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);
111 /* Not representable in UTF-EBCDIC */
112 flags |= UNICODE_DISALLOW_FE_FF;
115 /* The first problematic code point is the first surrogate */
116 if (uv >= UNICODE_SURROGATE_FIRST
117 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
119 if (UNICODE_IS_SURROGATE(uv)) {
120 if (flags & UNICODE_WARN_SURROGATE) {
121 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
122 "UTF-16 surrogate U+%04"UVXf, uv);
124 if (flags & UNICODE_DISALLOW_SURROGATE) {
128 else if (UNICODE_IS_SUPER(uv)) {
129 if (flags & UNICODE_WARN_SUPER
130 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
132 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
133 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
135 if (flags & UNICODE_DISALLOW_SUPER
136 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
139 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
145 else if (UNICODE_IS_NONCHAR(uv)) {
146 if (flags & UNICODE_WARN_NONCHAR) {
147 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
148 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
151 if (flags & UNICODE_DISALLOW_NONCHAR) {
159 STRLEN len = OFFUNISKIP(uv);
162 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
163 uv >>= UTF_ACCUMULATION_SHIFT;
165 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
168 #else /* Non loop style */
170 *d++ = (U8)(( uv >> 6) | 0xc0);
171 *d++ = (U8)(( uv & 0x3f) | 0x80);
175 *d++ = (U8)(( uv >> 12) | 0xe0);
176 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
177 *d++ = (U8)(( uv & 0x3f) | 0x80);
181 *d++ = (U8)(( uv >> 18) | 0xf0);
182 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
184 *d++ = (U8)(( uv & 0x3f) | 0x80);
187 if (uv < 0x4000000) {
188 *d++ = (U8)(( uv >> 24) | 0xf8);
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);
195 if (uv < 0x80000000) {
196 *d++ = (U8)(( uv >> 30) | 0xfc);
197 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
198 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
199 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
201 *d++ = (U8)(( uv & 0x3f) | 0x80);
205 if (uv < UTF8_QUAD_MAX)
208 *d++ = 0xfe; /* Can't match U+FEFF! */
209 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
210 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
211 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
212 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
214 *d++ = (U8)(( uv & 0x3f) | 0x80);
219 *d++ = 0xff; /* Can't match U+FFFE! */
220 *d++ = 0x80; /* 6 Reserved bits */
221 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
222 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
223 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
231 *d++ = (U8)(( uv & 0x3f) | 0x80);
235 #endif /* Non loop style */
238 =for apidoc uvchr_to_utf8
240 Adds the UTF-8 representation of the native code point C<uv> to the end
241 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
242 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
243 the byte after the end of the new character. In other words,
245 d = uvchr_to_utf8(d, uv);
247 is the recommended wide native character-aware way of saying
251 This function accepts any UV as input. To forbid or warn on non-Unicode code
252 points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
257 /* This is also a macro */
258 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
261 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
263 return uvchr_to_utf8(d, uv);
267 =for apidoc uvchr_to_utf8_flags
269 Adds the UTF-8 representation of the native code point C<uv> to the end
270 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
271 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
272 the byte after the end of the new character. In other words,
274 d = uvchr_to_utf8_flags(d, uv, flags);
278 d = uvchr_to_utf8_flags(d, uv, 0);
280 This is the Unicode-aware way of saying
284 This function will convert to UTF-8 (and not warn) even code points that aren't
285 legal Unicode or are problematic, unless C<flags> contains one or more of the
288 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
289 the function will raise a warning, provided UTF8 warnings are enabled. If instead
290 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
291 If both flags are set, the function will both warn and return NULL.
293 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
294 affect how the function handles a Unicode non-character. And likewise, the
295 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
297 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
298 even less portable) can be warned and/or disallowed even if other above-Unicode
299 code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
302 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
303 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
309 /* This is also a macro */
310 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
313 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
315 return uvchr_to_utf8_flags(d, uv, flags);
319 =for apidoc is_utf8_char_buf
321 This is identical to the macro L</isUTF8_CHAR>.
326 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
329 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
331 return isUTF8_CHAR(buf, buf_end);
335 =for apidoc is_utf8_string
337 Returns true if the first C<len> bytes of string C<s> form a valid
338 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
339 using C<strlen(s)> (which means if you use this option, that C<s> can't have
340 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
341 that all characters being ASCII constitute 'a valid UTF-8 string'.
343 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
349 Perl_is_utf8_string(const U8 *s, STRLEN len)
351 const U8* const send = s + (len ? len : strlen((const char *)s));
354 PERL_ARGS_ASSERT_IS_UTF8_STRING;
357 STRLEN len = isUTF8_CHAR(x, send);
358 if (UNLIKELY(! len)) {
368 Implemented as a macro in utf8.h
370 =for apidoc is_utf8_string_loc
372 Like L</is_utf8_string> but stores the location of the failure (in the
373 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
374 "utf8ness success") in the C<ep>.
376 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
378 =for apidoc is_utf8_string_loclen
380 Like L</is_utf8_string>() but stores the location of the failure (in the
381 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
382 "utf8ness success") in the C<ep>, and the number of UTF-8
383 encoded characters in the C<el>.
385 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
391 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
393 const U8* const send = s + (len ? len : strlen((const char *)s));
397 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
400 STRLEN len = isUTF8_CHAR(x, send);
401 if (UNLIKELY(! len)) {
419 =for apidoc utf8n_to_uvchr
421 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
422 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
424 Bottom level UTF-8 decode routine.
425 Returns the native code point value of the first character in the string C<s>,
426 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
427 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
428 the length, in bytes, of that character.
430 The value of C<flags> determines the behavior when C<s> does not point to a
431 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
432 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
433 next possible position in C<s> that could begin a non-malformed character.
434 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
436 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
437 individual types of malformations, such as the sequence being overlong (that
438 is, when there is a shorter sequence that can express the same code point;
439 overlong sequences are expressly forbidden in the UTF-8 standard due to
440 potential security issues). Another malformation example is the first byte of
441 a character not being a legal first byte. See F<utf8.h> for the list of such
442 flags. For allowed 0 length strings, this function returns 0; for allowed
443 overlong sequences, the computed code point is returned; for all other allowed
444 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
445 determinable reasonable value.
447 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
448 flags) malformation is found. If this flag is set, the routine assumes that
449 the caller will raise a warning, and this function will silently just set
450 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
452 Note that this API requires disambiguation between successful decoding a C<NUL>
453 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
454 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
455 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
458 Certain code points are considered problematic. These are Unicode surrogates,
459 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
460 By default these are considered regular code points, but certain situations
461 warrant special handling for them. If C<flags> contains
462 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
463 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
464 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
465 maximum) can be set to disallow these categories individually.
467 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
468 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
469 for their respective categories, but otherwise the code points are considered
470 valid (not malformations). To get a category to both be treated as a
471 malformation and raise a warning, specify both the WARN and DISALLOW flags.
472 (But note that warnings are not raised if lexically disabled nor if
473 UTF8_CHECK_ONLY is also specified.)
475 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
476 the others that are above the Unicode legal maximum. There are several
477 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
478 not representable at all on EBCDIC platforms, and the original UTF-8
479 specification never went above this number (the current 0x10FFFF limit was
480 imposed later). (The smaller ones, those that fit into 32 bits, are
481 representable by a UV on ASCII platforms, but not by an IV, which means that
482 the number of operations that can be performed on them is quite restricted.)
483 The UTF-8 encoding on ASCII platforms for these large code points begins with a
484 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
485 be treated as malformations, while allowing smaller above-Unicode code points.
486 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
487 including these, as malformations.)
488 Similarly, UTF8_WARN_FE_FF acts just like
489 the other WARN flags, but applies just to these code points.
491 All other code points corresponding to Unicode characters, including private
492 use and those yet to be assigned, are never considered malformed and never
499 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
501 const U8 * const s0 = s;
502 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
507 UV outlier_ret = 0; /* return value when input is in error or problematic
509 UV pack_warn = 0; /* Save result of packWARN() for later */
510 bool unexpected_non_continuation = FALSE;
511 bool overflowed = FALSE;
512 bool do_overlong_test = TRUE; /* May have to skip this test */
514 const char* const malformed_text = "Malformed UTF-8 character";
516 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
518 /* The order of malformation tests here is important. We should consume as
519 * few bytes as possible in order to not skip any valid character. This is
520 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
521 * http://unicode.org/reports/tr36 for more discussion as to why. For
522 * example, once we've done a UTF8SKIP, we can tell the expected number of
523 * bytes, and could fail right off the bat if the input parameters indicate
524 * that there are too few available. But it could be that just that first
525 * byte is garbled, and the intended character occupies fewer bytes. If we
526 * blindly assumed that the first byte is correct, and skipped based on
527 * that number, we could skip over a valid input character. So instead, we
528 * always examine the sequence byte-by-byte.
530 * We also should not consume too few bytes, otherwise someone could inject
531 * things. For example, an input could be deliberately designed to
532 * overflow, and if this code bailed out immediately upon discovering that,
533 * returning to the caller C<*retlen> pointing to the very next byte (one
534 * which is actually part of of the overflowing sequence), that could look
535 * legitimate to the caller, which could discard the initial partial
536 * sequence and process the rest, inappropriately */
538 /* Zero length strings, if allowed, of necessity are zero */
539 if (UNLIKELY(curlen == 0)) {
544 if (flags & UTF8_ALLOW_EMPTY) {
547 if (! (flags & UTF8_CHECK_ONLY)) {
548 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
553 expectlen = UTF8SKIP(s);
555 /* A well-formed UTF-8 character, as the vast majority of calls to this
556 * function will be for, has this expected length. For efficiency, set
557 * things up here to return it. It will be overriden only in those rare
558 * cases where a malformation is found */
563 /* An invariant is trivially well-formed */
564 if (UTF8_IS_INVARIANT(uv)) {
568 /* A continuation character can't start a valid sequence */
569 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
570 if (flags & UTF8_ALLOW_CONTINUATION) {
574 return UNICODE_REPLACEMENT;
577 if (! (flags & UTF8_CHECK_ONLY)) {
578 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
584 /* Here is not a continuation byte, nor an invariant. The only thing left
585 * is a start byte (possibly for an overlong) */
588 uv = NATIVE_UTF8_TO_I8(uv);
591 /* Remove the leading bits that indicate the number of bytes in the
592 * character's whole UTF-8 sequence, leaving just the bits that are part of
594 uv &= UTF_START_MASK(expectlen);
596 /* Now, loop through the remaining bytes in the character's sequence,
597 * accumulating each into the working value as we go. Be sure to not look
598 * past the end of the input string */
599 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
601 for (s = s0 + 1; s < send; s++) {
602 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
603 #ifndef EBCDIC /* Can't overflow in EBCDIC */
604 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
606 /* The original implementors viewed this malformation as more
607 * serious than the others (though I, khw, don't understand
608 * why, since other malformations also give very very wrong
609 * results), so there is no way to turn off checking for it.
610 * Set a flag, but keep going in the loop, so that we absorb
611 * the rest of the bytes that comprise the character. */
613 overflow_byte = *s; /* Save for warning message's use */
616 uv = UTF8_ACCUMULATE(uv, *s);
619 /* Here, found a non-continuation before processing all expected
620 * bytes. This byte begins a new character, so quit, even if
621 * allowing this malformation. */
622 unexpected_non_continuation = TRUE;
625 } /* End of loop through the character's bytes */
627 /* Save how many bytes were actually in the character */
630 /* The loop above finds two types of malformations: non-continuation and/or
631 * overflow. The non-continuation malformation is really a too-short
632 * malformation, as it means that the current character ended before it was
633 * expected to (being terminated prematurely by the beginning of the next
634 * character, whereas in the too-short malformation there just are too few
635 * bytes available to hold the character. In both cases, the check below
636 * that we have found the expected number of bytes would fail if executed.)
637 * Thus the non-continuation malformation is really unnecessary, being a
638 * subset of the too-short malformation. But there may be existing
639 * applications that are expecting the non-continuation type, so we retain
640 * it, and return it in preference to the too-short malformation. (If this
641 * code were being written from scratch, the two types might be collapsed
642 * into one.) I, khw, am also giving priority to returning the
643 * non-continuation and too-short malformations over overflow when multiple
644 * ones are present. I don't know of any real reason to prefer one over
645 * the other, except that it seems to me that multiple-byte errors trumps
646 * errors from a single byte */
647 if (UNLIKELY(unexpected_non_continuation)) {
648 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
649 if (! (flags & UTF8_CHECK_ONLY)) {
651 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
654 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));
659 uv = UNICODE_REPLACEMENT;
661 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
662 * as what the original expectations were. */
663 do_overlong_test = FALSE;
668 else if (UNLIKELY(curlen < expectlen)) {
669 if (! (flags & UTF8_ALLOW_SHORT)) {
670 if (! (flags & UTF8_CHECK_ONLY)) {
671 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));
675 uv = UNICODE_REPLACEMENT;
676 do_overlong_test = FALSE;
682 #ifndef EBCDIC /* EBCDIC can't overflow */
683 if (UNLIKELY(overflowed)) {
684 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
690 && expectlen > (STRLEN) OFFUNISKIP(uv)
691 && ! (flags & UTF8_ALLOW_LONG))
693 /* The overlong malformation has lower precedence than the others.
694 * Note that if this malformation is allowed, we return the actual
695 * value, instead of the replacement character. This is because this
696 * value is actually well-defined. */
697 if (! (flags & UTF8_CHECK_ONLY)) {
698 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));
703 /* Here, the input is considered to be well-formed, but it still could be a
704 * problematic code point that is not allowed by the input parameters. */
705 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
706 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
707 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
709 if (UNICODE_IS_SURROGATE(uv)) {
711 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
712 * generation of the sv, since no warnings are raised under CHECK */
713 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
714 && ckWARN_d(WARN_SURROGATE))
716 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
717 pack_warn = packWARN(WARN_SURROGATE);
719 if (flags & UTF8_DISALLOW_SURROGATE) {
723 else if ((uv > PERL_UNICODE_MAX)) {
724 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
725 && ckWARN_d(WARN_NON_UNICODE))
727 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
728 pack_warn = packWARN(WARN_NON_UNICODE);
730 #ifndef EBCDIC /* EBCDIC always allows FE, FF */
732 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
733 * points. We test for these after the regular SUPER ones, and
734 * before possibly bailing out, so that the more dire warning
735 * overrides the regular one, if applicable */
736 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
737 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
739 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
741 && ckWARN_d(WARN_UTF8))
743 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
744 pack_warn = packWARN(WARN_UTF8);
746 if (flags & UTF8_DISALLOW_FE_FF) {
751 if (flags & UTF8_DISALLOW_SUPER) {
755 else if (UNICODE_IS_NONCHAR(uv)) {
756 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
757 && ckWARN_d(WARN_NONCHAR))
759 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
760 pack_warn = packWARN(WARN_NONCHAR);
762 if (flags & UTF8_DISALLOW_NONCHAR) {
768 outlier_ret = uv; /* Note we don't bother to convert to native,
769 as all the outlier code points are the same
770 in both ASCII and EBCDIC */
774 /* Here, this is not considered a malformed character, so drop through
778 return UNI_TO_NATIVE(uv);
780 /* There are three cases which get to beyond this point. In all 3 cases:
781 * <sv> if not null points to a string to print as a warning.
782 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
784 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
785 * This is done by initializing it to 0, and changing it only
788 * 1) The input is valid but problematic, and to be warned about. The
789 * return value is the resultant code point; <*retlen> is set to
790 * <curlen>, the number of bytes that comprise the code point.
791 * <pack_warn> contains the result of packWARN() for the warning
792 * types. The entry point for this case is the label <do_warn>;
793 * 2) The input is a valid code point but disallowed by the parameters to
794 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
795 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
796 * comprise the code point. <pack_warn> contains the result of
797 * packWARN() for the warning types. The entry point for this case is
798 * the label <disallowed>.
799 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
800 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
801 * bytes that comprise the malformation. All such malformations are
802 * assumed to be warning type <utf8>. The entry point for this case
803 * is the label <malformed>.
808 if (sv && ckWARN_d(WARN_UTF8)) {
809 pack_warn = packWARN(WARN_UTF8);
814 if (flags & UTF8_CHECK_ONLY) {
816 *retlen = ((STRLEN) -1);
822 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
823 if warnings are to be raised. */
824 const char * const string = SvPVX_const(sv);
827 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
829 Perl_warner(aTHX_ pack_warn, "%s", string);
840 =for apidoc utf8_to_uvchr_buf
842 Returns the native code point of the first character in the string C<s> which
843 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
844 C<*retlen> will be set to the length, in bytes, of that character.
846 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
847 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
848 NULL) to -1. If those warnings are off, the computed value, if well-defined
849 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
850 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
851 the next possible position in C<s> that could begin a non-malformed character.
852 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
860 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
864 return utf8n_to_uvchr(s, send - s, retlen,
865 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
868 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
869 * there are no malformations in the input UTF-8 string C<s>. surrogates,
870 * non-character code points, and non-Unicode code points are allowed. */
873 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
875 UV expectlen = UTF8SKIP(s);
876 const U8* send = s + expectlen;
879 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
886 /* An invariant is trivially returned */
887 if (expectlen == 1) {
892 uv = NATIVE_UTF8_TO_I8(uv);
895 /* Remove the leading bits that indicate the number of bytes, leaving just
896 * the bits that are part of the value */
897 uv &= UTF_START_MASK(expectlen);
899 /* Now, loop through the remaining bytes, accumulating each into the
900 * working total as we go. (I khw tried unrolling the loop for up to 4
901 * bytes, but there was no performance improvement) */
902 for (++s; s < send; s++) {
903 uv = UTF8_ACCUMULATE(uv, *s);
906 return UNI_TO_NATIVE(uv);
911 =for apidoc utf8_to_uvuni_buf
913 Only in very rare circumstances should code need to be dealing in Unicode
914 (as opposed to native) code points. In those few cases, use
915 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
917 Returns the Unicode (not-native) code point of the first character in the
919 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
920 C<retlen> will be set to the length, in bytes, of that character.
922 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
923 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
924 NULL) to -1. If those warnings are off, the computed value if well-defined (or
925 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
926 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
927 next possible position in C<s> that could begin a non-malformed character.
928 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
934 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
936 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
940 /* Call the low level routine asking for checks */
941 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
942 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
946 =for apidoc utf8_length
948 Return the length of the UTF-8 char encoded string C<s> in characters.
949 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
950 up past C<e>, croaks.
956 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
960 PERL_ARGS_ASSERT_UTF8_LENGTH;
962 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
963 * the bitops (especially ~) can create illegal UTF-8.
964 * In other words: in Perl UTF-8 is not just for Unicode. */
967 goto warn_and_return;
977 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
978 "%s in %s", unees, OP_DESC(PL_op));
980 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
987 =for apidoc utf8_distance
989 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
992 WARNING: use only if you *know* that the pointers point inside the
999 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1001 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1003 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1007 =for apidoc utf8_hop
1009 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1010 forward or backward.
1012 WARNING: do not use the following unless you *know* C<off> is within
1013 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1014 on the first byte of character or just after the last byte of a character.
1020 Perl_utf8_hop(const U8 *s, I32 off)
1022 PERL_ARGS_ASSERT_UTF8_HOP;
1024 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1025 * the bitops (especially ~) can create illegal UTF-8.
1026 * In other words: in Perl UTF-8 is not just for Unicode. */
1035 while (UTF8_IS_CONTINUATION(*s))
1043 =for apidoc bytes_cmp_utf8
1045 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1046 sequence of characters (stored as UTF-8)
1047 in C<u>, C<ulen>. Returns 0 if they are
1048 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1049 if the first string is greater than the second string.
1051 -1 or +1 is returned if the shorter string was identical to the start of the
1052 longer string. -2 or +2 is returned if
1053 there was a difference between characters
1060 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1062 const U8 *const bend = b + blen;
1063 const U8 *const uend = u + ulen;
1065 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1067 while (b < bend && u < uend) {
1069 if (!UTF8_IS_INVARIANT(c)) {
1070 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1073 if (UTF8_IS_CONTINUATION(c1)) {
1074 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1076 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1077 "Malformed UTF-8 character "
1078 "(unexpected non-continuation byte 0x%02x"
1079 ", immediately after start byte 0x%02x)"
1080 /* Dear diag.t, it's in the pod. */
1082 PL_op ? " in " : "",
1083 PL_op ? OP_DESC(PL_op) : "");
1088 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1089 "%s in %s", unees, OP_DESC(PL_op));
1091 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1092 return -2; /* Really want to return undef :-) */
1099 return *b < c ? -2 : +2;
1104 if (b == bend && u == uend)
1107 return b < bend ? +1 : -1;
1111 =for apidoc utf8_to_bytes
1113 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1114 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1115 updates C<len> to contain the new length.
1116 Returns zero on failure, setting C<len> to -1.
1118 If you need a copy of the string, see L</bytes_from_utf8>.
1124 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1126 U8 * const save = s;
1127 U8 * const send = s + *len;
1130 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1131 PERL_UNUSED_CONTEXT;
1133 /* ensure valid UTF-8 and chars < 256 before updating string */
1135 if (! UTF8_IS_INVARIANT(*s)) {
1136 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1137 *len = ((STRLEN) -1);
1148 if (! UTF8_IS_INVARIANT(c)) {
1149 /* Then it is two-byte encoded */
1150 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1161 =for apidoc bytes_from_utf8
1163 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1164 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1165 the newly-created string, and updates C<len> to contain the new
1166 length. Returns the original string if no conversion occurs, C<len>
1167 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1168 0 if C<s> is converted or consisted entirely of characters that are invariant
1169 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1175 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1178 const U8 *start = s;
1182 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1183 PERL_UNUSED_CONTEXT;
1187 /* ensure valid UTF-8 and chars < 256 before converting string */
1188 for (send = s + *len; s < send;) {
1189 if (! UTF8_IS_INVARIANT(*s)) {
1190 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1201 Newx(d, (*len) - count + 1, U8);
1202 s = start; start = d;
1205 if (! UTF8_IS_INVARIANT(c)) {
1206 /* Then it is two-byte encoded */
1207 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1218 =for apidoc bytes_to_utf8
1220 Converts a string C<s> of length C<len> bytes from the native encoding into
1222 Returns a pointer to the newly-created string, and sets C<len> to
1223 reflect the new length in bytes.
1225 A C<NUL> character will be written after the end of the string.
1227 If you want to convert to UTF-8 from encodings other than
1228 the native (Latin1 or EBCDIC),
1229 see L</sv_recode_to_utf8>().
1234 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1235 likewise need duplication. */
1238 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1240 const U8 * const send = s + (*len);
1244 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1245 PERL_UNUSED_CONTEXT;
1247 Newx(d, (*len) * 2 + 1, U8);
1251 append_utf8_from_native_byte(*s, &d);
1260 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1262 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1263 * We optimize for native, for obvious reasons. */
1266 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1271 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1274 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1279 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1281 if (UNI_IS_INVARIANT(uv)) {
1282 *d++ = LATIN1_TO_NATIVE((U8) uv);
1285 if (uv <= MAX_UTF8_TWO_BYTE) {
1286 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1287 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1290 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1291 #define LAST_HIGH_SURROGATE 0xDBFF
1292 #define FIRST_LOW_SURROGATE 0xDC00
1293 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1294 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1296 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1298 UV low = (p[0] << 8) + p[1];
1300 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1301 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1302 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1303 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1305 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1306 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1309 d = uvoffuni_to_utf8_flags(d, uv, 0);
1312 *d++ = (U8)(( uv >> 12) | 0xe0);
1313 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1314 *d++ = (U8)(( uv & 0x3f) | 0x80);
1318 *d++ = (U8)(( uv >> 18) | 0xf0);
1319 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1320 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1321 *d++ = (U8)(( uv & 0x3f) | 0x80);
1326 *newlen = d - dstart;
1330 /* Note: this one is slightly destructive of the source. */
1333 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1336 U8* const send = s + bytelen;
1338 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1341 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1345 const U8 tmp = s[0];
1350 return utf16_to_utf8(p, d, bytelen, newlen);
1354 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1356 U8 tmpbuf[UTF8_MAXBYTES+1];
1357 uvchr_to_utf8(tmpbuf, c);
1358 return _is_utf8_FOO(classnum, tmpbuf);
1361 /* Internal function so we can deprecate the external one, and call
1362 this one from other deprecated functions in this file */
1365 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1367 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1371 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1375 Perl__is_uni_perl_idcont(pTHX_ UV c)
1377 U8 tmpbuf[UTF8_MAXBYTES+1];
1378 uvchr_to_utf8(tmpbuf, c);
1379 return _is_utf8_perl_idcont(tmpbuf);
1383 Perl__is_uni_perl_idstart(pTHX_ UV c)
1385 U8 tmpbuf[UTF8_MAXBYTES+1];
1386 uvchr_to_utf8(tmpbuf, c);
1387 return _is_utf8_perl_idstart(tmpbuf);
1391 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1393 /* We have the latin1-range values compiled into the core, so just use
1394 * those, converting the result to utf8. The only difference between upper
1395 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1396 * either "SS" or "Ss". Which one to use is passed into the routine in
1397 * 'S_or_s' to avoid a test */
1399 UV converted = toUPPER_LATIN1_MOD(c);
1401 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1403 assert(S_or_s == 'S' || S_or_s == 's');
1405 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1406 characters in this range */
1407 *p = (U8) converted;
1412 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1413 * which it maps to one of them, so as to only have to have one check for
1414 * it in the main case */
1415 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1417 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1418 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1421 converted = GREEK_CAPITAL_LETTER_MU;
1423 case LATIN_SMALL_LETTER_SHARP_S:
1429 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1430 assert(0); /* NOTREACHED */
1434 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1435 *p = UTF8_TWO_BYTE_LO(converted);
1441 /* Call the function to convert a UTF-8 encoded character to the specified case.
1442 * Note that there may be more than one character in the result.
1443 * INP is a pointer to the first byte of the input character
1444 * OUTP will be set to the first byte of the string of changed characters. It
1445 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1446 * LENP will be set to the length in bytes of the string of changed characters
1448 * The functions return the ordinal of the first character in the string of OUTP */
1449 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1450 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1451 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1453 /* This additionally has the input parameter SPECIALS, which if non-zero will
1454 * cause this to use the SPECIALS hash for folding (meaning get full case
1455 * folding); otherwise, when zero, this implies a simple case fold */
1456 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1459 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1461 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1462 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1463 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1464 * the changed version may be longer than the original character.
1466 * The ordinal of the first character of the changed version is returned
1467 * (but note, as explained above, that there may be more.) */
1469 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1472 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1475 uvchr_to_utf8(p, c);
1476 return CALL_UPPER_CASE(p, p, lenp);
1480 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1482 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1485 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1488 uvchr_to_utf8(p, c);
1489 return CALL_TITLE_CASE(p, p, lenp);
1493 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
1495 /* We have the latin1-range values compiled into the core, so just use
1496 * those, converting the result to utf8. Since the result is always just
1497 * one character, we allow <p> to be NULL */
1499 U8 converted = toLOWER_LATIN1(c);
1502 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1507 /* Result is known to always be < 256, so can use the EIGHT_BIT
1509 *p = UTF8_EIGHT_BIT_HI(converted);
1510 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
1518 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1520 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1523 return to_lower_latin1((U8) c, p, lenp);
1526 uvchr_to_utf8(p, c);
1527 return CALL_LOWER_CASE(p, p, lenp);
1531 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1533 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1534 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1535 * FOLD_FLAGS_FULL iff full folding is to be used;
1537 * Not to be used for locale folds
1542 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1543 PERL_UNUSED_CONTEXT;
1545 assert (! (flags & FOLD_FLAGS_LOCALE));
1547 if (c == MICRO_SIGN) {
1548 converted = GREEK_SMALL_LETTER_MU;
1550 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1552 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1553 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1554 * under those circumstances. */
1555 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1556 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1557 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1559 return LATIN_SMALL_LETTER_LONG_S;
1568 else { /* In this range the fold of all other characters is their lower
1570 converted = toLOWER_LATIN1(c);
1573 if (UVCHR_IS_INVARIANT(converted)) {
1574 *p = (U8) converted;
1578 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1579 *p = UTF8_TWO_BYTE_LO(converted);
1587 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1590 /* Not currently externally documented, and subject to change
1591 * <flags> bits meanings:
1592 * FOLD_FLAGS_FULL iff full folding is to be used;
1593 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1594 * locale are to be used.
1595 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1598 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1600 /* Tread a UTF-8 locale as not being in locale at all */
1601 if (IN_UTF8_CTYPE_LOCALE) {
1602 flags &= ~FOLD_FLAGS_LOCALE;
1606 UV result = _to_fold_latin1((U8) c, p, lenp,
1607 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1608 /* It is illegal for the fold to cross the 255/256 boundary under
1609 * locale; in this case return the original */
1610 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1615 /* If no special needs, just use the macro */
1616 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1617 uvchr_to_utf8(p, c);
1618 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1620 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1621 the special flags. */
1622 U8 utf8_c[UTF8_MAXBYTES + 1];
1623 uvchr_to_utf8(utf8_c, c);
1624 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1628 PERL_STATIC_INLINE bool
1629 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1630 const char *const swashname, SV* const invlist)
1632 /* returns a boolean giving whether or not the UTF8-encoded character that
1633 * starts at <p> is in the swash indicated by <swashname>. <swash>
1634 * contains a pointer to where the swash indicated by <swashname>
1635 * is to be stored; which this routine will do, so that future calls will
1636 * look at <*swash> and only generate a swash if it is not null. <invlist>
1637 * is NULL or an inversion list that defines the swash. If not null, it
1638 * saves time during initialization of the swash.
1640 * Note that it is assumed that the buffer length of <p> is enough to
1641 * contain all the bytes that comprise the character. Thus, <*p> should
1642 * have been checked before this call for mal-formedness enough to assure
1645 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1647 /* The API should have included a length for the UTF-8 character in <p>,
1648 * but it doesn't. We therefore assume that p has been validated at least
1649 * as far as there being enough bytes available in it to accommodate the
1650 * character without reading beyond the end, and pass that number on to the
1651 * validating routine */
1652 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1653 if (ckWARN_d(WARN_UTF8)) {
1654 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1655 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1656 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1657 what the malformation is */
1658 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1664 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1665 *swash = _core_swash_init("utf8",
1667 /* Only use the name if there is no inversion
1668 * list; otherwise will go out to disk */
1669 (invlist) ? "" : swashname,
1671 &PL_sv_undef, 1, 0, invlist, &flags);
1674 return swash_fetch(*swash, p, TRUE) != 0;
1678 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1680 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1682 assert(classnum < _FIRST_NON_SWASH_CC);
1684 return is_utf8_common(p,
1685 &PL_utf8_swash_ptrs[classnum],
1686 swash_property_names[classnum],
1687 PL_XPosix_ptrs[classnum]);
1691 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1695 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1697 if (! PL_utf8_perl_idstart) {
1698 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1700 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
1704 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1706 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1710 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1714 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1718 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1720 if (! PL_utf8_perl_idcont) {
1721 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1723 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
1727 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1729 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1731 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1735 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1737 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1739 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1743 Perl__is_utf8_mark(pTHX_ const U8 *p)
1745 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1747 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1751 =for apidoc to_utf8_case
1753 C<p> contains the pointer to the UTF-8 string encoding
1754 the character that is being converted. This routine assumes that the character
1755 at C<p> is well-formed.
1757 C<ustrp> is a pointer to the character buffer to put the
1758 conversion result to. C<lenp> is a pointer to the length
1761 C<swashp> is a pointer to the swash to use.
1763 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1764 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
1765 but not always, a multicharacter mapping), is tried first.
1767 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1768 any special mappings; C<""> means to use the special mappings. Values other
1769 than these two are treated as the name of the hash containing the special
1770 mappings, like C<"utf8::ToSpecLower">.
1772 C<normal> is a string like "ToLower" which means the swash
1778 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1779 SV **swashp, const char *normal, const char *special)
1782 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1784 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1786 /* Note that swash_fetch() doesn't output warnings for these because it
1787 * assumes we will */
1788 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1789 if (uv1 <= UNICODE_SURROGATE_LAST) {
1790 if (ckWARN_d(WARN_SURROGATE)) {
1791 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1792 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1793 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1796 else if (UNICODE_IS_SUPER(uv1)) {
1797 if (ckWARN_d(WARN_NON_UNICODE)) {
1798 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1799 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1800 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1804 /* Note that non-characters are perfectly legal, so no warning should
1808 if (!*swashp) /* load on-demand */
1809 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1812 /* It might be "special" (sometimes, but not always,
1813 * a multicharacter mapping) */
1817 /* If passed in the specials name, use that; otherwise use any
1818 * given in the swash */
1819 if (*special != '\0') {
1820 hv = get_hv(special, 0);
1823 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1825 hv = MUTABLE_HV(SvRV(*svp));
1830 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1835 s = SvPV_const(*svp, len);
1838 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1840 Copy(s, ustrp, len, U8);
1845 if (!len && *swashp) {
1846 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
1849 /* It was "normal" (a single character mapping). */
1850 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1858 return valid_utf8_to_uvchr(ustrp, 0);
1861 /* Here, there was no mapping defined, which means that the code point maps
1862 * to itself. Return the inputs */
1864 if (p != ustrp) { /* Don't copy onto itself */
1865 Copy(p, ustrp, len, U8);
1876 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1878 /* This is called when changing the case of a utf8-encoded character above
1879 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1880 * result contains a character that crosses the 255/256 boundary, disallow
1881 * the change, and return the original code point. See L<perlfunc/lc> for
1884 * p points to the original string whose case was changed; assumed
1885 * by this routine to be well-formed
1886 * result the code point of the first character in the changed-case string
1887 * ustrp points to the changed-case string (<result> represents its first char)
1888 * lenp points to the length of <ustrp> */
1890 UV original; /* To store the first code point of <p> */
1892 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1894 assert(UTF8_IS_ABOVE_LATIN1(*p));
1896 /* We know immediately if the first character in the string crosses the
1897 * boundary, so can skip */
1900 /* Look at every character in the result; if any cross the
1901 * boundary, the whole thing is disallowed */
1902 U8* s = ustrp + UTF8SKIP(ustrp);
1903 U8* e = ustrp + *lenp;
1905 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
1911 /* Here, no characters crossed, result is ok as-is */
1917 /* Failed, have to return the original */
1918 original = valid_utf8_to_uvchr(p, lenp);
1919 Copy(p, ustrp, *lenp, char);
1924 =for apidoc to_utf8_upper
1926 Instead use L</toUPPER_utf8>.
1930 /* Not currently externally documented, and subject to change:
1931 * <flags> is set iff iff the rules from the current underlying locale are to
1935 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
1939 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
1941 if (flags && IN_UTF8_CTYPE_LOCALE) {
1945 if (UTF8_IS_INVARIANT(*p)) {
1947 result = toUPPER_LC(*p);
1950 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
1953 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
1955 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
1956 result = toUPPER_LC(c);
1959 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
1963 else { /* utf8, ord above 255 */
1964 result = CALL_UPPER_CASE(p, ustrp, lenp);
1967 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
1972 /* Here, used locale rules. Convert back to utf8 */
1973 if (UTF8_IS_INVARIANT(result)) {
1974 *ustrp = (U8) result;
1978 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1979 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
1987 =for apidoc to_utf8_title
1989 Instead use L</toTITLE_utf8>.
1993 /* Not currently externally documented, and subject to change:
1994 * <flags> is set iff the rules from the current underlying locale are to be
1995 * used. Since titlecase is not defined in POSIX, for other than a
1996 * UTF-8 locale, uppercase is used instead for code points < 256.
2000 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2004 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2006 if (flags && IN_UTF8_CTYPE_LOCALE) {
2010 if (UTF8_IS_INVARIANT(*p)) {
2012 result = toUPPER_LC(*p);
2015 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2018 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2020 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2021 result = toUPPER_LC(c);
2024 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2028 else { /* utf8, ord above 255 */
2029 result = CALL_TITLE_CASE(p, ustrp, lenp);
2032 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2037 /* Here, used locale rules. Convert back to utf8 */
2038 if (UTF8_IS_INVARIANT(result)) {
2039 *ustrp = (U8) result;
2043 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2044 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2052 =for apidoc to_utf8_lower
2054 Instead use L</toLOWER_utf8>.
2058 /* Not currently externally documented, and subject to change:
2059 * <flags> is set iff iff the rules from the current underlying locale are to
2064 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2068 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2070 if (flags && IN_UTF8_CTYPE_LOCALE) {
2074 if (UTF8_IS_INVARIANT(*p)) {
2076 result = toLOWER_LC(*p);
2079 return to_lower_latin1(*p, ustrp, lenp);
2082 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2084 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2085 result = toLOWER_LC(c);
2088 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2092 else { /* utf8, ord above 255 */
2093 result = CALL_LOWER_CASE(p, ustrp, lenp);
2096 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2102 /* Here, used locale rules. Convert back to utf8 */
2103 if (UTF8_IS_INVARIANT(result)) {
2104 *ustrp = (U8) result;
2108 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2109 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2117 =for apidoc to_utf8_fold
2119 Instead use L</toFOLD_utf8>.
2123 /* Not currently externally documented, and subject to change,
2125 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2126 * locale are to be used.
2127 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2128 * otherwise simple folds
2129 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2134 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2138 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2140 /* These are mutually exclusive */
2141 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2143 assert(p != ustrp); /* Otherwise overwrites */
2145 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2146 flags &= ~FOLD_FLAGS_LOCALE;
2149 if (UTF8_IS_INVARIANT(*p)) {
2150 if (flags & FOLD_FLAGS_LOCALE) {
2151 result = toFOLD_LC(*p);
2154 return _to_fold_latin1(*p, ustrp, lenp,
2155 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2158 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2159 if (flags & FOLD_FLAGS_LOCALE) {
2160 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2161 result = toFOLD_LC(c);
2164 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2166 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2169 else { /* utf8, ord above 255 */
2170 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2172 if (flags & FOLD_FLAGS_LOCALE) {
2174 /* Special case these two characters, as what normally gets
2175 * returned under locale doesn't work */
2176 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2177 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2178 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2182 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2183 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2184 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2186 goto return_ligature_st;
2188 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2190 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2194 /* This is called when changing the case of a utf8-encoded
2195 * character above the ASCII range, and the result should not
2196 * contain an ASCII character. */
2198 UV original; /* To store the first code point of <p> */
2200 /* Look at every character in the result; if any cross the
2201 * boundary, the whole thing is disallowed */
2203 U8* e = ustrp + *lenp;
2206 /* Crossed, have to return the original */
2207 original = valid_utf8_to_uvchr(p, lenp);
2209 /* But in these instances, there is an alternative we can
2210 * return that is valid */
2211 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2212 || original == LATIN_SMALL_LETTER_SHARP_S)
2216 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2217 goto return_ligature_st;
2219 Copy(p, ustrp, *lenp, char);
2225 /* Here, no characters crossed, result is ok as-is */
2230 /* Here, used locale rules. Convert back to utf8 */
2231 if (UTF8_IS_INVARIANT(result)) {
2232 *ustrp = (U8) result;
2236 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2237 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2244 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2245 * folds to a string of two of these characters. By returning this
2246 * instead, then, e.g.,
2247 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2250 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2251 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2253 return LATIN_SMALL_LETTER_LONG_S;
2256 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2257 * have the other one fold to it */
2259 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2260 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2261 return LATIN_SMALL_LIGATURE_ST;
2265 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2266 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2267 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2271 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2273 PERL_ARGS_ASSERT_SWASH_INIT;
2275 /* Returns a copy of a swash initiated by the called function. This is the
2276 * public interface, and returning a copy prevents others from doing
2277 * mischief on the original */
2279 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2283 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2286 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2287 * use the following define */
2289 #define CORE_SWASH_INIT_RETURN(x) \
2290 PL_curpm= old_PL_curpm; \
2293 /* Initialize and return a swash, creating it if necessary. It does this
2294 * by calling utf8_heavy.pl in the general case. The returned value may be
2295 * the swash's inversion list instead if the input parameters allow it.
2296 * Which is returned should be immaterial to callers, as the only
2297 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2298 * and swash_to_invlist() handle both these transparently.
2300 * This interface should only be used by functions that won't destroy or
2301 * adversely change the swash, as doing so affects all other uses of the
2302 * swash in the program; the general public should use 'Perl_swash_init'
2305 * pkg is the name of the package that <name> should be in.
2306 * name is the name of the swash to find. Typically it is a Unicode
2307 * property name, including user-defined ones
2308 * listsv is a string to initialize the swash with. It must be of the form
2309 * documented as the subroutine return value in
2310 * L<perlunicode/User-Defined Character Properties>
2311 * minbits is the number of bits required to represent each data element.
2312 * It is '1' for binary properties.
2313 * none I (khw) do not understand this one, but it is used only in tr///.
2314 * invlist is an inversion list to initialize the swash with (or NULL)
2315 * flags_p if non-NULL is the address of various input and output flag bits
2316 * to the routine, as follows: ('I' means is input to the routine;
2317 * 'O' means output from the routine. Only flags marked O are
2318 * meaningful on return.)
2319 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2320 * came from a user-defined property. (I O)
2321 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2322 * when the swash cannot be located, to simply return NULL. (I)
2323 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2324 * return of an inversion list instead of a swash hash if this routine
2325 * thinks that would result in faster execution of swash_fetch() later
2328 * Thus there are three possible inputs to find the swash: <name>,
2329 * <listsv>, and <invlist>. At least one must be specified. The result
2330 * will be the union of the specified ones, although <listsv>'s various
2331 * actions can intersect, etc. what <name> gives. To avoid going out to
2332 * disk at all, <invlist> should specify completely what the swash should
2333 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2335 * <invlist> is only valid for binary properties */
2337 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2339 SV* retval = &PL_sv_undef;
2340 HV* swash_hv = NULL;
2341 const int invlist_swash_boundary =
2342 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2343 ? 512 /* Based on some benchmarking, but not extensive, see commit
2345 : -1; /* Never return just an inversion list */
2347 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2348 assert(! invlist || minbits == 1);
2350 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2351 that triggered the swash init and the swash init perl logic itself.
2354 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2356 if (listsv != &PL_sv_undef || strNE(name, "")) {
2358 const size_t pkg_len = strlen(pkg);
2359 const size_t name_len = strlen(name);
2360 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2364 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2366 PUSHSTACKi(PERLSI_MAGIC);
2369 /* We might get here via a subroutine signature which uses a utf8
2370 * parameter name, at which point PL_subname will have been set
2371 * but not yet used. */
2372 save_item(PL_subname);
2373 if (PL_parser && PL_parser->error_count)
2374 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2375 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2376 if (!method) { /* demand load utf8 */
2378 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2379 GvSV(PL_errgv) = NULL;
2380 #ifndef NO_TAINT_SUPPORT
2381 /* It is assumed that callers of this routine are not passing in
2382 * any user derived data. */
2383 SAVEBOOL(TAINT_get);
2386 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2389 /* Not ERRSV, as there is no need to vivify a scalar we are
2390 about to discard. */
2391 SV * const errsv = GvSV(PL_errgv);
2392 if (!SvTRUE(errsv)) {
2393 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2394 SvREFCNT_dec(errsv);
2402 mPUSHp(pkg, pkg_len);
2403 mPUSHp(name, name_len);
2408 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2409 GvSV(PL_errgv) = NULL;
2410 /* If we already have a pointer to the method, no need to use
2411 * call_method() to repeat the lookup. */
2413 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2414 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2416 retval = *PL_stack_sp--;
2417 SvREFCNT_inc(retval);
2420 /* Not ERRSV. See above. */
2421 SV * const errsv = GvSV(PL_errgv);
2422 if (!SvTRUE(errsv)) {
2423 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2424 SvREFCNT_dec(errsv);
2429 if (IN_PERL_COMPILETIME) {
2430 CopHINTS_set(PL_curcop, PL_hints);
2432 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2435 /* If caller wants to handle missing properties, let them */
2436 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2437 CORE_SWASH_INIT_RETURN(NULL);
2440 "Can't find Unicode property definition \"%"SVf"\"",
2442 NOT_REACHED; /* NOTREACHED */
2444 } /* End of calling the module to find the swash */
2446 /* If this operation fetched a swash, and we will need it later, get it */
2447 if (retval != &PL_sv_undef
2448 && (minbits == 1 || (flags_p
2450 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2452 swash_hv = MUTABLE_HV(SvRV(retval));
2454 /* If we don't already know that there is a user-defined component to
2455 * this swash, and the user has indicated they wish to know if there is
2456 * one (by passing <flags_p>), find out */
2457 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2458 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2459 if (user_defined && SvUV(*user_defined)) {
2460 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2465 /* Make sure there is an inversion list for binary properties */
2467 SV** swash_invlistsvp = NULL;
2468 SV* swash_invlist = NULL;
2469 bool invlist_in_swash_is_valid = FALSE;
2470 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2471 an unclaimed reference count */
2473 /* If this operation fetched a swash, get its already existing
2474 * inversion list, or create one for it */
2477 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2478 if (swash_invlistsvp) {
2479 swash_invlist = *swash_invlistsvp;
2480 invlist_in_swash_is_valid = TRUE;
2483 swash_invlist = _swash_to_invlist(retval);
2484 swash_invlist_unclaimed = TRUE;
2488 /* If an inversion list was passed in, have to include it */
2491 /* Any fetched swash will by now have an inversion list in it;
2492 * otherwise <swash_invlist> will be NULL, indicating that we
2493 * didn't fetch a swash */
2494 if (swash_invlist) {
2496 /* Add the passed-in inversion list, which invalidates the one
2497 * already stored in the swash */
2498 invlist_in_swash_is_valid = FALSE;
2499 _invlist_union(invlist, swash_invlist, &swash_invlist);
2503 /* Here, there is no swash already. Set up a minimal one, if
2504 * we are going to return a swash */
2505 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2507 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2509 swash_invlist = invlist;
2513 /* Here, we have computed the union of all the passed-in data. It may
2514 * be that there was an inversion list in the swash which didn't get
2515 * touched; otherwise save the computed one */
2516 if (! invlist_in_swash_is_valid
2517 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2519 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2521 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2523 /* We just stole a reference count. */
2524 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2525 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2528 SvREADONLY_on(swash_invlist);
2530 /* Use the inversion list stand-alone if small enough */
2531 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2532 SvREFCNT_dec(retval);
2533 if (!swash_invlist_unclaimed)
2534 SvREFCNT_inc_simple_void_NN(swash_invlist);
2535 retval = newRV_noinc(swash_invlist);
2539 CORE_SWASH_INIT_RETURN(retval);
2540 #undef CORE_SWASH_INIT_RETURN
2544 /* This API is wrong for special case conversions since we may need to
2545 * return several Unicode characters for a single Unicode character
2546 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2547 * the lower-level routine, and it is similarly broken for returning
2548 * multiple values. --jhi
2549 * For those, you should use to_utf8_case() instead */
2550 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2553 * Returns the value of property/mapping C<swash> for the first character
2554 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2555 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2556 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2558 * A "swash" is a hash which contains initially the keys/values set up by
2559 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2560 * property for all possible code points. Things are stored in a compact form
2561 * (see utf8_heavy.pl) so that calculation is required to find the actual
2562 * property value for a given code point. As code points are looked up, new
2563 * key/value pairs are added to the hash, so that the calculation doesn't have
2564 * to ever be re-done. Further, each calculation is done, not just for the
2565 * desired one, but for a whole block of code points adjacent to that one.
2566 * For binary properties on ASCII machines, the block is usually for 64 code
2567 * points, starting with a code point evenly divisible by 64. Thus if the
2568 * property value for code point 257 is requested, the code goes out and
2569 * calculates the property values for all 64 code points between 256 and 319,
2570 * and stores these as a single 64-bit long bit vector, called a "swatch",
2571 * under the key for code point 256. The key is the UTF-8 encoding for code
2572 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2573 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2574 * for code point 258 is then requested, this code realizes that it would be
2575 * stored under the key for 256, and would find that value and extract the
2576 * relevant bit, offset from 256.
2578 * Non-binary properties are stored in as many bits as necessary to represent
2579 * their values (32 currently, though the code is more general than that), not
2580 * as single bits, but the principal is the same: the value for each key is a
2581 * vector that encompasses the property values for all code points whose UTF-8
2582 * representations are represented by the key. That is, for all code points
2583 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2587 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2589 HV *const hv = MUTABLE_HV(SvRV(swash));
2594 const U8 *tmps = NULL;
2598 PERL_ARGS_ASSERT_SWASH_FETCH;
2600 /* If it really isn't a hash, it isn't really swash; must be an inversion
2602 if (SvTYPE(hv) != SVt_PVHV) {
2603 return _invlist_contains_cp((SV*)hv,
2605 ? valid_utf8_to_uvchr(ptr, NULL)
2609 /* We store the values in a "swatch" which is a vec() value in a swash
2610 * hash. Code points 0-255 are a single vec() stored with key length
2611 * (klen) 0. All other code points have a UTF-8 representation
2612 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2613 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2614 * length for them is the length of the encoded char - 1. ptr[klen] is the
2615 * final byte in the sequence representing the character */
2616 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2621 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2624 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
2627 klen = UTF8SKIP(ptr) - 1;
2629 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2630 * the vec is the final byte in the sequence. (In EBCDIC this is
2631 * converted to I8 to get consecutive values.) To help you visualize
2633 * Straight 1047 After final byte
2634 * UTF-8 UTF-EBCDIC I8 transform
2635 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2636 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2638 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2639 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2641 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2642 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2644 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2645 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2647 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2648 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2650 * (There are no discontinuities in the elided (...) entries.)
2651 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2652 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2653 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2654 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2655 * index into the vec() swatch (after subtracting 0x80, which we
2656 * actually do with an '&').
2657 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2658 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2659 * dicontinuities which go away by transforming it into I8, and we
2660 * effectively subtract 0xA0 to get the index. */
2661 needents = (1 << UTF_ACCUMULATION_SHIFT);
2662 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2666 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2667 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2668 * it's nothing to sniff at.) Pity we usually come through at least
2669 * two function calls to get here...
2671 * NB: this code assumes that swatches are never modified, once generated!
2674 if (hv == PL_last_swash_hv &&
2675 klen == PL_last_swash_klen &&
2676 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2678 tmps = PL_last_swash_tmps;
2679 slen = PL_last_swash_slen;
2682 /* Try our second-level swatch cache, kept in a hash. */
2683 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2685 /* If not cached, generate it via swatch_get */
2686 if (!svp || !SvPOK(*svp)
2687 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2690 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2691 swatch = swatch_get(swash,
2692 code_point & ~((UV)needents - 1),
2695 else { /* For the first 256 code points, the swatch has a key of
2697 swatch = swatch_get(swash, 0, needents);
2700 if (IN_PERL_COMPILETIME)
2701 CopHINTS_set(PL_curcop, PL_hints);
2703 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2705 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2706 || (slen << 3) < needents)
2707 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2708 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2709 svp, tmps, (UV)slen, (UV)needents);
2712 PL_last_swash_hv = hv;
2713 assert(klen <= sizeof(PL_last_swash_key));
2714 PL_last_swash_klen = (U8)klen;
2715 /* FIXME change interpvar.h? */
2716 PL_last_swash_tmps = (U8 *) tmps;
2717 PL_last_swash_slen = slen;
2719 Copy(ptr, PL_last_swash_key, klen, U8);
2722 switch ((int)((slen << 3) / needents)) {
2724 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2726 return ((UV) tmps[off]);
2730 ((UV) tmps[off ] << 8) +
2731 ((UV) tmps[off + 1]);
2735 ((UV) tmps[off ] << 24) +
2736 ((UV) tmps[off + 1] << 16) +
2737 ((UV) tmps[off + 2] << 8) +
2738 ((UV) tmps[off + 3]);
2740 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2741 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2742 NORETURN_FUNCTION_END;
2745 /* Read a single line of the main body of the swash input text. These are of
2748 * where each number is hex. The first two numbers form the minimum and
2749 * maximum of a range, and the third is the value associated with the range.
2750 * Not all swashes should have a third number
2752 * On input: l points to the beginning of the line to be examined; it points
2753 * to somewhere in the string of the whole input text, and is
2754 * terminated by a \n or the null string terminator.
2755 * lend points to the null terminator of that string
2756 * wants_value is non-zero if the swash expects a third number
2757 * typestr is the name of the swash's mapping, like 'ToLower'
2758 * On output: *min, *max, and *val are set to the values read from the line.
2759 * returns a pointer just beyond the line examined. If there was no
2760 * valid min number on the line, returns lend+1
2764 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2765 const bool wants_value, const U8* const typestr)
2767 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2768 STRLEN numlen; /* Length of the number */
2769 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2770 | PERL_SCAN_DISALLOW_PREFIX
2771 | PERL_SCAN_SILENT_NON_PORTABLE;
2773 /* nl points to the next \n in the scan */
2774 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2776 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2778 /* Get the first number on the line: the range minimum */
2780 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2781 *max = *min; /* So can never return without setting max */
2782 if (numlen) /* If found a hex number, position past it */
2784 else if (nl) { /* Else, go handle next line, if any */
2785 return nl + 1; /* 1 is length of "\n" */
2787 else { /* Else, no next line */
2788 return lend + 1; /* to LIST's end at which \n is not found */
2791 /* The max range value follows, separated by a BLANK */
2794 flags = PERL_SCAN_SILENT_ILLDIGIT
2795 | PERL_SCAN_DISALLOW_PREFIX
2796 | PERL_SCAN_SILENT_NON_PORTABLE;
2798 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2801 else /* If no value here, it is a single element range */
2804 /* Non-binary tables have a third entry: what the first element of the
2805 * range maps to. The map for those currently read here is in hex */
2809 flags = PERL_SCAN_SILENT_ILLDIGIT
2810 | PERL_SCAN_DISALLOW_PREFIX
2811 | PERL_SCAN_SILENT_NON_PORTABLE;
2813 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2822 /* diag_listed_as: To%s: illegal mapping '%s' */
2823 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2829 *val = 0; /* bits == 1, then any val should be ignored */
2831 else { /* Nothing following range min, should be single element with no
2836 /* diag_listed_as: To%s: illegal mapping '%s' */
2837 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2841 *val = 0; /* bits == 1, then val should be ignored */
2844 /* Position to next line if any, or EOF */
2854 * Returns a swatch (a bit vector string) for a code point sequence
2855 * that starts from the value C<start> and comprises the number C<span>.
2856 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2857 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2860 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
2863 U8 *l, *lend, *x, *xend, *s, *send;
2864 STRLEN lcur, xcur, scur;
2865 HV *const hv = MUTABLE_HV(SvRV(swash));
2866 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
2868 SV** listsvp = NULL; /* The string containing the main body of the table */
2869 SV** extssvp = NULL;
2870 SV** invert_it_svp = NULL;
2873 STRLEN octets; /* if bits == 1, then octets == 0 */
2875 UV end = start + span;
2877 if (invlistsvp == NULL) {
2878 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2879 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2880 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2881 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2882 listsvp = hv_fetchs(hv, "LIST", FALSE);
2883 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2885 bits = SvUV(*bitssvp);
2886 none = SvUV(*nonesvp);
2887 typestr = (U8*)SvPV_nolen(*typesvp);
2893 octets = bits >> 3; /* if bits == 1, then octets == 0 */
2895 PERL_ARGS_ASSERT_SWATCH_GET;
2897 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
2898 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
2902 /* If overflowed, use the max possible */
2908 /* create and initialize $swatch */
2909 scur = octets ? (span * octets) : (span + 7) / 8;
2910 swatch = newSV(scur);
2912 s = (U8*)SvPVX(swatch);
2913 if (octets && none) {
2914 const U8* const e = s + scur;
2917 *s++ = (U8)(none & 0xff);
2918 else if (bits == 16) {
2919 *s++ = (U8)((none >> 8) & 0xff);
2920 *s++ = (U8)( none & 0xff);
2922 else if (bits == 32) {
2923 *s++ = (U8)((none >> 24) & 0xff);
2924 *s++ = (U8)((none >> 16) & 0xff);
2925 *s++ = (U8)((none >> 8) & 0xff);
2926 *s++ = (U8)( none & 0xff);
2932 (void)memzero((U8*)s, scur + 1);
2934 SvCUR_set(swatch, scur);
2935 s = (U8*)SvPVX(swatch);
2937 if (invlistsvp) { /* If has an inversion list set up use that */
2938 _invlist_populate_swatch(*invlistsvp, start, end, s);
2942 /* read $swash->{LIST} */
2943 l = (U8*)SvPV(*listsvp, lcur);
2946 UV min, max, val, upper;
2947 l = swash_scan_list_line(l, lend, &min, &max, &val,
2948 cBOOL(octets), typestr);
2953 /* If looking for something beyond this range, go try the next one */
2957 /* <end> is generally 1 beyond where we want to set things, but at the
2958 * platform's infinity, where we can't go any higher, we want to
2959 * include the code point at <end> */
2962 : (max != UV_MAX || end != UV_MAX)
2969 if (!none || val < none) {
2974 for (key = min; key <= upper; key++) {
2976 /* offset must be non-negative (start <= min <= key < end) */
2977 offset = octets * (key - start);
2979 s[offset] = (U8)(val & 0xff);
2980 else if (bits == 16) {
2981 s[offset ] = (U8)((val >> 8) & 0xff);
2982 s[offset + 1] = (U8)( val & 0xff);
2984 else if (bits == 32) {
2985 s[offset ] = (U8)((val >> 24) & 0xff);
2986 s[offset + 1] = (U8)((val >> 16) & 0xff);
2987 s[offset + 2] = (U8)((val >> 8) & 0xff);
2988 s[offset + 3] = (U8)( val & 0xff);
2991 if (!none || val < none)
2995 else { /* bits == 1, then val should be ignored */
3000 for (key = min; key <= upper; key++) {
3001 const STRLEN offset = (STRLEN)(key - start);
3002 s[offset >> 3] |= 1 << (offset & 7);
3007 /* Invert if the data says it should be. Assumes that bits == 1 */
3008 if (invert_it_svp && SvUV(*invert_it_svp)) {
3010 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3011 * be 0, and their inversion should also be 0, as we don't succeed any
3012 * Unicode property matches for non-Unicode code points */
3013 if (start <= PERL_UNICODE_MAX) {
3015 /* The code below assumes that we never cross the
3016 * Unicode/above-Unicode boundary in a range, as otherwise we would
3017 * have to figure out where to stop flipping the bits. Since this
3018 * boundary is divisible by a large power of 2, and swatches comes
3019 * in small powers of 2, this should be a valid assumption */
3020 assert(start + span - 1 <= PERL_UNICODE_MAX);
3030 /* read $swash->{EXTRAS}
3031 * This code also copied to swash_to_invlist() below */
3032 x = (U8*)SvPV(*extssvp, xcur);
3040 SV **otherbitssvp, *other;
3044 const U8 opc = *x++;
3048 nl = (U8*)memchr(x, '\n', xend - x);
3050 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3052 x = nl + 1; /* 1 is length of "\n" */
3056 x = xend; /* to EXTRAS' end at which \n is not found */
3063 namelen = nl - namestr;
3067 namelen = xend - namestr;
3071 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3072 otherhv = MUTABLE_HV(SvRV(*othersvp));
3073 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3074 otherbits = (STRLEN)SvUV(*otherbitssvp);
3075 if (bits < otherbits)
3076 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3077 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3079 /* The "other" swatch must be destroyed after. */
3080 other = swatch_get(*othersvp, start, span);
3081 o = (U8*)SvPV(other, olen);
3084 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3086 s = (U8*)SvPV(swatch, slen);
3087 if (bits == 1 && otherbits == 1) {
3089 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3090 "mismatch, slen=%"UVuf", olen=%"UVuf,
3091 (UV)slen, (UV)olen);
3115 STRLEN otheroctets = otherbits >> 3;
3117 U8* const send = s + slen;
3122 if (otherbits == 1) {
3123 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3127 STRLEN vlen = otheroctets;
3135 if (opc == '+' && otherval)
3136 NOOP; /* replace with otherval */
3137 else if (opc == '!' && !otherval)
3139 else if (opc == '-' && otherval)
3141 else if (opc == '&' && !otherval)
3144 s += octets; /* no replacement */
3149 *s++ = (U8)( otherval & 0xff);
3150 else if (bits == 16) {
3151 *s++ = (U8)((otherval >> 8) & 0xff);
3152 *s++ = (U8)( otherval & 0xff);
3154 else if (bits == 32) {
3155 *s++ = (U8)((otherval >> 24) & 0xff);
3156 *s++ = (U8)((otherval >> 16) & 0xff);
3157 *s++ = (U8)((otherval >> 8) & 0xff);
3158 *s++ = (U8)( otherval & 0xff);
3162 sv_free(other); /* through with it! */
3168 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3171 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3172 * Can't be used on a property that is subject to user override, as it
3173 * relies on the value of SPECIALS in the swash which would be set by
3174 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3175 * for overridden properties
3177 * Returns a hash which is the inversion and closure of a swash mapping.
3178 * For example, consider the input lines:
3183 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3184 * 006C. The value for each key is an array. For 006C, the array would
3185 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3186 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3188 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3189 * keys are only code points that are folded-to, so it isn't a full closure.
3191 * Essentially, for any code point, it gives all the code points that map to
3192 * it, or the list of 'froms' for that point.
3194 * Currently it ignores any additions or deletions from other swashes,
3195 * looking at just the main body of the swash, and if there are SPECIALS
3196 * in the swash, at that hash
3198 * The specials hash can be extra code points, and most likely consists of
3199 * maps from single code points to multiple ones (each expressed as a string
3200 * of utf8 characters). This function currently returns only 1-1 mappings.
3201 * However consider this possible input in the specials hash:
3202 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3203 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3205 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3206 * currently handle. But it also means that FB05 and FB06 are equivalent in
3207 * a 1-1 mapping which we should handle, and this relationship may not be in
3208 * the main table. Therefore this function examines all the multi-char
3209 * sequences and adds the 1-1 mappings that come out of that. */
3213 HV *const hv = MUTABLE_HV(SvRV(swash));
3215 /* The string containing the main body of the table. This will have its
3216 * assertion fail if the swash has been converted to its inversion list */
3217 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3219 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3220 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3221 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3222 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3223 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3224 const STRLEN bits = SvUV(*bitssvp);
3225 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3226 const UV none = SvUV(*nonesvp);
3227 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3231 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3233 /* Must have at least 8 bits to get the mappings */
3234 if (bits != 8 && bits != 16 && bits != 32) {
3235 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3239 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3240 mapping to more than one character */
3242 /* Construct an inverse mapping hash for the specials */
3243 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3244 HV * specials_inverse = newHV();
3245 char *char_from; /* the lhs of the map */
3246 I32 from_len; /* its byte length */
3247 char *char_to; /* the rhs of the map */
3248 I32 to_len; /* its byte length */
3249 SV *sv_to; /* and in a sv */
3250 AV* from_list; /* list of things that map to each 'to' */
3252 hv_iterinit(specials_hv);
3254 /* The keys are the characters (in utf8) that map to the corresponding
3255 * utf8 string value. Iterate through the list creating the inverse
3257 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3259 if (! SvPOK(sv_to)) {
3260 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3261 "unexpectedly is not a string, flags=%lu",
3262 (unsigned long)SvFLAGS(sv_to));
3264 /*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)));*/
3266 /* Each key in the inverse list is a mapped-to value, and the key's
3267 * hash value is a list of the strings (each in utf8) that map to
3268 * it. Those strings are all one character long */
3269 if ((listp = hv_fetch(specials_inverse,
3273 from_list = (AV*) *listp;
3275 else { /* No entry yet for it: create one */
3276 from_list = newAV();
3277 if (! hv_store(specials_inverse,
3280 (SV*) from_list, 0))
3282 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3286 /* Here have the list associated with this 'to' (perhaps newly
3287 * created and empty). Just add to it. Note that we ASSUME that
3288 * the input is guaranteed to not have duplications, so we don't
3289 * check for that. Duplications just slow down execution time. */
3290 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3293 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3294 * it looking for cases like the FB05/FB06 examples above. There would
3295 * be an entry in the hash like
3296 * 'st' => [ FB05, FB06 ]
3297 * In this example we will create two lists that get stored in the
3298 * returned hash, 'ret':
3299 * FB05 => [ FB05, FB06 ]
3300 * FB06 => [ FB05, FB06 ]
3302 * Note that there is nothing to do if the array only has one element.
3303 * (In the normal 1-1 case handled below, we don't have to worry about
3304 * two lists, as everything gets tied to the single list that is
3305 * generated for the single character 'to'. But here, we are omitting
3306 * that list, ('st' in the example), so must have multiple lists.) */
3307 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3308 &char_to, &to_len)))
3310 if (av_tindex(from_list) > 0) {
3313 /* We iterate over all combinations of i,j to place each code
3314 * point on each list */
3315 for (i = 0; i <= av_tindex(from_list); i++) {
3317 AV* i_list = newAV();
3318 SV** entryp = av_fetch(from_list, i, FALSE);
3319 if (entryp == NULL) {
3320 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3322 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3323 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3325 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3326 (SV*) i_list, FALSE))
3328 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3331 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3332 for (j = 0; j <= av_tindex(from_list); j++) {
3333 entryp = av_fetch(from_list, j, FALSE);
3334 if (entryp == NULL) {
3335 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3338 /* When i==j this adds itself to the list */
3339 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3340 (U8*) SvPVX(*entryp),
3341 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3343 /*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));*/
3348 SvREFCNT_dec(specials_inverse); /* done with it */
3349 } /* End of specials */
3351 /* read $swash->{LIST} */
3352 l = (U8*)SvPV(*listsvp, lcur);
3355 /* Go through each input line */
3359 l = swash_scan_list_line(l, lend, &min, &max, &val,
3360 cBOOL(octets), typestr);
3365 /* Each element in the range is to be inverted */
3366 for (inverse = min; inverse <= max; inverse++) {
3370 bool found_key = FALSE;
3371 bool found_inverse = FALSE;
3373 /* The key is the inverse mapping */
3374 char key[UTF8_MAXBYTES+1];
3375 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3376 STRLEN key_len = key_end - key;
3378 /* Get the list for the map */
3379 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3380 list = (AV*) *listp;
3382 else { /* No entry yet for it: create one */
3384 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3385 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3389 /* Look through list to see if this inverse mapping already is
3390 * listed, or if there is a mapping to itself already */
3391 for (i = 0; i <= av_tindex(list); i++) {
3392 SV** entryp = av_fetch(list, i, FALSE);
3395 if (entryp == NULL) {
3396 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3400 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3404 if (uv == inverse) {
3405 found_inverse = TRUE;
3408 /* No need to continue searching if found everything we are
3410 if (found_key && found_inverse) {
3415 /* Make sure there is a mapping to itself on the list */
3417 av_push(list, newSVuv(val));
3418 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3422 /* Simply add the value to the list */
3423 if (! found_inverse) {
3424 av_push(list, newSVuv(inverse));
3425 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3428 /* swatch_get() increments the value of val for each element in the
3429 * range. That makes more compact tables possible. You can
3430 * express the capitalization, for example, of all consecutive
3431 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3432 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3433 * and it's not documented; it appears to be used only in
3434 * implementing tr//; I copied the semantics from swatch_get(), just
3436 if (!none || val < none) {
3446 Perl__swash_to_invlist(pTHX_ SV* const swash)
3449 /* Subject to change or removal. For use only in one place in regcomp.c.
3450 * Ownership is given to one reference count in the returned SV* */
3455 HV *const hv = MUTABLE_HV(SvRV(swash));
3456 UV elements = 0; /* Number of elements in the inversion list */
3466 STRLEN octets; /* if bits == 1, then octets == 0 */
3472 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3474 /* If not a hash, it must be the swash's inversion list instead */
3475 if (SvTYPE(hv) != SVt_PVHV) {
3476 return SvREFCNT_inc_simple_NN((SV*) hv);
3479 /* The string containing the main body of the table */
3480 listsvp = hv_fetchs(hv, "LIST", FALSE);
3481 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3482 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3483 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3484 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3486 typestr = (U8*)SvPV_nolen(*typesvp);
3487 bits = SvUV(*bitssvp);
3488 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3490 /* read $swash->{LIST} */
3491 if (SvPOK(*listsvp)) {
3492 l = (U8*)SvPV(*listsvp, lcur);
3495 /* LIST legitimately doesn't contain a string during compilation phases
3496 * of Perl itself, before the Unicode tables are generated. In this
3497 * case, just fake things up by creating an empty list */
3504 if (*l == 'V') { /* Inversion list format */
3505 const char *after_atou = (char *) lend;
3507 UV* other_elements_ptr;
3509 /* The first number is a count of the rest */
3511 elements = grok_atou((const char *)l, &after_atou);
3512 if (elements == 0) {
3513 invlist = _new_invlist(0);
3516 while (isSPACE(*l)) l++;
3517 l = (U8 *) after_atou;
3519 /* Get the 0th element, which is needed to setup the inversion list */
3520 while (isSPACE(*l)) l++;
3521 element0 = (UV) grok_atou((const char *)l, &after_atou);
3522 l = (U8 *) after_atou;
3523 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3526 /* Then just populate the rest of the input */
3527 while (elements-- > 0) {
3529 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3531 while (isSPACE(*l)) l++;
3532 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3533 l = (U8 *) after_atou;
3539 /* Scan the input to count the number of lines to preallocate array
3540 * size based on worst possible case, which is each line in the input
3541 * creates 2 elements in the inversion list: 1) the beginning of a
3542 * range in the list; 2) the beginning of a range not in the list. */
3543 while ((loc = (strchr(loc, '\n'))) != NULL) {
3548 /* If the ending is somehow corrupt and isn't a new line, add another
3549 * element for the final range that isn't in the inversion list */
3550 if (! (*lend == '\n'
3551 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3556 invlist = _new_invlist(elements);
3558 /* Now go through the input again, adding each range to the list */
3561 UV val; /* Not used by this function */
3563 l = swash_scan_list_line(l, lend, &start, &end, &val,
3564 cBOOL(octets), typestr);
3570 invlist = _add_range_to_invlist(invlist, start, end);
3574 /* Invert if the data says it should be */
3575 if (invert_it_svp && SvUV(*invert_it_svp)) {
3576 _invlist_invert(invlist);
3579 /* This code is copied from swatch_get()
3580 * read $swash->{EXTRAS} */
3581 x = (U8*)SvPV(*extssvp, xcur);
3589 SV **otherbitssvp, *other;
3592 const U8 opc = *x++;
3596 nl = (U8*)memchr(x, '\n', xend - x);
3598 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3600 x = nl + 1; /* 1 is length of "\n" */
3604 x = xend; /* to EXTRAS' end at which \n is not found */
3611 namelen = nl - namestr;
3615 namelen = xend - namestr;
3619 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3620 otherhv = MUTABLE_HV(SvRV(*othersvp));
3621 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3622 otherbits = (STRLEN)SvUV(*otherbitssvp);
3624 if (bits != otherbits || bits != 1) {
3625 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3626 "properties, bits=%"UVuf", otherbits=%"UVuf,
3627 (UV)bits, (UV)otherbits);
3630 /* The "other" swatch must be destroyed after. */
3631 other = _swash_to_invlist((SV *)*othersvp);
3633 /* End of code copied from swatch_get() */
3636 _invlist_union(invlist, other, &invlist);
3639 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3642 _invlist_subtract(invlist, other, &invlist);
3645 _invlist_intersection(invlist, other, &invlist);
3650 sv_free(other); /* through with it! */
3653 SvREADONLY_on(invlist);
3658 Perl__get_swash_invlist(pTHX_ SV* const swash)
3662 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3664 if (! SvROK(swash)) {
3668 /* If it really isn't a hash, it isn't really swash; must be an inversion
3670 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3674 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3683 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3685 /* May change: warns if surrogates, non-character code points, or
3686 * non-Unicode code points are in s which has length len bytes. Returns
3687 * TRUE if none found; FALSE otherwise. The only other validity check is
3688 * to make sure that this won't exceed the string's length */
3690 const U8* const e = s + len;
3693 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3696 if (UTF8SKIP(s) > len) {
3697 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3698 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3701 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3703 if (UTF8_IS_SUPER(s)) {
3704 if (ckWARN_d(WARN_NON_UNICODE)) {
3705 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3706 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3707 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3711 else if (UTF8_IS_SURROGATE(s)) {
3712 if (ckWARN_d(WARN_SURROGATE)) {
3713 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3714 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3715 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3720 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3721 && (ckWARN_d(WARN_NONCHAR)))
3723 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3724 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3725 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3736 =for apidoc pv_uni_display
3738 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3739 length C<len>, the displayable version being at most C<pvlim> bytes long
3740 (if longer, the rest is truncated and "..." will be appended).
3742 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
3743 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3744 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3745 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3746 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3747 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3749 The pointer to the PV of the C<dsv> is returned.
3753 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3758 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3762 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3764 /* This serves double duty as a flag and a character to print after
3765 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3769 if (pvlim && SvCUR(dsv) >= pvlim) {
3773 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3775 const unsigned char c = (unsigned char)u & 0xFF;
3776 if (flags & UNI_DISPLAY_BACKSLASH) {
3793 const char string = ok;
3794 sv_catpvs(dsv, "\\");
3795 sv_catpvn(dsv, &string, 1);
3798 /* isPRINT() is the locale-blind version. */
3799 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3800 const char string = c;
3801 sv_catpvn(dsv, &string, 1);
3806 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
3809 sv_catpvs(dsv, "...");
3815 =for apidoc sv_uni_display
3817 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
3818 the displayable version being at most C<pvlim> bytes long
3819 (if longer, the rest is truncated and "..." will be appended).
3821 The C<flags> argument is as in L</pv_uni_display>().
3823 The pointer to the PV of the C<dsv> is returned.
3828 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
3830 const char * const ptr =
3831 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
3833 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
3835 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
3836 SvCUR(ssv), pvlim, flags);
3840 =for apidoc foldEQ_utf8
3842 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
3843 of which may be in UTF-8) are the same case-insensitively; false otherwise.
3844 How far into the strings to compare is determined by other input parameters.
3846 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
3847 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
3848 with respect to C<s2>.
3850 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
3851 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
3852 scan will not be considered to be a match unless the goal is reached, and
3853 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
3856 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
3857 considered an end pointer to the position 1 byte past the maximum point
3858 in C<s1> beyond which scanning will not continue under any circumstances.
3859 (This routine assumes that UTF-8 encoded input strings are not malformed;
3860 malformed input can cause it to read past C<pe1>).
3861 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
3862 is less than C<s1>+C<l1>, the match will never be successful because it can
3864 get as far as its goal (and in fact is asserted against). Correspondingly for
3865 C<pe2> with respect to C<s2>.
3867 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
3868 C<l2> must be non-zero), and if both do, both have to be
3869 reached for a successful match. Also, if the fold of a character is multiple
3870 characters, all of them must be matched (see tr21 reference below for
3873 Upon a successful match, if C<pe1> is non-NULL,
3874 it will be set to point to the beginning of the I<next> character of C<s1>
3875 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
3877 For case-insensitiveness, the "casefolding" of Unicode is used
3878 instead of upper/lowercasing both the characters, see
3879 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
3883 /* A flags parameter has been added which may change, and hence isn't
3884 * externally documented. Currently it is:
3885 * 0 for as-documented above
3886 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
3887 ASCII one, to not match
3888 * FOLDEQ_LOCALE is set iff the rules from the current underlying
3889 * locale are to be used.
3890 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
3891 * routine. This allows that step to be skipped.
3892 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
3895 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)
3897 const U8 *p1 = (const U8*)s1; /* Point to current char */
3898 const U8 *p2 = (const U8*)s2;
3899 const U8 *g1 = NULL; /* goal for s1 */
3900 const U8 *g2 = NULL;
3901 const U8 *e1 = NULL; /* Don't scan s1 past this */
3902 U8 *f1 = NULL; /* Point to current folded */
3903 const U8 *e2 = NULL;
3905 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
3906 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
3907 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
3909 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
3911 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
3912 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
3913 /* The algorithm is to trial the folds without regard to the flags on
3914 * the first line of the above assert(), and then see if the result
3915 * violates them. This means that the inputs can't be pre-folded to a
3916 * violating result, hence the assert. This could be changed, with the
3917 * addition of extra tests here for the already-folded case, which would
3918 * slow it down. That cost is more than any possible gain for when these
3919 * flags are specified, as the flags indicate /il or /iaa matching which
3920 * is less common than /iu, and I (khw) also believe that real-world /il
3921 * and /iaa matches are most likely to involve code points 0-255, and this
3922 * function only under rare conditions gets called for 0-255. */
3924 if (IN_UTF8_CTYPE_LOCALE) {
3925 flags &= ~FOLDEQ_LOCALE;
3933 g1 = (const U8*)s1 + l1;
3941 g2 = (const U8*)s2 + l2;
3944 /* Must have at least one goal */
3949 /* Will never match if goal is out-of-bounds */
3950 assert(! e1 || e1 >= g1);
3952 /* Here, there isn't an end pointer, or it is beyond the goal. We
3953 * only go as far as the goal */
3957 assert(e1); /* Must have an end for looking at s1 */
3960 /* Same for goal for s2 */
3962 assert(! e2 || e2 >= g2);
3969 /* If both operands are already folded, we could just do a memEQ on the
3970 * whole strings at once, but it would be better if the caller realized
3971 * this and didn't even call us */
3973 /* Look through both strings, a character at a time */
3974 while (p1 < e1 && p2 < e2) {
3976 /* If at the beginning of a new character in s1, get its fold to use
3977 * and the length of the fold. (exception: locale rules just get the
3978 * character to a single byte) */
3980 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
3985 /* If in locale matching, we use two sets of rules, depending
3986 * on if the code point is above or below 255. Here, we test
3987 * for and handle locale rules */
3988 if ((flags & FOLDEQ_LOCALE)
3989 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
3991 /* There is no mixing of code points above and below 255. */
3992 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
3996 /* We handle locale rules by converting, if necessary, the
3997 * code point to a single byte. */
3998 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4002 *foldbuf1 = TWO_BYTE_UTF8_TO_NATIVE(*p1, *(p1 + 1));
4006 else if (isASCII(*p1)) { /* Note, that here won't be both
4007 ASCII and using locale rules */
4009 /* If trying to mix non- with ASCII, and not supposed to,
4011 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4015 *foldbuf1 = toFOLD(*p1);
4018 to_utf8_fold(p1, foldbuf1, &n1);
4020 else { /* Not utf8, get utf8 fold */
4021 to_uni_fold(*p1, foldbuf1, &n1);
4027 if (n2 == 0) { /* Same for s2 */
4028 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4033 if ((flags & FOLDEQ_LOCALE)
4034 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4036 /* Here, the next char in s2 is < 256. We've already
4037 * worked on s1, and if it isn't also < 256, can't match */
4038 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4041 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4045 *foldbuf2 = TWO_BYTE_UTF8_TO_NATIVE(*p2, *(p2 + 1));
4048 /* Use another function to handle locale rules. We've made
4049 * sure that both characters to compare are single bytes */
4050 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4055 else if (isASCII(*p2)) {
4056 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4060 *foldbuf2 = toFOLD(*p2);
4063 to_utf8_fold(p2, foldbuf2, &n2);
4066 to_uni_fold(*p2, foldbuf2, &n2);
4072 /* Here f1 and f2 point to the beginning of the strings to compare.
4073 * These strings are the folds of the next character from each input
4074 * string, stored in utf8. */
4076 /* While there is more to look for in both folds, see if they
4077 * continue to match */
4079 U8 fold_length = UTF8SKIP(f1);
4080 if (fold_length != UTF8SKIP(f2)
4081 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4082 function call for single
4084 || memNE((char*)f1, (char*)f2, fold_length))
4086 return 0; /* mismatch */
4089 /* Here, they matched, advance past them */
4096 /* When reach the end of any fold, advance the input past it */
4098 p1 += u1 ? UTF8SKIP(p1) : 1;
4101 p2 += u2 ? UTF8SKIP(p2) : 1;
4103 } /* End of loop through both strings */
4105 /* A match is defined by each scan that specified an explicit length
4106 * reaching its final goal, and the other not having matched a partial
4107 * character (which can happen when the fold of a character is more than one
4109 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4113 /* Successful match. Set output pointers */
4123 /* XXX The next two functions should likely be moved to mathoms.c once all
4124 * occurrences of them are removed from the core; some cpan-upstream modules
4128 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4130 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4132 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4136 =for apidoc utf8n_to_uvuni
4138 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4140 This function was useful for code that wanted to handle both EBCDIC and
4141 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4142 distinctions between the platforms have mostly been made invisible to most
4143 code, so this function is quite unlikely to be what you want. If you do need
4144 this precise functionality, use instead
4145 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4146 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4152 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4154 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4156 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4160 =for apidoc uvuni_to_utf8_flags
4162 Instead you almost certainly want to use L</uvchr_to_utf8> or
4163 L</uvchr_to_utf8_flags>>.
4165 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4166 which itself, while not deprecated, should be used only in isolated
4167 circumstances. These functions were useful for code that wanted to handle
4168 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4169 v5.20, the distinctions between the platforms have mostly been made invisible
4170 to most code, so this function is quite unlikely to be what you want.
4176 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4178 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4180 return uvoffuni_to_utf8_flags(d, uv, flags);
4185 * c-indentation-style: bsd
4187 * indent-tabs-mode: nil
4190 * ex: set ts=8 sts=4 sw=4 et: