, and <invlist>. At least one must be specified. The result
2820 * will be the union of the specified ones, although <listsv>'s various
2821 * actions can intersect, etc. what <name> gives.
2823 * <invlist> is only valid for binary properties */
2826 SV* retval = &PL_sv_undef;
2827 HV* swash_hv = NULL;
2828 const int invlist_swash_boundary =
2829 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2830 ? 512 /* Based on some benchmarking, but not extensive, see commit
2832 : -1; /* Never return just an inversion list */
2834 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2835 assert(! invlist || minbits == 1);
2837 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2839 if (listsv != &PL_sv_undef || strNE(name, "")) {
2841 const size_t pkg_len = strlen(pkg);
2842 const size_t name_len = strlen(name);
2843 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2847 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2849 PUSHSTACKi(PERLSI_MAGIC);
2853 /* We might get here via a subroutine signature which uses a utf8
2854 * parameter name, at which point PL_subname will have been set
2855 * but not yet used. */
2856 save_item(PL_subname);
2857 if (PL_parser && PL_parser->error_count)
2858 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2859 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2860 if (!method) { /* demand load utf8 */
2862 errsv_save = newSVsv(ERRSV);
2863 /* It is assumed that callers of this routine are not passing in
2864 * any user derived data. */
2865 /* Need to do this after save_re_context() as it will set
2866 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2867 * in Perl_magic_get). Even line to create errsv_save can turn on
2869 SAVEBOOL(PL_tainted);
2871 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2874 sv_setsv(ERRSV, errsv_save);
2875 SvREFCNT_dec(errsv_save);
2881 mPUSHp(pkg, pkg_len);
2882 mPUSHp(name, name_len);
2887 errsv_save = newSVsv(ERRSV);
2888 /* If we already have a pointer to the method, no need to use
2889 * call_method() to repeat the lookup. */
2890 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
2891 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2893 retval = *PL_stack_sp--;
2894 SvREFCNT_inc(retval);
2897 sv_setsv(ERRSV, errsv_save);
2898 SvREFCNT_dec(errsv_save);
2901 if (IN_PERL_COMPILETIME) {
2902 CopHINTS_set(PL_curcop, PL_hints);
2904 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2907 /* If caller wants to handle missing properties, let them */
2908 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2912 "Can't find Unicode property definition \"%"SVf"\"",
2914 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2916 } /* End of calling the module to find the swash */
2918 /* If this operation fetched a swash, and we will need it later, get it */
2919 if (retval != &PL_sv_undef
2920 && (minbits == 1 || (flags_p
2922 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2924 swash_hv = MUTABLE_HV(SvRV(retval));
2926 /* If we don't already know that there is a user-defined component to
2927 * this swash, and the user has indicated they wish to know if there is
2928 * one (by passing <flags_p>), find out */
2929 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2930 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2931 if (user_defined && SvUV(*user_defined)) {
2932 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2937 /* Make sure there is an inversion list for binary properties */
2939 SV** swash_invlistsvp = NULL;
2940 SV* swash_invlist = NULL;
2941 bool invlist_in_swash_is_valid = FALSE;
2943 /* If this operation fetched a swash, get its already existing
2944 * inversion list, or create one for it */
2947 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2948 if (swash_invlistsvp) {
2949 swash_invlist = *swash_invlistsvp;
2950 invlist_in_swash_is_valid = TRUE;
2953 swash_invlist = _swash_to_invlist(retval);
2957 /* If an inversion list was passed in, have to include it */
2960 /* Any fetched swash will by now have an inversion list in it;
2961 * otherwise <swash_invlist> will be NULL, indicating that we
2962 * didn't fetch a swash */
2963 if (swash_invlist) {
2965 /* Add the passed-in inversion list, which invalidates the one
2966 * already stored in the swash */
2967 invlist_in_swash_is_valid = FALSE;
2968 _invlist_union(invlist, swash_invlist, &swash_invlist);
2972 /* Here, there is no swash already. Set up a minimal one, if
2973 * we are going to return a swash */
2974 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2976 retval = newRV_inc(MUTABLE_SV(swash_hv));
2978 swash_invlist = invlist;
2982 /* Here, we have computed the union of all the passed-in data. It may
2983 * be that there was an inversion list in the swash which didn't get
2984 * touched; otherwise save the one computed one */
2985 if (! invlist_in_swash_is_valid
2986 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2988 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2990 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2994 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2995 SvREFCNT_dec(retval);
2996 retval = newRV_inc(swash_invlist);
3004 /* This API is wrong for special case conversions since we may need to
3005 * return several Unicode characters for a single Unicode character
3006 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3007 * the lower-level routine, and it is similarly broken for returning
3008 * multiple values. --jhi
3009 * For those, you should use to_utf8_case() instead */
3010 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3013 * Returns the value of property/mapping C<swash> for the first character
3014 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3015 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3016 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3018 * A "swash" is a hash which contains initially the keys/values set up by
3019 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3020 * property for all possible code points. Things are stored in a compact form
3021 * (see utf8_heavy.pl) so that calculation is required to find the actual
3022 * property value for a given code point. As code points are looked up, new
3023 * key/value pairs are added to the hash, so that the calculation doesn't have
3024 * to ever be re-done. Further, each calculation is done, not just for the
3025 * desired one, but for a whole block of code points adjacent to that one.
3026 * For binary properties on ASCII machines, the block is usually for 64 code
3027 * points, starting with a code point evenly divisible by 64. Thus if the
3028 * property value for code point 257 is requested, the code goes out and
3029 * calculates the property values for all 64 code points between 256 and 319,
3030 * and stores these as a single 64-bit long bit vector, called a "swatch",
3031 * under the key for code point 256. The key is the UTF-8 encoding for code
3032 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3033 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3034 * for code point 258 is then requested, this code realizes that it would be
3035 * stored under the key for 256, and would find that value and extract the
3036 * relevant bit, offset from 256.
3038 * Non-binary properties are stored in as many bits as necessary to represent
3039 * their values (32 currently, though the code is more general than that), not
3040 * as single bits, but the principal is the same: the value for each key is a
3041 * vector that encompasses the property values for all code points whose UTF-8
3042 * representations are represented by the key. That is, for all code points
3043 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3047 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3050 HV *const hv = MUTABLE_HV(SvRV(swash));
3055 const U8 *tmps = NULL;
3059 const UV c = NATIVE_TO_ASCII(*ptr);
3061 PERL_ARGS_ASSERT_SWASH_FETCH;
3063 /* If it really isn't a hash, it isn't really swash; must be an inversion
3065 if (SvTYPE(hv) != SVt_PVHV) {
3066 return _invlist_contains_cp((SV*)hv,
3068 ? valid_utf8_to_uvchr(ptr, NULL)
3072 /* Convert to utf8 if not already */
3073 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
3074 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3075 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3078 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
3079 * then the "swatch" is a vec() for all the chars which start
3081 * So the key in the hash (klen) is length of encoded char -1
3083 klen = UTF8SKIP(ptr) - 1;
3087 /* If char is invariant then swatch is for all the invariant chars
3088 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
3090 needents = UTF_CONTINUATION_MARK;
3091 off = NATIVE_TO_UTF(ptr[klen]);
3094 /* If char is encoded then swatch is for the prefix */
3095 needents = (1 << UTF_ACCUMULATION_SHIFT);
3096 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3100 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3101 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3102 * it's nothing to sniff at.) Pity we usually come through at least
3103 * two function calls to get here...
3105 * NB: this code assumes that swatches are never modified, once generated!
3108 if (hv == PL_last_swash_hv &&
3109 klen == PL_last_swash_klen &&
3110 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3112 tmps = PL_last_swash_tmps;
3113 slen = PL_last_swash_slen;
3116 /* Try our second-level swatch cache, kept in a hash. */
3117 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3119 /* If not cached, generate it via swatch_get */
3120 if (!svp || !SvPOK(*svp)
3121 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
3122 /* We use utf8n_to_uvuni() as we want an index into
3123 Unicode tables, not a native character number.
3125 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
3127 0 : UTF8_ALLOW_ANY);
3128 swatch = swatch_get(swash,
3129 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
3130 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
3133 if (IN_PERL_COMPILETIME)
3134 CopHINTS_set(PL_curcop, PL_hints);
3136 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3138 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3139 || (slen << 3) < needents)
3140 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3141 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3142 svp, tmps, (UV)slen, (UV)needents);
3145 PL_last_swash_hv = hv;
3146 assert(klen <= sizeof(PL_last_swash_key));
3147 PL_last_swash_klen = (U8)klen;
3148 /* FIXME change interpvar.h? */
3149 PL_last_swash_tmps = (U8 *) tmps;
3150 PL_last_swash_slen = slen;
3152 Copy(ptr, PL_last_swash_key, klen, U8);
3155 switch ((int)((slen << 3) / needents)) {
3157 bit = 1 << (off & 7);
3159 return (tmps[off] & bit) != 0;
3164 return (tmps[off] << 8) + tmps[off + 1] ;
3167 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3169 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3170 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
3171 NORETURN_FUNCTION_END;
3174 /* Read a single line of the main body of the swash input text. These are of
3177 * where each number is hex. The first two numbers form the minimum and
3178 * maximum of a range, and the third is the value associated with the range.
3179 * Not all swashes should have a third number
3181 * On input: l points to the beginning of the line to be examined; it points
3182 * to somewhere in the string of the whole input text, and is
3183 * terminated by a \n or the null string terminator.
3184 * lend points to the null terminator of that string
3185 * wants_value is non-zero if the swash expects a third number
3186 * typestr is the name of the swash's mapping, like 'ToLower'
3187 * On output: *min, *max, and *val are set to the values read from the line.
3188 * returns a pointer just beyond the line examined. If there was no
3189 * valid min number on the line, returns lend+1
3193 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3194 const bool wants_value, const U8* const typestr)
3196 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3197 STRLEN numlen; /* Length of the number */
3198 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3199 | PERL_SCAN_DISALLOW_PREFIX
3200 | PERL_SCAN_SILENT_NON_PORTABLE;
3202 /* nl points to the next \n in the scan */
3203 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3205 /* Get the first number on the line: the range minimum */
3207 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3208 if (numlen) /* If found a hex number, position past it */
3210 else if (nl) { /* Else, go handle next line, if any */
3211 return nl + 1; /* 1 is length of "\n" */
3213 else { /* Else, no next line */
3214 return lend + 1; /* to LIST's end at which \n is not found */
3217 /* The max range value follows, separated by a BLANK */
3220 flags = PERL_SCAN_SILENT_ILLDIGIT
3221 | PERL_SCAN_DISALLOW_PREFIX
3222 | PERL_SCAN_SILENT_NON_PORTABLE;
3224 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3227 else /* If no value here, it is a single element range */
3230 /* Non-binary tables have a third entry: what the first element of the
3236 /* The ToLc, etc table mappings are not in hex, and must be
3237 * corrected by adding the code point to them */
3239 char *after_strtol = (char *) lend;
3240 *val = Strtol((char *)l, &after_strtol, 10);
3241 l = (U8 *) after_strtol;
3243 else { /* Other tables are in hex, and are the correct result
3245 flags = PERL_SCAN_SILENT_ILLDIGIT
3246 | PERL_SCAN_DISALLOW_PREFIX
3247 | PERL_SCAN_SILENT_NON_PORTABLE;
3249 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3259 /* diag_listed_as: To%s: illegal mapping '%s' */
3260 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3266 *val = 0; /* bits == 1, then any val should be ignored */
3268 else { /* Nothing following range min, should be single element with no
3274 /* diag_listed_as: To%s: illegal mapping '%s' */
3275 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3279 *val = 0; /* bits == 1, then val should be ignored */
3282 /* Position to next line if any, or EOF */
3292 * Returns a swatch (a bit vector string) for a code point sequence
3293 * that starts from the value C<start> and comprises the number C<span>.
3294 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3295 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3298 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3301 U8 *l, *lend, *x, *xend, *s, *send;
3302 STRLEN lcur, xcur, scur;
3303 HV *const hv = MUTABLE_HV(SvRV(swash));
3304 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3306 SV** listsvp = NULL; /* The string containing the main body of the table */
3307 SV** extssvp = NULL;
3308 SV** invert_it_svp = NULL;
3311 STRLEN octets; /* if bits == 1, then octets == 0 */
3313 UV end = start + span;
3315 if (invlistsvp == NULL) {
3316 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3317 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3318 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3319 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3320 listsvp = hv_fetchs(hv, "LIST", FALSE);
3321 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3323 bits = SvUV(*bitssvp);
3324 none = SvUV(*nonesvp);
3325 typestr = (U8*)SvPV_nolen(*typesvp);
3331 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3333 PERL_ARGS_ASSERT_SWATCH_GET;
3335 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3336 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3340 /* If overflowed, use the max possible */
3346 /* create and initialize $swatch */
3347 scur = octets ? (span * octets) : (span + 7) / 8;
3348 swatch = newSV(scur);
3350 s = (U8*)SvPVX(swatch);
3351 if (octets && none) {
3352 const U8* const e = s + scur;
3355 *s++ = (U8)(none & 0xff);
3356 else if (bits == 16) {
3357 *s++ = (U8)((none >> 8) & 0xff);
3358 *s++ = (U8)( none & 0xff);
3360 else if (bits == 32) {
3361 *s++ = (U8)((none >> 24) & 0xff);
3362 *s++ = (U8)((none >> 16) & 0xff);
3363 *s++ = (U8)((none >> 8) & 0xff);
3364 *s++ = (U8)( none & 0xff);
3370 (void)memzero((U8*)s, scur + 1);
3372 SvCUR_set(swatch, scur);
3373 s = (U8*)SvPVX(swatch);
3375 if (invlistsvp) { /* If has an inversion list set up use that */
3376 _invlist_populate_swatch(*invlistsvp, start, end, s);
3380 /* read $swash->{LIST} */
3381 l = (U8*)SvPV(*listsvp, lcur);
3384 UV min, max, val, upper;
3385 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3386 cBOOL(octets), typestr);
3391 /* If looking for something beyond this range, go try the next one */
3395 /* <end> is generally 1 beyond where we want to set things, but at the
3396 * platform's infinity, where we can't go any higher, we want to
3397 * include the code point at <end> */
3400 : (max != UV_MAX || end != UV_MAX)
3407 if (!none || val < none) {
3412 for (key = min; key <= upper; key++) {
3414 /* offset must be non-negative (start <= min <= key < end) */
3415 offset = octets * (key - start);
3417 s[offset] = (U8)(val & 0xff);
3418 else if (bits == 16) {
3419 s[offset ] = (U8)((val >> 8) & 0xff);
3420 s[offset + 1] = (U8)( val & 0xff);
3422 else if (bits == 32) {
3423 s[offset ] = (U8)((val >> 24) & 0xff);
3424 s[offset + 1] = (U8)((val >> 16) & 0xff);
3425 s[offset + 2] = (U8)((val >> 8) & 0xff);
3426 s[offset + 3] = (U8)( val & 0xff);
3429 if (!none || val < none)
3433 else { /* bits == 1, then val should be ignored */
3438 for (key = min; key <= upper; key++) {
3439 const STRLEN offset = (STRLEN)(key - start);
3440 s[offset >> 3] |= 1 << (offset & 7);
3445 /* Invert if the data says it should be. Assumes that bits == 1 */
3446 if (invert_it_svp && SvUV(*invert_it_svp)) {
3448 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3449 * be 0, and their inversion should also be 0, as we don't succeed any
3450 * Unicode property matches for non-Unicode code points */
3451 if (start <= PERL_UNICODE_MAX) {
3453 /* The code below assumes that we never cross the
3454 * Unicode/above-Unicode boundary in a range, as otherwise we would
3455 * have to figure out where to stop flipping the bits. Since this
3456 * boundary is divisible by a large power of 2, and swatches comes
3457 * in small powers of 2, this should be a valid assumption */
3458 assert(start + span - 1 <= PERL_UNICODE_MAX);
3468 /* read $swash->{EXTRAS}
3469 * This code also copied to swash_to_invlist() below */
3470 x = (U8*)SvPV(*extssvp, xcur);
3478 SV **otherbitssvp, *other;
3482 const U8 opc = *x++;
3486 nl = (U8*)memchr(x, '\n', xend - x);
3488 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3490 x = nl + 1; /* 1 is length of "\n" */
3494 x = xend; /* to EXTRAS' end at which \n is not found */
3501 namelen = nl - namestr;
3505 namelen = xend - namestr;
3509 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3510 otherhv = MUTABLE_HV(SvRV(*othersvp));
3511 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3512 otherbits = (STRLEN)SvUV(*otherbitssvp);
3513 if (bits < otherbits)
3514 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3515 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3517 /* The "other" swatch must be destroyed after. */
3518 other = swatch_get(*othersvp, start, span);
3519 o = (U8*)SvPV(other, olen);
3522 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3524 s = (U8*)SvPV(swatch, slen);
3525 if (bits == 1 && otherbits == 1) {
3527 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3528 "mismatch, slen=%"UVuf", olen=%"UVuf,
3529 (UV)slen, (UV)olen);
3553 STRLEN otheroctets = otherbits >> 3;
3555 U8* const send = s + slen;
3560 if (otherbits == 1) {
3561 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3565 STRLEN vlen = otheroctets;
3573 if (opc == '+' && otherval)
3574 NOOP; /* replace with otherval */
3575 else if (opc == '!' && !otherval)
3577 else if (opc == '-' && otherval)
3579 else if (opc == '&' && !otherval)
3582 s += octets; /* no replacement */
3587 *s++ = (U8)( otherval & 0xff);
3588 else if (bits == 16) {
3589 *s++ = (U8)((otherval >> 8) & 0xff);
3590 *s++ = (U8)( otherval & 0xff);
3592 else if (bits == 32) {
3593 *s++ = (U8)((otherval >> 24) & 0xff);
3594 *s++ = (U8)((otherval >> 16) & 0xff);
3595 *s++ = (U8)((otherval >> 8) & 0xff);
3596 *s++ = (U8)( otherval & 0xff);
3600 sv_free(other); /* through with it! */
3606 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3609 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3610 * Can't be used on a property that is subject to user override, as it
3611 * relies on the value of SPECIALS in the swash which would be set by
3612 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3613 * for overridden properties
3615 * Returns a hash which is the inversion and closure of a swash mapping.
3616 * For example, consider the input lines:
3621 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3622 * 006C. The value for each key is an array. For 006C, the array would
3623 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3624 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3626 * Essentially, for any code point, it gives all the code points that map to
3627 * it, or the list of 'froms' for that point.
3629 * Currently it ignores any additions or deletions from other swashes,
3630 * looking at just the main body of the swash, and if there are SPECIALS
3631 * in the swash, at that hash
3633 * The specials hash can be extra code points, and most likely consists of
3634 * maps from single code points to multiple ones (each expressed as a string
3635 * of utf8 characters). This function currently returns only 1-1 mappings.
3636 * However consider this possible input in the specials hash:
3637 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3638 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3640 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3641 * currently handle. But it also means that FB05 and FB06 are equivalent in
3642 * a 1-1 mapping which we should handle, and this relationship may not be in
3643 * the main table. Therefore this function examines all the multi-char
3644 * sequences and adds the 1-1 mappings that come out of that. */
3648 HV *const hv = MUTABLE_HV(SvRV(swash));
3650 /* The string containing the main body of the table */
3651 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3653 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3654 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3655 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3656 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3657 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3658 const STRLEN bits = SvUV(*bitssvp);
3659 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3660 const UV none = SvUV(*nonesvp);
3661 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3665 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3667 /* Must have at least 8 bits to get the mappings */
3668 if (bits != 8 && bits != 16 && bits != 32) {
3669 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3673 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3674 mapping to more than one character */
3676 /* Construct an inverse mapping hash for the specials */
3677 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3678 HV * specials_inverse = newHV();
3679 char *char_from; /* the lhs of the map */
3680 I32 from_len; /* its byte length */
3681 char *char_to; /* the rhs of the map */
3682 I32 to_len; /* its byte length */
3683 SV *sv_to; /* and in a sv */
3684 AV* from_list; /* list of things that map to each 'to' */
3686 hv_iterinit(specials_hv);
3688 /* The keys are the characters (in utf8) that map to the corresponding
3689 * utf8 string value. Iterate through the list creating the inverse
3691 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3693 if (! SvPOK(sv_to)) {
3694 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3695 "unexpectedly is not a string, flags=%lu",
3696 (unsigned long)SvFLAGS(sv_to));
3698 /*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)));*/
3700 /* Each key in the inverse list is a mapped-to value, and the key's
3701 * hash value is a list of the strings (each in utf8) that map to
3702 * it. Those strings are all one character long */
3703 if ((listp = hv_fetch(specials_inverse,
3707 from_list = (AV*) *listp;
3709 else { /* No entry yet for it: create one */
3710 from_list = newAV();
3711 if (! hv_store(specials_inverse,
3714 (SV*) from_list, 0))
3716 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3720 /* Here have the list associated with this 'to' (perhaps newly
3721 * created and empty). Just add to it. Note that we ASSUME that
3722 * the input is guaranteed to not have duplications, so we don't
3723 * check for that. Duplications just slow down execution time. */
3724 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3727 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3728 * it looking for cases like the FB05/FB06 examples above. There would
3729 * be an entry in the hash like
3730 * 'st' => [ FB05, FB06 ]
3731 * In this example we will create two lists that get stored in the
3732 * returned hash, 'ret':
3733 * FB05 => [ FB05, FB06 ]
3734 * FB06 => [ FB05, FB06 ]
3736 * Note that there is nothing to do if the array only has one element.
3737 * (In the normal 1-1 case handled below, we don't have to worry about
3738 * two lists, as everything gets tied to the single list that is
3739 * generated for the single character 'to'. But here, we are omitting
3740 * that list, ('st' in the example), so must have multiple lists.) */
3741 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3742 &char_to, &to_len)))
3744 if (av_len(from_list) > 0) {
3747 /* We iterate over all combinations of i,j to place each code
3748 * point on each list */
3749 for (i = 0; i <= av_len(from_list); i++) {
3751 AV* i_list = newAV();
3752 SV** entryp = av_fetch(from_list, i, FALSE);
3753 if (entryp == NULL) {
3754 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3756 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3757 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3759 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3760 (SV*) i_list, FALSE))
3762 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3765 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3766 for (j = 0; j <= av_len(from_list); j++) {
3767 entryp = av_fetch(from_list, j, FALSE);
3768 if (entryp == NULL) {
3769 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3772 /* When i==j this adds itself to the list */
3773 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3774 (U8*) SvPVX(*entryp),
3775 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3777 /*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));*/
3782 SvREFCNT_dec(specials_inverse); /* done with it */
3783 } /* End of specials */
3785 /* read $swash->{LIST} */
3786 l = (U8*)SvPV(*listsvp, lcur);
3789 /* Go through each input line */
3793 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3794 cBOOL(octets), typestr);
3799 /* Each element in the range is to be inverted */
3800 for (inverse = min; inverse <= max; inverse++) {
3804 bool found_key = FALSE;
3805 bool found_inverse = FALSE;
3807 /* The key is the inverse mapping */
3808 char key[UTF8_MAXBYTES+1];
3809 char* key_end = (char *) uvuni_to_utf8((U8*) key, val);
3810 STRLEN key_len = key_end - key;
3812 /* Get the list for the map */
3813 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3814 list = (AV*) *listp;
3816 else { /* No entry yet for it: create one */
3818 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3819 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3823 /* Look through list to see if this inverse mapping already is
3824 * listed, or if there is a mapping to itself already */
3825 for (i = 0; i <= av_len(list); i++) {
3826 SV** entryp = av_fetch(list, i, FALSE);
3828 if (entryp == NULL) {
3829 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3832 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3833 if (SvUV(entry) == val) {
3836 if (SvUV(entry) == inverse) {
3837 found_inverse = TRUE;
3840 /* No need to continue searching if found everything we are
3842 if (found_key && found_inverse) {
3847 /* Make sure there is a mapping to itself on the list */
3849 av_push(list, newSVuv(val));
3850 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3854 /* Simply add the value to the list */
3855 if (! found_inverse) {
3856 av_push(list, newSVuv(inverse));
3857 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3860 /* swatch_get() increments the value of val for each element in the
3861 * range. That makes more compact tables possible. You can
3862 * express the capitalization, for example, of all consecutive
3863 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3864 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3865 * and it's not documented; it appears to be used only in
3866 * implementing tr//; I copied the semantics from swatch_get(), just
3868 if (!none || val < none) {
3878 Perl__swash_to_invlist(pTHX_ SV* const swash)
3881 /* Subject to change or removal. For use only in one place in regcomp.c */
3886 HV *const hv = MUTABLE_HV(SvRV(swash));
3887 UV elements = 0; /* Number of elements in the inversion list */
3890 /* The string containing the main body of the table */
3891 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3892 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3893 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3894 SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3895 SV** const invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3897 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3898 const STRLEN bits = SvUV(*bitssvp);
3899 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3905 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3907 /* read $swash->{LIST} */
3908 if (SvPOK(*listsvp)) {
3909 l = (U8*)SvPV(*listsvp, lcur);
3912 /* LIST legitimately doesn't contain a string during compilation phases
3913 * of Perl itself, before the Unicode tables are generated. In this
3914 * case, just fake things up by creating an empty list */
3921 /* Scan the input to count the number of lines to preallocate array size
3922 * based on worst possible case, which is each line in the input creates 2
3923 * elements in the inversion list: 1) the beginning of a range in the list;
3924 * 2) the beginning of a range not in the list. */
3925 while ((loc = (strchr(loc, '\n'))) != NULL) {
3930 /* If the ending is somehow corrupt and isn't a new line, add another
3931 * element for the final range that isn't in the inversion list */
3932 if (! (*lend == '\n'
3933 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3938 invlist = _new_invlist(elements);
3940 /* Now go through the input again, adding each range to the list */
3943 UV val; /* Not used by this function */
3945 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
3946 cBOOL(octets), typestr);
3952 invlist = _add_range_to_invlist(invlist, start, end);
3955 /* Invert if the data says it should be */
3956 if (invert_it_svp && SvUV(*invert_it_svp)) {
3957 _invlist_invert_prop(invlist);
3960 /* This code is copied from swatch_get()
3961 * read $swash->{EXTRAS} */
3962 x = (U8*)SvPV(*extssvp, xcur);
3970 SV **otherbitssvp, *other;
3973 const U8 opc = *x++;
3977 nl = (U8*)memchr(x, '\n', xend - x);
3979 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3981 x = nl + 1; /* 1 is length of "\n" */
3985 x = xend; /* to EXTRAS' end at which \n is not found */
3992 namelen = nl - namestr;
3996 namelen = xend - namestr;
4000 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4001 otherhv = MUTABLE_HV(SvRV(*othersvp));
4002 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4003 otherbits = (STRLEN)SvUV(*otherbitssvp);
4005 if (bits != otherbits || bits != 1) {
4006 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4007 "properties, bits=%"UVuf", otherbits=%"UVuf,
4008 (UV)bits, (UV)otherbits);
4011 /* The "other" swatch must be destroyed after. */
4012 other = _swash_to_invlist((SV *)*othersvp);
4014 /* End of code copied from swatch_get() */
4017 _invlist_union(invlist, other, &invlist);
4020 _invlist_invert(other);
4021 _invlist_union(invlist, other, &invlist);
4024 _invlist_subtract(invlist, other, &invlist);
4027 _invlist_intersection(invlist, other, &invlist);
4032 sv_free(other); /* through with it! */
4039 Perl__get_swash_invlist(pTHX_ SV* const swash)
4043 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4045 if (! SvROK(swash)) {
4049 /* If it really isn't a hash, it isn't really swash; must be an inversion
4051 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4055 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4064 =for apidoc uvchr_to_utf8
4066 Adds the UTF-8 representation of the Native code point C<uv> to the end
4067 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
4068 bytes available. The return value is the pointer to the byte after the
4069 end of the new character. In other words,
4071 d = uvchr_to_utf8(d, uv);
4073 is the recommended wide native character-aware way of saying
4080 /* On ASCII machines this is normally a macro but we want a
4081 real function in case XS code wants it
4084 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
4086 PERL_ARGS_ASSERT_UVCHR_TO_UTF8;
4088 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), 0);
4092 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4094 PERL_ARGS_ASSERT_UVCHR_TO_UTF8_FLAGS;
4096 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), flags);
4100 =for apidoc utf8n_to_uvchr
4102 Returns the native character value of the first character in the string
4104 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
4105 length, in bytes, of that character.
4107 C<length> and C<flags> are the same as L</utf8n_to_uvuni>().
4111 /* On ASCII machines this is normally a macro but we want
4112 a real function in case XS code wants it
4115 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen,
4118 const UV uv = Perl_utf8n_to_uvuni(aTHX_ s, curlen, retlen, flags);
4120 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
4122 return UNI_TO_NATIVE(uv);
4126 Perl_check_utf8_print(pTHX_ register const U8* s, const STRLEN len)
4128 /* May change: warns if surrogates, non-character code points, or
4129 * non-Unicode code points are in s which has length len bytes. Returns
4130 * TRUE if none found; FALSE otherwise. The only other validity check is
4131 * to make sure that this won't exceed the string's length */
4133 const U8* const e = s + len;
4136 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4139 if (UTF8SKIP(s) > len) {
4140 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4141 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4144 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
4146 if (UTF8_IS_SUPER(s)) {
4147 if (ckWARN_d(WARN_NON_UNICODE)) {
4148 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4149 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
4150 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
4154 else if (UTF8_IS_SURROGATE(s)) {
4155 if (ckWARN_d(WARN_SURROGATE)) {
4156 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4157 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4158 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
4163 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
4164 && (ckWARN_d(WARN_NONCHAR)))
4166 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4167 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
4168 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
4179 =for apidoc pv_uni_display
4181 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4182 length C<len>, the displayable version being at most C<pvlim> bytes long
4183 (if longer, the rest is truncated and "..." will be appended).
4185 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
4186 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
4187 to display the \\[nrfta\\] as the backslashed versions (like '\n')
4188 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
4189 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
4190 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
4192 The pointer to the PV of the C<dsv> is returned.
4196 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4201 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4205 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4207 /* This serves double duty as a flag and a character to print after
4208 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4212 if (pvlim && SvCUR(dsv) >= pvlim) {
4216 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4218 const unsigned char c = (unsigned char)u & 0xFF;
4219 if (flags & UNI_DISPLAY_BACKSLASH) {
4236 const char string = ok;
4237 sv_catpvs(dsv, "\\");
4238 sv_catpvn(dsv, &string, 1);
4241 /* isPRINT() is the locale-blind version. */
4242 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4243 const char string = c;
4244 sv_catpvn(dsv, &string, 1);
4249 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4252 sv_catpvs(dsv, "...");
4258 =for apidoc sv_uni_display
4260 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4261 the displayable version being at most C<pvlim> bytes long
4262 (if longer, the rest is truncated and "..." will be appended).
4264 The C<flags> argument is as in L</pv_uni_display>().
4266 The pointer to the PV of the C<dsv> is returned.
4271 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4273 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4275 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)SvPVX_const(ssv),
4276 SvCUR(ssv), pvlim, flags);
4280 =for apidoc foldEQ_utf8
4282 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4283 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4284 How far into the strings to compare is determined by other input parameters.
4286 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4287 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4288 with respect to C<s2>.
4290 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4291 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4292 scan will not be considered to be a match unless the goal is reached, and
4293 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4296 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4297 considered an end pointer beyond which scanning of C<s1> will not continue under
4298 any circumstances. This means that if both C<l1> and C<pe1> are specified, and
4300 is less than C<s1>+C<l1>, the match will never be successful because it can
4302 get as far as its goal (and in fact is asserted against). Correspondingly for
4303 C<pe2> with respect to C<s2>.
4305 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4306 C<l2> must be non-zero), and if both do, both have to be
4307 reached for a successful match. Also, if the fold of a character is multiple
4308 characters, all of them must be matched (see tr21 reference below for
4311 Upon a successful match, if C<pe1> is non-NULL,
4312 it will be set to point to the beginning of the I<next> character of C<s1>
4313 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4315 For case-insensitiveness, the "casefolding" of Unicode is used
4316 instead of upper/lowercasing both the characters, see
4317 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4321 /* A flags parameter has been added which may change, and hence isn't
4322 * externally documented. Currently it is:
4323 * 0 for as-documented above
4324 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4325 ASCII one, to not match
4326 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
4327 * points below 256; unicode rules for above 255; and
4328 * folds that cross those boundaries are disallowed,
4329 * like the NOMIX_ASCII option
4330 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4331 * routine. This allows that step to be skipped.
4332 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4335 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, register UV l1, bool u1, const char *s2, char **pe2, register UV l2, bool u2, U32 flags)
4338 const U8 *p1 = (const U8*)s1; /* Point to current char */
4339 const U8 *p2 = (const U8*)s2;
4340 const U8 *g1 = NULL; /* goal for s1 */
4341 const U8 *g2 = NULL;
4342 const U8 *e1 = NULL; /* Don't scan s1 past this */
4343 U8 *f1 = NULL; /* Point to current folded */
4344 const U8 *e2 = NULL;
4346 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4347 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4348 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4350 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4352 /* The algorithm requires that input with the flags on the first line of
4353 * the assert not be pre-folded. */
4354 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
4355 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4362 g1 = (const U8*)s1 + l1;
4370 g2 = (const U8*)s2 + l2;
4373 /* Must have at least one goal */
4378 /* Will never match if goal is out-of-bounds */
4379 assert(! e1 || e1 >= g1);
4381 /* Here, there isn't an end pointer, or it is beyond the goal. We
4382 * only go as far as the goal */
4386 assert(e1); /* Must have an end for looking at s1 */
4389 /* Same for goal for s2 */
4391 assert(! e2 || e2 >= g2);
4398 /* If both operands are already folded, we could just do a memEQ on the
4399 * whole strings at once, but it would be better if the caller realized
4400 * this and didn't even call us */
4402 /* Look through both strings, a character at a time */
4403 while (p1 < e1 && p2 < e2) {
4405 /* If at the beginning of a new character in s1, get its fold to use
4406 * and the length of the fold. (exception: locale rules just get the
4407 * character to a single byte) */
4409 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4415 /* If in locale matching, we use two sets of rules, depending
4416 * on if the code point is above or below 255. Here, we test
4417 * for and handle locale rules */
4418 if ((flags & FOLDEQ_UTF8_LOCALE)
4419 && (! u1 || UTF8_IS_INVARIANT(*p1)
4420 || UTF8_IS_DOWNGRADEABLE_START(*p1)))
4422 /* There is no mixing of code points above and below 255. */
4423 if (u2 && (! UTF8_IS_INVARIANT(*p2)
4424 && ! UTF8_IS_DOWNGRADEABLE_START(*p2)))
4429 /* We handle locale rules by converting, if necessary, the
4430 * code point to a single byte. */
4431 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4435 *foldbuf1 = TWO_BYTE_UTF8_TO_UNI(*p1, *(p1 + 1));
4439 else if (isASCII(*p1)) { /* Note, that here won't be both
4440 ASCII and using locale rules */
4442 /* If trying to mix non- with ASCII, and not supposed to,
4444 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4448 *foldbuf1 = toLOWER(*p1); /* Folds in the ASCII range are
4452 to_utf8_fold(p1, foldbuf1, &n1);
4454 else { /* Not utf8, get utf8 fold */
4455 to_uni_fold(NATIVE_TO_UNI(*p1), foldbuf1, &n1);
4461 if (n2 == 0) { /* Same for s2 */
4462 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4467 if ((flags & FOLDEQ_UTF8_LOCALE)
4468 && (! u2 || UTF8_IS_INVARIANT(*p2) || UTF8_IS_DOWNGRADEABLE_START(*p2)))
4470 /* Here, the next char in s2 is < 256. We've already
4471 * worked on s1, and if it isn't also < 256, can't match */
4472 if (u1 && (! UTF8_IS_INVARIANT(*p1)
4473 && ! UTF8_IS_DOWNGRADEABLE_START(*p1)))
4477 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4481 *foldbuf2 = TWO_BYTE_UTF8_TO_UNI(*p2, *(p2 + 1));
4484 /* Use another function to handle locale rules. We've made
4485 * sure that both characters to compare are single bytes */
4486 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4491 else if (isASCII(*p2)) {
4492 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4496 *foldbuf2 = toLOWER(*p2);
4499 to_utf8_fold(p2, foldbuf2, &n2);
4502 to_uni_fold(NATIVE_TO_UNI(*p2), foldbuf2, &n2);
4508 /* Here f1 and f2 point to the beginning of the strings to compare.
4509 * These strings are the folds of the next character from each input
4510 * string, stored in utf8. */
4512 /* While there is more to look for in both folds, see if they
4513 * continue to match */
4515 U8 fold_length = UTF8SKIP(f1);
4516 if (fold_length != UTF8SKIP(f2)
4517 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4518 function call for single
4520 || memNE((char*)f1, (char*)f2, fold_length))
4522 return 0; /* mismatch */
4525 /* Here, they matched, advance past them */
4532 /* When reach the end of any fold, advance the input past it */
4534 p1 += u1 ? UTF8SKIP(p1) : 1;
4537 p2 += u2 ? UTF8SKIP(p2) : 1;
4539 } /* End of loop through both strings */
4541 /* A match is defined by each scan that specified an explicit length
4542 * reaching its final goal, and the other not having matched a partial
4543 * character (which can happen when the fold of a character is more than one
4545 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4549 /* Successful match. Set output pointers */
4561 * c-indentation-style: bsd
4563 * indent-tabs-mode: nil
4566 * ex: set ts=8 sts=4 sw=4 et: