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)
492 const U8 * const s0 = s;
493 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
498 UV outlier_ret = 0; /* return value when input is in error or problematic
500 UV pack_warn = 0; /* Save result of packWARN() for later */
501 bool unexpected_non_continuation = FALSE;
502 bool overflowed = FALSE;
503 bool do_overlong_test = TRUE; /* May have to skip this test */
505 const char* const malformed_text = "Malformed UTF-8 character";
507 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
509 /* The order of malformation tests here is important. We should consume as
510 * few bytes as possible in order to not skip any valid character. This is
511 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
512 * http://unicode.org/reports/tr36 for more discussion as to why. For
513 * example, once we've done a UTF8SKIP, we can tell the expected number of
514 * bytes, and could fail right off the bat if the input parameters indicate
515 * that there are too few available. But it could be that just that first
516 * byte is garbled, and the intended character occupies fewer bytes. If we
517 * blindly assumed that the first byte is correct, and skipped based on
518 * that number, we could skip over a valid input character. So instead, we
519 * always examine the sequence byte-by-byte.
521 * We also should not consume too few bytes, otherwise someone could inject
522 * things. For example, an input could be deliberately designed to
523 * overflow, and if this code bailed out immediately upon discovering that,
524 * returning to the caller C<*retlen> pointing to the very next byte (one
525 * which is actually part of of the overflowing sequence), that could look
526 * legitimate to the caller, which could discard the initial partial
527 * sequence and process the rest, inappropriately */
529 /* Zero length strings, if allowed, of necessity are zero */
530 if (UNLIKELY(curlen == 0)) {
535 if (flags & UTF8_ALLOW_EMPTY) {
538 if (! (flags & UTF8_CHECK_ONLY)) {
539 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
544 expectlen = UTF8SKIP(s);
546 /* A well-formed UTF-8 character, as the vast majority of calls to this
547 * function will be for, has this expected length. For efficiency, set
548 * things up here to return it. It will be overriden only in those rare
549 * cases where a malformation is found */
554 /* An invariant is trivially well-formed */
555 if (UTF8_IS_INVARIANT(uv)) {
559 /* A continuation character can't start a valid sequence */
560 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
561 if (flags & UTF8_ALLOW_CONTINUATION) {
565 return UNICODE_REPLACEMENT;
568 if (! (flags & UTF8_CHECK_ONLY)) {
569 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
575 /* Here is not a continuation byte, nor an invariant. The only thing left
576 * is a start byte (possibly for an overlong) */
579 uv = NATIVE_UTF8_TO_I8(uv);
582 /* Remove the leading bits that indicate the number of bytes in the
583 * character's whole UTF-8 sequence, leaving just the bits that are part of
585 uv &= UTF_START_MASK(expectlen);
587 /* Now, loop through the remaining bytes in the character's sequence,
588 * accumulating each into the working value as we go. Be sure to not look
589 * past the end of the input string */
590 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
592 for (s = s0 + 1; s < send; s++) {
593 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
594 #ifndef EBCDIC /* Can't overflow in EBCDIC */
595 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
597 /* The original implementors viewed this malformation as more
598 * serious than the others (though I, khw, don't understand
599 * why, since other malformations also give very very wrong
600 * results), so there is no way to turn off checking for it.
601 * Set a flag, but keep going in the loop, so that we absorb
602 * the rest of the bytes that comprise the character. */
604 overflow_byte = *s; /* Save for warning message's use */
607 uv = UTF8_ACCUMULATE(uv, *s);
610 /* Here, found a non-continuation before processing all expected
611 * bytes. This byte begins a new character, so quit, even if
612 * allowing this malformation. */
613 unexpected_non_continuation = TRUE;
616 } /* End of loop through the character's bytes */
618 /* Save how many bytes were actually in the character */
621 /* The loop above finds two types of malformations: non-continuation and/or
622 * overflow. The non-continuation malformation is really a too-short
623 * malformation, as it means that the current character ended before it was
624 * expected to (being terminated prematurely by the beginning of the next
625 * character, whereas in the too-short malformation there just are too few
626 * bytes available to hold the character. In both cases, the check below
627 * that we have found the expected number of bytes would fail if executed.)
628 * Thus the non-continuation malformation is really unnecessary, being a
629 * subset of the too-short malformation. But there may be existing
630 * applications that are expecting the non-continuation type, so we retain
631 * it, and return it in preference to the too-short malformation. (If this
632 * code were being written from scratch, the two types might be collapsed
633 * into one.) I, khw, am also giving priority to returning the
634 * non-continuation and too-short malformations over overflow when multiple
635 * ones are present. I don't know of any real reason to prefer one over
636 * the other, except that it seems to me that multiple-byte errors trumps
637 * errors from a single byte */
638 if (UNLIKELY(unexpected_non_continuation)) {
639 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
640 if (! (flags & UTF8_CHECK_ONLY)) {
642 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
645 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));
650 uv = UNICODE_REPLACEMENT;
652 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
653 * as what the original expectations were. */
654 do_overlong_test = FALSE;
659 else if (UNLIKELY(curlen < expectlen)) {
660 if (! (flags & UTF8_ALLOW_SHORT)) {
661 if (! (flags & UTF8_CHECK_ONLY)) {
662 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));
666 uv = UNICODE_REPLACEMENT;
667 do_overlong_test = FALSE;
673 #ifndef EBCDIC /* EBCDIC can't overflow */
674 if (UNLIKELY(overflowed)) {
675 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
681 && expectlen > (STRLEN) OFFUNISKIP(uv)
682 && ! (flags & UTF8_ALLOW_LONG))
684 /* The overlong malformation has lower precedence than the others.
685 * Note that if this malformation is allowed, we return the actual
686 * value, instead of the replacement character. This is because this
687 * value is actually well-defined. */
688 if (! (flags & UTF8_CHECK_ONLY)) {
689 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));
694 /* Here, the input is considered to be well-formed, but it still could be a
695 * problematic code point that is not allowed by the input parameters. */
696 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
697 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
698 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
700 if (UNICODE_IS_SURROGATE(uv)) {
702 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
703 * generation of the sv, since no warnings are raised under CHECK */
704 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
705 && ckWARN_d(WARN_SURROGATE))
707 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
708 pack_warn = packWARN(WARN_SURROGATE);
710 if (flags & UTF8_DISALLOW_SURROGATE) {
714 else if ((uv > PERL_UNICODE_MAX)) {
715 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
716 && ckWARN_d(WARN_NON_UNICODE))
718 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
719 pack_warn = packWARN(WARN_NON_UNICODE);
721 #ifndef EBCDIC /* EBCDIC always allows FE, FF */
723 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
724 * points. We test for these after the regular SUPER ones, and
725 * before possibly bailing out, so that the more dire warning
726 * overrides the regular one, if applicable */
727 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
728 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
730 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
732 && ckWARN_d(WARN_UTF8))
734 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
735 pack_warn = packWARN(WARN_UTF8);
737 if (flags & UTF8_DISALLOW_FE_FF) {
742 if (flags & UTF8_DISALLOW_SUPER) {
746 else if (UNICODE_IS_NONCHAR(uv)) {
747 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
748 && ckWARN_d(WARN_NONCHAR))
750 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
751 pack_warn = packWARN(WARN_NONCHAR);
753 if (flags & UTF8_DISALLOW_NONCHAR) {
759 outlier_ret = uv; /* Note we don't bother to convert to native,
760 as all the outlier code points are the same
761 in both ASCII and EBCDIC */
765 /* Here, this is not considered a malformed character, so drop through
769 return UNI_TO_NATIVE(uv);
771 /* There are three cases which get to beyond this point. In all 3 cases:
772 * <sv> if not null points to a string to print as a warning.
773 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
775 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
776 * This is done by initializing it to 0, and changing it only
779 * 1) The input is valid but problematic, and to be warned about. The
780 * return value is the resultant code point; <*retlen> is set to
781 * <curlen>, the number of bytes that comprise the code point.
782 * <pack_warn> contains the result of packWARN() for the warning
783 * types. The entry point for this case is the label <do_warn>;
784 * 2) The input is a valid code point but disallowed by the parameters to
785 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
786 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
787 * comprise the code point. <pack_warn> contains the result of
788 * packWARN() for the warning types. The entry point for this case is
789 * the label <disallowed>.
790 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
791 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
792 * bytes that comprise the malformation. All such malformations are
793 * assumed to be warning type <utf8>. The entry point for this case
794 * is the label <malformed>.
799 if (sv && ckWARN_d(WARN_UTF8)) {
800 pack_warn = packWARN(WARN_UTF8);
805 if (flags & UTF8_CHECK_ONLY) {
807 *retlen = ((STRLEN) -1);
813 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
814 if warnings are to be raised. */
815 const char * const string = SvPVX_const(sv);
818 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
820 Perl_warner(aTHX_ pack_warn, "%s", string);
831 =for apidoc utf8_to_uvchr_buf
833 Returns the native code point of the first character in the string C<s> which
834 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
835 C<*retlen> will be set to the length, in bytes, of that character.
837 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
838 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
839 NULL) to -1. If those warnings are off, the computed value, if well-defined
840 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
841 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
842 the next possible position in C<s> that could begin a non-malformed character.
843 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
851 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
855 return utf8n_to_uvchr(s, send - s, retlen,
856 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
859 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
860 * there are no malformations in the input UTF-8 string C<s>. surrogates,
861 * non-character code points, and non-Unicode code points are allowed. */
864 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
866 UV expectlen = UTF8SKIP(s);
867 const U8* send = s + expectlen;
870 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)
951 PERL_ARGS_ASSERT_UTF8_LENGTH;
953 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
954 * the bitops (especially ~) can create illegal UTF-8.
955 * In other words: in Perl UTF-8 is not just for Unicode. */
958 goto warn_and_return;
968 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
969 "%s in %s", unees, OP_DESC(PL_op));
971 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
978 =for apidoc utf8_distance
980 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
983 WARNING: use only if you *know* that the pointers point inside the
990 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
992 PERL_ARGS_ASSERT_UTF8_DISTANCE;
994 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1000 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1001 forward or backward.
1003 WARNING: do not use the following unless you *know* C<off> is within
1004 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1005 on the first byte of character or just after the last byte of a character.
1011 Perl_utf8_hop(const U8 *s, I32 off)
1013 PERL_ARGS_ASSERT_UTF8_HOP;
1015 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1016 * the bitops (especially ~) can create illegal UTF-8.
1017 * In other words: in Perl UTF-8 is not just for Unicode. */
1026 while (UTF8_IS_CONTINUATION(*s))
1034 =for apidoc bytes_cmp_utf8
1036 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1037 sequence of characters (stored as UTF-8)
1038 in C<u>, C<ulen>. Returns 0 if they are
1039 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1040 if the first string is greater than the second string.
1042 -1 or +1 is returned if the shorter string was identical to the start of the
1043 longer string. -2 or +2 is returned if
1044 there was a difference between characters
1051 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1053 const U8 *const bend = b + blen;
1054 const U8 *const uend = u + ulen;
1056 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1058 while (b < bend && u < uend) {
1060 if (!UTF8_IS_INVARIANT(c)) {
1061 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1064 if (UTF8_IS_CONTINUATION(c1)) {
1065 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1067 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1068 "Malformed UTF-8 character "
1069 "(unexpected non-continuation byte 0x%02x"
1070 ", immediately after start byte 0x%02x)"
1071 /* Dear diag.t, it's in the pod. */
1073 PL_op ? " in " : "",
1074 PL_op ? OP_DESC(PL_op) : "");
1079 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1080 "%s in %s", unees, OP_DESC(PL_op));
1082 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1083 return -2; /* Really want to return undef :-) */
1090 return *b < c ? -2 : +2;
1095 if (b == bend && u == uend)
1098 return b < bend ? +1 : -1;
1102 =for apidoc utf8_to_bytes
1104 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1105 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1106 updates C<len> to contain the new length.
1107 Returns zero on failure, setting C<len> to -1.
1109 If you need a copy of the string, see L</bytes_from_utf8>.
1115 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1117 U8 * const save = s;
1118 U8 * const send = s + *len;
1121 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1122 PERL_UNUSED_CONTEXT;
1124 /* ensure valid UTF-8 and chars < 256 before updating string */
1126 if (! UTF8_IS_INVARIANT(*s)) {
1127 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1128 *len = ((STRLEN) -1);
1139 if (! UTF8_IS_INVARIANT(c)) {
1140 /* Then it is two-byte encoded */
1141 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1152 =for apidoc bytes_from_utf8
1154 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1155 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1156 the newly-created string, and updates C<len> to contain the new
1157 length. Returns the original string if no conversion occurs, C<len>
1158 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1159 0 if C<s> is converted or consisted entirely of characters that are invariant
1160 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1166 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1169 const U8 *start = s;
1173 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1174 PERL_UNUSED_CONTEXT;
1178 /* ensure valid UTF-8 and chars < 256 before converting string */
1179 for (send = s + *len; s < send;) {
1180 if (! UTF8_IS_INVARIANT(*s)) {
1181 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1192 Newx(d, (*len) - count + 1, U8);
1193 s = start; start = d;
1196 if (! UTF8_IS_INVARIANT(c)) {
1197 /* Then it is two-byte encoded */
1198 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1209 =for apidoc bytes_to_utf8
1211 Converts a string C<s> of length C<len> bytes from the native encoding into
1213 Returns a pointer to the newly-created string, and sets C<len> to
1214 reflect the new length in bytes.
1216 A C<NUL> character will be written after the end of the string.
1218 If you want to convert to UTF-8 from encodings other than
1219 the native (Latin1 or EBCDIC),
1220 see L</sv_recode_to_utf8>().
1225 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1226 likewise need duplication. */
1229 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1231 const U8 * const send = s + (*len);
1235 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1236 PERL_UNUSED_CONTEXT;
1238 Newx(d, (*len) * 2 + 1, U8);
1242 append_utf8_from_native_byte(*s, &d);
1251 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1253 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1254 * We optimize for native, for obvious reasons. */
1257 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1262 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1265 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1270 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1272 if (UNI_IS_INVARIANT(uv)) {
1273 *d++ = LATIN1_TO_NATIVE((U8) uv);
1276 if (uv <= MAX_UTF8_TWO_BYTE) {
1277 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1278 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1281 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1282 #define LAST_HIGH_SURROGATE 0xDBFF
1283 #define FIRST_LOW_SURROGATE 0xDC00
1284 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1285 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1287 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1289 UV low = (p[0] << 8) + p[1];
1291 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1292 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1293 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1294 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1296 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1297 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1300 d = uvoffuni_to_utf8_flags(d, uv, 0);
1303 *d++ = (U8)(( uv >> 12) | 0xe0);
1304 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1305 *d++ = (U8)(( uv & 0x3f) | 0x80);
1309 *d++ = (U8)(( uv >> 18) | 0xf0);
1310 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1311 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1312 *d++ = (U8)(( uv & 0x3f) | 0x80);
1317 *newlen = d - dstart;
1321 /* Note: this one is slightly destructive of the source. */
1324 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1327 U8* const send = s + bytelen;
1329 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1332 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1336 const U8 tmp = s[0];
1341 return utf16_to_utf8(p, d, bytelen, newlen);
1345 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1347 U8 tmpbuf[UTF8_MAXBYTES+1];
1348 uvchr_to_utf8(tmpbuf, c);
1349 return _is_utf8_FOO(classnum, tmpbuf);
1352 /* Internal function so we can deprecate the external one, and call
1353 this one from other deprecated functions in this file */
1356 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1358 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1362 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1366 Perl__is_uni_perl_idcont(pTHX_ UV c)
1368 U8 tmpbuf[UTF8_MAXBYTES+1];
1369 uvchr_to_utf8(tmpbuf, c);
1370 return _is_utf8_perl_idcont(tmpbuf);
1374 Perl__is_uni_perl_idstart(pTHX_ UV c)
1376 U8 tmpbuf[UTF8_MAXBYTES+1];
1377 uvchr_to_utf8(tmpbuf, c);
1378 return _is_utf8_perl_idstart(tmpbuf);
1382 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1384 /* We have the latin1-range values compiled into the core, so just use
1385 * those, converting the result to utf8. The only difference between upper
1386 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1387 * either "SS" or "Ss". Which one to use is passed into the routine in
1388 * 'S_or_s' to avoid a test */
1390 UV converted = toUPPER_LATIN1_MOD(c);
1392 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1394 assert(S_or_s == 'S' || S_or_s == 's');
1396 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1397 characters in this range */
1398 *p = (U8) converted;
1403 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1404 * which it maps to one of them, so as to only have to have one check for
1405 * it in the main case */
1406 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1408 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1409 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1412 converted = GREEK_CAPITAL_LETTER_MU;
1414 case LATIN_SMALL_LETTER_SHARP_S:
1420 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1421 assert(0); /* NOTREACHED */
1425 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1426 *p = UTF8_TWO_BYTE_LO(converted);
1432 /* Call the function to convert a UTF-8 encoded character to the specified case.
1433 * Note that there may be more than one character in the result.
1434 * INP is a pointer to the first byte of the input character
1435 * OUTP will be set to the first byte of the string of changed characters. It
1436 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1437 * LENP will be set to the length in bytes of the string of changed characters
1439 * The functions return the ordinal of the first character in the string of OUTP */
1440 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1441 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1442 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1444 /* This additionally has the input parameter SPECIALS, which if non-zero will
1445 * cause this to use the SPECIALS hash for folding (meaning get full case
1446 * folding); otherwise, when zero, this implies a simple case fold */
1447 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1450 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1452 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1453 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1454 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1455 * the changed version may be longer than the original character.
1457 * The ordinal of the first character of the changed version is returned
1458 * (but note, as explained above, that there may be more.) */
1460 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1463 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1466 uvchr_to_utf8(p, c);
1467 return CALL_UPPER_CASE(p, p, lenp);
1471 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1473 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1476 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1479 uvchr_to_utf8(p, c);
1480 return CALL_TITLE_CASE(p, p, lenp);
1484 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
1486 /* We have the latin1-range values compiled into the core, so just use
1487 * those, converting the result to utf8. Since the result is always just
1488 * one character, we allow <p> to be NULL */
1490 U8 converted = toLOWER_LATIN1(c);
1493 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1498 /* Result is known to always be < 256, so can use the EIGHT_BIT
1500 *p = UTF8_EIGHT_BIT_HI(converted);
1501 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
1509 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1511 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1514 return to_lower_latin1((U8) c, p, lenp);
1517 uvchr_to_utf8(p, c);
1518 return CALL_LOWER_CASE(p, p, lenp);
1522 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1524 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1525 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1526 * FOLD_FLAGS_FULL iff full folding is to be used;
1528 * Not to be used for locale folds
1533 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1534 PERL_UNUSED_CONTEXT;
1536 assert (! (flags & FOLD_FLAGS_LOCALE));
1538 if (c == MICRO_SIGN) {
1539 converted = GREEK_SMALL_LETTER_MU;
1541 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1543 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1544 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1545 * under those circumstances. */
1546 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1547 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1548 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1550 return LATIN_SMALL_LETTER_LONG_S;
1559 else { /* In this range the fold of all other characters is their lower
1561 converted = toLOWER_LATIN1(c);
1564 if (UVCHR_IS_INVARIANT(converted)) {
1565 *p = (U8) converted;
1569 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1570 *p = UTF8_TWO_BYTE_LO(converted);
1578 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1581 /* Not currently externally documented, and subject to change
1582 * <flags> bits meanings:
1583 * FOLD_FLAGS_FULL iff full folding is to be used;
1584 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1585 * locale are to be used.
1586 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1589 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1591 /* Tread a UTF-8 locale as not being in locale at all */
1592 if (IN_UTF8_CTYPE_LOCALE) {
1593 flags &= ~FOLD_FLAGS_LOCALE;
1597 UV result = _to_fold_latin1((U8) c, p, lenp,
1598 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1599 /* It is illegal for the fold to cross the 255/256 boundary under
1600 * locale; in this case return the original */
1601 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1606 /* If no special needs, just use the macro */
1607 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1608 uvchr_to_utf8(p, c);
1609 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1611 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1612 the special flags. */
1613 U8 utf8_c[UTF8_MAXBYTES + 1];
1614 uvchr_to_utf8(utf8_c, c);
1615 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1619 PERL_STATIC_INLINE bool
1620 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1621 const char *const swashname, SV* const invlist)
1623 /* returns a boolean giving whether or not the UTF8-encoded character that
1624 * starts at <p> is in the swash indicated by <swashname>. <swash>
1625 * contains a pointer to where the swash indicated by <swashname>
1626 * is to be stored; which this routine will do, so that future calls will
1627 * look at <*swash> and only generate a swash if it is not null. <invlist>
1628 * is NULL or an inversion list that defines the swash. If not null, it
1629 * saves time during initialization of the swash.
1631 * Note that it is assumed that the buffer length of <p> is enough to
1632 * contain all the bytes that comprise the character. Thus, <*p> should
1633 * have been checked before this call for mal-formedness enough to assure
1636 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1638 /* The API should have included a length for the UTF-8 character in <p>,
1639 * but it doesn't. We therefore assume that p has been validated at least
1640 * as far as there being enough bytes available in it to accommodate the
1641 * character without reading beyond the end, and pass that number on to the
1642 * validating routine */
1643 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1644 if (ckWARN_d(WARN_UTF8)) {
1645 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1646 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1647 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1648 what the malformation is */
1649 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1655 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1656 *swash = _core_swash_init("utf8",
1658 /* Only use the name if there is no inversion
1659 * list; otherwise will go out to disk */
1660 (invlist) ? "" : swashname,
1662 &PL_sv_undef, 1, 0, invlist, &flags);
1665 return swash_fetch(*swash, p, TRUE) != 0;
1669 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1671 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1673 assert(classnum < _FIRST_NON_SWASH_CC);
1675 return is_utf8_common(p,
1676 &PL_utf8_swash_ptrs[classnum],
1677 swash_property_names[classnum],
1678 PL_XPosix_ptrs[classnum]);
1682 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1686 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1688 if (! PL_utf8_perl_idstart) {
1689 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1691 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
1695 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1697 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1701 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1705 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1709 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1711 if (! PL_utf8_perl_idcont) {
1712 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1714 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
1718 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1720 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1722 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1726 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1728 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1730 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1734 Perl__is_utf8_mark(pTHX_ const U8 *p)
1736 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1738 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1742 =for apidoc to_utf8_case
1744 C<p> contains the pointer to the UTF-8 string encoding
1745 the character that is being converted. This routine assumes that the character
1746 at C<p> is well-formed.
1748 C<ustrp> is a pointer to the character buffer to put the
1749 conversion result to. C<lenp> is a pointer to the length
1752 C<swashp> is a pointer to the swash to use.
1754 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1755 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
1756 but not always, a multicharacter mapping), is tried first.
1758 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1759 any special mappings; C<""> means to use the special mappings. Values other
1760 than these two are treated as the name of the hash containing the special
1761 mappings, like C<"utf8::ToSpecLower">.
1763 C<normal> is a string like "ToLower" which means the swash
1769 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1770 SV **swashp, const char *normal, const char *special)
1773 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1775 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1777 /* Note that swash_fetch() doesn't output warnings for these because it
1778 * assumes we will */
1779 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1780 if (uv1 <= UNICODE_SURROGATE_LAST) {
1781 if (ckWARN_d(WARN_SURROGATE)) {
1782 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1783 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1784 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1787 else if (UNICODE_IS_SUPER(uv1)) {
1788 if (ckWARN_d(WARN_NON_UNICODE)) {
1789 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1790 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1791 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1795 /* Note that non-characters are perfectly legal, so no warning should
1799 if (!*swashp) /* load on-demand */
1800 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1803 /* It might be "special" (sometimes, but not always,
1804 * a multicharacter mapping) */
1808 /* If passed in the specials name, use that; otherwise use any
1809 * given in the swash */
1810 if (*special != '\0') {
1811 hv = get_hv(special, 0);
1814 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1816 hv = MUTABLE_HV(SvRV(*svp));
1821 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1826 s = SvPV_const(*svp, len);
1829 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1831 Copy(s, ustrp, len, U8);
1836 if (!len && *swashp) {
1837 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
1840 /* It was "normal" (a single character mapping). */
1841 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1849 return valid_utf8_to_uvchr(ustrp, 0);
1852 /* Here, there was no mapping defined, which means that the code point maps
1853 * to itself. Return the inputs */
1855 if (p != ustrp) { /* Don't copy onto itself */
1856 Copy(p, ustrp, len, U8);
1867 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1869 /* This is called when changing the case of a utf8-encoded character above
1870 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1871 * result contains a character that crosses the 255/256 boundary, disallow
1872 * the change, and return the original code point. See L<perlfunc/lc> for
1875 * p points to the original string whose case was changed; assumed
1876 * by this routine to be well-formed
1877 * result the code point of the first character in the changed-case string
1878 * ustrp points to the changed-case string (<result> represents its first char)
1879 * lenp points to the length of <ustrp> */
1881 UV original; /* To store the first code point of <p> */
1883 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1885 assert(UTF8_IS_ABOVE_LATIN1(*p));
1887 /* We know immediately if the first character in the string crosses the
1888 * boundary, so can skip */
1891 /* Look at every character in the result; if any cross the
1892 * boundary, the whole thing is disallowed */
1893 U8* s = ustrp + UTF8SKIP(ustrp);
1894 U8* e = ustrp + *lenp;
1896 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
1902 /* Here, no characters crossed, result is ok as-is */
1908 /* Failed, have to return the original */
1909 original = valid_utf8_to_uvchr(p, lenp);
1910 Copy(p, ustrp, *lenp, char);
1915 =for apidoc to_utf8_upper
1917 Instead use L</toUPPER_utf8>.
1921 /* Not currently externally documented, and subject to change:
1922 * <flags> is set iff iff the rules from the current underlying locale are to
1926 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
1930 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
1932 if (flags && IN_UTF8_CTYPE_LOCALE) {
1936 if (UTF8_IS_INVARIANT(*p)) {
1938 result = toUPPER_LC(*p);
1941 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
1944 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
1946 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
1947 result = toUPPER_LC(c);
1950 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
1954 else { /* utf8, ord above 255 */
1955 result = CALL_UPPER_CASE(p, ustrp, lenp);
1958 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
1963 /* Here, used locale rules. Convert back to utf8 */
1964 if (UTF8_IS_INVARIANT(result)) {
1965 *ustrp = (U8) result;
1969 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1970 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
1978 =for apidoc to_utf8_title
1980 Instead use L</toTITLE_utf8>.
1984 /* Not currently externally documented, and subject to change:
1985 * <flags> is set iff the rules from the current underlying locale are to be
1986 * used. Since titlecase is not defined in POSIX, for other than a
1987 * UTF-8 locale, uppercase is used instead for code points < 256.
1991 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
1995 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
1997 if (flags && IN_UTF8_CTYPE_LOCALE) {
2001 if (UTF8_IS_INVARIANT(*p)) {
2003 result = toUPPER_LC(*p);
2006 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2009 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2011 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2012 result = toUPPER_LC(c);
2015 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2019 else { /* utf8, ord above 255 */
2020 result = CALL_TITLE_CASE(p, ustrp, lenp);
2023 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2028 /* Here, used locale rules. Convert back to utf8 */
2029 if (UTF8_IS_INVARIANT(result)) {
2030 *ustrp = (U8) result;
2034 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2035 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2043 =for apidoc to_utf8_lower
2045 Instead use L</toLOWER_utf8>.
2049 /* Not currently externally documented, and subject to change:
2050 * <flags> is set iff iff the rules from the current underlying locale are to
2055 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2059 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2061 if (flags && IN_UTF8_CTYPE_LOCALE) {
2065 if (UTF8_IS_INVARIANT(*p)) {
2067 result = toLOWER_LC(*p);
2070 return to_lower_latin1(*p, ustrp, lenp);
2073 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2075 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2076 result = toLOWER_LC(c);
2079 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2083 else { /* utf8, ord above 255 */
2084 result = CALL_LOWER_CASE(p, ustrp, lenp);
2087 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2093 /* Here, used locale rules. Convert back to utf8 */
2094 if (UTF8_IS_INVARIANT(result)) {
2095 *ustrp = (U8) result;
2099 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2100 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2108 =for apidoc to_utf8_fold
2110 Instead use L</toFOLD_utf8>.
2114 /* Not currently externally documented, and subject to change,
2116 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2117 * locale are to be used.
2118 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2119 * otherwise simple folds
2120 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2125 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2129 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2131 /* These are mutually exclusive */
2132 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2134 assert(p != ustrp); /* Otherwise overwrites */
2136 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2137 flags &= ~FOLD_FLAGS_LOCALE;
2140 if (UTF8_IS_INVARIANT(*p)) {
2141 if (flags & FOLD_FLAGS_LOCALE) {
2142 result = toFOLD_LC(*p);
2145 return _to_fold_latin1(*p, ustrp, lenp,
2146 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2149 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2150 if (flags & FOLD_FLAGS_LOCALE) {
2151 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2152 result = toFOLD_LC(c);
2155 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2157 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2160 else { /* utf8, ord above 255 */
2161 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2163 if (flags & FOLD_FLAGS_LOCALE) {
2165 /* Special case these two characters, as what normally gets
2166 * returned under locale doesn't work */
2167 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2168 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2169 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2173 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2174 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2175 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2177 goto return_ligature_st;
2179 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2181 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2185 /* This is called when changing the case of a utf8-encoded
2186 * character above the ASCII range, and the result should not
2187 * contain an ASCII character. */
2189 UV original; /* To store the first code point of <p> */
2191 /* Look at every character in the result; if any cross the
2192 * boundary, the whole thing is disallowed */
2194 U8* e = ustrp + *lenp;
2197 /* Crossed, have to return the original */
2198 original = valid_utf8_to_uvchr(p, lenp);
2200 /* But in these instances, there is an alternative we can
2201 * return that is valid */
2202 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2203 || original == LATIN_SMALL_LETTER_SHARP_S)
2207 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2208 goto return_ligature_st;
2210 Copy(p, ustrp, *lenp, char);
2216 /* Here, no characters crossed, result is ok as-is */
2221 /* Here, used locale rules. Convert back to utf8 */
2222 if (UTF8_IS_INVARIANT(result)) {
2223 *ustrp = (U8) result;
2227 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2228 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2235 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2236 * folds to a string of two of these characters. By returning this
2237 * instead, then, e.g.,
2238 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2241 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2242 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2244 return LATIN_SMALL_LETTER_LONG_S;
2247 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2248 * have the other one fold to it */
2250 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2251 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2252 return LATIN_SMALL_LIGATURE_ST;
2256 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2257 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2258 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2262 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2264 PERL_ARGS_ASSERT_SWASH_INIT;
2266 /* Returns a copy of a swash initiated by the called function. This is the
2267 * public interface, and returning a copy prevents others from doing
2268 * mischief on the original */
2270 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2274 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2276 /* Initialize and return a swash, creating it if necessary. It does this
2277 * by calling utf8_heavy.pl in the general case. The returned value may be
2278 * the swash's inversion list instead if the input parameters allow it.
2279 * Which is returned should be immaterial to callers, as the only
2280 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2281 * and swash_to_invlist() handle both these transparently.
2283 * This interface should only be used by functions that won't destroy or
2284 * adversely change the swash, as doing so affects all other uses of the
2285 * swash in the program; the general public should use 'Perl_swash_init'
2288 * pkg is the name of the package that <name> should be in.
2289 * name is the name of the swash to find. Typically it is a Unicode
2290 * property name, including user-defined ones
2291 * listsv is a string to initialize the swash with. It must be of the form
2292 * documented as the subroutine return value in
2293 * L<perlunicode/User-Defined Character Properties>
2294 * minbits is the number of bits required to represent each data element.
2295 * It is '1' for binary properties.
2296 * none I (khw) do not understand this one, but it is used only in tr///.
2297 * invlist is an inversion list to initialize the swash with (or NULL)
2298 * flags_p if non-NULL is the address of various input and output flag bits
2299 * to the routine, as follows: ('I' means is input to the routine;
2300 * 'O' means output from the routine. Only flags marked O are
2301 * meaningful on return.)
2302 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2303 * came from a user-defined property. (I O)
2304 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2305 * when the swash cannot be located, to simply return NULL. (I)
2306 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2307 * return of an inversion list instead of a swash hash if this routine
2308 * thinks that would result in faster execution of swash_fetch() later
2311 * Thus there are three possible inputs to find the swash: <name>,
2312 * <listsv>, and <invlist>. At least one must be specified. The result
2313 * will be the union of the specified ones, although <listsv>'s various
2314 * actions can intersect, etc. what <name> gives. To avoid going out to
2315 * disk at all, <invlist> should specify completely what the swash should
2316 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2318 * <invlist> is only valid for binary properties */
2320 SV* retval = &PL_sv_undef;
2321 HV* swash_hv = NULL;
2322 const int invlist_swash_boundary =
2323 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2324 ? 512 /* Based on some benchmarking, but not extensive, see commit
2326 : -1; /* Never return just an inversion list */
2328 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2329 assert(! invlist || minbits == 1);
2331 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2333 if (listsv != &PL_sv_undef || strNE(name, "")) {
2335 const size_t pkg_len = strlen(pkg);
2336 const size_t name_len = strlen(name);
2337 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2341 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2343 PUSHSTACKi(PERLSI_MAGIC);
2347 /* We might get here via a subroutine signature which uses a utf8
2348 * parameter name, at which point PL_subname will have been set
2349 * but not yet used. */
2350 save_item(PL_subname);
2351 if (PL_parser && PL_parser->error_count)
2352 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2353 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2354 if (!method) { /* demand load utf8 */
2356 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2357 GvSV(PL_errgv) = NULL;
2358 /* It is assumed that callers of this routine are not passing in
2359 * any user derived data. */
2360 /* Need to do this after save_re_context() as it will set
2361 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2362 * in Perl_magic_get). Even line to create errsv_save can turn on
2364 #ifndef NO_TAINT_SUPPORT
2365 SAVEBOOL(TAINT_get);
2368 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2371 /* Not ERRSV, as there is no need to vivify a scalar we are
2372 about to discard. */
2373 SV * const errsv = GvSV(PL_errgv);
2374 if (!SvTRUE(errsv)) {
2375 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2376 SvREFCNT_dec(errsv);
2384 mPUSHp(pkg, pkg_len);
2385 mPUSHp(name, name_len);
2390 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2391 GvSV(PL_errgv) = NULL;
2392 /* If we already have a pointer to the method, no need to use
2393 * call_method() to repeat the lookup. */
2395 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2396 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2398 retval = *PL_stack_sp--;
2399 SvREFCNT_inc(retval);
2402 /* Not ERRSV. See above. */
2403 SV * const errsv = GvSV(PL_errgv);
2404 if (!SvTRUE(errsv)) {
2405 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2406 SvREFCNT_dec(errsv);
2411 if (IN_PERL_COMPILETIME) {
2412 CopHINTS_set(PL_curcop, PL_hints);
2414 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2417 /* If caller wants to handle missing properties, let them */
2418 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2422 "Can't find Unicode property definition \"%"SVf"\"",
2424 NOT_REACHED; /* NOTREACHED */
2426 } /* End of calling the module to find the swash */
2428 /* If this operation fetched a swash, and we will need it later, get it */
2429 if (retval != &PL_sv_undef
2430 && (minbits == 1 || (flags_p
2432 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2434 swash_hv = MUTABLE_HV(SvRV(retval));
2436 /* If we don't already know that there is a user-defined component to
2437 * this swash, and the user has indicated they wish to know if there is
2438 * one (by passing <flags_p>), find out */
2439 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2440 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2441 if (user_defined && SvUV(*user_defined)) {
2442 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2447 /* Make sure there is an inversion list for binary properties */
2449 SV** swash_invlistsvp = NULL;
2450 SV* swash_invlist = NULL;
2451 bool invlist_in_swash_is_valid = FALSE;
2452 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2453 an unclaimed reference count */
2455 /* If this operation fetched a swash, get its already existing
2456 * inversion list, or create one for it */
2459 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2460 if (swash_invlistsvp) {
2461 swash_invlist = *swash_invlistsvp;
2462 invlist_in_swash_is_valid = TRUE;
2465 swash_invlist = _swash_to_invlist(retval);
2466 swash_invlist_unclaimed = TRUE;
2470 /* If an inversion list was passed in, have to include it */
2473 /* Any fetched swash will by now have an inversion list in it;
2474 * otherwise <swash_invlist> will be NULL, indicating that we
2475 * didn't fetch a swash */
2476 if (swash_invlist) {
2478 /* Add the passed-in inversion list, which invalidates the one
2479 * already stored in the swash */
2480 invlist_in_swash_is_valid = FALSE;
2481 _invlist_union(invlist, swash_invlist, &swash_invlist);
2485 /* Here, there is no swash already. Set up a minimal one, if
2486 * we are going to return a swash */
2487 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2489 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2491 swash_invlist = invlist;
2495 /* Here, we have computed the union of all the passed-in data. It may
2496 * be that there was an inversion list in the swash which didn't get
2497 * touched; otherwise save the computed one */
2498 if (! invlist_in_swash_is_valid
2499 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2501 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2503 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2505 /* We just stole a reference count. */
2506 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2507 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2510 SvREADONLY_on(swash_invlist);
2512 /* Use the inversion list stand-alone if small enough */
2513 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2514 SvREFCNT_dec(retval);
2515 if (!swash_invlist_unclaimed)
2516 SvREFCNT_inc_simple_void_NN(swash_invlist);
2517 retval = newRV_noinc(swash_invlist);
2525 /* This API is wrong for special case conversions since we may need to
2526 * return several Unicode characters for a single Unicode character
2527 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2528 * the lower-level routine, and it is similarly broken for returning
2529 * multiple values. --jhi
2530 * For those, you should use to_utf8_case() instead */
2531 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2534 * Returns the value of property/mapping C<swash> for the first character
2535 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2536 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2537 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2539 * A "swash" is a hash which contains initially the keys/values set up by
2540 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2541 * property for all possible code points. Things are stored in a compact form
2542 * (see utf8_heavy.pl) so that calculation is required to find the actual
2543 * property value for a given code point. As code points are looked up, new
2544 * key/value pairs are added to the hash, so that the calculation doesn't have
2545 * to ever be re-done. Further, each calculation is done, not just for the
2546 * desired one, but for a whole block of code points adjacent to that one.
2547 * For binary properties on ASCII machines, the block is usually for 64 code
2548 * points, starting with a code point evenly divisible by 64. Thus if the
2549 * property value for code point 257 is requested, the code goes out and
2550 * calculates the property values for all 64 code points between 256 and 319,
2551 * and stores these as a single 64-bit long bit vector, called a "swatch",
2552 * under the key for code point 256. The key is the UTF-8 encoding for code
2553 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2554 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2555 * for code point 258 is then requested, this code realizes that it would be
2556 * stored under the key for 256, and would find that value and extract the
2557 * relevant bit, offset from 256.
2559 * Non-binary properties are stored in as many bits as necessary to represent
2560 * their values (32 currently, though the code is more general than that), not
2561 * as single bits, but the principal is the same: the value for each key is a
2562 * vector that encompasses the property values for all code points whose UTF-8
2563 * representations are represented by the key. That is, for all code points
2564 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2568 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2570 HV *const hv = MUTABLE_HV(SvRV(swash));
2575 const U8 *tmps = NULL;
2579 PERL_ARGS_ASSERT_SWASH_FETCH;
2581 /* If it really isn't a hash, it isn't really swash; must be an inversion
2583 if (SvTYPE(hv) != SVt_PVHV) {
2584 return _invlist_contains_cp((SV*)hv,
2586 ? valid_utf8_to_uvchr(ptr, NULL)
2590 /* We store the values in a "swatch" which is a vec() value in a swash
2591 * hash. Code points 0-255 are a single vec() stored with key length
2592 * (klen) 0. All other code points have a UTF-8 representation
2593 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2594 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2595 * length for them is the length of the encoded char - 1. ptr[klen] is the
2596 * final byte in the sequence representing the character */
2597 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2602 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2605 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
2608 klen = UTF8SKIP(ptr) - 1;
2610 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2611 * the vec is the final byte in the sequence. (In EBCDIC this is
2612 * converted to I8 to get consecutive values.) To help you visualize
2614 * Straight 1047 After final byte
2615 * UTF-8 UTF-EBCDIC I8 transform
2616 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2617 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2619 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2620 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2622 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2623 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2625 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2626 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2628 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2629 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2631 * (There are no discontinuities in the elided (...) entries.)
2632 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2633 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2634 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2635 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2636 * index into the vec() swatch (after subtracting 0x80, which we
2637 * actually do with an '&').
2638 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2639 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2640 * dicontinuities which go away by transforming it into I8, and we
2641 * effectively subtract 0xA0 to get the index. */
2642 needents = (1 << UTF_ACCUMULATION_SHIFT);
2643 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2647 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2648 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2649 * it's nothing to sniff at.) Pity we usually come through at least
2650 * two function calls to get here...
2652 * NB: this code assumes that swatches are never modified, once generated!
2655 if (hv == PL_last_swash_hv &&
2656 klen == PL_last_swash_klen &&
2657 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2659 tmps = PL_last_swash_tmps;
2660 slen = PL_last_swash_slen;
2663 /* Try our second-level swatch cache, kept in a hash. */
2664 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2666 /* If not cached, generate it via swatch_get */
2667 if (!svp || !SvPOK(*svp)
2668 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2671 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2672 swatch = swatch_get(swash,
2673 code_point & ~((UV)needents - 1),
2676 else { /* For the first 256 code points, the swatch has a key of
2678 swatch = swatch_get(swash, 0, needents);
2681 if (IN_PERL_COMPILETIME)
2682 CopHINTS_set(PL_curcop, PL_hints);
2684 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2686 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2687 || (slen << 3) < needents)
2688 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2689 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2690 svp, tmps, (UV)slen, (UV)needents);
2693 PL_last_swash_hv = hv;
2694 assert(klen <= sizeof(PL_last_swash_key));
2695 PL_last_swash_klen = (U8)klen;
2696 /* FIXME change interpvar.h? */
2697 PL_last_swash_tmps = (U8 *) tmps;
2698 PL_last_swash_slen = slen;
2700 Copy(ptr, PL_last_swash_key, klen, U8);
2703 switch ((int)((slen << 3) / needents)) {
2705 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2707 return ((UV) tmps[off]);
2711 ((UV) tmps[off ] << 8) +
2712 ((UV) tmps[off + 1]);
2716 ((UV) tmps[off ] << 24) +
2717 ((UV) tmps[off + 1] << 16) +
2718 ((UV) tmps[off + 2] << 8) +
2719 ((UV) tmps[off + 3]);
2721 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2722 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2723 NORETURN_FUNCTION_END;
2726 /* Read a single line of the main body of the swash input text. These are of
2729 * where each number is hex. The first two numbers form the minimum and
2730 * maximum of a range, and the third is the value associated with the range.
2731 * Not all swashes should have a third number
2733 * On input: l points to the beginning of the line to be examined; it points
2734 * to somewhere in the string of the whole input text, and is
2735 * terminated by a \n or the null string terminator.
2736 * lend points to the null terminator of that string
2737 * wants_value is non-zero if the swash expects a third number
2738 * typestr is the name of the swash's mapping, like 'ToLower'
2739 * On output: *min, *max, and *val are set to the values read from the line.
2740 * returns a pointer just beyond the line examined. If there was no
2741 * valid min number on the line, returns lend+1
2745 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2746 const bool wants_value, const U8* const typestr)
2748 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2749 STRLEN numlen; /* Length of the number */
2750 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2751 | PERL_SCAN_DISALLOW_PREFIX
2752 | PERL_SCAN_SILENT_NON_PORTABLE;
2754 /* nl points to the next \n in the scan */
2755 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2757 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2759 /* Get the first number on the line: the range minimum */
2761 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2762 *max = *min; /* So can never return without setting max */
2763 if (numlen) /* If found a hex number, position past it */
2765 else if (nl) { /* Else, go handle next line, if any */
2766 return nl + 1; /* 1 is length of "\n" */
2768 else { /* Else, no next line */
2769 return lend + 1; /* to LIST's end at which \n is not found */
2772 /* The max range value follows, separated by a BLANK */
2775 flags = PERL_SCAN_SILENT_ILLDIGIT
2776 | PERL_SCAN_DISALLOW_PREFIX
2777 | PERL_SCAN_SILENT_NON_PORTABLE;
2779 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2782 else /* If no value here, it is a single element range */
2785 /* Non-binary tables have a third entry: what the first element of the
2786 * range maps to. The map for those currently read here is in hex */
2790 flags = PERL_SCAN_SILENT_ILLDIGIT
2791 | PERL_SCAN_DISALLOW_PREFIX
2792 | PERL_SCAN_SILENT_NON_PORTABLE;
2794 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2803 /* diag_listed_as: To%s: illegal mapping '%s' */
2804 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2810 *val = 0; /* bits == 1, then any val should be ignored */
2812 else { /* Nothing following range min, should be single element with no
2817 /* diag_listed_as: To%s: illegal mapping '%s' */
2818 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2822 *val = 0; /* bits == 1, then val should be ignored */
2825 /* Position to next line if any, or EOF */
2835 * Returns a swatch (a bit vector string) for a code point sequence
2836 * that starts from the value C<start> and comprises the number C<span>.
2837 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2838 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2841 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
2844 U8 *l, *lend, *x, *xend, *s, *send;
2845 STRLEN lcur, xcur, scur;
2846 HV *const hv = MUTABLE_HV(SvRV(swash));
2847 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
2849 SV** listsvp = NULL; /* The string containing the main body of the table */
2850 SV** extssvp = NULL;
2851 SV** invert_it_svp = NULL;
2854 STRLEN octets; /* if bits == 1, then octets == 0 */
2856 UV end = start + span;
2858 if (invlistsvp == NULL) {
2859 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2860 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2861 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2862 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2863 listsvp = hv_fetchs(hv, "LIST", FALSE);
2864 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2866 bits = SvUV(*bitssvp);
2867 none = SvUV(*nonesvp);
2868 typestr = (U8*)SvPV_nolen(*typesvp);
2874 octets = bits >> 3; /* if bits == 1, then octets == 0 */
2876 PERL_ARGS_ASSERT_SWATCH_GET;
2878 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
2879 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
2883 /* If overflowed, use the max possible */
2889 /* create and initialize $swatch */
2890 scur = octets ? (span * octets) : (span + 7) / 8;
2891 swatch = newSV(scur);
2893 s = (U8*)SvPVX(swatch);
2894 if (octets && none) {
2895 const U8* const e = s + scur;
2898 *s++ = (U8)(none & 0xff);
2899 else if (bits == 16) {
2900 *s++ = (U8)((none >> 8) & 0xff);
2901 *s++ = (U8)( none & 0xff);
2903 else if (bits == 32) {
2904 *s++ = (U8)((none >> 24) & 0xff);
2905 *s++ = (U8)((none >> 16) & 0xff);
2906 *s++ = (U8)((none >> 8) & 0xff);
2907 *s++ = (U8)( none & 0xff);
2913 (void)memzero((U8*)s, scur + 1);
2915 SvCUR_set(swatch, scur);
2916 s = (U8*)SvPVX(swatch);
2918 if (invlistsvp) { /* If has an inversion list set up use that */
2919 _invlist_populate_swatch(*invlistsvp, start, end, s);
2923 /* read $swash->{LIST} */
2924 l = (U8*)SvPV(*listsvp, lcur);
2927 UV min, max, val, upper;
2928 l = swash_scan_list_line(l, lend, &min, &max, &val,
2929 cBOOL(octets), typestr);
2934 /* If looking for something beyond this range, go try the next one */
2938 /* <end> is generally 1 beyond where we want to set things, but at the
2939 * platform's infinity, where we can't go any higher, we want to
2940 * include the code point at <end> */
2943 : (max != UV_MAX || end != UV_MAX)
2950 if (!none || val < none) {
2955 for (key = min; key <= upper; key++) {
2957 /* offset must be non-negative (start <= min <= key < end) */
2958 offset = octets * (key - start);
2960 s[offset] = (U8)(val & 0xff);
2961 else if (bits == 16) {
2962 s[offset ] = (U8)((val >> 8) & 0xff);
2963 s[offset + 1] = (U8)( val & 0xff);
2965 else if (bits == 32) {
2966 s[offset ] = (U8)((val >> 24) & 0xff);
2967 s[offset + 1] = (U8)((val >> 16) & 0xff);
2968 s[offset + 2] = (U8)((val >> 8) & 0xff);
2969 s[offset + 3] = (U8)( val & 0xff);
2972 if (!none || val < none)
2976 else { /* bits == 1, then val should be ignored */
2981 for (key = min; key <= upper; key++) {
2982 const STRLEN offset = (STRLEN)(key - start);
2983 s[offset >> 3] |= 1 << (offset & 7);
2988 /* Invert if the data says it should be. Assumes that bits == 1 */
2989 if (invert_it_svp && SvUV(*invert_it_svp)) {
2991 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
2992 * be 0, and their inversion should also be 0, as we don't succeed any
2993 * Unicode property matches for non-Unicode code points */
2994 if (start <= PERL_UNICODE_MAX) {
2996 /* The code below assumes that we never cross the
2997 * Unicode/above-Unicode boundary in a range, as otherwise we would
2998 * have to figure out where to stop flipping the bits. Since this
2999 * boundary is divisible by a large power of 2, and swatches comes
3000 * in small powers of 2, this should be a valid assumption */
3001 assert(start + span - 1 <= PERL_UNICODE_MAX);
3011 /* read $swash->{EXTRAS}
3012 * This code also copied to swash_to_invlist() below */
3013 x = (U8*)SvPV(*extssvp, xcur);
3021 SV **otherbitssvp, *other;
3025 const U8 opc = *x++;
3029 nl = (U8*)memchr(x, '\n', xend - x);
3031 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3033 x = nl + 1; /* 1 is length of "\n" */
3037 x = xend; /* to EXTRAS' end at which \n is not found */
3044 namelen = nl - namestr;
3048 namelen = xend - namestr;
3052 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3053 otherhv = MUTABLE_HV(SvRV(*othersvp));
3054 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3055 otherbits = (STRLEN)SvUV(*otherbitssvp);
3056 if (bits < otherbits)
3057 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3058 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3060 /* The "other" swatch must be destroyed after. */
3061 other = swatch_get(*othersvp, start, span);
3062 o = (U8*)SvPV(other, olen);
3065 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3067 s = (U8*)SvPV(swatch, slen);
3068 if (bits == 1 && otherbits == 1) {
3070 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3071 "mismatch, slen=%"UVuf", olen=%"UVuf,
3072 (UV)slen, (UV)olen);
3096 STRLEN otheroctets = otherbits >> 3;
3098 U8* const send = s + slen;
3103 if (otherbits == 1) {
3104 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3108 STRLEN vlen = otheroctets;
3116 if (opc == '+' && otherval)
3117 NOOP; /* replace with otherval */
3118 else if (opc == '!' && !otherval)
3120 else if (opc == '-' && otherval)
3122 else if (opc == '&' && !otherval)
3125 s += octets; /* no replacement */
3130 *s++ = (U8)( otherval & 0xff);
3131 else if (bits == 16) {
3132 *s++ = (U8)((otherval >> 8) & 0xff);
3133 *s++ = (U8)( otherval & 0xff);
3135 else if (bits == 32) {
3136 *s++ = (U8)((otherval >> 24) & 0xff);
3137 *s++ = (U8)((otherval >> 16) & 0xff);
3138 *s++ = (U8)((otherval >> 8) & 0xff);
3139 *s++ = (U8)( otherval & 0xff);
3143 sv_free(other); /* through with it! */
3149 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3152 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3153 * Can't be used on a property that is subject to user override, as it
3154 * relies on the value of SPECIALS in the swash which would be set by
3155 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3156 * for overridden properties
3158 * Returns a hash which is the inversion and closure of a swash mapping.
3159 * For example, consider the input lines:
3164 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3165 * 006C. The value for each key is an array. For 006C, the array would
3166 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3167 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3169 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3170 * keys are only code points that are folded-to, so it isn't a full closure.
3172 * Essentially, for any code point, it gives all the code points that map to
3173 * it, or the list of 'froms' for that point.
3175 * Currently it ignores any additions or deletions from other swashes,
3176 * looking at just the main body of the swash, and if there are SPECIALS
3177 * in the swash, at that hash
3179 * The specials hash can be extra code points, and most likely consists of
3180 * maps from single code points to multiple ones (each expressed as a string
3181 * of utf8 characters). This function currently returns only 1-1 mappings.
3182 * However consider this possible input in the specials hash:
3183 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3184 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3186 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3187 * currently handle. But it also means that FB05 and FB06 are equivalent in
3188 * a 1-1 mapping which we should handle, and this relationship may not be in
3189 * the main table. Therefore this function examines all the multi-char
3190 * sequences and adds the 1-1 mappings that come out of that. */
3194 HV *const hv = MUTABLE_HV(SvRV(swash));
3196 /* The string containing the main body of the table. This will have its
3197 * assertion fail if the swash has been converted to its inversion list */
3198 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3200 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3201 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3202 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3203 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3204 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3205 const STRLEN bits = SvUV(*bitssvp);
3206 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3207 const UV none = SvUV(*nonesvp);
3208 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3212 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3214 /* Must have at least 8 bits to get the mappings */
3215 if (bits != 8 && bits != 16 && bits != 32) {
3216 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3220 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3221 mapping to more than one character */
3223 /* Construct an inverse mapping hash for the specials */
3224 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3225 HV * specials_inverse = newHV();
3226 char *char_from; /* the lhs of the map */
3227 I32 from_len; /* its byte length */
3228 char *char_to; /* the rhs of the map */
3229 I32 to_len; /* its byte length */
3230 SV *sv_to; /* and in a sv */
3231 AV* from_list; /* list of things that map to each 'to' */
3233 hv_iterinit(specials_hv);
3235 /* The keys are the characters (in utf8) that map to the corresponding
3236 * utf8 string value. Iterate through the list creating the inverse
3238 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3240 if (! SvPOK(sv_to)) {
3241 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3242 "unexpectedly is not a string, flags=%lu",
3243 (unsigned long)SvFLAGS(sv_to));
3245 /*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)));*/
3247 /* Each key in the inverse list is a mapped-to value, and the key's
3248 * hash value is a list of the strings (each in utf8) that map to
3249 * it. Those strings are all one character long */
3250 if ((listp = hv_fetch(specials_inverse,
3254 from_list = (AV*) *listp;
3256 else { /* No entry yet for it: create one */
3257 from_list = newAV();
3258 if (! hv_store(specials_inverse,
3261 (SV*) from_list, 0))
3263 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3267 /* Here have the list associated with this 'to' (perhaps newly
3268 * created and empty). Just add to it. Note that we ASSUME that
3269 * the input is guaranteed to not have duplications, so we don't
3270 * check for that. Duplications just slow down execution time. */
3271 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3274 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3275 * it looking for cases like the FB05/FB06 examples above. There would
3276 * be an entry in the hash like
3277 * 'st' => [ FB05, FB06 ]
3278 * In this example we will create two lists that get stored in the
3279 * returned hash, 'ret':
3280 * FB05 => [ FB05, FB06 ]
3281 * FB06 => [ FB05, FB06 ]
3283 * Note that there is nothing to do if the array only has one element.
3284 * (In the normal 1-1 case handled below, we don't have to worry about
3285 * two lists, as everything gets tied to the single list that is
3286 * generated for the single character 'to'. But here, we are omitting
3287 * that list, ('st' in the example), so must have multiple lists.) */
3288 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3289 &char_to, &to_len)))
3291 if (av_tindex(from_list) > 0) {
3294 /* We iterate over all combinations of i,j to place each code
3295 * point on each list */
3296 for (i = 0; i <= av_tindex(from_list); i++) {
3298 AV* i_list = newAV();
3299 SV** entryp = av_fetch(from_list, i, FALSE);
3300 if (entryp == NULL) {
3301 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3303 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3304 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3306 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3307 (SV*) i_list, FALSE))
3309 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3312 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3313 for (j = 0; j <= av_tindex(from_list); j++) {
3314 entryp = av_fetch(from_list, j, FALSE);
3315 if (entryp == NULL) {
3316 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3319 /* When i==j this adds itself to the list */
3320 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3321 (U8*) SvPVX(*entryp),
3322 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3324 /*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));*/
3329 SvREFCNT_dec(specials_inverse); /* done with it */
3330 } /* End of specials */
3332 /* read $swash->{LIST} */
3333 l = (U8*)SvPV(*listsvp, lcur);
3336 /* Go through each input line */
3340 l = swash_scan_list_line(l, lend, &min, &max, &val,
3341 cBOOL(octets), typestr);
3346 /* Each element in the range is to be inverted */
3347 for (inverse = min; inverse <= max; inverse++) {
3351 bool found_key = FALSE;
3352 bool found_inverse = FALSE;
3354 /* The key is the inverse mapping */
3355 char key[UTF8_MAXBYTES+1];
3356 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3357 STRLEN key_len = key_end - key;
3359 /* Get the list for the map */
3360 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3361 list = (AV*) *listp;
3363 else { /* No entry yet for it: create one */
3365 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3366 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3370 /* Look through list to see if this inverse mapping already is
3371 * listed, or if there is a mapping to itself already */
3372 for (i = 0; i <= av_tindex(list); i++) {
3373 SV** entryp = av_fetch(list, i, FALSE);
3376 if (entryp == NULL) {
3377 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3381 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3385 if (uv == inverse) {
3386 found_inverse = TRUE;
3389 /* No need to continue searching if found everything we are
3391 if (found_key && found_inverse) {
3396 /* Make sure there is a mapping to itself on the list */
3398 av_push(list, newSVuv(val));
3399 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3403 /* Simply add the value to the list */
3404 if (! found_inverse) {
3405 av_push(list, newSVuv(inverse));
3406 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3409 /* swatch_get() increments the value of val for each element in the
3410 * range. That makes more compact tables possible. You can
3411 * express the capitalization, for example, of all consecutive
3412 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3413 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3414 * and it's not documented; it appears to be used only in
3415 * implementing tr//; I copied the semantics from swatch_get(), just
3417 if (!none || val < none) {
3427 Perl__swash_to_invlist(pTHX_ SV* const swash)
3430 /* Subject to change or removal. For use only in one place in regcomp.c.
3431 * Ownership is given to one reference count in the returned SV* */
3436 HV *const hv = MUTABLE_HV(SvRV(swash));
3437 UV elements = 0; /* Number of elements in the inversion list */
3447 STRLEN octets; /* if bits == 1, then octets == 0 */
3453 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3455 /* If not a hash, it must be the swash's inversion list instead */
3456 if (SvTYPE(hv) != SVt_PVHV) {
3457 return SvREFCNT_inc_simple_NN((SV*) hv);
3460 /* The string containing the main body of the table */
3461 listsvp = hv_fetchs(hv, "LIST", FALSE);
3462 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3463 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3464 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3465 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3467 typestr = (U8*)SvPV_nolen(*typesvp);
3468 bits = SvUV(*bitssvp);
3469 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3471 /* read $swash->{LIST} */
3472 if (SvPOK(*listsvp)) {
3473 l = (U8*)SvPV(*listsvp, lcur);
3476 /* LIST legitimately doesn't contain a string during compilation phases
3477 * of Perl itself, before the Unicode tables are generated. In this
3478 * case, just fake things up by creating an empty list */
3485 if (*l == 'V') { /* Inversion list format */
3486 const char *after_atou = (char *) lend;
3488 UV* other_elements_ptr;
3490 /* The first number is a count of the rest */
3492 elements = grok_atou((const char *)l, &after_atou);
3493 if (elements == 0) {
3494 invlist = _new_invlist(0);
3497 while (isSPACE(*l)) l++;
3498 l = (U8 *) after_atou;
3500 /* Get the 0th element, which is needed to setup the inversion list */
3501 while (isSPACE(*l)) l++;
3502 element0 = (UV) grok_atou((const char *)l, &after_atou);
3503 l = (U8 *) after_atou;
3504 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3507 /* Then just populate the rest of the input */
3508 while (elements-- > 0) {
3510 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3512 while (isSPACE(*l)) l++;
3513 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3514 l = (U8 *) after_atou;
3520 /* Scan the input to count the number of lines to preallocate array
3521 * size based on worst possible case, which is each line in the input
3522 * creates 2 elements in the inversion list: 1) the beginning of a
3523 * range in the list; 2) the beginning of a range not in the list. */
3524 while ((loc = (strchr(loc, '\n'))) != NULL) {
3529 /* If the ending is somehow corrupt and isn't a new line, add another
3530 * element for the final range that isn't in the inversion list */
3531 if (! (*lend == '\n'
3532 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3537 invlist = _new_invlist(elements);
3539 /* Now go through the input again, adding each range to the list */
3542 UV val; /* Not used by this function */
3544 l = swash_scan_list_line(l, lend, &start, &end, &val,
3545 cBOOL(octets), typestr);
3551 invlist = _add_range_to_invlist(invlist, start, end);
3555 /* Invert if the data says it should be */
3556 if (invert_it_svp && SvUV(*invert_it_svp)) {
3557 _invlist_invert(invlist);
3560 /* This code is copied from swatch_get()
3561 * read $swash->{EXTRAS} */
3562 x = (U8*)SvPV(*extssvp, xcur);
3570 SV **otherbitssvp, *other;
3573 const U8 opc = *x++;
3577 nl = (U8*)memchr(x, '\n', xend - x);
3579 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3581 x = nl + 1; /* 1 is length of "\n" */
3585 x = xend; /* to EXTRAS' end at which \n is not found */
3592 namelen = nl - namestr;
3596 namelen = xend - namestr;
3600 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3601 otherhv = MUTABLE_HV(SvRV(*othersvp));
3602 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3603 otherbits = (STRLEN)SvUV(*otherbitssvp);
3605 if (bits != otherbits || bits != 1) {
3606 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3607 "properties, bits=%"UVuf", otherbits=%"UVuf,
3608 (UV)bits, (UV)otherbits);
3611 /* The "other" swatch must be destroyed after. */
3612 other = _swash_to_invlist((SV *)*othersvp);
3614 /* End of code copied from swatch_get() */
3617 _invlist_union(invlist, other, &invlist);
3620 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3623 _invlist_subtract(invlist, other, &invlist);
3626 _invlist_intersection(invlist, other, &invlist);
3631 sv_free(other); /* through with it! */
3634 SvREADONLY_on(invlist);
3639 Perl__get_swash_invlist(pTHX_ SV* const swash)
3643 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3645 if (! SvROK(swash)) {
3649 /* If it really isn't a hash, it isn't really swash; must be an inversion
3651 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3655 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3664 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3666 /* May change: warns if surrogates, non-character code points, or
3667 * non-Unicode code points are in s which has length len bytes. Returns
3668 * TRUE if none found; FALSE otherwise. The only other validity check is
3669 * to make sure that this won't exceed the string's length */
3671 const U8* const e = s + len;
3674 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3677 if (UTF8SKIP(s) > len) {
3678 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3679 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3682 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3684 if (UTF8_IS_SUPER(s)) {
3685 if (ckWARN_d(WARN_NON_UNICODE)) {
3686 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3687 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3688 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3692 else if (UTF8_IS_SURROGATE(s)) {
3693 if (ckWARN_d(WARN_SURROGATE)) {
3694 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3695 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3696 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3701 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3702 && (ckWARN_d(WARN_NONCHAR)))
3704 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3705 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3706 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3717 =for apidoc pv_uni_display
3719 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3720 length C<len>, the displayable version being at most C<pvlim> bytes long
3721 (if longer, the rest is truncated and "..." will be appended).
3723 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
3724 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3725 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3726 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3727 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3728 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3730 The pointer to the PV of the C<dsv> is returned.
3734 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3739 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3743 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3745 /* This serves double duty as a flag and a character to print after
3746 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3750 if (pvlim && SvCUR(dsv) >= pvlim) {
3754 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3756 const unsigned char c = (unsigned char)u & 0xFF;
3757 if (flags & UNI_DISPLAY_BACKSLASH) {
3774 const char string = ok;
3775 sv_catpvs(dsv, "\\");
3776 sv_catpvn(dsv, &string, 1);
3779 /* isPRINT() is the locale-blind version. */
3780 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3781 const char string = c;
3782 sv_catpvn(dsv, &string, 1);
3787 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
3790 sv_catpvs(dsv, "...");
3796 =for apidoc sv_uni_display
3798 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
3799 the displayable version being at most C<pvlim> bytes long
3800 (if longer, the rest is truncated and "..." will be appended).
3802 The C<flags> argument is as in L</pv_uni_display>().
3804 The pointer to the PV of the C<dsv> is returned.
3809 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
3811 const char * const ptr =
3812 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
3814 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
3816 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
3817 SvCUR(ssv), pvlim, flags);
3821 =for apidoc foldEQ_utf8
3823 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
3824 of which may be in UTF-8) are the same case-insensitively; false otherwise.
3825 How far into the strings to compare is determined by other input parameters.
3827 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
3828 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
3829 with respect to C<s2>.
3831 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
3832 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
3833 scan will not be considered to be a match unless the goal is reached, and
3834 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
3837 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
3838 considered an end pointer to the position 1 byte past the maximum point
3839 in C<s1> beyond which scanning will not continue under any circumstances.
3840 (This routine assumes that UTF-8 encoded input strings are not malformed;
3841 malformed input can cause it to read past C<pe1>).
3842 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
3843 is less than C<s1>+C<l1>, the match will never be successful because it can
3845 get as far as its goal (and in fact is asserted against). Correspondingly for
3846 C<pe2> with respect to C<s2>.
3848 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
3849 C<l2> must be non-zero), and if both do, both have to be
3850 reached for a successful match. Also, if the fold of a character is multiple
3851 characters, all of them must be matched (see tr21 reference below for
3854 Upon a successful match, if C<pe1> is non-NULL,
3855 it will be set to point to the beginning of the I<next> character of C<s1>
3856 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
3858 For case-insensitiveness, the "casefolding" of Unicode is used
3859 instead of upper/lowercasing both the characters, see
3860 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
3864 /* A flags parameter has been added which may change, and hence isn't
3865 * externally documented. Currently it is:
3866 * 0 for as-documented above
3867 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
3868 ASCII one, to not match
3869 * FOLDEQ_LOCALE is set iff the rules from the current underlying
3870 * locale are to be used.
3871 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
3872 * routine. This allows that step to be skipped.
3873 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
3876 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)
3878 const U8 *p1 = (const U8*)s1; /* Point to current char */
3879 const U8 *p2 = (const U8*)s2;
3880 const U8 *g1 = NULL; /* goal for s1 */
3881 const U8 *g2 = NULL;
3882 const U8 *e1 = NULL; /* Don't scan s1 past this */
3883 U8 *f1 = NULL; /* Point to current folded */
3884 const U8 *e2 = NULL;
3886 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
3887 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
3888 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
3890 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
3892 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
3893 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
3894 /* The algorithm is to trial the folds without regard to the flags on
3895 * the first line of the above assert(), and then see if the result
3896 * violates them. This means that the inputs can't be pre-folded to a
3897 * violating result, hence the assert. This could be changed, with the
3898 * addition of extra tests here for the already-folded case, which would
3899 * slow it down. That cost is more than any possible gain for when these
3900 * flags are specified, as the flags indicate /il or /iaa matching which
3901 * is less common than /iu, and I (khw) also believe that real-world /il
3902 * and /iaa matches are most likely to involve code points 0-255, and this
3903 * function only under rare conditions gets called for 0-255. */
3905 if (IN_UTF8_CTYPE_LOCALE) {
3906 flags &= ~FOLDEQ_LOCALE;
3914 g1 = (const U8*)s1 + l1;
3922 g2 = (const U8*)s2 + l2;
3925 /* Must have at least one goal */
3930 /* Will never match if goal is out-of-bounds */
3931 assert(! e1 || e1 >= g1);
3933 /* Here, there isn't an end pointer, or it is beyond the goal. We
3934 * only go as far as the goal */
3938 assert(e1); /* Must have an end for looking at s1 */
3941 /* Same for goal for s2 */
3943 assert(! e2 || e2 >= g2);
3950 /* If both operands are already folded, we could just do a memEQ on the
3951 * whole strings at once, but it would be better if the caller realized
3952 * this and didn't even call us */
3954 /* Look through both strings, a character at a time */
3955 while (p1 < e1 && p2 < e2) {
3957 /* If at the beginning of a new character in s1, get its fold to use
3958 * and the length of the fold. (exception: locale rules just get the
3959 * character to a single byte) */
3961 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
3966 /* If in locale matching, we use two sets of rules, depending
3967 * on if the code point is above or below 255. Here, we test
3968 * for and handle locale rules */
3969 if ((flags & FOLDEQ_LOCALE)
3970 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
3972 /* There is no mixing of code points above and below 255. */
3973 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
3977 /* We handle locale rules by converting, if necessary, the
3978 * code point to a single byte. */
3979 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
3983 *foldbuf1 = TWO_BYTE_UTF8_TO_NATIVE(*p1, *(p1 + 1));
3987 else if (isASCII(*p1)) { /* Note, that here won't be both
3988 ASCII and using locale rules */
3990 /* If trying to mix non- with ASCII, and not supposed to,
3992 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
3996 *foldbuf1 = toFOLD(*p1);
3999 to_utf8_fold(p1, foldbuf1, &n1);
4001 else { /* Not utf8, get utf8 fold */
4002 to_uni_fold(*p1, foldbuf1, &n1);
4008 if (n2 == 0) { /* Same for s2 */
4009 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4014 if ((flags & FOLDEQ_LOCALE)
4015 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4017 /* Here, the next char in s2 is < 256. We've already
4018 * worked on s1, and if it isn't also < 256, can't match */
4019 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4022 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4026 *foldbuf2 = TWO_BYTE_UTF8_TO_NATIVE(*p2, *(p2 + 1));
4029 /* Use another function to handle locale rules. We've made
4030 * sure that both characters to compare are single bytes */
4031 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4036 else if (isASCII(*p2)) {
4037 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4041 *foldbuf2 = toFOLD(*p2);
4044 to_utf8_fold(p2, foldbuf2, &n2);
4047 to_uni_fold(*p2, foldbuf2, &n2);
4053 /* Here f1 and f2 point to the beginning of the strings to compare.
4054 * These strings are the folds of the next character from each input
4055 * string, stored in utf8. */
4057 /* While there is more to look for in both folds, see if they
4058 * continue to match */
4060 U8 fold_length = UTF8SKIP(f1);
4061 if (fold_length != UTF8SKIP(f2)
4062 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4063 function call for single
4065 || memNE((char*)f1, (char*)f2, fold_length))
4067 return 0; /* mismatch */
4070 /* Here, they matched, advance past them */
4077 /* When reach the end of any fold, advance the input past it */
4079 p1 += u1 ? UTF8SKIP(p1) : 1;
4082 p2 += u2 ? UTF8SKIP(p2) : 1;
4084 } /* End of loop through both strings */
4086 /* A match is defined by each scan that specified an explicit length
4087 * reaching its final goal, and the other not having matched a partial
4088 * character (which can happen when the fold of a character is more than one
4090 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4094 /* Successful match. Set output pointers */
4104 /* XXX The next two functions should likely be moved to mathoms.c once all
4105 * occurrences of them are removed from the core; some cpan-upstream modules
4109 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4111 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4113 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4117 =for apidoc utf8n_to_uvuni
4119 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4121 This function was useful for code that wanted to handle both EBCDIC and
4122 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4123 distinctions between the platforms have mostly been made invisible to most
4124 code, so this function is quite unlikely to be what you want. If you do need
4125 this precise functionality, use instead
4126 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4127 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4133 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4135 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4137 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4141 =for apidoc uvuni_to_utf8_flags
4143 Instead you almost certainly want to use L</uvchr_to_utf8> or
4144 L</uvchr_to_utf8_flags>>.
4146 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4147 which itself, while not deprecated, should be used only in isolated
4148 circumstances. These functions were useful for code that wanted to handle
4149 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4150 v5.20, the distinctions between the platforms have mostly been made invisible
4151 to most code, so this function is quite unlikely to be what you want.
4157 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4159 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4161 return uvoffuni_to_utf8_flags(d, uv, flags);
4166 * c-indentation-style: bsd
4168 * indent-tabs-mode: nil
4171 * ex: set ts=8 sts=4 sw=4 et: