3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "inline_invlist.c"
35 #include "charclass_invlists.h"
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
41 =head1 Unicode Support
42 These are various utility functions for manipulating UTF8-encoded
43 strings. For the uninitiated, this is a method of representing arbitrary
44 Unicode characters as a variable number of bytes, in such a way that
45 characters in the ASCII range are unmodified, and a zero byte never appears
46 within non-zero characters.
52 =for apidoc is_ascii_string
54 Returns true if the first C<len> bytes of the string C<s> are the same whether
55 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
56 is, if they are invariant. On ASCII-ish machines, only ASCII characters
57 fit this definition, hence the function's name.
59 If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
60 use this option, that C<s> can't have embedded C<NUL> characters and has to
61 have a terminating C<NUL> byte).
63 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
69 Perl_is_ascii_string(const U8 *s, STRLEN len)
71 const U8* const send = s + (len ? len : strlen((const char *)s));
74 PERL_ARGS_ASSERT_IS_ASCII_STRING;
76 for (; x < send; ++x) {
77 if (!UTF8_IS_INVARIANT(*x))
85 =for apidoc uvoffuni_to_utf8_flags
87 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
88 Instead, B<Almost all code should use L</uvchr_to_utf8> or
89 L</uvchr_to_utf8_flags>>.
91 This function is like them, but the input is a strict Unicode
92 (as opposed to native) code point. Only in very rare circumstances should code
93 not be using the native code point.
95 For details, see the description for L</uvchr_to_utf8_flags>>.
101 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
103 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
105 if (UNI_IS_INVARIANT(uv)) {
106 *d++ = (U8) LATIN1_TO_NATIVE(uv);
110 /* The first problematic code point is the first surrogate */
111 if (uv >= UNICODE_SURROGATE_FIRST
112 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
114 if (UNICODE_IS_SURROGATE(uv)) {
115 if (flags & UNICODE_WARN_SURROGATE) {
116 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
117 "UTF-16 surrogate U+%04"UVXf, uv);
119 if (flags & UNICODE_DISALLOW_SURROGATE) {
123 else if (UNICODE_IS_SUPER(uv)) {
124 if (flags & UNICODE_WARN_SUPER
125 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
127 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
128 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
130 if (flags & UNICODE_DISALLOW_SUPER
131 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
136 else if (UNICODE_IS_NONCHAR(uv)) {
137 if (flags & UNICODE_WARN_NONCHAR) {
138 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
139 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
142 if (flags & UNICODE_DISALLOW_NONCHAR) {
150 STRLEN len = OFFUNISKIP(uv);
153 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
154 uv >>= UTF_ACCUMULATION_SHIFT;
156 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
159 #else /* Non loop style */
161 *d++ = (U8)(( uv >> 6) | 0xc0);
162 *d++ = (U8)(( uv & 0x3f) | 0x80);
166 *d++ = (U8)(( uv >> 12) | 0xe0);
167 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
168 *d++ = (U8)(( uv & 0x3f) | 0x80);
172 *d++ = (U8)(( uv >> 18) | 0xf0);
173 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
174 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
175 *d++ = (U8)(( uv & 0x3f) | 0x80);
178 if (uv < 0x4000000) {
179 *d++ = (U8)(( uv >> 24) | 0xf8);
180 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
181 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
182 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
183 *d++ = (U8)(( uv & 0x3f) | 0x80);
186 if (uv < 0x80000000) {
187 *d++ = (U8)(( uv >> 30) | 0xfc);
188 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
189 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
191 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
192 *d++ = (U8)(( uv & 0x3f) | 0x80);
196 if (uv < UTF8_QUAD_MAX)
199 *d++ = 0xfe; /* Can't match U+FEFF! */
200 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
205 *d++ = (U8)(( uv & 0x3f) | 0x80);
210 *d++ = 0xff; /* Can't match U+FFFE! */
211 *d++ = 0x80; /* 6 Reserved bits */
212 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
213 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
221 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
222 *d++ = (U8)(( uv & 0x3f) | 0x80);
226 #endif /* Non loop style */
229 =for apidoc uvchr_to_utf8
231 Adds the UTF-8 representation of the native code point C<uv> to the end
232 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
233 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
234 the byte after the end of the new character. In other words,
236 d = uvchr_to_utf8(d, uv);
238 is the recommended wide native character-aware way of saying
242 This function accepts any UV as input. To forbid or warn on non-Unicode code
243 points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
248 /* This is also a macro */
249 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
252 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
254 return uvchr_to_utf8(d, uv);
258 =for apidoc uvchr_to_utf8_flags
260 Adds the UTF-8 representation of the native code point C<uv> to the end
261 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
262 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
263 the byte after the end of the new character. In other words,
265 d = uvchr_to_utf8_flags(d, uv, flags);
269 d = uvchr_to_utf8_flags(d, uv, 0);
271 This is the Unicode-aware way of saying
275 This function will convert to UTF-8 (and not warn) even code points that aren't
276 legal Unicode or are problematic, unless C<flags> contains one or more of the
279 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
280 the function will raise a warning, provided UTF8 warnings are enabled. If instead
281 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
282 If both flags are set, the function will both warn and return NULL.
284 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
285 affect how the function handles a Unicode non-character. And likewise, the
286 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
288 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
289 even less portable) can be warned and/or disallowed even if other above-Unicode
290 code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
293 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
294 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
300 /* This is also a macro */
301 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
304 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
306 return uvchr_to_utf8_flags(d, uv, flags);
310 =for apidoc is_utf8_char_buf
312 This is identical to the macro L</isUTF8_CHAR>.
317 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
320 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
322 return isUTF8_CHAR(buf, buf_end);
326 =for apidoc is_utf8_string
328 Returns true if the first C<len> bytes of string C<s> form a valid
329 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
330 using C<strlen(s)> (which means if you use this option, that C<s> can't have
331 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
332 that all characters being ASCII constitute 'a valid UTF-8 string'.
334 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
340 Perl_is_utf8_string(const U8 *s, STRLEN len)
342 const U8* const send = s + (len ? len : strlen((const char *)s));
345 PERL_ARGS_ASSERT_IS_UTF8_STRING;
348 STRLEN len = isUTF8_CHAR(x, send);
349 if (UNLIKELY(! len)) {
359 Implemented as a macro in utf8.h
361 =for apidoc is_utf8_string_loc
363 Like L</is_utf8_string> but stores the location of the failure (in the
364 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
365 "utf8ness success") in the C<ep>.
367 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
369 =for apidoc is_utf8_string_loclen
371 Like L</is_utf8_string>() but stores the location of the failure (in the
372 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
373 "utf8ness success") in the C<ep>, and the number of UTF-8
374 encoded characters in the C<el>.
376 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
382 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
384 const U8* const send = s + (len ? len : strlen((const char *)s));
388 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
391 STRLEN len = isUTF8_CHAR(x, send);
392 if (UNLIKELY(! len)) {
410 =for apidoc utf8n_to_uvchr
412 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
413 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
415 Bottom level UTF-8 decode routine.
416 Returns the native code point value of the first character in the string C<s>,
417 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
418 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
419 the length, in bytes, of that character.
421 The value of C<flags> determines the behavior when C<s> does not point to a
422 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
423 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
424 next possible position in C<s> that could begin a non-malformed character.
425 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
427 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
428 individual types of malformations, such as the sequence being overlong (that
429 is, when there is a shorter sequence that can express the same code point;
430 overlong sequences are expressly forbidden in the UTF-8 standard due to
431 potential security issues). Another malformation example is the first byte of
432 a character not being a legal first byte. See F<utf8.h> for the list of such
433 flags. For allowed 0 length strings, this function returns 0; for allowed
434 overlong sequences, the computed code point is returned; for all other allowed
435 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
436 determinable reasonable value.
438 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
439 flags) malformation is found. If this flag is set, the routine assumes that
440 the caller will raise a warning, and this function will silently just set
441 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
443 Note that this API requires disambiguation between successful decoding a C<NUL>
444 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
445 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
446 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
449 Certain code points are considered problematic. These are Unicode surrogates,
450 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
451 By default these are considered regular code points, but certain situations
452 warrant special handling for them. If C<flags> contains
453 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
454 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
455 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
456 maximum) can be set to disallow these categories individually.
458 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
459 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
460 for their respective categories, but otherwise the code points are considered
461 valid (not malformations). To get a category to both be treated as a
462 malformation and raise a warning, specify both the WARN and DISALLOW flags.
463 (But note that warnings are not raised if lexically disabled nor if
464 UTF8_CHECK_ONLY is also specified.)
466 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
467 the others that are above the Unicode legal maximum. There are several
468 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
469 not representable at all on EBCDIC platforms, and the original UTF-8
470 specification never went above this number (the current 0x10FFFF limit was
471 imposed later). (The smaller ones, those that fit into 32 bits, are
472 representable by a UV on ASCII platforms, but not by an IV, which means that
473 the number of operations that can be performed on them is quite restricted.)
474 The UTF-8 encoding on ASCII platforms for these large code points begins with a
475 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
476 be treated as malformations, while allowing smaller above-Unicode code points.
477 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
478 including these, as malformations.)
479 Similarly, UTF8_WARN_FE_FF acts just like
480 the other WARN flags, but applies just to these code points.
482 All other code points corresponding to Unicode characters, including private
483 use and those yet to be assigned, are never considered malformed and never
490 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
493 const U8 * const s0 = s;
494 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
499 UV outlier_ret = 0; /* return value when input is in error or problematic
501 UV pack_warn = 0; /* Save result of packWARN() for later */
502 bool unexpected_non_continuation = FALSE;
503 bool overflowed = FALSE;
504 bool do_overlong_test = TRUE; /* May have to skip this test */
506 const char* const malformed_text = "Malformed UTF-8 character";
508 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
510 /* The order of malformation tests here is important. We should consume as
511 * few bytes as possible in order to not skip any valid character. This is
512 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
513 * http://unicode.org/reports/tr36 for more discussion as to why. For
514 * example, once we've done a UTF8SKIP, we can tell the expected number of
515 * bytes, and could fail right off the bat if the input parameters indicate
516 * that there are too few available. But it could be that just that first
517 * byte is garbled, and the intended character occupies fewer bytes. If we
518 * blindly assumed that the first byte is correct, and skipped based on
519 * that number, we could skip over a valid input character. So instead, we
520 * always examine the sequence byte-by-byte.
522 * We also should not consume too few bytes, otherwise someone could inject
523 * things. For example, an input could be deliberately designed to
524 * overflow, and if this code bailed out immediately upon discovering that,
525 * returning to the caller C<*retlen> pointing to the very next byte (one
526 * which is actually part of of the overflowing sequence), that could look
527 * legitimate to the caller, which could discard the initial partial
528 * sequence and process the rest, inappropriately */
530 /* Zero length strings, if allowed, of necessity are zero */
531 if (UNLIKELY(curlen == 0)) {
536 if (flags & UTF8_ALLOW_EMPTY) {
539 if (! (flags & UTF8_CHECK_ONLY)) {
540 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
545 expectlen = UTF8SKIP(s);
547 /* A well-formed UTF-8 character, as the vast majority of calls to this
548 * function will be for, has this expected length. For efficiency, set
549 * things up here to return it. It will be overriden only in those rare
550 * cases where a malformation is found */
555 /* An invariant is trivially well-formed */
556 if (UTF8_IS_INVARIANT(uv)) {
560 /* A continuation character can't start a valid sequence */
561 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
562 if (flags & UTF8_ALLOW_CONTINUATION) {
566 return UNICODE_REPLACEMENT;
569 if (! (flags & UTF8_CHECK_ONLY)) {
570 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
576 /* Here is not a continuation byte, nor an invariant. The only thing left
577 * is a start byte (possibly for an overlong) */
580 uv = NATIVE_UTF8_TO_I8(uv);
583 /* Remove the leading bits that indicate the number of bytes in the
584 * character's whole UTF-8 sequence, leaving just the bits that are part of
586 uv &= UTF_START_MASK(expectlen);
588 /* Now, loop through the remaining bytes in the character's sequence,
589 * accumulating each into the working value as we go. Be sure to not look
590 * past the end of the input string */
591 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
593 for (s = s0 + 1; s < send; s++) {
594 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
595 #ifndef EBCDIC /* Can't overflow in EBCDIC */
596 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
598 /* The original implementors viewed this malformation as more
599 * serious than the others (though I, khw, don't understand
600 * why, since other malformations also give very very wrong
601 * results), so there is no way to turn off checking for it.
602 * Set a flag, but keep going in the loop, so that we absorb
603 * the rest of the bytes that comprise the character. */
605 overflow_byte = *s; /* Save for warning message's use */
608 uv = UTF8_ACCUMULATE(uv, *s);
611 /* Here, found a non-continuation before processing all expected
612 * bytes. This byte begins a new character, so quit, even if
613 * allowing this malformation. */
614 unexpected_non_continuation = TRUE;
617 } /* End of loop through the character's bytes */
619 /* Save how many bytes were actually in the character */
622 /* The loop above finds two types of malformations: non-continuation and/or
623 * overflow. The non-continuation malformation is really a too-short
624 * malformation, as it means that the current character ended before it was
625 * expected to (being terminated prematurely by the beginning of the next
626 * character, whereas in the too-short malformation there just are too few
627 * bytes available to hold the character. In both cases, the check below
628 * that we have found the expected number of bytes would fail if executed.)
629 * Thus the non-continuation malformation is really unnecessary, being a
630 * subset of the too-short malformation. But there may be existing
631 * applications that are expecting the non-continuation type, so we retain
632 * it, and return it in preference to the too-short malformation. (If this
633 * code were being written from scratch, the two types might be collapsed
634 * into one.) I, khw, am also giving priority to returning the
635 * non-continuation and too-short malformations over overflow when multiple
636 * ones are present. I don't know of any real reason to prefer one over
637 * the other, except that it seems to me that multiple-byte errors trumps
638 * errors from a single byte */
639 if (UNLIKELY(unexpected_non_continuation)) {
640 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
641 if (! (flags & UTF8_CHECK_ONLY)) {
643 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
646 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));
651 uv = UNICODE_REPLACEMENT;
653 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
654 * as what the original expectations were. */
655 do_overlong_test = FALSE;
660 else if (UNLIKELY(curlen < expectlen)) {
661 if (! (flags & UTF8_ALLOW_SHORT)) {
662 if (! (flags & UTF8_CHECK_ONLY)) {
663 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));
667 uv = UNICODE_REPLACEMENT;
668 do_overlong_test = FALSE;
674 #ifndef EBCDIC /* EBCDIC can't overflow */
675 if (UNLIKELY(overflowed)) {
676 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
682 && expectlen > (STRLEN) OFFUNISKIP(uv)
683 && ! (flags & UTF8_ALLOW_LONG))
685 /* The overlong malformation has lower precedence than the others.
686 * Note that if this malformation is allowed, we return the actual
687 * value, instead of the replacement character. This is because this
688 * value is actually well-defined. */
689 if (! (flags & UTF8_CHECK_ONLY)) {
690 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", OFFUNISKIP(uv), *s0));
695 /* Here, the input is considered to be well-formed, but it still could be a
696 * problematic code point that is not allowed by the input parameters. */
697 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
698 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
699 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
701 if (UNICODE_IS_SURROGATE(uv)) {
703 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
704 * generation of the sv, since no warnings are raised under CHECK */
705 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
706 && ckWARN_d(WARN_SURROGATE))
708 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
709 pack_warn = packWARN(WARN_SURROGATE);
711 if (flags & UTF8_DISALLOW_SURROGATE) {
715 else if ((uv > PERL_UNICODE_MAX)) {
716 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
717 && ckWARN_d(WARN_NON_UNICODE))
719 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
720 pack_warn = packWARN(WARN_NON_UNICODE);
722 #ifndef EBCDIC /* EBCDIC always allows FE, FF */
724 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
725 * points. We test for these after the regular SUPER ones, and
726 * before possibly bailing out, so that the more dire warning
727 * overrides the regular one, if applicable */
728 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
729 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
731 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
733 && ckWARN_d(WARN_UTF8))
735 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
736 pack_warn = packWARN(WARN_UTF8);
738 if (flags & UTF8_DISALLOW_FE_FF) {
743 if (flags & UTF8_DISALLOW_SUPER) {
747 else if (UNICODE_IS_NONCHAR(uv)) {
748 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
749 && ckWARN_d(WARN_NONCHAR))
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
752 pack_warn = packWARN(WARN_NONCHAR);
754 if (flags & UTF8_DISALLOW_NONCHAR) {
760 outlier_ret = uv; /* Note we don't bother to convert to native,
761 as all the outlier code points are the same
762 in both ASCII and EBCDIC */
766 /* Here, this is not considered a malformed character, so drop through
770 return UNI_TO_NATIVE(uv);
772 /* There are three cases which get to beyond this point. In all 3 cases:
773 * <sv> if not null points to a string to print as a warning.
774 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
776 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
777 * This is done by initializing it to 0, and changing it only
780 * 1) The input is valid but problematic, and to be warned about. The
781 * return value is the resultant code point; <*retlen> is set to
782 * <curlen>, the number of bytes that comprise the code point.
783 * <pack_warn> contains the result of packWARN() for the warning
784 * types. The entry point for this case is the label <do_warn>;
785 * 2) The input is a valid code point but disallowed by the parameters to
786 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
787 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
788 * comprise the code point. <pack_warn> contains the result of
789 * packWARN() for the warning types. The entry point for this case is
790 * the label <disallowed>.
791 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
792 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
793 * bytes that comprise the malformation. All such malformations are
794 * assumed to be warning type <utf8>. The entry point for this case
795 * is the label <malformed>.
800 if (sv && ckWARN_d(WARN_UTF8)) {
801 pack_warn = packWARN(WARN_UTF8);
806 if (flags & UTF8_CHECK_ONLY) {
808 *retlen = ((STRLEN) -1);
814 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
815 if warnings are to be raised. */
816 const char * const string = SvPVX_const(sv);
819 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
821 Perl_warner(aTHX_ pack_warn, "%s", string);
832 =for apidoc utf8_to_uvchr_buf
834 Returns the native code point of the first character in the string C<s> which
835 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
836 C<*retlen> will be set to the length, in bytes, of that character.
838 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
839 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
840 NULL) to -1. If those warnings are off, the computed value, if well-defined
841 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
842 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
843 the next possible position in C<s> that could begin a non-malformed character.
844 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
852 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
856 return utf8n_to_uvchr(s, send - s, retlen,
857 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
860 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
861 * there are no malformations in the input UTF-8 string C<s>. surrogates,
862 * non-character code points, and non-Unicode code points are allowed. */
865 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
867 UV expectlen = UTF8SKIP(s);
868 const U8* send = s + expectlen;
871 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
877 /* An invariant is trivially returned */
878 if (expectlen == 1) {
883 uv = NATIVE_UTF8_TO_I8(uv);
886 /* Remove the leading bits that indicate the number of bytes, leaving just
887 * the bits that are part of the value */
888 uv &= UTF_START_MASK(expectlen);
890 /* Now, loop through the remaining bytes, accumulating each into the
891 * working total as we go. (I khw tried unrolling the loop for up to 4
892 * bytes, but there was no performance improvement) */
893 for (++s; s < send; s++) {
894 uv = UTF8_ACCUMULATE(uv, *s);
897 return UNI_TO_NATIVE(uv);
902 =for apidoc utf8_to_uvuni_buf
904 Only in very rare circumstances should code need to be dealing in Unicode
905 (as opposed to native) code points. In those few cases, use
906 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
908 Returns the Unicode (not-native) code point of the first character in the
910 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
911 C<retlen> will be set to the length, in bytes, of that character.
913 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
914 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
915 NULL) to -1. If those warnings are off, the computed value if well-defined (or
916 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
917 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
918 next possible position in C<s> that could begin a non-malformed character.
919 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
925 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
927 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
931 /* Call the low level routine asking for checks */
932 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
933 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
937 =for apidoc utf8_length
939 Return the length of the UTF-8 char encoded string C<s> in characters.
940 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
941 up past C<e>, croaks.
947 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
952 PERL_ARGS_ASSERT_UTF8_LENGTH;
954 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
955 * the bitops (especially ~) can create illegal UTF-8.
956 * In other words: in Perl UTF-8 is not just for Unicode. */
959 goto warn_and_return;
969 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
970 "%s in %s", unees, OP_DESC(PL_op));
972 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
979 =for apidoc utf8_distance
981 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
984 WARNING: use only if you *know* that the pointers point inside the
991 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
993 PERL_ARGS_ASSERT_UTF8_DISTANCE;
995 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1001 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1002 forward or backward.
1004 WARNING: do not use the following unless you *know* C<off> is within
1005 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1006 on the first byte of character or just after the last byte of a character.
1012 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1014 PERL_ARGS_ASSERT_UTF8_HOP;
1016 PERL_UNUSED_CONTEXT;
1017 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1018 * the bitops (especially ~) can create illegal UTF-8.
1019 * In other words: in Perl UTF-8 is not just for Unicode. */
1028 while (UTF8_IS_CONTINUATION(*s))
1036 =for apidoc bytes_cmp_utf8
1038 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1039 sequence of characters (stored as UTF-8)
1040 in C<u>, C<ulen>. Returns 0 if they are
1041 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1042 if the first string is greater than the second string.
1044 -1 or +1 is returned if the shorter string was identical to the start of the
1045 longer string. -2 or +2 is returned if
1046 there was a difference between characters
1053 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1055 const U8 *const bend = b + blen;
1056 const U8 *const uend = u + ulen;
1058 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1060 PERL_UNUSED_CONTEXT;
1062 while (b < bend && u < uend) {
1064 if (!UTF8_IS_INVARIANT(c)) {
1065 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1068 if (UTF8_IS_CONTINUATION(c1)) {
1069 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1071 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1072 "Malformed UTF-8 character "
1073 "(unexpected non-continuation byte 0x%02x"
1074 ", immediately after start byte 0x%02x)"
1075 /* Dear diag.t, it's in the pod. */
1077 PL_op ? " in " : "",
1078 PL_op ? OP_DESC(PL_op) : "");
1083 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1084 "%s in %s", unees, OP_DESC(PL_op));
1086 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1087 return -2; /* Really want to return undef :-) */
1094 return *b < c ? -2 : +2;
1099 if (b == bend && u == uend)
1102 return b < bend ? +1 : -1;
1106 =for apidoc utf8_to_bytes
1108 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1109 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1110 updates C<len> to contain the new length.
1111 Returns zero on failure, setting C<len> to -1.
1113 If you need a copy of the string, see L</bytes_from_utf8>.
1119 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1121 U8 * const save = s;
1122 U8 * const send = s + *len;
1125 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1127 /* ensure valid UTF-8 and chars < 256 before updating string */
1129 if (! UTF8_IS_INVARIANT(*s)) {
1130 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1131 *len = ((STRLEN) -1);
1142 if (! UTF8_IS_INVARIANT(c)) {
1143 /* Then it is two-byte encoded */
1144 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1155 =for apidoc bytes_from_utf8
1157 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1158 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1159 the newly-created string, and updates C<len> to contain the new
1160 length. Returns the original string if no conversion occurs, C<len>
1161 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1162 0 if C<s> is converted or consisted entirely of characters that are invariant
1163 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1169 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1172 const U8 *start = s;
1176 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1178 PERL_UNUSED_CONTEXT;
1182 /* ensure valid UTF-8 and chars < 256 before converting string */
1183 for (send = s + *len; s < send;) {
1184 if (! UTF8_IS_INVARIANT(*s)) {
1185 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1196 Newx(d, (*len) - count + 1, U8);
1197 s = start; start = d;
1200 if (! UTF8_IS_INVARIANT(c)) {
1201 /* Then it is two-byte encoded */
1202 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1213 =for apidoc bytes_to_utf8
1215 Converts a string C<s> of length C<len> bytes from the native encoding into
1217 Returns a pointer to the newly-created string, and sets C<len> to
1218 reflect the new length in bytes.
1220 A C<NUL> character will be written after the end of the string.
1222 If you want to convert to UTF-8 from encodings other than
1223 the native (Latin1 or EBCDIC),
1224 see L</sv_recode_to_utf8>().
1229 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1230 likewise need duplication. */
1233 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1235 const U8 * const send = s + (*len);
1239 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1240 PERL_UNUSED_CONTEXT;
1242 Newx(d, (*len) * 2 + 1, U8);
1246 append_utf8_from_native_byte(*s, &d);
1255 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1257 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1258 * We optimize for native, for obvious reasons. */
1261 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1266 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1269 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1274 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1276 if (UNI_IS_INVARIANT(uv)) {
1277 *d++ = LATIN1_TO_NATIVE((U8) uv);
1280 if (uv <= MAX_UTF8_TWO_BYTE) {
1281 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1282 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1285 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1286 #define LAST_HIGH_SURROGATE 0xDBFF
1287 #define FIRST_LOW_SURROGATE 0xDC00
1288 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1289 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1291 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1293 UV low = (p[0] << 8) + p[1];
1295 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1296 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1297 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1298 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1300 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1301 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1304 d = uvoffuni_to_utf8_flags(d, uv, 0);
1307 *d++ = (U8)(( uv >> 12) | 0xe0);
1308 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1309 *d++ = (U8)(( uv & 0x3f) | 0x80);
1313 *d++ = (U8)(( uv >> 18) | 0xf0);
1314 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1315 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1316 *d++ = (U8)(( uv & 0x3f) | 0x80);
1321 *newlen = d - dstart;
1325 /* Note: this one is slightly destructive of the source. */
1328 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1331 U8* const send = s + bytelen;
1333 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1336 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1340 const U8 tmp = s[0];
1345 return utf16_to_utf8(p, d, bytelen, newlen);
1349 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1351 U8 tmpbuf[UTF8_MAXBYTES+1];
1352 uvchr_to_utf8(tmpbuf, c);
1353 return _is_utf8_FOO(classnum, tmpbuf);
1356 /* Internal function so we can deprecate the external one, and call
1357 this one from other deprecated functions in this file */
1360 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1363 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1367 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1371 Perl__is_uni_perl_idcont(pTHX_ UV c)
1373 U8 tmpbuf[UTF8_MAXBYTES+1];
1374 uvchr_to_utf8(tmpbuf, c);
1375 return _is_utf8_perl_idcont(tmpbuf);
1379 Perl__is_uni_perl_idstart(pTHX_ UV c)
1381 U8 tmpbuf[UTF8_MAXBYTES+1];
1382 uvchr_to_utf8(tmpbuf, c);
1383 return _is_utf8_perl_idstart(tmpbuf);
1387 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1389 /* We have the latin1-range values compiled into the core, so just use
1390 * those, converting the result to utf8. The only difference between upper
1391 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1392 * either "SS" or "Ss". Which one to use is passed into the routine in
1393 * 'S_or_s' to avoid a test */
1395 UV converted = toUPPER_LATIN1_MOD(c);
1397 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1399 assert(S_or_s == 'S' || S_or_s == 's');
1401 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1402 characters in this range */
1403 *p = (U8) converted;
1408 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1409 * which it maps to one of them, so as to only have to have one check for
1410 * it in the main case */
1411 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1413 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1414 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1417 converted = GREEK_CAPITAL_LETTER_MU;
1419 case LATIN_SMALL_LETTER_SHARP_S:
1425 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1426 assert(0); /* NOTREACHED */
1430 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1431 *p = UTF8_TWO_BYTE_LO(converted);
1437 /* Call the function to convert a UTF-8 encoded character to the specified case.
1438 * Note that there may be more than one character in the result.
1439 * INP is a pointer to the first byte of the input character
1440 * OUTP will be set to the first byte of the string of changed characters. It
1441 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1442 * LENP will be set to the length in bytes of the string of changed characters
1444 * The functions return the ordinal of the first character in the string of OUTP */
1445 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1446 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1447 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1449 /* This additionally has the input parameter SPECIALS, which if non-zero will
1450 * cause this to use the SPECIALS hash for folding (meaning get full case
1451 * folding); otherwise, when zero, this implies a simple case fold */
1452 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1455 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1459 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1460 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1461 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1462 * the changed version may be longer than the original character.
1464 * The ordinal of the first character of the changed version is returned
1465 * (but note, as explained above, that there may be more.) */
1467 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1470 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1473 uvchr_to_utf8(p, c);
1474 return CALL_UPPER_CASE(p, p, lenp);
1478 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1482 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1485 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1488 uvchr_to_utf8(p, c);
1489 return CALL_TITLE_CASE(p, p, lenp);
1493 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1495 /* We have the latin1-range values compiled into the core, so just use
1496 * those, converting the result to utf8. Since the result is always just
1497 * one character, we allow <p> to be NULL */
1499 U8 converted = toLOWER_LATIN1(c);
1502 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1507 *p = UTF8_TWO_BYTE_HI(converted);
1508 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1516 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1520 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1523 return to_lower_latin1((U8) c, p, lenp);
1526 uvchr_to_utf8(p, c);
1527 return CALL_LOWER_CASE(p, p, lenp);
1531 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1533 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1534 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1535 * FOLD_FLAGS_FULL iff full folding is to be used;
1537 * Not to be used for locale folds
1542 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1544 assert (! (flags & FOLD_FLAGS_LOCALE));
1546 if (c == MICRO_SIGN) {
1547 converted = GREEK_SMALL_LETTER_MU;
1549 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1551 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1552 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1553 * under those circumstances. */
1554 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1555 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1556 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1558 return LATIN_SMALL_LETTER_LONG_S;
1567 else { /* In this range the fold of all other characters is their lower
1569 converted = toLOWER_LATIN1(c);
1572 if (UVCHR_IS_INVARIANT(converted)) {
1573 *p = (U8) converted;
1577 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1578 *p = UTF8_TWO_BYTE_LO(converted);
1586 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1589 /* Not currently externally documented, and subject to change
1590 * <flags> bits meanings:
1591 * FOLD_FLAGS_FULL iff full folding is to be used;
1592 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1593 * locale are to be used.
1594 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1597 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1599 /* Tread a UTF-8 locale as not being in locale at all */
1600 if (IN_UTF8_CTYPE_LOCALE) {
1601 flags &= ~FOLD_FLAGS_LOCALE;
1605 UV result = _to_fold_latin1((U8) c, p, lenp,
1606 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1607 /* It is illegal for the fold to cross the 255/256 boundary under
1608 * locale; in this case return the original */
1609 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1614 /* If no special needs, just use the macro */
1615 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1616 uvchr_to_utf8(p, c);
1617 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1619 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1620 the special flags. */
1621 U8 utf8_c[UTF8_MAXBYTES + 1];
1622 uvchr_to_utf8(utf8_c, c);
1623 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1627 PERL_STATIC_INLINE bool
1628 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1629 const char *const swashname, SV* const invlist)
1631 /* returns a boolean giving whether or not the UTF8-encoded character that
1632 * starts at <p> is in the swash indicated by <swashname>. <swash>
1633 * contains a pointer to where the swash indicated by <swashname>
1634 * is to be stored; which this routine will do, so that future calls will
1635 * look at <*swash> and only generate a swash if it is not null. <invlist>
1636 * is NULL or an inversion list that defines the swash. If not null, it
1637 * saves time during initialization of the swash.
1639 * Note that it is assumed that the buffer length of <p> is enough to
1640 * contain all the bytes that comprise the character. Thus, <*p> should
1641 * have been checked before this call for mal-formedness enough to assure
1646 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1648 /* The API should have included a length for the UTF-8 character in <p>,
1649 * but it doesn't. We therefore assume that p has been validated at least
1650 * as far as there being enough bytes available in it to accommodate the
1651 * character without reading beyond the end, and pass that number on to the
1652 * validating routine */
1653 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1654 if (ckWARN_d(WARN_UTF8)) {
1655 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1656 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1657 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1658 what the malformation is */
1659 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1665 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1666 *swash = _core_swash_init("utf8",
1668 /* Only use the name if there is no inversion
1669 * list; otherwise will go out to disk */
1670 (invlist) ? "" : swashname,
1672 &PL_sv_undef, 1, 0, invlist, &flags);
1675 return swash_fetch(*swash, p, TRUE) != 0;
1679 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1683 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1685 assert(classnum < _FIRST_NON_SWASH_CC);
1687 return is_utf8_common(p,
1688 &PL_utf8_swash_ptrs[classnum],
1689 swash_property_names[classnum],
1690 PL_XPosix_ptrs[classnum]);
1694 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1699 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1701 if (! PL_utf8_perl_idstart) {
1702 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1704 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
1708 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1712 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1716 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1720 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1725 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1727 if (! PL_utf8_perl_idcont) {
1728 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1730 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
1734 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1738 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1740 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1744 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1748 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1750 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1754 Perl__is_utf8_mark(pTHX_ const U8 *p)
1758 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1760 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1764 =for apidoc to_utf8_case
1766 C<p> contains the pointer to the UTF-8 string encoding
1767 the character that is being converted. This routine assumes that the character
1768 at C<p> is well-formed.
1770 C<ustrp> is a pointer to the character buffer to put the
1771 conversion result to. C<lenp> is a pointer to the length
1774 C<swashp> is a pointer to the swash to use.
1776 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1777 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
1778 but not always, a multicharacter mapping), is tried first.
1780 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1781 any special mappings; C<""> means to use the special mappings. Values other
1782 than these two are treated as the name of the hash containing the special
1783 mappings, like C<"utf8::ToSpecLower">.
1785 C<normal> is a string like "ToLower" which means the swash
1791 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1792 SV **swashp, const char *normal, const char *special)
1796 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1798 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1800 /* Note that swash_fetch() doesn't output warnings for these because it
1801 * assumes we will */
1802 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1803 if (uv1 <= UNICODE_SURROGATE_LAST) {
1804 if (ckWARN_d(WARN_SURROGATE)) {
1805 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1806 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1807 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1810 else if (UNICODE_IS_SUPER(uv1)) {
1811 if (ckWARN_d(WARN_NON_UNICODE)) {
1812 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1813 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1814 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1818 /* Note that non-characters are perfectly legal, so no warning should
1822 if (!*swashp) /* load on-demand */
1823 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1826 /* It might be "special" (sometimes, but not always,
1827 * a multicharacter mapping) */
1831 /* If passed in the specials name, use that; otherwise use any
1832 * given in the swash */
1833 if (*special != '\0') {
1834 hv = get_hv(special, 0);
1837 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1839 hv = MUTABLE_HV(SvRV(*svp));
1844 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1849 s = SvPV_const(*svp, len);
1852 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1854 Copy(s, ustrp, len, U8);
1859 if (!len && *swashp) {
1860 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
1863 /* It was "normal" (a single character mapping). */
1864 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1872 return valid_utf8_to_uvchr(ustrp, 0);
1875 /* Here, there was no mapping defined, which means that the code point maps
1876 * to itself. Return the inputs */
1878 if (p != ustrp) { /* Don't copy onto itself */
1879 Copy(p, ustrp, len, U8);
1890 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1892 /* This is called when changing the case of a utf8-encoded character above
1893 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1894 * result contains a character that crosses the 255/256 boundary, disallow
1895 * the change, and return the original code point. See L<perlfunc/lc> for
1898 * p points to the original string whose case was changed; assumed
1899 * by this routine to be well-formed
1900 * result the code point of the first character in the changed-case string
1901 * ustrp points to the changed-case string (<result> represents its first char)
1902 * lenp points to the length of <ustrp> */
1904 UV original; /* To store the first code point of <p> */
1906 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1908 assert(UTF8_IS_ABOVE_LATIN1(*p));
1910 /* We know immediately if the first character in the string crosses the
1911 * boundary, so can skip */
1914 /* Look at every character in the result; if any cross the
1915 * boundary, the whole thing is disallowed */
1916 U8* s = ustrp + UTF8SKIP(ustrp);
1917 U8* e = ustrp + *lenp;
1919 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
1925 /* Here, no characters crossed, result is ok as-is */
1931 /* Failed, have to return the original */
1932 original = valid_utf8_to_uvchr(p, lenp);
1933 Copy(p, ustrp, *lenp, char);
1938 =for apidoc to_utf8_upper
1940 Instead use L</toUPPER_utf8>.
1944 /* Not currently externally documented, and subject to change:
1945 * <flags> is set iff iff the rules from the current underlying locale are to
1949 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
1955 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
1957 if (flags && IN_UTF8_CTYPE_LOCALE) {
1961 if (UTF8_IS_INVARIANT(*p)) {
1963 result = toUPPER_LC(*p);
1966 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
1969 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
1971 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
1972 result = toUPPER_LC(c);
1975 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
1979 else { /* utf8, ord above 255 */
1980 result = CALL_UPPER_CASE(p, ustrp, lenp);
1983 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
1988 /* Here, used locale rules. Convert back to utf8 */
1989 if (UTF8_IS_INVARIANT(result)) {
1990 *ustrp = (U8) result;
1994 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1995 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2003 =for apidoc to_utf8_title
2005 Instead use L</toTITLE_utf8>.
2009 /* Not currently externally documented, and subject to change:
2010 * <flags> is set iff the rules from the current underlying locale are to be
2011 * used. Since titlecase is not defined in POSIX, for other than a
2012 * UTF-8 locale, uppercase is used instead for code points < 256.
2016 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2022 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2024 if (flags && IN_UTF8_CTYPE_LOCALE) {
2028 if (UTF8_IS_INVARIANT(*p)) {
2030 result = toUPPER_LC(*p);
2033 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2036 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2038 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2039 result = toUPPER_LC(c);
2042 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2046 else { /* utf8, ord above 255 */
2047 result = CALL_TITLE_CASE(p, ustrp, lenp);
2050 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2055 /* Here, used locale rules. Convert back to utf8 */
2056 if (UTF8_IS_INVARIANT(result)) {
2057 *ustrp = (U8) result;
2061 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2062 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2070 =for apidoc to_utf8_lower
2072 Instead use L</toLOWER_utf8>.
2076 /* Not currently externally documented, and subject to change:
2077 * <flags> is set iff iff the rules from the current underlying locale are to
2082 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2088 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2090 if (flags && IN_UTF8_CTYPE_LOCALE) {
2094 if (UTF8_IS_INVARIANT(*p)) {
2096 result = toLOWER_LC(*p);
2099 return to_lower_latin1(*p, ustrp, lenp);
2102 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2104 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2105 result = toLOWER_LC(c);
2108 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2112 else { /* utf8, ord above 255 */
2113 result = CALL_LOWER_CASE(p, ustrp, lenp);
2116 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2122 /* Here, used locale rules. Convert back to utf8 */
2123 if (UTF8_IS_INVARIANT(result)) {
2124 *ustrp = (U8) result;
2128 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2129 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2137 =for apidoc to_utf8_fold
2139 Instead use L</toFOLD_utf8>.
2143 /* Not currently externally documented, and subject to change,
2145 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2146 * locale are to be used.
2147 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2148 * otherwise simple folds
2149 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2154 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2160 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2162 /* These are mutually exclusive */
2163 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2165 assert(p != ustrp); /* Otherwise overwrites */
2167 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2168 flags &= ~FOLD_FLAGS_LOCALE;
2171 if (UTF8_IS_INVARIANT(*p)) {
2172 if (flags & FOLD_FLAGS_LOCALE) {
2173 result = toFOLD_LC(*p);
2176 return _to_fold_latin1(*p, ustrp, lenp,
2177 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2180 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2181 if (flags & FOLD_FLAGS_LOCALE) {
2182 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2183 result = toFOLD_LC(c);
2186 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2188 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2191 else { /* utf8, ord above 255 */
2192 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2194 if (flags & FOLD_FLAGS_LOCALE) {
2196 /* Special case these two characters, as what normally gets
2197 * returned under locale doesn't work */
2198 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2199 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2200 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2204 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2205 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2206 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2208 goto return_ligature_st;
2210 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2212 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2216 /* This is called when changing the case of a utf8-encoded
2217 * character above the ASCII range, and the result should not
2218 * contain an ASCII character. */
2220 UV original; /* To store the first code point of <p> */
2222 /* Look at every character in the result; if any cross the
2223 * boundary, the whole thing is disallowed */
2225 U8* e = ustrp + *lenp;
2228 /* Crossed, have to return the original */
2229 original = valid_utf8_to_uvchr(p, lenp);
2231 /* But in these instances, there is an alternative we can
2232 * return that is valid */
2233 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2234 || original == LATIN_SMALL_LETTER_SHARP_S)
2238 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2239 goto return_ligature_st;
2241 Copy(p, ustrp, *lenp, char);
2247 /* Here, no characters crossed, result is ok as-is */
2252 /* Here, used locale rules. Convert back to utf8 */
2253 if (UTF8_IS_INVARIANT(result)) {
2254 *ustrp = (U8) result;
2258 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2259 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2266 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2267 * folds to a string of two of these characters. By returning this
2268 * instead, then, e.g.,
2269 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2272 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2273 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2275 return LATIN_SMALL_LETTER_LONG_S;
2278 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2279 * have the other one fold to it */
2281 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2282 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2283 return LATIN_SMALL_LIGATURE_ST;
2287 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2288 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2289 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2293 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2295 PERL_ARGS_ASSERT_SWASH_INIT;
2297 /* Returns a copy of a swash initiated by the called function. This is the
2298 * public interface, and returning a copy prevents others from doing
2299 * mischief on the original */
2301 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2305 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2307 /* Initialize and return a swash, creating it if necessary. It does this
2308 * by calling utf8_heavy.pl in the general case. The returned value may be
2309 * the swash's inversion list instead if the input parameters allow it.
2310 * Which is returned should be immaterial to callers, as the only
2311 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2312 * and swash_to_invlist() handle both these transparently.
2314 * This interface should only be used by functions that won't destroy or
2315 * adversely change the swash, as doing so affects all other uses of the
2316 * swash in the program; the general public should use 'Perl_swash_init'
2319 * pkg is the name of the package that <name> should be in.
2320 * name is the name of the swash to find. Typically it is a Unicode
2321 * property name, including user-defined ones
2322 * listsv is a string to initialize the swash with. It must be of the form
2323 * documented as the subroutine return value in
2324 * L<perlunicode/User-Defined Character Properties>
2325 * minbits is the number of bits required to represent each data element.
2326 * It is '1' for binary properties.
2327 * none I (khw) do not understand this one, but it is used only in tr///.
2328 * invlist is an inversion list to initialize the swash with (or NULL)
2329 * flags_p if non-NULL is the address of various input and output flag bits
2330 * to the routine, as follows: ('I' means is input to the routine;
2331 * 'O' means output from the routine. Only flags marked O are
2332 * meaningful on return.)
2333 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2334 * came from a user-defined property. (I O)
2335 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2336 * when the swash cannot be located, to simply return NULL. (I)
2337 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2338 * return of an inversion list instead of a swash hash if this routine
2339 * thinks that would result in faster execution of swash_fetch() later
2342 * Thus there are three possible inputs to find the swash: <name>,
2343 * <listsv>, and <invlist>. At least one must be specified. The result
2344 * will be the union of the specified ones, although <listsv>'s various
2345 * actions can intersect, etc. what <name> gives. To avoid going out to
2346 * disk at all, <invlist> should specify completely what the swash should
2347 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2349 * <invlist> is only valid for binary properties */
2352 SV* retval = &PL_sv_undef;
2353 HV* swash_hv = NULL;
2354 const int invlist_swash_boundary =
2355 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2356 ? 512 /* Based on some benchmarking, but not extensive, see commit
2358 : -1; /* Never return just an inversion list */
2360 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2361 assert(! invlist || minbits == 1);
2363 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2365 if (listsv != &PL_sv_undef || strNE(name, "")) {
2367 const size_t pkg_len = strlen(pkg);
2368 const size_t name_len = strlen(name);
2369 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2373 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2375 PUSHSTACKi(PERLSI_MAGIC);
2379 /* We might get here via a subroutine signature which uses a utf8
2380 * parameter name, at which point PL_subname will have been set
2381 * but not yet used. */
2382 save_item(PL_subname);
2383 if (PL_parser && PL_parser->error_count)
2384 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2385 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2386 if (!method) { /* demand load utf8 */
2388 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2389 GvSV(PL_errgv) = NULL;
2390 /* It is assumed that callers of this routine are not passing in
2391 * any user derived data. */
2392 /* Need to do this after save_re_context() as it will set
2393 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2394 * in Perl_magic_get). Even line to create errsv_save can turn on
2396 #ifndef NO_TAINT_SUPPORT
2397 SAVEBOOL(TAINT_get);
2400 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2403 /* Not ERRSV, as there is no need to vivify a scalar we are
2404 about to discard. */
2405 SV * const errsv = GvSV(PL_errgv);
2406 if (!SvTRUE(errsv)) {
2407 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2408 SvREFCNT_dec(errsv);
2416 mPUSHp(pkg, pkg_len);
2417 mPUSHp(name, name_len);
2422 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2423 GvSV(PL_errgv) = NULL;
2424 /* If we already have a pointer to the method, no need to use
2425 * call_method() to repeat the lookup. */
2427 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2428 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2430 retval = *PL_stack_sp--;
2431 SvREFCNT_inc(retval);
2434 /* Not ERRSV. See above. */
2435 SV * const errsv = GvSV(PL_errgv);
2436 if (!SvTRUE(errsv)) {
2437 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2438 SvREFCNT_dec(errsv);
2443 if (IN_PERL_COMPILETIME) {
2444 CopHINTS_set(PL_curcop, PL_hints);
2446 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2449 /* If caller wants to handle missing properties, let them */
2450 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2454 "Can't find Unicode property definition \"%"SVf"\"",
2456 NOT_REACHED; /* NOTREACHED */
2458 } /* End of calling the module to find the swash */
2460 /* If this operation fetched a swash, and we will need it later, get it */
2461 if (retval != &PL_sv_undef
2462 && (minbits == 1 || (flags_p
2464 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2466 swash_hv = MUTABLE_HV(SvRV(retval));
2468 /* If we don't already know that there is a user-defined component to
2469 * this swash, and the user has indicated they wish to know if there is
2470 * one (by passing <flags_p>), find out */
2471 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2472 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2473 if (user_defined && SvUV(*user_defined)) {
2474 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2479 /* Make sure there is an inversion list for binary properties */
2481 SV** swash_invlistsvp = NULL;
2482 SV* swash_invlist = NULL;
2483 bool invlist_in_swash_is_valid = FALSE;
2484 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2485 an unclaimed reference count */
2487 /* If this operation fetched a swash, get its already existing
2488 * inversion list, or create one for it */
2491 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2492 if (swash_invlistsvp) {
2493 swash_invlist = *swash_invlistsvp;
2494 invlist_in_swash_is_valid = TRUE;
2497 swash_invlist = _swash_to_invlist(retval);
2498 swash_invlist_unclaimed = TRUE;
2502 /* If an inversion list was passed in, have to include it */
2505 /* Any fetched swash will by now have an inversion list in it;
2506 * otherwise <swash_invlist> will be NULL, indicating that we
2507 * didn't fetch a swash */
2508 if (swash_invlist) {
2510 /* Add the passed-in inversion list, which invalidates the one
2511 * already stored in the swash */
2512 invlist_in_swash_is_valid = FALSE;
2513 _invlist_union(invlist, swash_invlist, &swash_invlist);
2517 /* Here, there is no swash already. Set up a minimal one, if
2518 * we are going to return a swash */
2519 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2521 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2523 swash_invlist = invlist;
2527 /* Here, we have computed the union of all the passed-in data. It may
2528 * be that there was an inversion list in the swash which didn't get
2529 * touched; otherwise save the computed one */
2530 if (! invlist_in_swash_is_valid
2531 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2533 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2535 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2537 /* We just stole a reference count. */
2538 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2539 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2542 SvREADONLY_on(swash_invlist);
2544 /* Use the inversion list stand-alone if small enough */
2545 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2546 SvREFCNT_dec(retval);
2547 if (!swash_invlist_unclaimed)
2548 SvREFCNT_inc_simple_void_NN(swash_invlist);
2549 retval = newRV_noinc(swash_invlist);
2557 /* This API is wrong for special case conversions since we may need to
2558 * return several Unicode characters for a single Unicode character
2559 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2560 * the lower-level routine, and it is similarly broken for returning
2561 * multiple values. --jhi
2562 * For those, you should use to_utf8_case() instead */
2563 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2566 * Returns the value of property/mapping C<swash> for the first character
2567 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2568 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2569 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2571 * A "swash" is a hash which contains initially the keys/values set up by
2572 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2573 * property for all possible code points. Things are stored in a compact form
2574 * (see utf8_heavy.pl) so that calculation is required to find the actual
2575 * property value for a given code point. As code points are looked up, new
2576 * key/value pairs are added to the hash, so that the calculation doesn't have
2577 * to ever be re-done. Further, each calculation is done, not just for the
2578 * desired one, but for a whole block of code points adjacent to that one.
2579 * For binary properties on ASCII machines, the block is usually for 64 code
2580 * points, starting with a code point evenly divisible by 64. Thus if the
2581 * property value for code point 257 is requested, the code goes out and
2582 * calculates the property values for all 64 code points between 256 and 319,
2583 * and stores these as a single 64-bit long bit vector, called a "swatch",
2584 * under the key for code point 256. The key is the UTF-8 encoding for code
2585 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2586 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2587 * for code point 258 is then requested, this code realizes that it would be
2588 * stored under the key for 256, and would find that value and extract the
2589 * relevant bit, offset from 256.
2591 * Non-binary properties are stored in as many bits as necessary to represent
2592 * their values (32 currently, though the code is more general than that), not
2593 * as single bits, but the principal is the same: the value for each key is a
2594 * vector that encompasses the property values for all code points whose UTF-8
2595 * representations are represented by the key. That is, for all code points
2596 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2600 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2603 HV *const hv = MUTABLE_HV(SvRV(swash));
2608 const U8 *tmps = NULL;
2612 PERL_ARGS_ASSERT_SWASH_FETCH;
2614 /* If it really isn't a hash, it isn't really swash; must be an inversion
2616 if (SvTYPE(hv) != SVt_PVHV) {
2617 return _invlist_contains_cp((SV*)hv,
2619 ? valid_utf8_to_uvchr(ptr, NULL)
2623 /* We store the values in a "swatch" which is a vec() value in a swash
2624 * hash. Code points 0-255 are a single vec() stored with key length
2625 * (klen) 0. All other code points have a UTF-8 representation
2626 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2627 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2628 * length for them is the length of the encoded char - 1. ptr[klen] is the
2629 * final byte in the sequence representing the character */
2630 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2635 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2638 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
2641 klen = UTF8SKIP(ptr) - 1;
2643 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2644 * the vec is the final byte in the sequence. (In EBCDIC this is
2645 * converted to I8 to get consecutive values.) To help you visualize
2647 * Straight 1047 After final byte
2648 * UTF-8 UTF-EBCDIC I8 transform
2649 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2650 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2652 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2653 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2655 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2656 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2658 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2659 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2661 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2662 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2664 * (There are no discontinuities in the elided (...) entries.)
2665 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2666 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2667 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2668 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2669 * index into the vec() swatch (after subtracting 0x80, which we
2670 * actually do with an '&').
2671 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2672 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2673 * dicontinuities which go away by transforming it into I8, and we
2674 * effectively subtract 0xA0 to get the index. */
2675 needents = (1 << UTF_ACCUMULATION_SHIFT);
2676 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2680 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2681 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2682 * it's nothing to sniff at.) Pity we usually come through at least
2683 * two function calls to get here...
2685 * NB: this code assumes that swatches are never modified, once generated!
2688 if (hv == PL_last_swash_hv &&
2689 klen == PL_last_swash_klen &&
2690 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2692 tmps = PL_last_swash_tmps;
2693 slen = PL_last_swash_slen;
2696 /* Try our second-level swatch cache, kept in a hash. */
2697 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2699 /* If not cached, generate it via swatch_get */
2700 if (!svp || !SvPOK(*svp)
2701 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2704 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2705 swatch = swatch_get(swash,
2706 code_point & ~((UV)needents - 1),
2709 else { /* For the first 256 code points, the swatch has a key of
2711 swatch = swatch_get(swash, 0, needents);
2714 if (IN_PERL_COMPILETIME)
2715 CopHINTS_set(PL_curcop, PL_hints);
2717 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2719 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2720 || (slen << 3) < needents)
2721 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2722 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2723 svp, tmps, (UV)slen, (UV)needents);
2726 PL_last_swash_hv = hv;
2727 assert(klen <= sizeof(PL_last_swash_key));
2728 PL_last_swash_klen = (U8)klen;
2729 /* FIXME change interpvar.h? */
2730 PL_last_swash_tmps = (U8 *) tmps;
2731 PL_last_swash_slen = slen;
2733 Copy(ptr, PL_last_swash_key, klen, U8);
2736 switch ((int)((slen << 3) / needents)) {
2738 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2740 return ((UV) tmps[off]);
2744 ((UV) tmps[off ] << 8) +
2745 ((UV) tmps[off + 1]);
2749 ((UV) tmps[off ] << 24) +
2750 ((UV) tmps[off + 1] << 16) +
2751 ((UV) tmps[off + 2] << 8) +
2752 ((UV) tmps[off + 3]);
2754 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2755 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2756 NORETURN_FUNCTION_END;
2759 /* Read a single line of the main body of the swash input text. These are of
2762 * where each number is hex. The first two numbers form the minimum and
2763 * maximum of a range, and the third is the value associated with the range.
2764 * Not all swashes should have a third number
2766 * On input: l points to the beginning of the line to be examined; it points
2767 * to somewhere in the string of the whole input text, and is
2768 * terminated by a \n or the null string terminator.
2769 * lend points to the null terminator of that string
2770 * wants_value is non-zero if the swash expects a third number
2771 * typestr is the name of the swash's mapping, like 'ToLower'
2772 * On output: *min, *max, and *val are set to the values read from the line.
2773 * returns a pointer just beyond the line examined. If there was no
2774 * valid min number on the line, returns lend+1
2778 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2779 const bool wants_value, const U8* const typestr)
2781 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2782 STRLEN numlen; /* Length of the number */
2783 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2784 | PERL_SCAN_DISALLOW_PREFIX
2785 | PERL_SCAN_SILENT_NON_PORTABLE;
2787 /* nl points to the next \n in the scan */
2788 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2790 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2792 /* Get the first number on the line: the range minimum */
2794 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2795 *max = *min; /* So can never return without setting max */
2796 if (numlen) /* If found a hex number, position past it */
2798 else if (nl) { /* Else, go handle next line, if any */
2799 return nl + 1; /* 1 is length of "\n" */
2801 else { /* Else, no next line */
2802 return lend + 1; /* to LIST's end at which \n is not found */
2805 /* The max range value follows, separated by a BLANK */
2808 flags = PERL_SCAN_SILENT_ILLDIGIT
2809 | PERL_SCAN_DISALLOW_PREFIX
2810 | PERL_SCAN_SILENT_NON_PORTABLE;
2812 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2815 else /* If no value here, it is a single element range */
2818 /* Non-binary tables have a third entry: what the first element of the
2819 * range maps to. The map for those currently read here is in hex */
2823 flags = PERL_SCAN_SILENT_ILLDIGIT
2824 | PERL_SCAN_DISALLOW_PREFIX
2825 | PERL_SCAN_SILENT_NON_PORTABLE;
2827 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2836 /* diag_listed_as: To%s: illegal mapping '%s' */
2837 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2843 *val = 0; /* bits == 1, then any val should be ignored */
2845 else { /* Nothing following range min, should be single element with no
2850 /* diag_listed_as: To%s: illegal mapping '%s' */
2851 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2855 *val = 0; /* bits == 1, then val should be ignored */
2858 /* Position to next line if any, or EOF */
2868 * Returns a swatch (a bit vector string) for a code point sequence
2869 * that starts from the value C<start> and comprises the number C<span>.
2870 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2871 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2874 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
2877 U8 *l, *lend, *x, *xend, *s, *send;
2878 STRLEN lcur, xcur, scur;
2879 HV *const hv = MUTABLE_HV(SvRV(swash));
2880 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
2882 SV** listsvp = NULL; /* The string containing the main body of the table */
2883 SV** extssvp = NULL;
2884 SV** invert_it_svp = NULL;
2887 STRLEN octets; /* if bits == 1, then octets == 0 */
2889 UV end = start + span;
2891 if (invlistsvp == NULL) {
2892 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2893 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2894 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2895 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2896 listsvp = hv_fetchs(hv, "LIST", FALSE);
2897 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2899 bits = SvUV(*bitssvp);
2900 none = SvUV(*nonesvp);
2901 typestr = (U8*)SvPV_nolen(*typesvp);
2907 octets = bits >> 3; /* if bits == 1, then octets == 0 */
2909 PERL_ARGS_ASSERT_SWATCH_GET;
2911 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
2912 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
2916 /* If overflowed, use the max possible */
2922 /* create and initialize $swatch */
2923 scur = octets ? (span * octets) : (span + 7) / 8;
2924 swatch = newSV(scur);
2926 s = (U8*)SvPVX(swatch);
2927 if (octets && none) {
2928 const U8* const e = s + scur;
2931 *s++ = (U8)(none & 0xff);
2932 else if (bits == 16) {
2933 *s++ = (U8)((none >> 8) & 0xff);
2934 *s++ = (U8)( none & 0xff);
2936 else if (bits == 32) {
2937 *s++ = (U8)((none >> 24) & 0xff);
2938 *s++ = (U8)((none >> 16) & 0xff);
2939 *s++ = (U8)((none >> 8) & 0xff);
2940 *s++ = (U8)( none & 0xff);
2946 (void)memzero((U8*)s, scur + 1);
2948 SvCUR_set(swatch, scur);
2949 s = (U8*)SvPVX(swatch);
2951 if (invlistsvp) { /* If has an inversion list set up use that */
2952 _invlist_populate_swatch(*invlistsvp, start, end, s);
2956 /* read $swash->{LIST} */
2957 l = (U8*)SvPV(*listsvp, lcur);
2960 UV min, max, val, upper;
2961 l = swash_scan_list_line(l, lend, &min, &max, &val,
2962 cBOOL(octets), typestr);
2967 /* If looking for something beyond this range, go try the next one */
2971 /* <end> is generally 1 beyond where we want to set things, but at the
2972 * platform's infinity, where we can't go any higher, we want to
2973 * include the code point at <end> */
2976 : (max != UV_MAX || end != UV_MAX)
2983 if (!none || val < none) {
2988 for (key = min; key <= upper; key++) {
2990 /* offset must be non-negative (start <= min <= key < end) */
2991 offset = octets * (key - start);
2993 s[offset] = (U8)(val & 0xff);
2994 else if (bits == 16) {
2995 s[offset ] = (U8)((val >> 8) & 0xff);
2996 s[offset + 1] = (U8)( val & 0xff);
2998 else if (bits == 32) {
2999 s[offset ] = (U8)((val >> 24) & 0xff);
3000 s[offset + 1] = (U8)((val >> 16) & 0xff);
3001 s[offset + 2] = (U8)((val >> 8) & 0xff);
3002 s[offset + 3] = (U8)( val & 0xff);
3005 if (!none || val < none)
3009 else { /* bits == 1, then val should be ignored */
3014 for (key = min; key <= upper; key++) {
3015 const STRLEN offset = (STRLEN)(key - start);
3016 s[offset >> 3] |= 1 << (offset & 7);
3021 /* Invert if the data says it should be. Assumes that bits == 1 */
3022 if (invert_it_svp && SvUV(*invert_it_svp)) {
3024 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3025 * be 0, and their inversion should also be 0, as we don't succeed any
3026 * Unicode property matches for non-Unicode code points */
3027 if (start <= PERL_UNICODE_MAX) {
3029 /* The code below assumes that we never cross the
3030 * Unicode/above-Unicode boundary in a range, as otherwise we would
3031 * have to figure out where to stop flipping the bits. Since this
3032 * boundary is divisible by a large power of 2, and swatches comes
3033 * in small powers of 2, this should be a valid assumption */
3034 assert(start + span - 1 <= PERL_UNICODE_MAX);
3044 /* read $swash->{EXTRAS}
3045 * This code also copied to swash_to_invlist() below */
3046 x = (U8*)SvPV(*extssvp, xcur);
3054 SV **otherbitssvp, *other;
3058 const U8 opc = *x++;
3062 nl = (U8*)memchr(x, '\n', xend - x);
3064 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3066 x = nl + 1; /* 1 is length of "\n" */
3070 x = xend; /* to EXTRAS' end at which \n is not found */
3077 namelen = nl - namestr;
3081 namelen = xend - namestr;
3085 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3086 otherhv = MUTABLE_HV(SvRV(*othersvp));
3087 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3088 otherbits = (STRLEN)SvUV(*otherbitssvp);
3089 if (bits < otherbits)
3090 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3091 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3093 /* The "other" swatch must be destroyed after. */
3094 other = swatch_get(*othersvp, start, span);
3095 o = (U8*)SvPV(other, olen);
3098 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3100 s = (U8*)SvPV(swatch, slen);
3101 if (bits == 1 && otherbits == 1) {
3103 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3104 "mismatch, slen=%"UVuf", olen=%"UVuf,
3105 (UV)slen, (UV)olen);
3129 STRLEN otheroctets = otherbits >> 3;
3131 U8* const send = s + slen;
3136 if (otherbits == 1) {
3137 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3141 STRLEN vlen = otheroctets;
3149 if (opc == '+' && otherval)
3150 NOOP; /* replace with otherval */
3151 else if (opc == '!' && !otherval)
3153 else if (opc == '-' && otherval)
3155 else if (opc == '&' && !otherval)
3158 s += octets; /* no replacement */
3163 *s++ = (U8)( otherval & 0xff);
3164 else if (bits == 16) {
3165 *s++ = (U8)((otherval >> 8) & 0xff);
3166 *s++ = (U8)( otherval & 0xff);
3168 else if (bits == 32) {
3169 *s++ = (U8)((otherval >> 24) & 0xff);
3170 *s++ = (U8)((otherval >> 16) & 0xff);
3171 *s++ = (U8)((otherval >> 8) & 0xff);
3172 *s++ = (U8)( otherval & 0xff);
3176 sv_free(other); /* through with it! */
3182 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3185 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3186 * Can't be used on a property that is subject to user override, as it
3187 * relies on the value of SPECIALS in the swash which would be set by
3188 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3189 * for overridden properties
3191 * Returns a hash which is the inversion and closure of a swash mapping.
3192 * For example, consider the input lines:
3197 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3198 * 006C. The value for each key is an array. For 006C, the array would
3199 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3200 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3202 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3203 * keys are only code points that are folded-to, so it isn't a full closure.
3205 * Essentially, for any code point, it gives all the code points that map to
3206 * it, or the list of 'froms' for that point.
3208 * Currently it ignores any additions or deletions from other swashes,
3209 * looking at just the main body of the swash, and if there are SPECIALS
3210 * in the swash, at that hash
3212 * The specials hash can be extra code points, and most likely consists of
3213 * maps from single code points to multiple ones (each expressed as a string
3214 * of utf8 characters). This function currently returns only 1-1 mappings.
3215 * However consider this possible input in the specials hash:
3216 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3217 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3219 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3220 * currently handle. But it also means that FB05 and FB06 are equivalent in
3221 * a 1-1 mapping which we should handle, and this relationship may not be in
3222 * the main table. Therefore this function examines all the multi-char
3223 * sequences and adds the 1-1 mappings that come out of that. */
3227 HV *const hv = MUTABLE_HV(SvRV(swash));
3229 /* The string containing the main body of the table. This will have its
3230 * assertion fail if the swash has been converted to its inversion list */
3231 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3233 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3234 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3235 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3236 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3237 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3238 const STRLEN bits = SvUV(*bitssvp);
3239 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3240 const UV none = SvUV(*nonesvp);
3241 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3245 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3247 /* Must have at least 8 bits to get the mappings */
3248 if (bits != 8 && bits != 16 && bits != 32) {
3249 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3253 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3254 mapping to more than one character */
3256 /* Construct an inverse mapping hash for the specials */
3257 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3258 HV * specials_inverse = newHV();
3259 char *char_from; /* the lhs of the map */
3260 I32 from_len; /* its byte length */
3261 char *char_to; /* the rhs of the map */
3262 I32 to_len; /* its byte length */
3263 SV *sv_to; /* and in a sv */
3264 AV* from_list; /* list of things that map to each 'to' */
3266 hv_iterinit(specials_hv);
3268 /* The keys are the characters (in utf8) that map to the corresponding
3269 * utf8 string value. Iterate through the list creating the inverse
3271 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3273 if (! SvPOK(sv_to)) {
3274 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3275 "unexpectedly is not a string, flags=%lu",
3276 (unsigned long)SvFLAGS(sv_to));
3278 /*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)));*/
3280 /* Each key in the inverse list is a mapped-to value, and the key's
3281 * hash value is a list of the strings (each in utf8) that map to
3282 * it. Those strings are all one character long */
3283 if ((listp = hv_fetch(specials_inverse,
3287 from_list = (AV*) *listp;
3289 else { /* No entry yet for it: create one */
3290 from_list = newAV();
3291 if (! hv_store(specials_inverse,
3294 (SV*) from_list, 0))
3296 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3300 /* Here have the list associated with this 'to' (perhaps newly
3301 * created and empty). Just add to it. Note that we ASSUME that
3302 * the input is guaranteed to not have duplications, so we don't
3303 * check for that. Duplications just slow down execution time. */
3304 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3307 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3308 * it looking for cases like the FB05/FB06 examples above. There would
3309 * be an entry in the hash like
3310 * 'st' => [ FB05, FB06 ]
3311 * In this example we will create two lists that get stored in the
3312 * returned hash, 'ret':
3313 * FB05 => [ FB05, FB06 ]
3314 * FB06 => [ FB05, FB06 ]
3316 * Note that there is nothing to do if the array only has one element.
3317 * (In the normal 1-1 case handled below, we don't have to worry about
3318 * two lists, as everything gets tied to the single list that is
3319 * generated for the single character 'to'. But here, we are omitting
3320 * that list, ('st' in the example), so must have multiple lists.) */
3321 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3322 &char_to, &to_len)))
3324 if (av_tindex(from_list) > 0) {
3327 /* We iterate over all combinations of i,j to place each code
3328 * point on each list */
3329 for (i = 0; i <= av_tindex(from_list); i++) {
3331 AV* i_list = newAV();
3332 SV** entryp = av_fetch(from_list, i, FALSE);
3333 if (entryp == NULL) {
3334 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3336 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3337 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3339 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3340 (SV*) i_list, FALSE))
3342 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3345 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3346 for (j = 0; j <= av_tindex(from_list); j++) {
3347 entryp = av_fetch(from_list, j, FALSE);
3348 if (entryp == NULL) {
3349 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3352 /* When i==j this adds itself to the list */
3353 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3354 (U8*) SvPVX(*entryp),
3355 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3357 /*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));*/
3362 SvREFCNT_dec(specials_inverse); /* done with it */
3363 } /* End of specials */
3365 /* read $swash->{LIST} */
3366 l = (U8*)SvPV(*listsvp, lcur);
3369 /* Go through each input line */
3373 l = swash_scan_list_line(l, lend, &min, &max, &val,
3374 cBOOL(octets), typestr);
3379 /* Each element in the range is to be inverted */
3380 for (inverse = min; inverse <= max; inverse++) {
3384 bool found_key = FALSE;
3385 bool found_inverse = FALSE;
3387 /* The key is the inverse mapping */
3388 char key[UTF8_MAXBYTES+1];
3389 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3390 STRLEN key_len = key_end - key;
3392 /* Get the list for the map */
3393 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3394 list = (AV*) *listp;
3396 else { /* No entry yet for it: create one */
3398 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3399 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3403 /* Look through list to see if this inverse mapping already is
3404 * listed, or if there is a mapping to itself already */
3405 for (i = 0; i <= av_tindex(list); i++) {
3406 SV** entryp = av_fetch(list, i, FALSE);
3409 if (entryp == NULL) {
3410 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3414 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3418 if (uv == inverse) {
3419 found_inverse = TRUE;
3422 /* No need to continue searching if found everything we are
3424 if (found_key && found_inverse) {
3429 /* Make sure there is a mapping to itself on the list */
3431 av_push(list, newSVuv(val));
3432 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3436 /* Simply add the value to the list */
3437 if (! found_inverse) {
3438 av_push(list, newSVuv(inverse));
3439 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3442 /* swatch_get() increments the value of val for each element in the
3443 * range. That makes more compact tables possible. You can
3444 * express the capitalization, for example, of all consecutive
3445 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3446 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3447 * and it's not documented; it appears to be used only in
3448 * implementing tr//; I copied the semantics from swatch_get(), just
3450 if (!none || val < none) {
3460 Perl__swash_to_invlist(pTHX_ SV* const swash)
3463 /* Subject to change or removal. For use only in one place in regcomp.c.
3464 * Ownership is given to one reference count in the returned SV* */
3469 HV *const hv = MUTABLE_HV(SvRV(swash));
3470 UV elements = 0; /* Number of elements in the inversion list */
3480 STRLEN octets; /* if bits == 1, then octets == 0 */
3486 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3488 /* If not a hash, it must be the swash's inversion list instead */
3489 if (SvTYPE(hv) != SVt_PVHV) {
3490 return SvREFCNT_inc_simple_NN((SV*) hv);
3493 /* The string containing the main body of the table */
3494 listsvp = hv_fetchs(hv, "LIST", FALSE);
3495 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3496 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3497 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3498 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3500 typestr = (U8*)SvPV_nolen(*typesvp);
3501 bits = SvUV(*bitssvp);
3502 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3504 /* read $swash->{LIST} */
3505 if (SvPOK(*listsvp)) {
3506 l = (U8*)SvPV(*listsvp, lcur);
3509 /* LIST legitimately doesn't contain a string during compilation phases
3510 * of Perl itself, before the Unicode tables are generated. In this
3511 * case, just fake things up by creating an empty list */
3518 if (*l == 'V') { /* Inversion list format */
3519 char *after_strtol = (char *) lend;
3521 UV* other_elements_ptr;
3523 /* The first number is a count of the rest */
3525 elements = Strtoul((char *)l, &after_strtol, 10);
3526 if (elements == 0) {
3527 invlist = _new_invlist(0);
3530 l = (U8 *) after_strtol;
3532 /* Get the 0th element, which is needed to setup the inversion list */
3533 element0 = (UV) Strtoul((char *)l, &after_strtol, 10);
3534 l = (U8 *) after_strtol;
3535 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3538 /* Then just populate the rest of the input */
3539 while (elements-- > 0) {
3541 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3543 *other_elements_ptr++ = (UV) Strtoul((char *)l, &after_strtol, 10);
3544 l = (U8 *) after_strtol;
3550 /* Scan the input to count the number of lines to preallocate array
3551 * size based on worst possible case, which is each line in the input
3552 * creates 2 elements in the inversion list: 1) the beginning of a
3553 * range in the list; 2) the beginning of a range not in the list. */
3554 while ((loc = (strchr(loc, '\n'))) != NULL) {
3559 /* If the ending is somehow corrupt and isn't a new line, add another
3560 * element for the final range that isn't in the inversion list */
3561 if (! (*lend == '\n'
3562 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3567 invlist = _new_invlist(elements);
3569 /* Now go through the input again, adding each range to the list */
3572 UV val; /* Not used by this function */
3574 l = swash_scan_list_line(l, lend, &start, &end, &val,
3575 cBOOL(octets), typestr);
3581 invlist = _add_range_to_invlist(invlist, start, end);
3585 /* Invert if the data says it should be */
3586 if (invert_it_svp && SvUV(*invert_it_svp)) {
3587 _invlist_invert(invlist);
3590 /* This code is copied from swatch_get()
3591 * read $swash->{EXTRAS} */
3592 x = (U8*)SvPV(*extssvp, xcur);
3600 SV **otherbitssvp, *other;
3603 const U8 opc = *x++;
3607 nl = (U8*)memchr(x, '\n', xend - x);
3609 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3611 x = nl + 1; /* 1 is length of "\n" */
3615 x = xend; /* to EXTRAS' end at which \n is not found */
3622 namelen = nl - namestr;
3626 namelen = xend - namestr;
3630 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3631 otherhv = MUTABLE_HV(SvRV(*othersvp));
3632 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3633 otherbits = (STRLEN)SvUV(*otherbitssvp);
3635 if (bits != otherbits || bits != 1) {
3636 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3637 "properties, bits=%"UVuf", otherbits=%"UVuf,
3638 (UV)bits, (UV)otherbits);
3641 /* The "other" swatch must be destroyed after. */
3642 other = _swash_to_invlist((SV *)*othersvp);
3644 /* End of code copied from swatch_get() */
3647 _invlist_union(invlist, other, &invlist);
3650 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3653 _invlist_subtract(invlist, other, &invlist);
3656 _invlist_intersection(invlist, other, &invlist);
3661 sv_free(other); /* through with it! */
3664 SvREADONLY_on(invlist);
3669 Perl__get_swash_invlist(pTHX_ SV* const swash)
3673 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3675 if (! SvROK(swash)) {
3679 /* If it really isn't a hash, it isn't really swash; must be an inversion
3681 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3685 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3694 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3696 /* May change: warns if surrogates, non-character code points, or
3697 * non-Unicode code points are in s which has length len bytes. Returns
3698 * TRUE if none found; FALSE otherwise. The only other validity check is
3699 * to make sure that this won't exceed the string's length */
3701 const U8* const e = s + len;
3704 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3707 if (UTF8SKIP(s) > len) {
3708 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3709 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3712 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3714 if (UTF8_IS_SUPER(s)) {
3715 if (ckWARN_d(WARN_NON_UNICODE)) {
3716 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3717 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3718 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3722 else if (UTF8_IS_SURROGATE(s)) {
3723 if (ckWARN_d(WARN_SURROGATE)) {
3724 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3725 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3726 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3731 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3732 && (ckWARN_d(WARN_NONCHAR)))
3734 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3735 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3736 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3747 =for apidoc pv_uni_display
3749 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3750 length C<len>, the displayable version being at most C<pvlim> bytes long
3751 (if longer, the rest is truncated and "..." will be appended).
3753 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
3754 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3755 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3756 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3757 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3758 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3760 The pointer to the PV of the C<dsv> is returned.
3764 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3769 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3773 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3775 /* This serves double duty as a flag and a character to print after
3776 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3780 if (pvlim && SvCUR(dsv) >= pvlim) {
3784 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3786 const unsigned char c = (unsigned char)u & 0xFF;
3787 if (flags & UNI_DISPLAY_BACKSLASH) {
3804 const char string = ok;
3805 sv_catpvs(dsv, "\\");
3806 sv_catpvn(dsv, &string, 1);
3809 /* isPRINT() is the locale-blind version. */
3810 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3811 const char string = c;
3812 sv_catpvn(dsv, &string, 1);
3817 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
3820 sv_catpvs(dsv, "...");
3826 =for apidoc sv_uni_display
3828 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
3829 the displayable version being at most C<pvlim> bytes long
3830 (if longer, the rest is truncated and "..." will be appended).
3832 The C<flags> argument is as in L</pv_uni_display>().
3834 The pointer to the PV of the C<dsv> is returned.
3839 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
3841 const char * const ptr =
3842 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
3844 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
3846 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
3847 SvCUR(ssv), pvlim, flags);
3851 =for apidoc foldEQ_utf8
3853 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
3854 of which may be in UTF-8) are the same case-insensitively; false otherwise.
3855 How far into the strings to compare is determined by other input parameters.
3857 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
3858 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
3859 with respect to C<s2>.
3861 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
3862 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
3863 scan will not be considered to be a match unless the goal is reached, and
3864 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
3867 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
3868 considered an end pointer to the position 1 byte past the maximum point
3869 in C<s1> beyond which scanning will not continue under any circumstances.
3870 (This routine assumes that UTF-8 encoded input strings are not malformed;
3871 malformed input can cause it to read past C<pe1>).
3872 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
3873 is less than C<s1>+C<l1>, the match will never be successful because it can
3875 get as far as its goal (and in fact is asserted against). Correspondingly for
3876 C<pe2> with respect to C<s2>.
3878 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
3879 C<l2> must be non-zero), and if both do, both have to be
3880 reached for a successful match. Also, if the fold of a character is multiple
3881 characters, all of them must be matched (see tr21 reference below for
3884 Upon a successful match, if C<pe1> is non-NULL,
3885 it will be set to point to the beginning of the I<next> character of C<s1>
3886 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
3888 For case-insensitiveness, the "casefolding" of Unicode is used
3889 instead of upper/lowercasing both the characters, see
3890 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
3894 /* A flags parameter has been added which may change, and hence isn't
3895 * externally documented. Currently it is:
3896 * 0 for as-documented above
3897 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
3898 ASCII one, to not match
3899 * FOLDEQ_LOCALE is set iff the rules from the current underlying
3900 * locale are to be used.
3901 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
3902 * routine. This allows that step to be skipped.
3903 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
3906 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, const char *s2, char **pe2, UV l2, bool u2, U32 flags)
3909 const U8 *p1 = (const U8*)s1; /* Point to current char */
3910 const U8 *p2 = (const U8*)s2;
3911 const U8 *g1 = NULL; /* goal for s1 */
3912 const U8 *g2 = NULL;
3913 const U8 *e1 = NULL; /* Don't scan s1 past this */
3914 U8 *f1 = NULL; /* Point to current folded */
3915 const U8 *e2 = NULL;
3917 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
3918 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
3919 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
3921 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
3923 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
3924 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
3925 /* The algorithm is to trial the folds without regard to the flags on
3926 * the first line of the above assert(), and then see if the result
3927 * violates them. This means that the inputs can't be pre-folded to a
3928 * violating result, hence the assert. This could be changed, with the
3929 * addition of extra tests here for the already-folded case, which would
3930 * slow it down. That cost is more than any possible gain for when these
3931 * flags are specified, as the flags indicate /il or /iaa matching which
3932 * is less common than /iu, and I (khw) also believe that real-world /il
3933 * and /iaa matches are most likely to involve code points 0-255, and this
3934 * function only under rare conditions gets called for 0-255. */
3936 if (IN_UTF8_CTYPE_LOCALE) {
3937 flags &= ~FOLDEQ_LOCALE;
3945 g1 = (const U8*)s1 + l1;
3953 g2 = (const U8*)s2 + l2;
3956 /* Must have at least one goal */
3961 /* Will never match if goal is out-of-bounds */
3962 assert(! e1 || e1 >= g1);
3964 /* Here, there isn't an end pointer, or it is beyond the goal. We
3965 * only go as far as the goal */
3969 assert(e1); /* Must have an end for looking at s1 */
3972 /* Same for goal for s2 */
3974 assert(! e2 || e2 >= g2);
3981 /* If both operands are already folded, we could just do a memEQ on the
3982 * whole strings at once, but it would be better if the caller realized
3983 * this and didn't even call us */
3985 /* Look through both strings, a character at a time */
3986 while (p1 < e1 && p2 < e2) {
3988 /* If at the beginning of a new character in s1, get its fold to use
3989 * and the length of the fold. (exception: locale rules just get the
3990 * character to a single byte) */
3992 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
3997 /* If in locale matching, we use two sets of rules, depending
3998 * on if the code point is above or below 255. Here, we test
3999 * for and handle locale rules */
4000 if ((flags & FOLDEQ_LOCALE)
4001 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
4003 /* There is no mixing of code points above and below 255. */
4004 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
4008 /* We handle locale rules by converting, if necessary, the
4009 * code point to a single byte. */
4010 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4014 *foldbuf1 = TWO_BYTE_UTF8_TO_NATIVE(*p1, *(p1 + 1));
4018 else if (isASCII(*p1)) { /* Note, that here won't be both
4019 ASCII and using locale rules */
4021 /* If trying to mix non- with ASCII, and not supposed to,
4023 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4027 *foldbuf1 = toFOLD(*p1);
4030 to_utf8_fold(p1, foldbuf1, &n1);
4032 else { /* Not utf8, get utf8 fold */
4033 to_uni_fold(*p1, foldbuf1, &n1);
4039 if (n2 == 0) { /* Same for s2 */
4040 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4045 if ((flags & FOLDEQ_LOCALE)
4046 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4048 /* Here, the next char in s2 is < 256. We've already
4049 * worked on s1, and if it isn't also < 256, can't match */
4050 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4053 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4057 *foldbuf2 = TWO_BYTE_UTF8_TO_NATIVE(*p2, *(p2 + 1));
4060 /* Use another function to handle locale rules. We've made
4061 * sure that both characters to compare are single bytes */
4062 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4067 else if (isASCII(*p2)) {
4068 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4072 *foldbuf2 = toFOLD(*p2);
4075 to_utf8_fold(p2, foldbuf2, &n2);
4078 to_uni_fold(*p2, foldbuf2, &n2);
4084 /* Here f1 and f2 point to the beginning of the strings to compare.
4085 * These strings are the folds of the next character from each input
4086 * string, stored in utf8. */
4088 /* While there is more to look for in both folds, see if they
4089 * continue to match */
4091 U8 fold_length = UTF8SKIP(f1);
4092 if (fold_length != UTF8SKIP(f2)
4093 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4094 function call for single
4096 || memNE((char*)f1, (char*)f2, fold_length))
4098 return 0; /* mismatch */
4101 /* Here, they matched, advance past them */
4108 /* When reach the end of any fold, advance the input past it */
4110 p1 += u1 ? UTF8SKIP(p1) : 1;
4113 p2 += u2 ? UTF8SKIP(p2) : 1;
4115 } /* End of loop through both strings */
4117 /* A match is defined by each scan that specified an explicit length
4118 * reaching its final goal, and the other not having matched a partial
4119 * character (which can happen when the fold of a character is more than one
4121 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4125 /* Successful match. Set output pointers */
4135 /* XXX The next two functions should likely be moved to mathoms.c once all
4136 * occurrences of them are removed from the core; some cpan-upstream modules
4140 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4142 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4144 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4148 =for apidoc utf8n_to_uvuni
4150 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4152 This function was useful for code that wanted to handle both EBCDIC and
4153 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4154 distinctions between the platforms have mostly been made invisible to most
4155 code, so this function is quite unlikely to be what you want. If you do need
4156 this precise functionality, use instead
4157 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4158 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4164 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4166 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4168 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4172 =for apidoc uvuni_to_utf8_flags
4174 Instead you almost certainly want to use L</uvchr_to_utf8> or
4175 L</uvchr_to_utf8_flags>>.
4177 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4178 which itself, while not deprecated, should be used only in isolated
4179 circumstances. These functions were useful for code that wanted to handle
4180 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4181 v5.20, the distinctions between the platforms have mostly been made invisible
4182 to most code, so this function is quite unlikely to be what you want.
4188 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4190 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4192 return uvoffuni_to_utf8_flags(d, uv, flags);
4197 * c-indentation-style: bsd
4199 * indent-tabs-mode: nil
4202 * ex: set ts=8 sts=4 sw=4 et: