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"
36 static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
40 =head1 Unicode Support
41 These are various utility functions for manipulating UTF8-encoded
42 strings. For the uninitiated, this is a method of representing arbitrary
43 Unicode characters as a variable number of bytes, in such a way that
44 characters in the ASCII range are unmodified, and a zero byte never appears
45 within non-zero characters.
51 =for apidoc is_invariant_string
53 Returns true iff the first C<len> bytes of the string C<s> are the same
54 regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
55 EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
56 machines, all the ASCII characters and only the ASCII characters fit this
57 definition. On EBCDIC machines, the ASCII-range characters are invariant, but
58 so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
61 If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
62 use this option, that C<s> can't have embedded C<NUL> characters and has to
63 have a terminating C<NUL> byte).
65 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
71 Perl_is_invariant_string(const U8 *s, STRLEN len)
73 const U8* const send = s + (len ? len : strlen((const char *)s));
76 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
78 for (; x < send; ++x) {
79 if (!UTF8_IS_INVARIANT(*x))
87 =for apidoc uvoffuni_to_utf8_flags
89 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
90 Instead, B<Almost all code should use L</uvchr_to_utf8> or
91 L</uvchr_to_utf8_flags>>.
93 This function is like them, but the input is a strict Unicode
94 (as opposed to native) code point. Only in very rare circumstances should code
95 not be using the native code point.
97 For details, see the description for L</uvchr_to_utf8_flags>>.
103 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
105 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
107 if (UNI_IS_INVARIANT(uv)) {
108 *d++ = (U8) LATIN1_TO_NATIVE(uv);
113 /* Not representable in UTF-EBCDIC */
114 flags |= UNICODE_DISALLOW_FE_FF;
117 /* The first problematic code point is the first surrogate */
118 if (uv >= UNICODE_SURROGATE_FIRST
119 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
121 if (UNICODE_IS_SURROGATE(uv)) {
122 if (flags & UNICODE_WARN_SURROGATE) {
123 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
124 "UTF-16 surrogate U+%04"UVXf, uv);
126 if (flags & UNICODE_DISALLOW_SURROGATE) {
130 else if (UNICODE_IS_SUPER(uv)) {
131 if (flags & UNICODE_WARN_SUPER
132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
134 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
135 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
137 if (flags & UNICODE_DISALLOW_SUPER
138 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
141 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
142 NOT_REACHED; /* NOTREACHED */
147 else if (UNICODE_IS_NONCHAR(uv)) {
148 if (flags & UNICODE_WARN_NONCHAR) {
149 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
150 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
153 if (flags & UNICODE_DISALLOW_NONCHAR) {
161 STRLEN len = OFFUNISKIP(uv);
164 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
165 uv >>= UTF_ACCUMULATION_SHIFT;
167 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
170 #else /* Non loop style */
172 *d++ = (U8)(( uv >> 6) | 0xc0);
173 *d++ = (U8)(( uv & 0x3f) | 0x80);
177 *d++ = (U8)(( uv >> 12) | 0xe0);
178 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
179 *d++ = (U8)(( uv & 0x3f) | 0x80);
183 *d++ = (U8)(( uv >> 18) | 0xf0);
184 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
185 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
186 *d++ = (U8)(( uv & 0x3f) | 0x80);
189 if (uv < 0x4000000) {
190 *d++ = (U8)(( uv >> 24) | 0xf8);
191 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
192 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
193 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
194 *d++ = (U8)(( uv & 0x3f) | 0x80);
197 if (uv < 0x80000000) {
198 *d++ = (U8)(( uv >> 30) | 0xfc);
199 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
207 if (uv < UTF8_QUAD_MAX)
210 *d++ = 0xfe; /* Can't match U+FEFF! */
211 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
212 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
216 *d++ = (U8)(( uv & 0x3f) | 0x80);
221 *d++ = 0xff; /* Can't match U+FFFE! */
222 *d++ = 0x80; /* 6 Reserved bits */
223 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
224 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
233 *d++ = (U8)(( uv & 0x3f) | 0x80);
237 #endif /* Non loop style */
240 =for apidoc uvchr_to_utf8
242 Adds the UTF-8 representation of the native code point C<uv> to the end
243 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
244 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
245 the byte after the end of the new character. In other words,
247 d = uvchr_to_utf8(d, uv);
249 is the recommended wide native character-aware way of saying
253 This function accepts any UV as input. To forbid or warn on non-Unicode code
254 points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
259 /* This is also a macro */
260 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
263 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
265 return uvchr_to_utf8(d, uv);
269 =for apidoc uvchr_to_utf8_flags
271 Adds the UTF-8 representation of the native code point C<uv> to the end
272 of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
273 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
274 the byte after the end of the new character. In other words,
276 d = uvchr_to_utf8_flags(d, uv, flags);
280 d = uvchr_to_utf8_flags(d, uv, 0);
282 This is the Unicode-aware way of saying
286 This function will convert to UTF-8 (and not warn) even code points that aren't
287 legal Unicode or are problematic, unless C<flags> contains one or more of the
290 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
291 the function will raise a warning, provided UTF8 warnings are enabled. If instead
292 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
293 If both flags are set, the function will both warn and return NULL.
295 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
296 affect how the function handles a Unicode non-character. And likewise, the
297 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
299 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
300 even less portable) can be warned and/or disallowed even if other above-Unicode
301 code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
304 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
305 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
311 /* This is also a macro */
312 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
315 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
317 return uvchr_to_utf8_flags(d, uv, flags);
321 =for apidoc is_utf8_string
323 Returns true if the first C<len> bytes of string C<s> form a valid
324 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
325 using C<strlen(s)> (which means if you use this option, that C<s> can't have
326 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
327 that all characters being ASCII constitute 'a valid UTF-8 string'.
329 See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
335 Perl_is_utf8_string(const U8 *s, STRLEN len)
337 const U8* const send = s + (len ? len : strlen((const char *)s));
340 PERL_ARGS_ASSERT_IS_UTF8_STRING;
343 STRLEN len = isUTF8_CHAR(x, send);
344 if (UNLIKELY(! len)) {
354 Implemented as a macro in utf8.h
356 =for apidoc is_utf8_string_loc
358 Like L</is_utf8_string> but stores the location of the failure (in the
359 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
360 "utf8ness success") in the C<ep>.
362 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
364 =for apidoc is_utf8_string_loclen
366 Like L</is_utf8_string>() but stores the location of the failure (in the
367 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
368 "utf8ness success") in the C<ep>, and the number of UTF-8
369 encoded characters in the C<el>.
371 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
377 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
379 const U8* const send = s + (len ? len : strlen((const char *)s));
383 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
386 STRLEN len = isUTF8_CHAR(x, send);
387 if (UNLIKELY(! len)) {
405 =for apidoc utf8n_to_uvchr
407 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
408 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
410 Bottom level UTF-8 decode routine.
411 Returns the native code point value of the first character in the string C<s>,
412 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
413 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
414 the length, in bytes, of that character.
416 The value of C<flags> determines the behavior when C<s> does not point to a
417 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
418 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
419 next possible position in C<s> that could begin a non-malformed character.
420 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
422 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
423 individual types of malformations, such as the sequence being overlong (that
424 is, when there is a shorter sequence that can express the same code point;
425 overlong sequences are expressly forbidden in the UTF-8 standard due to
426 potential security issues). Another malformation example is the first byte of
427 a character not being a legal first byte. See F<utf8.h> for the list of such
428 flags. For allowed 0 length strings, this function returns 0; for allowed
429 overlong sequences, the computed code point is returned; for all other allowed
430 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
431 determinable reasonable value.
433 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
434 flags) malformation is found. If this flag is set, the routine assumes that
435 the caller will raise a warning, and this function will silently just set
436 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
438 Note that this API requires disambiguation between successful decoding a C<NUL>
439 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
440 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
441 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
444 Certain code points are considered problematic. These are Unicode surrogates,
445 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
446 By default these are considered regular code points, but certain situations
447 warrant special handling for them. If C<flags> contains
448 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
449 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
450 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
451 maximum) can be set to disallow these categories individually.
453 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
454 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
455 for their respective categories, but otherwise the code points are considered
456 valid (not malformations). To get a category to both be treated as a
457 malformation and raise a warning, specify both the WARN and DISALLOW flags.
458 (But note that warnings are not raised if lexically disabled nor if
459 UTF8_CHECK_ONLY is also specified.)
461 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
462 the others that are above the Unicode legal maximum. There are several
463 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
464 not representable at all on EBCDIC platforms, and the original UTF-8
465 specification never went above this number (the current 0x10FFFF limit was
466 imposed later). (The smaller ones, those that fit into 32 bits, are
467 representable by a UV on ASCII platforms, but not by an IV, which means that
468 the number of operations that can be performed on them is quite restricted.)
469 The UTF-8 encoding on ASCII platforms for these large code points begins with a
470 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
471 be treated as malformations, while allowing smaller above-Unicode code points.
472 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
473 including these, as malformations.)
474 Similarly, UTF8_WARN_FE_FF acts just like
475 the other WARN flags, but applies just to these code points.
477 All other code points corresponding to Unicode characters, including private
478 use and those yet to be assigned, are never considered malformed and never
485 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
487 const U8 * const s0 = s;
488 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
493 UV outlier_ret = 0; /* return value when input is in error or problematic
495 UV pack_warn = 0; /* Save result of packWARN() for later */
496 bool unexpected_non_continuation = FALSE;
497 bool overflowed = FALSE;
498 bool do_overlong_test = TRUE; /* May have to skip this test */
500 const char* const malformed_text = "Malformed UTF-8 character";
502 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
504 /* The order of malformation tests here is important. We should consume as
505 * few bytes as possible in order to not skip any valid character. This is
506 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
507 * http://unicode.org/reports/tr36 for more discussion as to why. For
508 * example, once we've done a UTF8SKIP, we can tell the expected number of
509 * bytes, and could fail right off the bat if the input parameters indicate
510 * that there are too few available. But it could be that just that first
511 * byte is garbled, and the intended character occupies fewer bytes. If we
512 * blindly assumed that the first byte is correct, and skipped based on
513 * that number, we could skip over a valid input character. So instead, we
514 * always examine the sequence byte-by-byte.
516 * We also should not consume too few bytes, otherwise someone could inject
517 * things. For example, an input could be deliberately designed to
518 * overflow, and if this code bailed out immediately upon discovering that,
519 * returning to the caller C<*retlen> pointing to the very next byte (one
520 * which is actually part of of the overflowing sequence), that could look
521 * legitimate to the caller, which could discard the initial partial
522 * sequence and process the rest, inappropriately */
524 /* Zero length strings, if allowed, of necessity are zero */
525 if (UNLIKELY(curlen == 0)) {
530 if (flags & UTF8_ALLOW_EMPTY) {
533 if (! (flags & UTF8_CHECK_ONLY)) {
534 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
539 expectlen = UTF8SKIP(s);
541 /* A well-formed UTF-8 character, as the vast majority of calls to this
542 * function will be for, has this expected length. For efficiency, set
543 * things up here to return it. It will be overriden only in those rare
544 * cases where a malformation is found */
549 /* An invariant is trivially well-formed */
550 if (UTF8_IS_INVARIANT(uv)) {
554 /* A continuation character can't start a valid sequence */
555 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
556 if (flags & UTF8_ALLOW_CONTINUATION) {
560 return UNICODE_REPLACEMENT;
563 if (! (flags & UTF8_CHECK_ONLY)) {
564 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
570 /* Here is not a continuation byte, nor an invariant. The only thing left
571 * is a start byte (possibly for an overlong) */
574 uv = NATIVE_UTF8_TO_I8(uv);
577 /* Remove the leading bits that indicate the number of bytes in the
578 * character's whole UTF-8 sequence, leaving just the bits that are part of
580 uv &= UTF_START_MASK(expectlen);
582 /* Now, loop through the remaining bytes in the character's sequence,
583 * accumulating each into the working value as we go. Be sure to not look
584 * past the end of the input string */
585 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
587 for (s = s0 + 1; s < send; s++) {
588 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
589 #ifndef EBCDIC /* Can't overflow in EBCDIC */
590 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
592 /* The original implementors viewed this malformation as more
593 * serious than the others (though I, khw, don't understand
594 * why, since other malformations also give very very wrong
595 * results), so there is no way to turn off checking for it.
596 * Set a flag, but keep going in the loop, so that we absorb
597 * the rest of the bytes that comprise the character. */
599 overflow_byte = *s; /* Save for warning message's use */
602 uv = UTF8_ACCUMULATE(uv, *s);
605 /* Here, found a non-continuation before processing all expected
606 * bytes. This byte begins a new character, so quit, even if
607 * allowing this malformation. */
608 unexpected_non_continuation = TRUE;
611 } /* End of loop through the character's bytes */
613 /* Save how many bytes were actually in the character */
616 /* The loop above finds two types of malformations: non-continuation and/or
617 * overflow. The non-continuation malformation is really a too-short
618 * malformation, as it means that the current character ended before it was
619 * expected to (being terminated prematurely by the beginning of the next
620 * character, whereas in the too-short malformation there just are too few
621 * bytes available to hold the character. In both cases, the check below
622 * that we have found the expected number of bytes would fail if executed.)
623 * Thus the non-continuation malformation is really unnecessary, being a
624 * subset of the too-short malformation. But there may be existing
625 * applications that are expecting the non-continuation type, so we retain
626 * it, and return it in preference to the too-short malformation. (If this
627 * code were being written from scratch, the two types might be collapsed
628 * into one.) I, khw, am also giving priority to returning the
629 * non-continuation and too-short malformations over overflow when multiple
630 * ones are present. I don't know of any real reason to prefer one over
631 * the other, except that it seems to me that multiple-byte errors trumps
632 * errors from a single byte */
633 if (UNLIKELY(unexpected_non_continuation)) {
634 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
635 if (! (flags & UTF8_CHECK_ONLY)) {
637 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
640 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));
645 uv = UNICODE_REPLACEMENT;
647 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
648 * as what the original expectations were. */
649 do_overlong_test = FALSE;
654 else if (UNLIKELY(curlen < expectlen)) {
655 if (! (flags & UTF8_ALLOW_SHORT)) {
656 if (! (flags & UTF8_CHECK_ONLY)) {
657 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));
661 uv = UNICODE_REPLACEMENT;
662 do_overlong_test = FALSE;
668 #ifndef EBCDIC /* EBCDIC can't overflow */
669 if (UNLIKELY(overflowed)) {
670 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
676 && expectlen > (STRLEN) OFFUNISKIP(uv)
677 && ! (flags & UTF8_ALLOW_LONG))
679 /* The overlong malformation has lower precedence than the others.
680 * Note that if this malformation is allowed, we return the actual
681 * value, instead of the replacement character. This is because this
682 * value is actually well-defined. */
683 if (! (flags & UTF8_CHECK_ONLY)) {
684 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));
689 /* Here, the input is considered to be well-formed, but it still could be a
690 * problematic code point that is not allowed by the input parameters. */
691 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
692 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
693 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
695 if (UNICODE_IS_SURROGATE(uv)) {
697 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
698 * generation of the sv, since no warnings are raised under CHECK */
699 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
700 && ckWARN_d(WARN_SURROGATE))
702 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
703 pack_warn = packWARN(WARN_SURROGATE);
705 if (flags & UTF8_DISALLOW_SURROGATE) {
709 else if ((uv > PERL_UNICODE_MAX)) {
710 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
711 && ckWARN_d(WARN_NON_UNICODE))
713 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
714 pack_warn = packWARN(WARN_NON_UNICODE);
716 #ifndef EBCDIC /* EBCDIC always allows FE, FF */
718 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
719 * points. We test for these after the regular SUPER ones, and
720 * before possibly bailing out, so that the more dire warning
721 * overrides the regular one, if applicable */
722 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
723 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
725 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
727 && ckWARN_d(WARN_UTF8))
729 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
730 pack_warn = packWARN(WARN_UTF8);
732 if (flags & UTF8_DISALLOW_FE_FF) {
737 if (flags & UTF8_DISALLOW_SUPER) {
741 else if (UNICODE_IS_NONCHAR(uv)) {
742 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
743 && ckWARN_d(WARN_NONCHAR))
745 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
746 pack_warn = packWARN(WARN_NONCHAR);
748 if (flags & UTF8_DISALLOW_NONCHAR) {
754 outlier_ret = uv; /* Note we don't bother to convert to native,
755 as all the outlier code points are the same
756 in both ASCII and EBCDIC */
760 /* Here, this is not considered a malformed character, so drop through
764 return UNI_TO_NATIVE(uv);
766 /* There are three cases which get to beyond this point. In all 3 cases:
767 * <sv> if not null points to a string to print as a warning.
768 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
770 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
771 * This is done by initializing it to 0, and changing it only
774 * 1) The input is valid but problematic, and to be warned about. The
775 * return value is the resultant code point; <*retlen> is set to
776 * <curlen>, the number of bytes that comprise the code point.
777 * <pack_warn> contains the result of packWARN() for the warning
778 * types. The entry point for this case is the label <do_warn>;
779 * 2) The input is a valid code point but disallowed by the parameters to
780 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
781 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
782 * comprise the code point. <pack_warn> contains the result of
783 * packWARN() for the warning types. The entry point for this case is
784 * the label <disallowed>.
785 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
786 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
787 * bytes that comprise the malformation. All such malformations are
788 * assumed to be warning type <utf8>. The entry point for this case
789 * is the label <malformed>.
794 if (sv && ckWARN_d(WARN_UTF8)) {
795 pack_warn = packWARN(WARN_UTF8);
800 if (flags & UTF8_CHECK_ONLY) {
802 *retlen = ((STRLEN) -1);
808 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
809 if warnings are to be raised. */
810 const char * const string = SvPVX_const(sv);
813 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
815 Perl_warner(aTHX_ pack_warn, "%s", string);
826 =for apidoc utf8_to_uvchr_buf
828 Returns the native code point of the first character in the string C<s> which
829 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
830 C<*retlen> will be set to the length, in bytes, of that character.
832 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
833 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
834 NULL) to -1. If those warnings are off, the computed value, if well-defined
835 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
836 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
837 the next possible position in C<s> that could begin a non-malformed character.
838 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
846 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
850 return utf8n_to_uvchr(s, send - s, retlen,
851 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
854 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
855 * there are no malformations in the input UTF-8 string C<s>. surrogates,
856 * non-character code points, and non-Unicode code points are allowed. */
859 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
861 UV expectlen = UTF8SKIP(s);
862 const U8* send = s + expectlen;
865 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
872 /* An invariant is trivially returned */
873 if (expectlen == 1) {
878 uv = NATIVE_UTF8_TO_I8(uv);
881 /* Remove the leading bits that indicate the number of bytes, leaving just
882 * the bits that are part of the value */
883 uv &= UTF_START_MASK(expectlen);
885 /* Now, loop through the remaining bytes, accumulating each into the
886 * working total as we go. (I khw tried unrolling the loop for up to 4
887 * bytes, but there was no performance improvement) */
888 for (++s; s < send; s++) {
889 uv = UTF8_ACCUMULATE(uv, *s);
892 return UNI_TO_NATIVE(uv);
897 =for apidoc utf8_to_uvuni_buf
899 Only in very rare circumstances should code need to be dealing in Unicode
900 (as opposed to native) code points. In those few cases, use
901 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
903 Returns the Unicode (not-native) code point of the first character in the
905 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
906 C<retlen> will be set to the length, in bytes, of that character.
908 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
909 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
910 NULL) to -1. If those warnings are off, the computed value if well-defined (or
911 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
912 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
913 next possible position in C<s> that could begin a non-malformed character.
914 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
920 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
922 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
926 /* Call the low level routine asking for checks */
927 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
928 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
932 =for apidoc utf8_length
934 Return the length of the UTF-8 char encoded string C<s> in characters.
935 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
936 up past C<e>, croaks.
942 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
946 PERL_ARGS_ASSERT_UTF8_LENGTH;
948 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
949 * the bitops (especially ~) can create illegal UTF-8.
950 * In other words: in Perl UTF-8 is not just for Unicode. */
953 goto warn_and_return;
963 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
964 "%s in %s", unees, OP_DESC(PL_op));
966 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
973 =for apidoc utf8_distance
975 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
978 WARNING: use only if you *know* that the pointers point inside the
985 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
987 PERL_ARGS_ASSERT_UTF8_DISTANCE;
989 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
995 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
998 WARNING: do not use the following unless you *know* C<off> is within
999 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1000 on the first byte of character or just after the last byte of a character.
1006 Perl_utf8_hop(const U8 *s, I32 off)
1008 PERL_ARGS_ASSERT_UTF8_HOP;
1010 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1011 * the bitops (especially ~) can create illegal UTF-8.
1012 * In other words: in Perl UTF-8 is not just for Unicode. */
1021 while (UTF8_IS_CONTINUATION(*s))
1029 =for apidoc bytes_cmp_utf8
1031 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1032 sequence of characters (stored as UTF-8)
1033 in C<u>, C<ulen>. Returns 0 if they are
1034 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1035 if the first string is greater than the second string.
1037 -1 or +1 is returned if the shorter string was identical to the start of the
1038 longer string. -2 or +2 is returned if
1039 there was a difference between characters
1046 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1048 const U8 *const bend = b + blen;
1049 const U8 *const uend = u + ulen;
1051 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1053 while (b < bend && u < uend) {
1055 if (!UTF8_IS_INVARIANT(c)) {
1056 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1059 if (UTF8_IS_CONTINUATION(c1)) {
1060 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1062 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1063 "Malformed UTF-8 character "
1064 "(unexpected non-continuation byte 0x%02x"
1065 ", immediately after start byte 0x%02x)"
1066 /* Dear diag.t, it's in the pod. */
1068 PL_op ? " in " : "",
1069 PL_op ? OP_DESC(PL_op) : "");
1074 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1075 "%s in %s", unees, OP_DESC(PL_op));
1077 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1078 return -2; /* Really want to return undef :-) */
1085 return *b < c ? -2 : +2;
1090 if (b == bend && u == uend)
1093 return b < bend ? +1 : -1;
1097 =for apidoc utf8_to_bytes
1099 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1100 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1101 updates C<len> to contain the new length.
1102 Returns zero on failure, setting C<len> to -1.
1104 If you need a copy of the string, see L</bytes_from_utf8>.
1110 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1112 U8 * const save = s;
1113 U8 * const send = s + *len;
1116 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1117 PERL_UNUSED_CONTEXT;
1119 /* ensure valid UTF-8 and chars < 256 before updating string */
1121 if (! UTF8_IS_INVARIANT(*s)) {
1122 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1123 *len = ((STRLEN) -1);
1134 if (! UTF8_IS_INVARIANT(c)) {
1135 /* Then it is two-byte encoded */
1136 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1147 =for apidoc bytes_from_utf8
1149 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1150 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1151 the newly-created string, and updates C<len> to contain the new
1152 length. Returns the original string if no conversion occurs, C<len>
1153 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1154 0 if C<s> is converted or consisted entirely of characters that are invariant
1155 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1161 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1164 const U8 *start = s;
1168 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1169 PERL_UNUSED_CONTEXT;
1173 /* ensure valid UTF-8 and chars < 256 before converting string */
1174 for (send = s + *len; s < send;) {
1175 if (! UTF8_IS_INVARIANT(*s)) {
1176 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1187 Newx(d, (*len) - count + 1, U8);
1188 s = start; start = d;
1191 if (! UTF8_IS_INVARIANT(c)) {
1192 /* Then it is two-byte encoded */
1193 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1204 =for apidoc bytes_to_utf8
1206 Converts a string C<s> of length C<len> bytes from the native encoding into
1208 Returns a pointer to the newly-created string, and sets C<len> to
1209 reflect the new length in bytes.
1211 A C<NUL> character will be written after the end of the string.
1213 If you want to convert to UTF-8 from encodings other than
1214 the native (Latin1 or EBCDIC),
1215 see L</sv_recode_to_utf8>().
1220 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1221 likewise need duplication. */
1224 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1226 const U8 * const send = s + (*len);
1230 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1231 PERL_UNUSED_CONTEXT;
1233 Newx(d, (*len) * 2 + 1, U8);
1237 append_utf8_from_native_byte(*s, &d);
1246 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1248 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1249 * We optimize for native, for obvious reasons. */
1252 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1257 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1260 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1265 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1267 if (UNI_IS_INVARIANT(uv)) {
1268 *d++ = LATIN1_TO_NATIVE((U8) uv);
1271 if (uv <= MAX_UTF8_TWO_BYTE) {
1272 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1273 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1276 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1277 #define LAST_HIGH_SURROGATE 0xDBFF
1278 #define FIRST_LOW_SURROGATE 0xDC00
1279 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1281 /* This assumes that most uses will be in the first Unicode plane, not
1282 * needing surrogates */
1283 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1284 && uv <= UNICODE_SURROGATE_LAST))
1286 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1287 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1290 UV low = (p[0] << 8) + p[1];
1291 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1292 || UNLIKELY(low > LAST_LOW_SURROGATE))
1294 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1297 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1298 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1302 d = uvoffuni_to_utf8_flags(d, uv, 0);
1305 *d++ = (U8)(( uv >> 12) | 0xe0);
1306 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1307 *d++ = (U8)(( uv & 0x3f) | 0x80);
1311 *d++ = (U8)(( uv >> 18) | 0xf0);
1312 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1313 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1314 *d++ = (U8)(( uv & 0x3f) | 0x80);
1319 *newlen = d - dstart;
1323 /* Note: this one is slightly destructive of the source. */
1326 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1329 U8* const send = s + bytelen;
1331 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1334 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1338 const U8 tmp = s[0];
1343 return utf16_to_utf8(p, d, bytelen, newlen);
1347 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1349 U8 tmpbuf[UTF8_MAXBYTES+1];
1350 uvchr_to_utf8(tmpbuf, c);
1351 return _is_utf8_FOO(classnum, tmpbuf);
1354 /* Internal function so we can deprecate the external one, and call
1355 this one from other deprecated functions in this file */
1358 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1360 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1364 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1368 Perl__is_uni_perl_idcont(pTHX_ UV c)
1370 U8 tmpbuf[UTF8_MAXBYTES+1];
1371 uvchr_to_utf8(tmpbuf, c);
1372 return _is_utf8_perl_idcont(tmpbuf);
1376 Perl__is_uni_perl_idstart(pTHX_ UV c)
1378 U8 tmpbuf[UTF8_MAXBYTES+1];
1379 uvchr_to_utf8(tmpbuf, c);
1380 return _is_utf8_perl_idstart(tmpbuf);
1384 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1386 /* We have the latin1-range values compiled into the core, so just use
1387 * those, converting the result to utf8. The only difference between upper
1388 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1389 * either "SS" or "Ss". Which one to use is passed into the routine in
1390 * 'S_or_s' to avoid a test */
1392 UV converted = toUPPER_LATIN1_MOD(c);
1394 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1396 assert(S_or_s == 'S' || S_or_s == 's');
1398 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1399 characters in this range */
1400 *p = (U8) converted;
1405 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1406 * which it maps to one of them, so as to only have to have one check for
1407 * it in the main case */
1408 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1410 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1411 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1414 converted = GREEK_CAPITAL_LETTER_MU;
1416 case LATIN_SMALL_LETTER_SHARP_S:
1422 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1423 NOT_REACHED; /* NOTREACHED */
1427 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1428 *p = UTF8_TWO_BYTE_LO(converted);
1434 /* Call the function to convert a UTF-8 encoded character to the specified case.
1435 * Note that there may be more than one character in the result.
1436 * INP is a pointer to the first byte of the input character
1437 * OUTP will be set to the first byte of the string of changed characters. It
1438 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1439 * LENP will be set to the length in bytes of the string of changed characters
1441 * The functions return the ordinal of the first character in the string of OUTP */
1442 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1443 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1444 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1446 /* This additionally has the input parameter SPECIALS, which if non-zero will
1447 * cause this to use the SPECIALS hash for folding (meaning get full case
1448 * folding); otherwise, when zero, this implies a simple case fold */
1449 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1452 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1454 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1455 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1456 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1457 * the changed version may be longer than the original character.
1459 * The ordinal of the first character of the changed version is returned
1460 * (but note, as explained above, that there may be more.) */
1462 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1465 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1468 uvchr_to_utf8(p, c);
1469 return CALL_UPPER_CASE(p, p, lenp);
1473 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1475 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1478 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1481 uvchr_to_utf8(p, c);
1482 return CALL_TITLE_CASE(p, p, lenp);
1486 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
1488 /* We have the latin1-range values compiled into the core, so just use
1489 * those, converting the result to utf8. Since the result is always just
1490 * one character, we allow <p> to be NULL */
1492 U8 converted = toLOWER_LATIN1(c);
1495 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1500 /* Result is known to always be < 256, so can use the EIGHT_BIT
1502 *p = UTF8_EIGHT_BIT_HI(converted);
1503 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
1511 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1513 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1516 return to_lower_latin1((U8) c, p, lenp);
1519 uvchr_to_utf8(p, c);
1520 return CALL_LOWER_CASE(p, p, lenp);
1524 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1526 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1527 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1528 * FOLD_FLAGS_FULL iff full folding is to be used;
1530 * Not to be used for locale folds
1535 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1536 PERL_UNUSED_CONTEXT;
1538 assert (! (flags & FOLD_FLAGS_LOCALE));
1540 if (c == MICRO_SIGN) {
1541 converted = GREEK_SMALL_LETTER_MU;
1543 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1545 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1546 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1547 * under those circumstances. */
1548 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1549 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1550 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1552 return LATIN_SMALL_LETTER_LONG_S;
1561 else { /* In this range the fold of all other characters is their lower
1563 converted = toLOWER_LATIN1(c);
1566 if (UVCHR_IS_INVARIANT(converted)) {
1567 *p = (U8) converted;
1571 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1572 *p = UTF8_TWO_BYTE_LO(converted);
1580 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1583 /* Not currently externally documented, and subject to change
1584 * <flags> bits meanings:
1585 * FOLD_FLAGS_FULL iff full folding is to be used;
1586 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1587 * locale are to be used.
1588 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1591 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1593 if (flags & FOLD_FLAGS_LOCALE) {
1594 /* Treat a UTF-8 locale as not being in locale at all */
1595 if (IN_UTF8_CTYPE_LOCALE) {
1596 flags &= ~FOLD_FLAGS_LOCALE;
1599 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1600 goto needs_full_generality;
1605 return _to_fold_latin1((U8) c, p, lenp,
1606 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1609 /* Here, above 255. If no special needs, just use the macro */
1610 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1611 uvchr_to_utf8(p, c);
1612 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1614 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1615 the special flags. */
1616 U8 utf8_c[UTF8_MAXBYTES + 1];
1618 needs_full_generality:
1619 uvchr_to_utf8(utf8_c, c);
1620 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1624 PERL_STATIC_INLINE bool
1625 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1626 const char *const swashname, SV* const invlist)
1628 /* returns a boolean giving whether or not the UTF8-encoded character that
1629 * starts at <p> is in the swash indicated by <swashname>. <swash>
1630 * contains a pointer to where the swash indicated by <swashname>
1631 * is to be stored; which this routine will do, so that future calls will
1632 * look at <*swash> and only generate a swash if it is not null. <invlist>
1633 * is NULL or an inversion list that defines the swash. If not null, it
1634 * saves time during initialization of the swash.
1636 * Note that it is assumed that the buffer length of <p> is enough to
1637 * contain all the bytes that comprise the character. Thus, <*p> should
1638 * have been checked before this call for mal-formedness enough to assure
1641 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1643 /* The API should have included a length for the UTF-8 character in <p>,
1644 * but it doesn't. We therefore assume that p has been validated at least
1645 * as far as there being enough bytes available in it to accommodate the
1646 * character without reading beyond the end, and pass that number on to the
1647 * validating routine */
1648 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1649 if (ckWARN_d(WARN_UTF8)) {
1650 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1651 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1652 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1653 what the malformation is */
1654 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1660 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1661 *swash = _core_swash_init("utf8",
1663 /* Only use the name if there is no inversion
1664 * list; otherwise will go out to disk */
1665 (invlist) ? "" : swashname,
1667 &PL_sv_undef, 1, 0, invlist, &flags);
1670 return swash_fetch(*swash, p, TRUE) != 0;
1674 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1676 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1678 assert(classnum < _FIRST_NON_SWASH_CC);
1680 return is_utf8_common(p,
1681 &PL_utf8_swash_ptrs[classnum],
1682 swash_property_names[classnum],
1683 PL_XPosix_ptrs[classnum]);
1687 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1691 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1693 if (! PL_utf8_perl_idstart) {
1694 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1696 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
1700 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1702 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1706 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1710 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1714 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1716 if (! PL_utf8_perl_idcont) {
1717 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1719 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
1723 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1725 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1727 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1731 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1733 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1735 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1739 Perl__is_utf8_mark(pTHX_ const U8 *p)
1741 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1743 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1747 =for apidoc to_utf8_case
1749 C<p> contains the pointer to the UTF-8 string encoding
1750 the character that is being converted. This routine assumes that the character
1751 at C<p> is well-formed.
1753 C<ustrp> is a pointer to the character buffer to put the
1754 conversion result to. C<lenp> is a pointer to the length
1757 C<swashp> is a pointer to the swash to use.
1759 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1760 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
1761 but not always, a multicharacter mapping), is tried first.
1763 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1764 any special mappings; C<""> means to use the special mappings. Values other
1765 than these two are treated as the name of the hash containing the special
1766 mappings, like C<"utf8::ToSpecLower">.
1768 C<normal> is a string like "ToLower" which means the swash
1774 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1775 SV **swashp, const char *normal, const char *special)
1778 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1780 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1782 /* Note that swash_fetch() doesn't output warnings for these because it
1783 * assumes we will */
1784 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1785 if (uv1 <= UNICODE_SURROGATE_LAST) {
1786 if (ckWARN_d(WARN_SURROGATE)) {
1787 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1788 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1789 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1792 else if (UNICODE_IS_SUPER(uv1)) {
1793 if (ckWARN_d(WARN_NON_UNICODE)) {
1794 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1795 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1796 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1800 /* Note that non-characters are perfectly legal, so no warning should
1804 if (!*swashp) /* load on-demand */
1805 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1808 /* It might be "special" (sometimes, but not always,
1809 * a multicharacter mapping) */
1813 /* If passed in the specials name, use that; otherwise use any
1814 * given in the swash */
1815 if (*special != '\0') {
1816 hv = get_hv(special, 0);
1819 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1821 hv = MUTABLE_HV(SvRV(*svp));
1826 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1831 s = SvPV_const(*svp, len);
1834 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1836 Copy(s, ustrp, len, U8);
1841 if (!len && *swashp) {
1842 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
1845 /* It was "normal" (a single character mapping). */
1846 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1854 return valid_utf8_to_uvchr(ustrp, 0);
1857 /* Here, there was no mapping defined, which means that the code point maps
1858 * to itself. Return the inputs */
1860 if (p != ustrp) { /* Don't copy onto itself */
1861 Copy(p, ustrp, len, U8);
1872 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1874 /* This is called when changing the case of a utf8-encoded character above
1875 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1876 * result contains a character that crosses the 255/256 boundary, disallow
1877 * the change, and return the original code point. See L<perlfunc/lc> for
1880 * p points to the original string whose case was changed; assumed
1881 * by this routine to be well-formed
1882 * result the code point of the first character in the changed-case string
1883 * ustrp points to the changed-case string (<result> represents its first char)
1884 * lenp points to the length of <ustrp> */
1886 UV original; /* To store the first code point of <p> */
1888 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1890 assert(UTF8_IS_ABOVE_LATIN1(*p));
1892 /* We know immediately if the first character in the string crosses the
1893 * boundary, so can skip */
1896 /* Look at every character in the result; if any cross the
1897 * boundary, the whole thing is disallowed */
1898 U8* s = ustrp + UTF8SKIP(ustrp);
1899 U8* e = ustrp + *lenp;
1901 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
1907 /* Here, no characters crossed, result is ok as-is, but we warn. */
1908 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
1914 /* Failed, have to return the original */
1915 original = valid_utf8_to_uvchr(p, lenp);
1917 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1918 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1919 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1920 "resolved to \"\\x{%"UVXf"}\".",
1924 Copy(p, ustrp, *lenp, char);
1929 =for apidoc to_utf8_upper
1931 Instead use L</toUPPER_utf8>.
1935 /* Not currently externally documented, and subject to change:
1936 * <flags> is set iff iff the rules from the current underlying locale are to
1940 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
1944 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
1947 /* Treat a UTF-8 locale as not being in locale at all */
1948 if (IN_UTF8_CTYPE_LOCALE) {
1952 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1956 if (UTF8_IS_INVARIANT(*p)) {
1958 result = toUPPER_LC(*p);
1961 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
1964 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
1966 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
1967 result = toUPPER_LC(c);
1970 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
1974 else { /* utf8, ord above 255 */
1975 result = CALL_UPPER_CASE(p, ustrp, lenp);
1978 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
1983 /* Here, used locale rules. Convert back to utf8 */
1984 if (UTF8_IS_INVARIANT(result)) {
1985 *ustrp = (U8) result;
1989 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1990 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
1998 =for apidoc to_utf8_title
2000 Instead use L</toTITLE_utf8>.
2004 /* Not currently externally documented, and subject to change:
2005 * <flags> is set iff the rules from the current underlying locale are to be
2006 * used. Since titlecase is not defined in POSIX, for other than a
2007 * UTF-8 locale, uppercase is used instead for code points < 256.
2011 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2015 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2018 /* Treat a UTF-8 locale as not being in locale at all */
2019 if (IN_UTF8_CTYPE_LOCALE) {
2023 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2027 if (UTF8_IS_INVARIANT(*p)) {
2029 result = toUPPER_LC(*p);
2032 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2035 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2037 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2038 result = toUPPER_LC(c);
2041 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2045 else { /* utf8, ord above 255 */
2046 result = CALL_TITLE_CASE(p, ustrp, lenp);
2049 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2054 /* Here, used locale rules. Convert back to utf8 */
2055 if (UTF8_IS_INVARIANT(result)) {
2056 *ustrp = (U8) result;
2060 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2061 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2069 =for apidoc to_utf8_lower
2071 Instead use L</toLOWER_utf8>.
2075 /* Not currently externally documented, and subject to change:
2076 * <flags> is set iff iff the rules from the current underlying locale are to
2081 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2085 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2088 /* Treat a UTF-8 locale as not being in locale at all */
2089 if (IN_UTF8_CTYPE_LOCALE) {
2093 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2097 if (UTF8_IS_INVARIANT(*p)) {
2099 result = toLOWER_LC(*p);
2102 return to_lower_latin1(*p, ustrp, lenp);
2105 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2107 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2108 result = toLOWER_LC(c);
2111 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2115 else { /* utf8, ord above 255 */
2116 result = CALL_LOWER_CASE(p, ustrp, lenp);
2119 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2125 /* Here, used locale rules. Convert back to utf8 */
2126 if (UTF8_IS_INVARIANT(result)) {
2127 *ustrp = (U8) result;
2131 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2132 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2140 =for apidoc to_utf8_fold
2142 Instead use L</toFOLD_utf8>.
2146 /* Not currently externally documented, and subject to change,
2148 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2149 * locale are to be used.
2150 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2151 * otherwise simple folds
2152 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2157 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2161 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2163 /* These are mutually exclusive */
2164 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2166 assert(p != ustrp); /* Otherwise overwrites */
2168 if (flags & FOLD_FLAGS_LOCALE) {
2169 /* Treat a UTF-8 locale as not being in locale at all */
2170 if (IN_UTF8_CTYPE_LOCALE) {
2171 flags &= ~FOLD_FLAGS_LOCALE;
2174 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2178 if (UTF8_IS_INVARIANT(*p)) {
2179 if (flags & FOLD_FLAGS_LOCALE) {
2180 result = toFOLD_LC(*p);
2183 return _to_fold_latin1(*p, ustrp, lenp,
2184 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2187 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2188 if (flags & FOLD_FLAGS_LOCALE) {
2189 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2190 result = toFOLD_LC(c);
2193 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2195 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2198 else { /* utf8, ord above 255 */
2199 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2201 if (flags & FOLD_FLAGS_LOCALE) {
2203 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2204 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2206 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2207 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2209 /* Special case these two characters, as what normally gets
2210 * returned under locale doesn't work */
2211 if (UTF8SKIP(p) == cap_sharp_s_len
2212 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
2214 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2215 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2216 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2217 "resolved to \"\\x{17F}\\x{17F}\".");
2220 else if (UTF8SKIP(p) == long_s_t_len
2221 && memEQ((char *) p, LONG_S_T, long_s_t_len))
2223 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2224 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2225 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2226 "resolved to \"\\x{FB06}\".");
2227 goto return_ligature_st;
2229 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2231 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2235 /* This is called when changing the case of a utf8-encoded
2236 * character above the ASCII range, and the result should not
2237 * contain an ASCII character. */
2239 UV original; /* To store the first code point of <p> */
2241 /* Look at every character in the result; if any cross the
2242 * boundary, the whole thing is disallowed */
2244 U8* e = ustrp + *lenp;
2247 /* Crossed, have to return the original */
2248 original = valid_utf8_to_uvchr(p, lenp);
2250 /* But in these instances, there is an alternative we can
2251 * return that is valid */
2252 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2253 || original == LATIN_SMALL_LETTER_SHARP_S)
2257 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2258 goto return_ligature_st;
2260 Copy(p, ustrp, *lenp, char);
2266 /* Here, no characters crossed, result is ok as-is */
2271 /* Here, used locale rules. Convert back to utf8 */
2272 if (UTF8_IS_INVARIANT(result)) {
2273 *ustrp = (U8) result;
2277 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2278 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2285 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2286 * folds to a string of two of these characters. By returning this
2287 * instead, then, e.g.,
2288 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2291 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2292 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2294 return LATIN_SMALL_LETTER_LONG_S;
2297 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2298 * have the other one fold to it */
2300 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2301 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2302 return LATIN_SMALL_LIGATURE_ST;
2306 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2307 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2308 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2312 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2314 PERL_ARGS_ASSERT_SWASH_INIT;
2316 /* Returns a copy of a swash initiated by the called function. This is the
2317 * public interface, and returning a copy prevents others from doing
2318 * mischief on the original */
2320 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2324 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2327 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2328 * use the following define */
2330 #define CORE_SWASH_INIT_RETURN(x) \
2331 PL_curpm= old_PL_curpm; \
2334 /* Initialize and return a swash, creating it if necessary. It does this
2335 * by calling utf8_heavy.pl in the general case. The returned value may be
2336 * the swash's inversion list instead if the input parameters allow it.
2337 * Which is returned should be immaterial to callers, as the only
2338 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2339 * and swash_to_invlist() handle both these transparently.
2341 * This interface should only be used by functions that won't destroy or
2342 * adversely change the swash, as doing so affects all other uses of the
2343 * swash in the program; the general public should use 'Perl_swash_init'
2346 * pkg is the name of the package that <name> should be in.
2347 * name is the name of the swash to find. Typically it is a Unicode
2348 * property name, including user-defined ones
2349 * listsv is a string to initialize the swash with. It must be of the form
2350 * documented as the subroutine return value in
2351 * L<perlunicode/User-Defined Character Properties>
2352 * minbits is the number of bits required to represent each data element.
2353 * It is '1' for binary properties.
2354 * none I (khw) do not understand this one, but it is used only in tr///.
2355 * invlist is an inversion list to initialize the swash with (or NULL)
2356 * flags_p if non-NULL is the address of various input and output flag bits
2357 * to the routine, as follows: ('I' means is input to the routine;
2358 * 'O' means output from the routine. Only flags marked O are
2359 * meaningful on return.)
2360 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2361 * came from a user-defined property. (I O)
2362 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2363 * when the swash cannot be located, to simply return NULL. (I)
2364 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2365 * return of an inversion list instead of a swash hash if this routine
2366 * thinks that would result in faster execution of swash_fetch() later
2369 * Thus there are three possible inputs to find the swash: <name>,
2370 * <listsv>, and <invlist>. At least one must be specified. The result
2371 * will be the union of the specified ones, although <listsv>'s various
2372 * actions can intersect, etc. what <name> gives. To avoid going out to
2373 * disk at all, <invlist> should specify completely what the swash should
2374 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2376 * <invlist> is only valid for binary properties */
2378 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2380 SV* retval = &PL_sv_undef;
2381 HV* swash_hv = NULL;
2382 const int invlist_swash_boundary =
2383 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2384 ? 512 /* Based on some benchmarking, but not extensive, see commit
2386 : -1; /* Never return just an inversion list */
2388 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2389 assert(! invlist || minbits == 1);
2391 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2392 that triggered the swash init and the swash init perl logic itself.
2395 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2397 if (listsv != &PL_sv_undef || strNE(name, "")) {
2399 const size_t pkg_len = strlen(pkg);
2400 const size_t name_len = strlen(name);
2401 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2405 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2407 PUSHSTACKi(PERLSI_MAGIC);
2410 /* We might get here via a subroutine signature which uses a utf8
2411 * parameter name, at which point PL_subname will have been set
2412 * but not yet used. */
2413 save_item(PL_subname);
2414 if (PL_parser && PL_parser->error_count)
2415 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2416 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2417 if (!method) { /* demand load utf8 */
2419 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2420 GvSV(PL_errgv) = NULL;
2421 #ifndef NO_TAINT_SUPPORT
2422 /* It is assumed that callers of this routine are not passing in
2423 * any user derived data. */
2424 SAVEBOOL(TAINT_get);
2427 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2430 /* Not ERRSV, as there is no need to vivify a scalar we are
2431 about to discard. */
2432 SV * const errsv = GvSV(PL_errgv);
2433 if (!SvTRUE(errsv)) {
2434 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2435 SvREFCNT_dec(errsv);
2443 mPUSHp(pkg, pkg_len);
2444 mPUSHp(name, name_len);
2449 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2450 GvSV(PL_errgv) = NULL;
2451 /* If we already have a pointer to the method, no need to use
2452 * call_method() to repeat the lookup. */
2454 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2455 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2457 retval = *PL_stack_sp--;
2458 SvREFCNT_inc(retval);
2461 /* Not ERRSV. See above. */
2462 SV * const errsv = GvSV(PL_errgv);
2463 if (!SvTRUE(errsv)) {
2464 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2465 SvREFCNT_dec(errsv);
2470 if (IN_PERL_COMPILETIME) {
2471 CopHINTS_set(PL_curcop, PL_hints);
2473 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2476 /* If caller wants to handle missing properties, let them */
2477 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2478 CORE_SWASH_INIT_RETURN(NULL);
2481 "Can't find Unicode property definition \"%"SVf"\"",
2483 NOT_REACHED; /* NOTREACHED */
2485 } /* End of calling the module to find the swash */
2487 /* If this operation fetched a swash, and we will need it later, get it */
2488 if (retval != &PL_sv_undef
2489 && (minbits == 1 || (flags_p
2491 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2493 swash_hv = MUTABLE_HV(SvRV(retval));
2495 /* If we don't already know that there is a user-defined component to
2496 * this swash, and the user has indicated they wish to know if there is
2497 * one (by passing <flags_p>), find out */
2498 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2499 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2500 if (user_defined && SvUV(*user_defined)) {
2501 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2506 /* Make sure there is an inversion list for binary properties */
2508 SV** swash_invlistsvp = NULL;
2509 SV* swash_invlist = NULL;
2510 bool invlist_in_swash_is_valid = FALSE;
2511 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2512 an unclaimed reference count */
2514 /* If this operation fetched a swash, get its already existing
2515 * inversion list, or create one for it */
2518 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2519 if (swash_invlistsvp) {
2520 swash_invlist = *swash_invlistsvp;
2521 invlist_in_swash_is_valid = TRUE;
2524 swash_invlist = _swash_to_invlist(retval);
2525 swash_invlist_unclaimed = TRUE;
2529 /* If an inversion list was passed in, have to include it */
2532 /* Any fetched swash will by now have an inversion list in it;
2533 * otherwise <swash_invlist> will be NULL, indicating that we
2534 * didn't fetch a swash */
2535 if (swash_invlist) {
2537 /* Add the passed-in inversion list, which invalidates the one
2538 * already stored in the swash */
2539 invlist_in_swash_is_valid = FALSE;
2540 _invlist_union(invlist, swash_invlist, &swash_invlist);
2544 /* Here, there is no swash already. Set up a minimal one, if
2545 * we are going to return a swash */
2546 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2548 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2550 swash_invlist = invlist;
2554 /* Here, we have computed the union of all the passed-in data. It may
2555 * be that there was an inversion list in the swash which didn't get
2556 * touched; otherwise save the computed one */
2557 if (! invlist_in_swash_is_valid
2558 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2560 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2562 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2564 /* We just stole a reference count. */
2565 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2566 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2569 SvREADONLY_on(swash_invlist);
2571 /* Use the inversion list stand-alone if small enough */
2572 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2573 SvREFCNT_dec(retval);
2574 if (!swash_invlist_unclaimed)
2575 SvREFCNT_inc_simple_void_NN(swash_invlist);
2576 retval = newRV_noinc(swash_invlist);
2580 CORE_SWASH_INIT_RETURN(retval);
2581 #undef CORE_SWASH_INIT_RETURN
2585 /* This API is wrong for special case conversions since we may need to
2586 * return several Unicode characters for a single Unicode character
2587 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2588 * the lower-level routine, and it is similarly broken for returning
2589 * multiple values. --jhi
2590 * For those, you should use to_utf8_case() instead */
2591 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2594 * Returns the value of property/mapping C<swash> for the first character
2595 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2596 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2597 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2599 * A "swash" is a hash which contains initially the keys/values set up by
2600 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2601 * property for all possible code points. Things are stored in a compact form
2602 * (see utf8_heavy.pl) so that calculation is required to find the actual
2603 * property value for a given code point. As code points are looked up, new
2604 * key/value pairs are added to the hash, so that the calculation doesn't have
2605 * to ever be re-done. Further, each calculation is done, not just for the
2606 * desired one, but for a whole block of code points adjacent to that one.
2607 * For binary properties on ASCII machines, the block is usually for 64 code
2608 * points, starting with a code point evenly divisible by 64. Thus if the
2609 * property value for code point 257 is requested, the code goes out and
2610 * calculates the property values for all 64 code points between 256 and 319,
2611 * and stores these as a single 64-bit long bit vector, called a "swatch",
2612 * under the key for code point 256. The key is the UTF-8 encoding for code
2613 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2614 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2615 * for code point 258 is then requested, this code realizes that it would be
2616 * stored under the key for 256, and would find that value and extract the
2617 * relevant bit, offset from 256.
2619 * Non-binary properties are stored in as many bits as necessary to represent
2620 * their values (32 currently, though the code is more general than that), not
2621 * as single bits, but the principal is the same: the value for each key is a
2622 * vector that encompasses the property values for all code points whose UTF-8
2623 * representations are represented by the key. That is, for all code points
2624 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2628 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2630 HV *const hv = MUTABLE_HV(SvRV(swash));
2635 const U8 *tmps = NULL;
2639 PERL_ARGS_ASSERT_SWASH_FETCH;
2641 /* If it really isn't a hash, it isn't really swash; must be an inversion
2643 if (SvTYPE(hv) != SVt_PVHV) {
2644 return _invlist_contains_cp((SV*)hv,
2646 ? valid_utf8_to_uvchr(ptr, NULL)
2650 /* We store the values in a "swatch" which is a vec() value in a swash
2651 * hash. Code points 0-255 are a single vec() stored with key length
2652 * (klen) 0. All other code points have a UTF-8 representation
2653 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2654 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2655 * length for them is the length of the encoded char - 1. ptr[klen] is the
2656 * final byte in the sequence representing the character */
2657 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2662 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2665 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
2668 klen = UTF8SKIP(ptr) - 1;
2670 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2671 * the vec is the final byte in the sequence. (In EBCDIC this is
2672 * converted to I8 to get consecutive values.) To help you visualize
2674 * Straight 1047 After final byte
2675 * UTF-8 UTF-EBCDIC I8 transform
2676 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2677 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2679 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2680 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2682 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2683 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2685 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2686 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2688 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2689 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2691 * (There are no discontinuities in the elided (...) entries.)
2692 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2693 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2694 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2695 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2696 * index into the vec() swatch (after subtracting 0x80, which we
2697 * actually do with an '&').
2698 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2699 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2700 * dicontinuities which go away by transforming it into I8, and we
2701 * effectively subtract 0xA0 to get the index. */
2702 needents = (1 << UTF_ACCUMULATION_SHIFT);
2703 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2707 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2708 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2709 * it's nothing to sniff at.) Pity we usually come through at least
2710 * two function calls to get here...
2712 * NB: this code assumes that swatches are never modified, once generated!
2715 if (hv == PL_last_swash_hv &&
2716 klen == PL_last_swash_klen &&
2717 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2719 tmps = PL_last_swash_tmps;
2720 slen = PL_last_swash_slen;
2723 /* Try our second-level swatch cache, kept in a hash. */
2724 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2726 /* If not cached, generate it via swatch_get */
2727 if (!svp || !SvPOK(*svp)
2728 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2731 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2732 swatch = swatch_get(swash,
2733 code_point & ~((UV)needents - 1),
2736 else { /* For the first 256 code points, the swatch has a key of
2738 swatch = swatch_get(swash, 0, needents);
2741 if (IN_PERL_COMPILETIME)
2742 CopHINTS_set(PL_curcop, PL_hints);
2744 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2746 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2747 || (slen << 3) < needents)
2748 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2749 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2750 svp, tmps, (UV)slen, (UV)needents);
2753 PL_last_swash_hv = hv;
2754 assert(klen <= sizeof(PL_last_swash_key));
2755 PL_last_swash_klen = (U8)klen;
2756 /* FIXME change interpvar.h? */
2757 PL_last_swash_tmps = (U8 *) tmps;
2758 PL_last_swash_slen = slen;
2760 Copy(ptr, PL_last_swash_key, klen, U8);
2763 switch ((int)((slen << 3) / needents)) {
2765 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2767 return ((UV) tmps[off]);
2771 ((UV) tmps[off ] << 8) +
2772 ((UV) tmps[off + 1]);
2776 ((UV) tmps[off ] << 24) +
2777 ((UV) tmps[off + 1] << 16) +
2778 ((UV) tmps[off + 2] << 8) +
2779 ((UV) tmps[off + 3]);
2781 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2782 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2783 NORETURN_FUNCTION_END;
2786 /* Read a single line of the main body of the swash input text. These are of
2789 * where each number is hex. The first two numbers form the minimum and
2790 * maximum of a range, and the third is the value associated with the range.
2791 * Not all swashes should have a third number
2793 * On input: l points to the beginning of the line to be examined; it points
2794 * to somewhere in the string of the whole input text, and is
2795 * terminated by a \n or the null string terminator.
2796 * lend points to the null terminator of that string
2797 * wants_value is non-zero if the swash expects a third number
2798 * typestr is the name of the swash's mapping, like 'ToLower'
2799 * On output: *min, *max, and *val are set to the values read from the line.
2800 * returns a pointer just beyond the line examined. If there was no
2801 * valid min number on the line, returns lend+1
2805 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2806 const bool wants_value, const U8* const typestr)
2808 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2809 STRLEN numlen; /* Length of the number */
2810 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2811 | PERL_SCAN_DISALLOW_PREFIX
2812 | PERL_SCAN_SILENT_NON_PORTABLE;
2814 /* nl points to the next \n in the scan */
2815 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2817 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2819 /* Get the first number on the line: the range minimum */
2821 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2822 *max = *min; /* So can never return without setting max */
2823 if (numlen) /* If found a hex number, position past it */
2825 else if (nl) { /* Else, go handle next line, if any */
2826 return nl + 1; /* 1 is length of "\n" */
2828 else { /* Else, no next line */
2829 return lend + 1; /* to LIST's end at which \n is not found */
2832 /* The max range value follows, separated by a BLANK */
2835 flags = PERL_SCAN_SILENT_ILLDIGIT
2836 | PERL_SCAN_DISALLOW_PREFIX
2837 | PERL_SCAN_SILENT_NON_PORTABLE;
2839 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2842 else /* If no value here, it is a single element range */
2845 /* Non-binary tables have a third entry: what the first element of the
2846 * range maps to. The map for those currently read here is in hex */
2850 flags = PERL_SCAN_SILENT_ILLDIGIT
2851 | PERL_SCAN_DISALLOW_PREFIX
2852 | PERL_SCAN_SILENT_NON_PORTABLE;
2854 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2863 /* diag_listed_as: To%s: illegal mapping '%s' */
2864 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2870 *val = 0; /* bits == 1, then any val should be ignored */
2872 else { /* Nothing following range min, should be single element with no
2877 /* diag_listed_as: To%s: illegal mapping '%s' */
2878 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2882 *val = 0; /* bits == 1, then val should be ignored */
2885 /* Position to next line if any, or EOF */
2895 * Returns a swatch (a bit vector string) for a code point sequence
2896 * that starts from the value C<start> and comprises the number C<span>.
2897 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2898 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2901 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
2904 U8 *l, *lend, *x, *xend, *s, *send;
2905 STRLEN lcur, xcur, scur;
2906 HV *const hv = MUTABLE_HV(SvRV(swash));
2907 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
2909 SV** listsvp = NULL; /* The string containing the main body of the table */
2910 SV** extssvp = NULL;
2911 SV** invert_it_svp = NULL;
2914 STRLEN octets; /* if bits == 1, then octets == 0 */
2916 UV end = start + span;
2918 if (invlistsvp == NULL) {
2919 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2920 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2921 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2922 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2923 listsvp = hv_fetchs(hv, "LIST", FALSE);
2924 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2926 bits = SvUV(*bitssvp);
2927 none = SvUV(*nonesvp);
2928 typestr = (U8*)SvPV_nolen(*typesvp);
2934 octets = bits >> 3; /* if bits == 1, then octets == 0 */
2936 PERL_ARGS_ASSERT_SWATCH_GET;
2938 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
2939 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
2943 /* If overflowed, use the max possible */
2949 /* create and initialize $swatch */
2950 scur = octets ? (span * octets) : (span + 7) / 8;
2951 swatch = newSV(scur);
2953 s = (U8*)SvPVX(swatch);
2954 if (octets && none) {
2955 const U8* const e = s + scur;
2958 *s++ = (U8)(none & 0xff);
2959 else if (bits == 16) {
2960 *s++ = (U8)((none >> 8) & 0xff);
2961 *s++ = (U8)( none & 0xff);
2963 else if (bits == 32) {
2964 *s++ = (U8)((none >> 24) & 0xff);
2965 *s++ = (U8)((none >> 16) & 0xff);
2966 *s++ = (U8)((none >> 8) & 0xff);
2967 *s++ = (U8)( none & 0xff);
2973 (void)memzero((U8*)s, scur + 1);
2975 SvCUR_set(swatch, scur);
2976 s = (U8*)SvPVX(swatch);
2978 if (invlistsvp) { /* If has an inversion list set up use that */
2979 _invlist_populate_swatch(*invlistsvp, start, end, s);
2983 /* read $swash->{LIST} */
2984 l = (U8*)SvPV(*listsvp, lcur);
2987 UV min, max, val, upper;
2988 l = swash_scan_list_line(l, lend, &min, &max, &val,
2989 cBOOL(octets), typestr);
2994 /* If looking for something beyond this range, go try the next one */
2998 /* <end> is generally 1 beyond where we want to set things, but at the
2999 * platform's infinity, where we can't go any higher, we want to
3000 * include the code point at <end> */
3003 : (max != UV_MAX || end != UV_MAX)
3010 if (!none || val < none) {
3015 for (key = min; key <= upper; key++) {
3017 /* offset must be non-negative (start <= min <= key < end) */
3018 offset = octets * (key - start);
3020 s[offset] = (U8)(val & 0xff);
3021 else if (bits == 16) {
3022 s[offset ] = (U8)((val >> 8) & 0xff);
3023 s[offset + 1] = (U8)( val & 0xff);
3025 else if (bits == 32) {
3026 s[offset ] = (U8)((val >> 24) & 0xff);
3027 s[offset + 1] = (U8)((val >> 16) & 0xff);
3028 s[offset + 2] = (U8)((val >> 8) & 0xff);
3029 s[offset + 3] = (U8)( val & 0xff);
3032 if (!none || val < none)
3036 else { /* bits == 1, then val should be ignored */
3041 for (key = min; key <= upper; key++) {
3042 const STRLEN offset = (STRLEN)(key - start);
3043 s[offset >> 3] |= 1 << (offset & 7);
3048 /* Invert if the data says it should be. Assumes that bits == 1 */
3049 if (invert_it_svp && SvUV(*invert_it_svp)) {
3051 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3052 * be 0, and their inversion should also be 0, as we don't succeed any
3053 * Unicode property matches for non-Unicode code points */
3054 if (start <= PERL_UNICODE_MAX) {
3056 /* The code below assumes that we never cross the
3057 * Unicode/above-Unicode boundary in a range, as otherwise we would
3058 * have to figure out where to stop flipping the bits. Since this
3059 * boundary is divisible by a large power of 2, and swatches comes
3060 * in small powers of 2, this should be a valid assumption */
3061 assert(start + span - 1 <= PERL_UNICODE_MAX);
3071 /* read $swash->{EXTRAS}
3072 * This code also copied to swash_to_invlist() below */
3073 x = (U8*)SvPV(*extssvp, xcur);
3081 SV **otherbitssvp, *other;
3085 const U8 opc = *x++;
3089 nl = (U8*)memchr(x, '\n', xend - x);
3091 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3093 x = nl + 1; /* 1 is length of "\n" */
3097 x = xend; /* to EXTRAS' end at which \n is not found */
3104 namelen = nl - namestr;
3108 namelen = xend - namestr;
3112 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3113 otherhv = MUTABLE_HV(SvRV(*othersvp));
3114 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3115 otherbits = (STRLEN)SvUV(*otherbitssvp);
3116 if (bits < otherbits)
3117 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3118 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3120 /* The "other" swatch must be destroyed after. */
3121 other = swatch_get(*othersvp, start, span);
3122 o = (U8*)SvPV(other, olen);
3125 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3127 s = (U8*)SvPV(swatch, slen);
3128 if (bits == 1 && otherbits == 1) {
3130 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3131 "mismatch, slen=%"UVuf", olen=%"UVuf,
3132 (UV)slen, (UV)olen);
3156 STRLEN otheroctets = otherbits >> 3;
3158 U8* const send = s + slen;
3163 if (otherbits == 1) {
3164 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3168 STRLEN vlen = otheroctets;
3176 if (opc == '+' && otherval)
3177 NOOP; /* replace with otherval */
3178 else if (opc == '!' && !otherval)
3180 else if (opc == '-' && otherval)
3182 else if (opc == '&' && !otherval)
3185 s += octets; /* no replacement */
3190 *s++ = (U8)( otherval & 0xff);
3191 else if (bits == 16) {
3192 *s++ = (U8)((otherval >> 8) & 0xff);
3193 *s++ = (U8)( otherval & 0xff);
3195 else if (bits == 32) {
3196 *s++ = (U8)((otherval >> 24) & 0xff);
3197 *s++ = (U8)((otherval >> 16) & 0xff);
3198 *s++ = (U8)((otherval >> 8) & 0xff);
3199 *s++ = (U8)( otherval & 0xff);
3203 sv_free(other); /* through with it! */
3209 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3212 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3213 * Can't be used on a property that is subject to user override, as it
3214 * relies on the value of SPECIALS in the swash which would be set by
3215 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3216 * for overridden properties
3218 * Returns a hash which is the inversion and closure of a swash mapping.
3219 * For example, consider the input lines:
3224 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3225 * 006C. The value for each key is an array. For 006C, the array would
3226 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3227 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3229 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3230 * keys are only code points that are folded-to, so it isn't a full closure.
3232 * Essentially, for any code point, it gives all the code points that map to
3233 * it, or the list of 'froms' for that point.
3235 * Currently it ignores any additions or deletions from other swashes,
3236 * looking at just the main body of the swash, and if there are SPECIALS
3237 * in the swash, at that hash
3239 * The specials hash can be extra code points, and most likely consists of
3240 * maps from single code points to multiple ones (each expressed as a string
3241 * of utf8 characters). This function currently returns only 1-1 mappings.
3242 * However consider this possible input in the specials hash:
3243 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3244 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3246 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3247 * currently handle. But it also means that FB05 and FB06 are equivalent in
3248 * a 1-1 mapping which we should handle, and this relationship may not be in
3249 * the main table. Therefore this function examines all the multi-char
3250 * sequences and adds the 1-1 mappings that come out of that. */
3254 HV *const hv = MUTABLE_HV(SvRV(swash));
3256 /* The string containing the main body of the table. This will have its
3257 * assertion fail if the swash has been converted to its inversion list */
3258 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3260 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3261 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3262 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3263 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3264 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3265 const STRLEN bits = SvUV(*bitssvp);
3266 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3267 const UV none = SvUV(*nonesvp);
3268 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3272 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3274 /* Must have at least 8 bits to get the mappings */
3275 if (bits != 8 && bits != 16 && bits != 32) {
3276 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3280 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3281 mapping to more than one character */
3283 /* Construct an inverse mapping hash for the specials */
3284 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3285 HV * specials_inverse = newHV();
3286 char *char_from; /* the lhs of the map */
3287 I32 from_len; /* its byte length */
3288 char *char_to; /* the rhs of the map */
3289 I32 to_len; /* its byte length */
3290 SV *sv_to; /* and in a sv */
3291 AV* from_list; /* list of things that map to each 'to' */
3293 hv_iterinit(specials_hv);
3295 /* The keys are the characters (in utf8) that map to the corresponding
3296 * utf8 string value. Iterate through the list creating the inverse
3298 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3300 if (! SvPOK(sv_to)) {
3301 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3302 "unexpectedly is not a string, flags=%lu",
3303 (unsigned long)SvFLAGS(sv_to));
3305 /*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)));*/
3307 /* Each key in the inverse list is a mapped-to value, and the key's
3308 * hash value is a list of the strings (each in utf8) that map to
3309 * it. Those strings are all one character long */
3310 if ((listp = hv_fetch(specials_inverse,
3314 from_list = (AV*) *listp;
3316 else { /* No entry yet for it: create one */
3317 from_list = newAV();
3318 if (! hv_store(specials_inverse,
3321 (SV*) from_list, 0))
3323 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3327 /* Here have the list associated with this 'to' (perhaps newly
3328 * created and empty). Just add to it. Note that we ASSUME that
3329 * the input is guaranteed to not have duplications, so we don't
3330 * check for that. Duplications just slow down execution time. */
3331 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3334 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3335 * it looking for cases like the FB05/FB06 examples above. There would
3336 * be an entry in the hash like
3337 * 'st' => [ FB05, FB06 ]
3338 * In this example we will create two lists that get stored in the
3339 * returned hash, 'ret':
3340 * FB05 => [ FB05, FB06 ]
3341 * FB06 => [ FB05, FB06 ]
3343 * Note that there is nothing to do if the array only has one element.
3344 * (In the normal 1-1 case handled below, we don't have to worry about
3345 * two lists, as everything gets tied to the single list that is
3346 * generated for the single character 'to'. But here, we are omitting
3347 * that list, ('st' in the example), so must have multiple lists.) */
3348 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3349 &char_to, &to_len)))
3351 if (av_tindex(from_list) > 0) {
3354 /* We iterate over all combinations of i,j to place each code
3355 * point on each list */
3356 for (i = 0; i <= av_tindex(from_list); i++) {
3358 AV* i_list = newAV();
3359 SV** entryp = av_fetch(from_list, i, FALSE);
3360 if (entryp == NULL) {
3361 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3363 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3364 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3366 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3367 (SV*) i_list, FALSE))
3369 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3372 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3373 for (j = 0; j <= av_tindex(from_list); j++) {
3374 entryp = av_fetch(from_list, j, FALSE);
3375 if (entryp == NULL) {
3376 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3379 /* When i==j this adds itself to the list */
3380 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3381 (U8*) SvPVX(*entryp),
3382 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3384 /*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));*/
3389 SvREFCNT_dec(specials_inverse); /* done with it */
3390 } /* End of specials */
3392 /* read $swash->{LIST} */
3393 l = (U8*)SvPV(*listsvp, lcur);
3396 /* Go through each input line */
3400 l = swash_scan_list_line(l, lend, &min, &max, &val,
3401 cBOOL(octets), typestr);
3406 /* Each element in the range is to be inverted */
3407 for (inverse = min; inverse <= max; inverse++) {
3411 bool found_key = FALSE;
3412 bool found_inverse = FALSE;
3414 /* The key is the inverse mapping */
3415 char key[UTF8_MAXBYTES+1];
3416 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3417 STRLEN key_len = key_end - key;
3419 /* Get the list for the map */
3420 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3421 list = (AV*) *listp;
3423 else { /* No entry yet for it: create one */
3425 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3426 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3430 /* Look through list to see if this inverse mapping already is
3431 * listed, or if there is a mapping to itself already */
3432 for (i = 0; i <= av_tindex(list); i++) {
3433 SV** entryp = av_fetch(list, i, FALSE);
3436 if (entryp == NULL) {
3437 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3441 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3445 if (uv == inverse) {
3446 found_inverse = TRUE;
3449 /* No need to continue searching if found everything we are
3451 if (found_key && found_inverse) {
3456 /* Make sure there is a mapping to itself on the list */
3458 av_push(list, newSVuv(val));
3459 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3463 /* Simply add the value to the list */
3464 if (! found_inverse) {
3465 av_push(list, newSVuv(inverse));
3466 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3469 /* swatch_get() increments the value of val for each element in the
3470 * range. That makes more compact tables possible. You can
3471 * express the capitalization, for example, of all consecutive
3472 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3473 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3474 * and it's not documented; it appears to be used only in
3475 * implementing tr//; I copied the semantics from swatch_get(), just
3477 if (!none || val < none) {
3487 Perl__swash_to_invlist(pTHX_ SV* const swash)
3490 /* Subject to change or removal. For use only in one place in regcomp.c.
3491 * Ownership is given to one reference count in the returned SV* */
3496 HV *const hv = MUTABLE_HV(SvRV(swash));
3497 UV elements = 0; /* Number of elements in the inversion list */
3507 STRLEN octets; /* if bits == 1, then octets == 0 */
3513 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3515 /* If not a hash, it must be the swash's inversion list instead */
3516 if (SvTYPE(hv) != SVt_PVHV) {
3517 return SvREFCNT_inc_simple_NN((SV*) hv);
3520 /* The string containing the main body of the table */
3521 listsvp = hv_fetchs(hv, "LIST", FALSE);
3522 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3523 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3524 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3525 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3527 typestr = (U8*)SvPV_nolen(*typesvp);
3528 bits = SvUV(*bitssvp);
3529 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3531 /* read $swash->{LIST} */
3532 if (SvPOK(*listsvp)) {
3533 l = (U8*)SvPV(*listsvp, lcur);
3536 /* LIST legitimately doesn't contain a string during compilation phases
3537 * of Perl itself, before the Unicode tables are generated. In this
3538 * case, just fake things up by creating an empty list */
3545 if (*l == 'V') { /* Inversion list format */
3546 const char *after_atou = (char *) lend;
3548 UV* other_elements_ptr;
3550 /* The first number is a count of the rest */
3552 elements = grok_atou((const char *)l, &after_atou);
3553 if (elements == 0) {
3554 invlist = _new_invlist(0);
3557 while (isSPACE(*l)) l++;
3558 l = (U8 *) after_atou;
3560 /* Get the 0th element, which is needed to setup the inversion list */
3561 while (isSPACE(*l)) l++;
3562 element0 = (UV) grok_atou((const char *)l, &after_atou);
3563 l = (U8 *) after_atou;
3564 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3567 /* Then just populate the rest of the input */
3568 while (elements-- > 0) {
3570 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3572 while (isSPACE(*l)) l++;
3573 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3574 l = (U8 *) after_atou;
3580 /* Scan the input to count the number of lines to preallocate array
3581 * size based on worst possible case, which is each line in the input
3582 * creates 2 elements in the inversion list: 1) the beginning of a
3583 * range in the list; 2) the beginning of a range not in the list. */
3584 while ((loc = (strchr(loc, '\n'))) != NULL) {
3589 /* If the ending is somehow corrupt and isn't a new line, add another
3590 * element for the final range that isn't in the inversion list */
3591 if (! (*lend == '\n'
3592 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3597 invlist = _new_invlist(elements);
3599 /* Now go through the input again, adding each range to the list */
3602 UV val; /* Not used by this function */
3604 l = swash_scan_list_line(l, lend, &start, &end, &val,
3605 cBOOL(octets), typestr);
3611 invlist = _add_range_to_invlist(invlist, start, end);
3615 /* Invert if the data says it should be */
3616 if (invert_it_svp && SvUV(*invert_it_svp)) {
3617 _invlist_invert(invlist);
3620 /* This code is copied from swatch_get()
3621 * read $swash->{EXTRAS} */
3622 x = (U8*)SvPV(*extssvp, xcur);
3630 SV **otherbitssvp, *other;
3633 const U8 opc = *x++;
3637 nl = (U8*)memchr(x, '\n', xend - x);
3639 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3641 x = nl + 1; /* 1 is length of "\n" */
3645 x = xend; /* to EXTRAS' end at which \n is not found */
3652 namelen = nl - namestr;
3656 namelen = xend - namestr;
3660 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3661 otherhv = MUTABLE_HV(SvRV(*othersvp));
3662 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3663 otherbits = (STRLEN)SvUV(*otherbitssvp);
3665 if (bits != otherbits || bits != 1) {
3666 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3667 "properties, bits=%"UVuf", otherbits=%"UVuf,
3668 (UV)bits, (UV)otherbits);
3671 /* The "other" swatch must be destroyed after. */
3672 other = _swash_to_invlist((SV *)*othersvp);
3674 /* End of code copied from swatch_get() */
3677 _invlist_union(invlist, other, &invlist);
3680 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3683 _invlist_subtract(invlist, other, &invlist);
3686 _invlist_intersection(invlist, other, &invlist);
3691 sv_free(other); /* through with it! */
3694 SvREADONLY_on(invlist);
3699 Perl__get_swash_invlist(pTHX_ SV* const swash)
3703 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3705 if (! SvROK(swash)) {
3709 /* If it really isn't a hash, it isn't really swash; must be an inversion
3711 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3715 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3724 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3726 /* May change: warns if surrogates, non-character code points, or
3727 * non-Unicode code points are in s which has length len bytes. Returns
3728 * TRUE if none found; FALSE otherwise. The only other validity check is
3729 * to make sure that this won't exceed the string's length */
3731 const U8* const e = s + len;
3734 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3737 if (UTF8SKIP(s) > len) {
3738 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3739 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3742 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3744 if (UTF8_IS_SUPER(s)) {
3745 if (ckWARN_d(WARN_NON_UNICODE)) {
3746 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3747 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3748 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3752 else if (UTF8_IS_SURROGATE(s)) {
3753 if (ckWARN_d(WARN_SURROGATE)) {
3754 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3755 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3756 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3761 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3762 && (ckWARN_d(WARN_NONCHAR)))
3764 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3765 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3766 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3777 =for apidoc pv_uni_display
3779 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3780 length C<len>, the displayable version being at most C<pvlim> bytes long
3781 (if longer, the rest is truncated and "..." will be appended).
3783 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
3784 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3785 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3786 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3787 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3788 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3790 The pointer to the PV of the C<dsv> is returned.
3794 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3799 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3803 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3805 /* This serves double duty as a flag and a character to print after
3806 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3810 if (pvlim && SvCUR(dsv) >= pvlim) {
3814 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3816 const unsigned char c = (unsigned char)u & 0xFF;
3817 if (flags & UNI_DISPLAY_BACKSLASH) {
3834 const char string = ok;
3835 sv_catpvs(dsv, "\\");
3836 sv_catpvn(dsv, &string, 1);
3839 /* isPRINT() is the locale-blind version. */
3840 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3841 const char string = c;
3842 sv_catpvn(dsv, &string, 1);
3847 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
3850 sv_catpvs(dsv, "...");
3856 =for apidoc sv_uni_display
3858 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
3859 the displayable version being at most C<pvlim> bytes long
3860 (if longer, the rest is truncated and "..." will be appended).
3862 The C<flags> argument is as in L</pv_uni_display>().
3864 The pointer to the PV of the C<dsv> is returned.
3869 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
3871 const char * const ptr =
3872 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
3874 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
3876 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
3877 SvCUR(ssv), pvlim, flags);
3881 =for apidoc foldEQ_utf8
3883 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
3884 of which may be in UTF-8) are the same case-insensitively; false otherwise.
3885 How far into the strings to compare is determined by other input parameters.
3887 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
3888 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
3889 with respect to C<s2>.
3891 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
3892 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
3893 scan will not be considered to be a match unless the goal is reached, and
3894 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
3897 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
3898 considered an end pointer to the position 1 byte past the maximum point
3899 in C<s1> beyond which scanning will not continue under any circumstances.
3900 (This routine assumes that UTF-8 encoded input strings are not malformed;
3901 malformed input can cause it to read past C<pe1>).
3902 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
3903 is less than C<s1>+C<l1>, the match will never be successful because it can
3905 get as far as its goal (and in fact is asserted against). Correspondingly for
3906 C<pe2> with respect to C<s2>.
3908 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
3909 C<l2> must be non-zero), and if both do, both have to be
3910 reached for a successful match. Also, if the fold of a character is multiple
3911 characters, all of them must be matched (see tr21 reference below for
3914 Upon a successful match, if C<pe1> is non-NULL,
3915 it will be set to point to the beginning of the I<next> character of C<s1>
3916 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
3918 For case-insensitiveness, the "casefolding" of Unicode is used
3919 instead of upper/lowercasing both the characters, see
3920 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
3924 /* A flags parameter has been added which may change, and hence isn't
3925 * externally documented. Currently it is:
3926 * 0 for as-documented above
3927 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
3928 ASCII one, to not match
3929 * FOLDEQ_LOCALE is set iff the rules from the current underlying
3930 * locale are to be used.
3931 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
3932 * routine. This allows that step to be skipped.
3933 * Currently, this requires s1 to be encoded as UTF-8
3934 * (u1 must be true), which is asserted for.
3935 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
3936 * cross certain boundaries. Hence, the caller should
3937 * let this function do the folding instead of
3938 * pre-folding. This code contains an assertion to
3939 * that effect. However, if the caller knows what
3940 * it's doing, it can pass this flag to indicate that,
3941 * and the assertion is skipped.
3942 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
3943 * FOLDEQ_S2_FOLDS_SANE
3946 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)
3948 const U8 *p1 = (const U8*)s1; /* Point to current char */
3949 const U8 *p2 = (const U8*)s2;
3950 const U8 *g1 = NULL; /* goal for s1 */
3951 const U8 *g2 = NULL;
3952 const U8 *e1 = NULL; /* Don't scan s1 past this */
3953 U8 *f1 = NULL; /* Point to current folded */
3954 const U8 *e2 = NULL;
3956 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
3957 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
3958 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
3959 U8 flags_for_folder = FOLD_FLAGS_FULL;
3961 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
3963 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
3964 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
3965 && !(flags & FOLDEQ_S1_FOLDS_SANE))
3966 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
3967 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
3968 /* The algorithm is to trial the folds without regard to the flags on
3969 * the first line of the above assert(), and then see if the result
3970 * violates them. This means that the inputs can't be pre-folded to a
3971 * violating result, hence the assert. This could be changed, with the
3972 * addition of extra tests here for the already-folded case, which would
3973 * slow it down. That cost is more than any possible gain for when these
3974 * flags are specified, as the flags indicate /il or /iaa matching which
3975 * is less common than /iu, and I (khw) also believe that real-world /il
3976 * and /iaa matches are most likely to involve code points 0-255, and this
3977 * function only under rare conditions gets called for 0-255. */
3979 if (flags & FOLDEQ_LOCALE) {
3980 if (IN_UTF8_CTYPE_LOCALE) {
3981 flags &= ~FOLDEQ_LOCALE;
3984 flags_for_folder |= FOLD_FLAGS_LOCALE;
3993 g1 = (const U8*)s1 + l1;
4001 g2 = (const U8*)s2 + l2;
4004 /* Must have at least one goal */
4009 /* Will never match if goal is out-of-bounds */
4010 assert(! e1 || e1 >= g1);
4012 /* Here, there isn't an end pointer, or it is beyond the goal. We
4013 * only go as far as the goal */
4017 assert(e1); /* Must have an end for looking at s1 */
4020 /* Same for goal for s2 */
4022 assert(! e2 || e2 >= g2);
4029 /* If both operands are already folded, we could just do a memEQ on the
4030 * whole strings at once, but it would be better if the caller realized
4031 * this and didn't even call us */
4033 /* Look through both strings, a character at a time */
4034 while (p1 < e1 && p2 < e2) {
4036 /* If at the beginning of a new character in s1, get its fold to use
4037 * and the length of the fold. */
4039 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4045 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4047 /* We have to forbid mixing ASCII with non-ASCII if the
4048 * flags so indicate. And, we can short circuit having to
4049 * call the general functions for this common ASCII case,
4050 * all of whose non-locale folds are also ASCII, and hence
4051 * UTF-8 invariants, so the UTF8ness of the strings is not
4053 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4057 *foldbuf1 = toFOLD(*p1);
4060 _to_utf8_fold_flags(p1, foldbuf1, &n1, flags_for_folder);
4062 else { /* Not utf8, get utf8 fold */
4063 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4069 if (n2 == 0) { /* Same for s2 */
4070 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4076 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4077 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4081 *foldbuf2 = toFOLD(*p2);
4084 _to_utf8_fold_flags(p2, foldbuf2, &n2, flags_for_folder);
4087 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4093 /* Here f1 and f2 point to the beginning of the strings to compare.
4094 * These strings are the folds of the next character from each input
4095 * string, stored in utf8. */
4097 /* While there is more to look for in both folds, see if they
4098 * continue to match */
4100 U8 fold_length = UTF8SKIP(f1);
4101 if (fold_length != UTF8SKIP(f2)
4102 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4103 function call for single
4105 || memNE((char*)f1, (char*)f2, fold_length))
4107 return 0; /* mismatch */
4110 /* Here, they matched, advance past them */
4117 /* When reach the end of any fold, advance the input past it */
4119 p1 += u1 ? UTF8SKIP(p1) : 1;
4122 p2 += u2 ? UTF8SKIP(p2) : 1;
4124 } /* End of loop through both strings */
4126 /* A match is defined by each scan that specified an explicit length
4127 * reaching its final goal, and the other not having matched a partial
4128 * character (which can happen when the fold of a character is more than one
4130 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4134 /* Successful match. Set output pointers */
4144 /* XXX The next two functions should likely be moved to mathoms.c once all
4145 * occurrences of them are removed from the core; some cpan-upstream modules
4149 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4151 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4153 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4157 =for apidoc utf8n_to_uvuni
4159 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4161 This function was useful for code that wanted to handle both EBCDIC and
4162 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4163 distinctions between the platforms have mostly been made invisible to most
4164 code, so this function is quite unlikely to be what you want. If you do need
4165 this precise functionality, use instead
4166 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4167 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4173 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4175 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4177 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4181 =for apidoc uvuni_to_utf8_flags
4183 Instead you almost certainly want to use L</uvchr_to_utf8> or
4184 L</uvchr_to_utf8_flags>>.
4186 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4187 which itself, while not deprecated, should be used only in isolated
4188 circumstances. These functions were useful for code that wanted to handle
4189 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4190 v5.20, the distinctions between the platforms have mostly been made invisible
4191 to most code, so this function is quite unlikely to be what you want.
4197 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4199 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4201 return uvoffuni_to_utf8_flags(d, uv, flags);
4206 * c-indentation-style: bsd
4208 * indent-tabs-mode: nil
4211 * ex: set ts=8 sts=4 sw=4 et: