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 "invlist_inline.h"
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 (OFFUNI_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 ( flags /* It's common to turn off all these */
119 && uv >= UNICODE_SURROGATE_FIRST
120 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
122 if (UNICODE_IS_SURROGATE(uv)) {
123 if (flags & UNICODE_WARN_SURROGATE) {
124 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
125 "UTF-16 surrogate U+%04"UVXf, uv);
127 if (flags & UNICODE_DISALLOW_SURROGATE) {
131 else if (UNICODE_IS_SUPER(uv)) {
132 if (flags & UNICODE_WARN_SUPER
133 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
135 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
136 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
138 if (flags & UNICODE_DISALLOW_SUPER
139 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
142 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
143 NOT_REACHED; /* NOTREACHED */
148 else if (UNICODE_IS_NONCHAR(uv)) {
149 if (flags & UNICODE_WARN_NONCHAR) {
150 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
151 "Unicode non-character U+%04"UVXf" is not recommended for open interchange",
154 if (flags & UNICODE_DISALLOW_NONCHAR) {
162 STRLEN len = OFFUNISKIP(uv);
165 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
166 uv >>= UTF_ACCUMULATION_SHIFT;
168 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
171 #else /* Non loop style */
173 *d++ = (U8)(( uv >> 6) | 0xc0);
174 *d++ = (U8)(( uv & 0x3f) | 0x80);
178 *d++ = (U8)(( uv >> 12) | 0xe0);
179 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
180 *d++ = (U8)(( uv & 0x3f) | 0x80);
184 *d++ = (U8)(( uv >> 18) | 0xf0);
185 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
186 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
187 *d++ = (U8)(( uv & 0x3f) | 0x80);
190 if (uv < 0x4000000) {
191 *d++ = (U8)(( uv >> 24) | 0xf8);
192 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
193 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
194 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
195 *d++ = (U8)(( uv & 0x3f) | 0x80);
198 if (uv < 0x80000000) {
199 *d++ = (U8)(( uv >> 30) | 0xfc);
200 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
204 *d++ = (U8)(( uv & 0x3f) | 0x80);
208 if (uv < UTF8_QUAD_MAX)
211 *d++ = 0xfe; /* Can't match U+FEFF! */
212 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
217 *d++ = (U8)(( uv & 0x3f) | 0x80);
222 *d++ = 0xff; /* Can't match U+FFFE! */
223 *d++ = 0x80; /* 6 Reserved bits */
224 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
225 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
233 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
234 *d++ = (U8)(( uv & 0x3f) | 0x80);
238 #endif /* Non loop style */
241 =for apidoc uvchr_to_utf8
243 Adds the UTF-8 representation of the native code point C<uv> to the end
244 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
245 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
246 the byte after the end of the new character. In other words,
248 d = uvchr_to_utf8(d, uv);
250 is the recommended wide native character-aware way of saying
254 This function accepts any UV as input. To forbid or warn on non-Unicode code
255 points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
260 /* This is also a macro */
261 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
264 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
266 return uvchr_to_utf8(d, uv);
270 =for apidoc uvchr_to_utf8_flags
272 Adds the UTF-8 representation of the native code point C<uv> to the end
273 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
274 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
275 the byte after the end of the new character. In other words,
277 d = uvchr_to_utf8_flags(d, uv, flags);
281 d = uvchr_to_utf8_flags(d, uv, 0);
283 This is the Unicode-aware way of saying
287 This function will convert to UTF-8 (and not warn) even code points that aren't
288 legal Unicode or are problematic, unless C<flags> contains one or more of the
291 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
292 the function will raise a warning, provided UTF8 warnings are enabled. If instead
293 C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return NULL.
294 If both flags are set, the function will both warn and return NULL.
296 The C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
297 affect how the function handles a Unicode non-character. And likewise, the
298 C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags affect the handling of
300 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
301 even less portable) can be warned and/or disallowed even if other above-Unicode
302 code points are accepted, by the C<UNICODE_WARN_FE_FF> and
303 C<UNICODE_DISALLOW_FE_FF> flags.
305 And finally, the flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all four of
306 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
312 /* This is also a macro */
313 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
316 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
318 return uvchr_to_utf8_flags(d, uv, flags);
322 =for apidoc is_utf8_string
324 Returns true if the first C<len> bytes of string C<s> form a valid
325 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
326 using C<strlen(s)> (which means if you use this option, that C<s> can't have
327 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
328 that all characters being ASCII constitute 'a valid UTF-8 string'.
330 See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
336 Perl_is_utf8_string(const U8 *s, STRLEN len)
338 const U8* const send = s + (len ? len : strlen((const char *)s));
341 PERL_ARGS_ASSERT_IS_UTF8_STRING;
344 STRLEN len = isUTF8_CHAR(x, send);
345 if (UNLIKELY(! len)) {
355 Implemented as a macro in utf8.h
357 =for apidoc is_utf8_string_loc
359 Like L</is_utf8_string> but stores the location of the failure (in the
360 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
361 "utf8ness success") in the C<ep>.
363 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
365 =for apidoc is_utf8_string_loclen
367 Like L</is_utf8_string>() but stores the location of the failure (in the
368 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
369 "utf8ness success") in the C<ep>, and the number of UTF-8
370 encoded characters in the C<el>.
372 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
378 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
380 const U8* const send = s + (len ? len : strlen((const char *)s));
384 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
387 STRLEN len = isUTF8_CHAR(x, send);
388 if (UNLIKELY(! len)) {
406 =for apidoc utf8n_to_uvchr
408 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
409 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
411 Bottom level UTF-8 decode routine.
412 Returns the native code point value of the first character in the string C<s>,
413 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
414 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
415 the length, in bytes, of that character.
417 The value of C<flags> determines the behavior when C<s> does not point to a
418 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
419 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
420 next possible position in C<s> that could begin a non-malformed character.
421 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
423 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
424 individual types of malformations, such as the sequence being overlong (that
425 is, when there is a shorter sequence that can express the same code point;
426 overlong sequences are expressly forbidden in the UTF-8 standard due to
427 potential security issues). Another malformation example is the first byte of
428 a character not being a legal first byte. See F<utf8.h> for the list of such
429 flags. For allowed 0 length strings, this function returns 0; for allowed
430 overlong sequences, the computed code point is returned; for all other allowed
431 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
432 determinable reasonable value.
434 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
435 flags) malformation is found. If this flag is set, the routine assumes that
436 the caller will raise a warning, and this function will silently just set
437 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
439 Note that this API requires disambiguation between successful decoding a C<NUL>
440 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
441 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
442 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
445 Certain code points are considered problematic. These are Unicode surrogates,
446 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
447 By default these are considered regular code points, but certain situations
448 warrant special handling for them. If C<flags> contains
449 C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
450 malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
451 C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
452 Unicode maximum) can be set to disallow these categories individually.
454 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
455 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
456 raised for their respective categories, but otherwise the code points are
457 considered valid (not malformations). To get a category to both be treated as
458 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
459 (But note that warnings are not raised if lexically disabled nor if
460 C<UTF8_CHECK_ONLY> is also specified.)
462 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
463 the others that are above the Unicode legal maximum. There are several
464 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
465 not representable at all on EBCDIC platforms, and the original UTF-8
466 specification never went above this number (the current 0x10FFFF limit was
467 imposed later). (The smaller ones, those that fit into 32 bits, are
468 representable by a UV on ASCII platforms, but not by an IV, which means that
469 the number of operations that can be performed on them is quite restricted.)
470 The UTF-8 encoding on ASCII platforms for these large code points begins with a
471 byte containing 0xFE or 0xFF. The C<UTF8_DISALLOW_FE_FF> flag will cause them to
472 be treated as malformations, while allowing smaller above-Unicode code points.
473 (Of course C<UTF8_DISALLOW_SUPER> will treat all above-Unicode code points,
474 including these, as malformations.)
475 Similarly, C<UTF8_WARN_FE_FF> acts just like
476 the other WARN flags, but applies just to these code points.
478 All other code points corresponding to Unicode characters, including private
479 use and those yet to be assigned, are never considered malformed and never
486 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
488 const U8 * const s0 = s;
489 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
494 UV outlier_ret = 0; /* return value when input is in error or problematic
496 UV pack_warn = 0; /* Save result of packWARN() for later */
497 bool unexpected_non_continuation = FALSE;
498 bool overflowed = FALSE;
499 bool do_overlong_test = TRUE; /* May have to skip this test */
501 const char* const malformed_text = "Malformed UTF-8 character";
503 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
505 /* The order of malformation tests here is important. We should consume as
506 * few bytes as possible in order to not skip any valid character. This is
507 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
508 * http://unicode.org/reports/tr36 for more discussion as to why. For
509 * example, once we've done a UTF8SKIP, we can tell the expected number of
510 * bytes, and could fail right off the bat if the input parameters indicate
511 * that there are too few available. But it could be that just that first
512 * byte is garbled, and the intended character occupies fewer bytes. If we
513 * blindly assumed that the first byte is correct, and skipped based on
514 * that number, we could skip over a valid input character. So instead, we
515 * always examine the sequence byte-by-byte.
517 * We also should not consume too few bytes, otherwise someone could inject
518 * things. For example, an input could be deliberately designed to
519 * overflow, and if this code bailed out immediately upon discovering that,
520 * returning to the caller C<*retlen> pointing to the very next byte (one
521 * which is actually part of of the overflowing sequence), that could look
522 * legitimate to the caller, which could discard the initial partial
523 * sequence and process the rest, inappropriately */
525 /* Zero length strings, if allowed, of necessity are zero */
526 if (UNLIKELY(curlen == 0)) {
531 if (flags & UTF8_ALLOW_EMPTY) {
534 if (! (flags & UTF8_CHECK_ONLY)) {
535 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
540 expectlen = UTF8SKIP(s);
542 /* A well-formed UTF-8 character, as the vast majority of calls to this
543 * function will be for, has this expected length. For efficiency, set
544 * things up here to return it. It will be overriden only in those rare
545 * cases where a malformation is found */
550 /* An invariant is trivially well-formed */
551 if (UTF8_IS_INVARIANT(uv)) {
555 /* A continuation character can't start a valid sequence */
556 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
557 if (flags & UTF8_ALLOW_CONTINUATION) {
561 return UNICODE_REPLACEMENT;
564 if (! (flags & UTF8_CHECK_ONLY)) {
565 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
571 /* Here is not a continuation byte, nor an invariant. The only thing left
572 * is a start byte (possibly for an overlong) */
575 uv = NATIVE_UTF8_TO_I8(uv);
578 /* Remove the leading bits that indicate the number of bytes in the
579 * character's whole UTF-8 sequence, leaving just the bits that are part of
581 uv &= UTF_START_MASK(expectlen);
583 /* Now, loop through the remaining bytes in the character's sequence,
584 * accumulating each into the working value as we go. Be sure to not look
585 * past the end of the input string */
586 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
588 for (s = s0 + 1; s < send; s++) {
589 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
590 #ifndef EBCDIC /* Can't overflow in EBCDIC */
591 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
593 /* The original implementors viewed this malformation as more
594 * serious than the others (though I, khw, don't understand
595 * why, since other malformations also give very very wrong
596 * results), so there is no way to turn off checking for it.
597 * Set a flag, but keep going in the loop, so that we absorb
598 * the rest of the bytes that comprise the character. */
600 overflow_byte = *s; /* Save for warning message's use */
603 uv = UTF8_ACCUMULATE(uv, *s);
606 /* Here, found a non-continuation before processing all expected
607 * bytes. This byte begins a new character, so quit, even if
608 * allowing this malformation. */
609 unexpected_non_continuation = TRUE;
612 } /* End of loop through the character's bytes */
614 /* Save how many bytes were actually in the character */
617 /* The loop above finds two types of malformations: non-continuation and/or
618 * overflow. The non-continuation malformation is really a too-short
619 * malformation, as it means that the current character ended before it was
620 * expected to (being terminated prematurely by the beginning of the next
621 * character, whereas in the too-short malformation there just are too few
622 * bytes available to hold the character. In both cases, the check below
623 * that we have found the expected number of bytes would fail if executed.)
624 * Thus the non-continuation malformation is really unnecessary, being a
625 * subset of the too-short malformation. But there may be existing
626 * applications that are expecting the non-continuation type, so we retain
627 * it, and return it in preference to the too-short malformation. (If this
628 * code were being written from scratch, the two types might be collapsed
629 * into one.) I, khw, am also giving priority to returning the
630 * non-continuation and too-short malformations over overflow when multiple
631 * ones are present. I don't know of any real reason to prefer one over
632 * the other, except that it seems to me that multiple-byte errors trumps
633 * errors from a single byte */
634 if (UNLIKELY(unexpected_non_continuation)) {
635 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
636 if (! (flags & UTF8_CHECK_ONLY)) {
638 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
641 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));
646 uv = UNICODE_REPLACEMENT;
648 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
649 * as what the original expectations were. */
650 do_overlong_test = FALSE;
655 else if (UNLIKELY(curlen < expectlen)) {
656 if (! (flags & UTF8_ALLOW_SHORT)) {
657 if (! (flags & UTF8_CHECK_ONLY)) {
658 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));
662 uv = UNICODE_REPLACEMENT;
663 do_overlong_test = FALSE;
669 #ifndef EBCDIC /* EBCDIC can't overflow */
670 if (UNLIKELY(overflowed)) {
671 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
677 && expectlen > (STRLEN) OFFUNISKIP(uv)
678 && ! (flags & UTF8_ALLOW_LONG))
680 /* The overlong malformation has lower precedence than the others.
681 * Note that if this malformation is allowed, we return the actual
682 * value, instead of the replacement character. This is because this
683 * value is actually well-defined. */
684 if (! (flags & UTF8_CHECK_ONLY)) {
685 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));
690 /* Here, the input is considered to be well-formed, but it still could be a
691 * problematic code point that is not allowed by the input parameters. */
692 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
693 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
694 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
696 if (UNICODE_IS_SURROGATE(uv)) {
698 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
699 * generation of the sv, since no warnings are raised under CHECK */
700 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
701 && ckWARN_d(WARN_SURROGATE))
703 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
704 pack_warn = packWARN(WARN_SURROGATE);
706 if (flags & UTF8_DISALLOW_SURROGATE) {
710 else if ((uv > PERL_UNICODE_MAX)) {
711 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
712 && ckWARN_d(WARN_NON_UNICODE))
714 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
715 pack_warn = packWARN(WARN_NON_UNICODE);
717 #ifndef EBCDIC /* Can never have the equivalent of FE nor FF on EBCDIC, since
718 not representable in UTF-EBCDIC */
720 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
721 * points. We test for these after the regular SUPER ones, and
722 * before possibly bailing out, so that the more dire warning
723 * overrides the regular one, if applicable */
724 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
725 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
727 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
729 && ckWARN_d(WARN_UTF8))
731 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
732 pack_warn = packWARN(WARN_UTF8);
734 if (flags & UTF8_DISALLOW_FE_FF) {
739 if (flags & UTF8_DISALLOW_SUPER) {
743 else if (UNICODE_IS_NONCHAR(uv)) {
744 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
745 && ckWARN_d(WARN_NONCHAR))
747 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
748 pack_warn = packWARN(WARN_NONCHAR);
750 if (flags & UTF8_DISALLOW_NONCHAR) {
756 outlier_ret = uv; /* Note we don't bother to convert to native,
757 as all the outlier code points are the same
758 in both ASCII and EBCDIC */
762 /* Here, this is not considered a malformed character, so drop through
766 return UNI_TO_NATIVE(uv);
768 /* There are three cases which get to beyond this point. In all 3 cases:
769 * <sv> if not null points to a string to print as a warning.
770 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
772 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
773 * This is done by initializing it to 0, and changing it only
776 * 1) The input is valid but problematic, and to be warned about. The
777 * return value is the resultant code point; <*retlen> is set to
778 * <curlen>, the number of bytes that comprise the code point.
779 * <pack_warn> contains the result of packWARN() for the warning
780 * types. The entry point for this case is the label <do_warn>;
781 * 2) The input is a valid code point but disallowed by the parameters to
782 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
783 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
784 * comprise the code point. <pack_warn> contains the result of
785 * packWARN() for the warning types. The entry point for this case is
786 * the label <disallowed>.
787 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
788 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
789 * bytes that comprise the malformation. All such malformations are
790 * assumed to be warning type <utf8>. The entry point for this case
791 * is the label <malformed>.
796 if (sv && ckWARN_d(WARN_UTF8)) {
797 pack_warn = packWARN(WARN_UTF8);
802 if (flags & UTF8_CHECK_ONLY) {
804 *retlen = ((STRLEN) -1);
810 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
811 if warnings are to be raised. */
812 const char * const string = SvPVX_const(sv);
815 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
817 Perl_warner(aTHX_ pack_warn, "%s", string);
828 =for apidoc utf8_to_uvchr_buf
830 Returns the native code point of the first character in the string C<s> which
831 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
832 C<*retlen> will be set to the length, in bytes, of that character.
834 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
835 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
836 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
837 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
838 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
839 the next possible position in C<s> that could begin a non-malformed character.
840 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
848 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
852 return utf8n_to_uvchr(s, send - s, retlen,
853 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
856 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
857 * there are no malformations in the input UTF-8 string C<s>. surrogates,
858 * non-character code points, and non-Unicode code points are allowed. */
861 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
863 UV expectlen = UTF8SKIP(s);
864 const U8* send = s + expectlen;
867 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
874 /* An invariant is trivially returned */
875 if (expectlen == 1) {
880 uv = NATIVE_UTF8_TO_I8(uv);
883 /* Remove the leading bits that indicate the number of bytes, leaving just
884 * the bits that are part of the value */
885 uv &= UTF_START_MASK(expectlen);
887 /* Now, loop through the remaining bytes, accumulating each into the
888 * working total as we go. (I khw tried unrolling the loop for up to 4
889 * bytes, but there was no performance improvement) */
890 for (++s; s < send; s++) {
891 uv = UTF8_ACCUMULATE(uv, *s);
894 return UNI_TO_NATIVE(uv);
899 =for apidoc utf8_to_uvuni_buf
901 Only in very rare circumstances should code need to be dealing in Unicode
902 (as opposed to native) code points. In those few cases, use
903 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
905 Returns the Unicode (not-native) code point of the first character in the
907 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
908 C<retlen> will be set to the length, in bytes, of that character.
910 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
911 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
912 NULL) to -1. If those warnings are off, the computed value if well-defined (or
913 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
914 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
915 next possible position in C<s> that could begin a non-malformed character.
916 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
922 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
924 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
928 /* Call the low level routine asking for checks */
929 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
930 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
934 =for apidoc utf8_length
936 Return the length of the UTF-8 char encoded string C<s> in characters.
937 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
938 up past C<e>, croaks.
944 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
948 PERL_ARGS_ASSERT_UTF8_LENGTH;
950 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
951 * the bitops (especially ~) can create illegal UTF-8.
952 * In other words: in Perl UTF-8 is not just for Unicode. */
955 goto warn_and_return;
965 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
966 "%s in %s", unees, OP_DESC(PL_op));
968 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
975 =for apidoc utf8_distance
977 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
980 WARNING: use only if you *know* that the pointers point inside the
987 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
989 PERL_ARGS_ASSERT_UTF8_DISTANCE;
991 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
997 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1000 WARNING: do not use the following unless you *know* C<off> is within
1001 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1002 on the first byte of character or just after the last byte of a character.
1008 Perl_utf8_hop(const U8 *s, I32 off)
1010 PERL_ARGS_ASSERT_UTF8_HOP;
1012 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1013 * the bitops (especially ~) can create illegal UTF-8.
1014 * In other words: in Perl UTF-8 is not just for Unicode. */
1023 while (UTF8_IS_CONTINUATION(*s))
1031 =for apidoc bytes_cmp_utf8
1033 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1034 sequence of characters (stored as UTF-8)
1035 in C<u>, C<ulen>. Returns 0 if they are
1036 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1037 if the first string is greater than the second string.
1039 -1 or +1 is returned if the shorter string was identical to the start of the
1040 longer string. -2 or +2 is returned if
1041 there was a difference between characters
1048 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1050 const U8 *const bend = b + blen;
1051 const U8 *const uend = u + ulen;
1053 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1055 while (b < bend && u < uend) {
1057 if (!UTF8_IS_INVARIANT(c)) {
1058 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1061 if (UTF8_IS_CONTINUATION(c1)) {
1062 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1064 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1065 "Malformed UTF-8 character "
1066 "(unexpected non-continuation byte 0x%02x"
1067 ", immediately after start byte 0x%02x)"
1068 /* Dear diag.t, it's in the pod. */
1070 PL_op ? " in " : "",
1071 PL_op ? OP_DESC(PL_op) : "");
1076 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1077 "%s in %s", unees, OP_DESC(PL_op));
1079 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1080 return -2; /* Really want to return undef :-) */
1087 return *b < c ? -2 : +2;
1092 if (b == bend && u == uend)
1095 return b < bend ? +1 : -1;
1099 =for apidoc utf8_to_bytes
1101 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1102 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1103 updates C<len> to contain the new length.
1104 Returns zero on failure, setting C<len> to -1.
1106 If you need a copy of the string, see L</bytes_from_utf8>.
1112 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1114 U8 * const save = s;
1115 U8 * const send = s + *len;
1118 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1119 PERL_UNUSED_CONTEXT;
1121 /* ensure valid UTF-8 and chars < 256 before updating string */
1123 if (! UTF8_IS_INVARIANT(*s)) {
1124 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1125 *len = ((STRLEN) -1);
1136 if (! UTF8_IS_INVARIANT(c)) {
1137 /* Then it is two-byte encoded */
1138 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1149 =for apidoc bytes_from_utf8
1151 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1152 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1153 the newly-created string, and updates C<len> to contain the new
1154 length. Returns the original string if no conversion occurs, C<len>
1155 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1156 0 if C<s> is converted or consisted entirely of characters that are invariant
1157 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
1163 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1166 const U8 *start = s;
1170 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1171 PERL_UNUSED_CONTEXT;
1175 /* ensure valid UTF-8 and chars < 256 before converting string */
1176 for (send = s + *len; s < send;) {
1177 if (! UTF8_IS_INVARIANT(*s)) {
1178 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1189 Newx(d, (*len) - count + 1, U8);
1190 s = start; start = d;
1193 if (! UTF8_IS_INVARIANT(c)) {
1194 /* Then it is two-byte encoded */
1195 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1206 =for apidoc bytes_to_utf8
1208 Converts a string C<s> of length C<len> bytes from the native encoding into
1210 Returns a pointer to the newly-created string, and sets C<len> to
1211 reflect the new length in bytes.
1213 A C<NUL> character will be written after the end of the string.
1215 If you want to convert to UTF-8 from encodings other than
1216 the native (Latin1 or EBCDIC),
1217 see L</sv_recode_to_utf8>().
1222 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1223 likewise need duplication. */
1226 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1228 const U8 * const send = s + (*len);
1232 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1233 PERL_UNUSED_CONTEXT;
1235 Newx(d, (*len) * 2 + 1, U8);
1239 append_utf8_from_native_byte(*s, &d);
1248 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1250 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1251 * We optimize for native, for obvious reasons. */
1254 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1259 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1262 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1267 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1269 if (OFFUNI_IS_INVARIANT(uv)) {
1270 *d++ = LATIN1_TO_NATIVE((U8) uv);
1273 if (uv <= MAX_UTF8_TWO_BYTE) {
1274 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1275 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1278 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1279 #define LAST_HIGH_SURROGATE 0xDBFF
1280 #define FIRST_LOW_SURROGATE 0xDC00
1281 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1283 /* This assumes that most uses will be in the first Unicode plane, not
1284 * needing surrogates */
1285 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1286 && uv <= UNICODE_SURROGATE_LAST))
1288 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1289 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1292 UV low = (p[0] << 8) + p[1];
1293 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1294 || UNLIKELY(low > LAST_LOW_SURROGATE))
1296 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1299 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1300 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1304 d = uvoffuni_to_utf8_flags(d, uv, 0);
1307 *d++ = (U8)(( uv >> 12) | 0xe0);
1308 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1309 *d++ = (U8)(( uv & 0x3f) | 0x80);
1313 *d++ = (U8)(( uv >> 18) | 0xf0);
1314 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1315 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1316 *d++ = (U8)(( uv & 0x3f) | 0x80);
1321 *newlen = d - dstart;
1325 /* Note: this one is slightly destructive of the source. */
1328 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1331 U8* const send = s + bytelen;
1333 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1336 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1340 const U8 tmp = s[0];
1345 return utf16_to_utf8(p, d, bytelen, newlen);
1349 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1351 U8 tmpbuf[UTF8_MAXBYTES+1];
1352 uvchr_to_utf8(tmpbuf, c);
1353 return _is_utf8_FOO(classnum, tmpbuf);
1356 /* Internal function so we can deprecate the external one, and call
1357 this one from other deprecated functions in this file */
1360 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1362 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1366 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1370 Perl__is_uni_perl_idcont(pTHX_ UV c)
1372 U8 tmpbuf[UTF8_MAXBYTES+1];
1373 uvchr_to_utf8(tmpbuf, c);
1374 return _is_utf8_perl_idcont(tmpbuf);
1378 Perl__is_uni_perl_idstart(pTHX_ UV c)
1380 U8 tmpbuf[UTF8_MAXBYTES+1];
1381 uvchr_to_utf8(tmpbuf, c);
1382 return _is_utf8_perl_idstart(tmpbuf);
1386 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1388 /* We have the latin1-range values compiled into the core, so just use
1389 * those, converting the result to UTF-8. The only difference between upper
1390 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1391 * either "SS" or "Ss". Which one to use is passed into the routine in
1392 * 'S_or_s' to avoid a test */
1394 UV converted = toUPPER_LATIN1_MOD(c);
1396 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1398 assert(S_or_s == 'S' || S_or_s == 's');
1400 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1401 characters in this range */
1402 *p = (U8) converted;
1407 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1408 * which it maps to one of them, so as to only have to have one check for
1409 * it in the main case */
1410 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1412 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1413 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1416 converted = GREEK_CAPITAL_LETTER_MU;
1418 #if UNICODE_MAJOR_VERSION > 2 \
1419 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1420 && UNICODE_DOT_DOT_VERSION >= 8)
1421 case LATIN_SMALL_LETTER_SHARP_S:
1428 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1429 NOT_REACHED; /* NOTREACHED */
1433 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1434 *p = UTF8_TWO_BYTE_LO(converted);
1440 /* Call the function to convert a UTF-8 encoded character to the specified case.
1441 * Note that there may be more than one character in the result.
1442 * INP is a pointer to the first byte of the input character
1443 * OUTP will be set to the first byte of the string of changed characters. It
1444 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1445 * LENP will be set to the length in bytes of the string of changed characters
1447 * The functions return the ordinal of the first character in the string of OUTP */
1448 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1449 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1450 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1452 /* This additionally has the input parameter SPECIALS, which if non-zero will
1453 * cause this to use the SPECIALS hash for folding (meaning get full case
1454 * folding); otherwise, when zero, this implies a simple case fold */
1455 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1458 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1460 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1461 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1462 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1463 * the changed version may be longer than the original character.
1465 * The ordinal of the first character of the changed version is returned
1466 * (but note, as explained above, that there may be more.) */
1468 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1471 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1474 uvchr_to_utf8(p, c);
1475 return CALL_UPPER_CASE(p, p, lenp);
1479 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1481 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1484 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1487 uvchr_to_utf8(p, c);
1488 return CALL_TITLE_CASE(p, p, lenp);
1492 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
1494 /* We have the latin1-range values compiled into the core, so just use
1495 * those, converting the result to UTF-8. Since the result is always just
1496 * one character, we allow <p> to be NULL */
1498 U8 converted = toLOWER_LATIN1(c);
1501 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1506 /* Result is known to always be < 256, so can use the EIGHT_BIT
1508 *p = UTF8_EIGHT_BIT_HI(converted);
1509 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
1517 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1519 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1522 return to_lower_latin1((U8) c, p, lenp);
1525 uvchr_to_utf8(p, c);
1526 return CALL_LOWER_CASE(p, p, lenp);
1530 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1532 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1533 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1534 * FOLD_FLAGS_FULL iff full folding is to be used;
1536 * Not to be used for locale folds
1541 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1542 PERL_UNUSED_CONTEXT;
1544 assert (! (flags & FOLD_FLAGS_LOCALE));
1546 if (c == MICRO_SIGN) {
1547 converted = GREEK_SMALL_LETTER_MU;
1549 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1550 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1551 || UNICODE_DOT_DOT_VERSION > 0)
1552 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1554 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1555 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1556 * under those circumstances. */
1557 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1558 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1559 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1561 return LATIN_SMALL_LETTER_LONG_S;
1571 else { /* In this range the fold of all other characters is their lower
1573 converted = toLOWER_LATIN1(c);
1576 if (UVCHR_IS_INVARIANT(converted)) {
1577 *p = (U8) converted;
1581 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1582 *p = UTF8_TWO_BYTE_LO(converted);
1590 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
1593 /* Not currently externally documented, and subject to change
1594 * <flags> bits meanings:
1595 * FOLD_FLAGS_FULL iff full folding is to be used;
1596 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1597 * locale are to be used.
1598 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1601 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1603 if (flags & FOLD_FLAGS_LOCALE) {
1604 /* Treat a UTF-8 locale as not being in locale at all */
1605 if (IN_UTF8_CTYPE_LOCALE) {
1606 flags &= ~FOLD_FLAGS_LOCALE;
1609 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1610 goto needs_full_generality;
1615 return _to_fold_latin1((U8) c, p, lenp,
1616 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1619 /* Here, above 255. If no special needs, just use the macro */
1620 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1621 uvchr_to_utf8(p, c);
1622 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1624 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1625 the special flags. */
1626 U8 utf8_c[UTF8_MAXBYTES + 1];
1628 needs_full_generality:
1629 uvchr_to_utf8(utf8_c, c);
1630 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1634 PERL_STATIC_INLINE bool
1635 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1636 const char *const swashname, SV* const invlist)
1638 /* returns a boolean giving whether or not the UTF8-encoded character that
1639 * starts at <p> is in the swash indicated by <swashname>. <swash>
1640 * contains a pointer to where the swash indicated by <swashname>
1641 * is to be stored; which this routine will do, so that future calls will
1642 * look at <*swash> and only generate a swash if it is not null. <invlist>
1643 * is NULL or an inversion list that defines the swash. If not null, it
1644 * saves time during initialization of the swash.
1646 * Note that it is assumed that the buffer length of <p> is enough to
1647 * contain all the bytes that comprise the character. Thus, <*p> should
1648 * have been checked before this call for mal-formedness enough to assure
1651 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1653 /* The API should have included a length for the UTF-8 character in <p>,
1654 * but it doesn't. We therefore assume that p has been validated at least
1655 * as far as there being enough bytes available in it to accommodate the
1656 * character without reading beyond the end, and pass that number on to the
1657 * validating routine */
1658 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1659 if (ckWARN_d(WARN_UTF8)) {
1660 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1661 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1662 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1663 what the malformation is */
1664 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1670 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1671 *swash = _core_swash_init("utf8",
1673 /* Only use the name if there is no inversion
1674 * list; otherwise will go out to disk */
1675 (invlist) ? "" : swashname,
1677 &PL_sv_undef, 1, 0, invlist, &flags);
1680 return swash_fetch(*swash, p, TRUE) != 0;
1684 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1686 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1688 assert(classnum < _FIRST_NON_SWASH_CC);
1690 return is_utf8_common(p,
1691 &PL_utf8_swash_ptrs[classnum],
1692 swash_property_names[classnum],
1693 PL_XPosix_ptrs[classnum]);
1697 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1701 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1703 if (! PL_utf8_perl_idstart) {
1704 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1706 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
1710 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1712 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1716 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1720 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1724 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1726 if (! PL_utf8_perl_idcont) {
1727 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1729 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
1733 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1735 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1737 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1741 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1743 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1745 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1749 Perl__is_utf8_mark(pTHX_ const U8 *p)
1751 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1753 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1757 =for apidoc to_utf8_case
1759 C<p> contains the pointer to the UTF-8 string encoding
1760 the character that is being converted. This routine assumes that the character
1761 at C<p> is well-formed.
1763 C<ustrp> is a pointer to the character buffer to put the
1764 conversion result to. C<lenp> is a pointer to the length
1767 C<swashp> is a pointer to the swash to use.
1769 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1770 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
1771 but not always, a multicharacter mapping), is tried first.
1773 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1774 any special mappings; C<""> means to use the special mappings. Values other
1775 than these two are treated as the name of the hash containing the special
1776 mappings, like C<"utf8::ToSpecLower">.
1778 C<normal> is a string like C<"ToLower"> which means the swash
1784 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1785 SV **swashp, const char *normal, const char *special)
1788 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1790 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1792 /* Note that swash_fetch() doesn't output warnings for these because it
1793 * assumes we will */
1794 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1795 if (uv1 <= UNICODE_SURROGATE_LAST) {
1796 if (ckWARN_d(WARN_SURROGATE)) {
1797 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1798 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1799 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1802 else if (UNICODE_IS_SUPER(uv1)) {
1803 if (ckWARN_d(WARN_NON_UNICODE)) {
1804 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1805 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1806 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1810 /* Note that non-characters are perfectly legal, so no warning should
1814 if (!*swashp) /* load on-demand */
1815 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1818 /* It might be "special" (sometimes, but not always,
1819 * a multicharacter mapping) */
1823 /* If passed in the specials name, use that; otherwise use any
1824 * given in the swash */
1825 if (*special != '\0') {
1826 hv = get_hv(special, 0);
1829 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1831 hv = MUTABLE_HV(SvRV(*svp));
1836 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
1841 s = SvPV_const(*svp, len);
1844 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1846 Copy(s, ustrp, len, U8);
1851 if (!len && *swashp) {
1852 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
1855 /* It was "normal" (a single character mapping). */
1856 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1864 return valid_utf8_to_uvchr(ustrp, 0);
1867 /* Here, there was no mapping defined, which means that the code point maps
1868 * to itself. Return the inputs */
1870 if (p != ustrp) { /* Don't copy onto itself */
1871 Copy(p, ustrp, len, U8);
1882 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1884 /* This is called when changing the case of a UTF-8-encoded character above
1885 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1886 * result contains a character that crosses the 255/256 boundary, disallow
1887 * the change, and return the original code point. See L<perlfunc/lc> for
1890 * p points to the original string whose case was changed; assumed
1891 * by this routine to be well-formed
1892 * result the code point of the first character in the changed-case string
1893 * ustrp points to the changed-case string (<result> represents its first char)
1894 * lenp points to the length of <ustrp> */
1896 UV original; /* To store the first code point of <p> */
1898 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1900 assert(UTF8_IS_ABOVE_LATIN1(*p));
1902 /* We know immediately if the first character in the string crosses the
1903 * boundary, so can skip */
1906 /* Look at every character in the result; if any cross the
1907 * boundary, the whole thing is disallowed */
1908 U8* s = ustrp + UTF8SKIP(ustrp);
1909 U8* e = ustrp + *lenp;
1911 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
1917 /* Here, no characters crossed, result is ok as-is, but we warn. */
1918 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
1924 /* Failed, have to return the original */
1925 original = valid_utf8_to_uvchr(p, lenp);
1927 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1928 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1929 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1930 "resolved to \"\\x{%"UVXf"}\".",
1934 Copy(p, ustrp, *lenp, char);
1939 =for apidoc to_utf8_upper
1941 Instead use L</toUPPER_utf8>.
1945 /* Not currently externally documented, and subject to change:
1946 * <flags> is set iff iff the rules from the current underlying locale are to
1950 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
1954 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
1957 /* Treat a UTF-8 locale as not being in locale at all */
1958 if (IN_UTF8_CTYPE_LOCALE) {
1962 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1966 if (UTF8_IS_INVARIANT(*p)) {
1968 result = toUPPER_LC(*p);
1971 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
1974 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
1976 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
1977 result = toUPPER_LC(c);
1980 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
1984 else { /* UTF-8, ord above 255 */
1985 result = CALL_UPPER_CASE(p, ustrp, lenp);
1988 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
1993 /* Here, used locale rules. Convert back to UTF-8 */
1994 if (UTF8_IS_INVARIANT(result)) {
1995 *ustrp = (U8) result;
1999 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2000 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2008 =for apidoc to_utf8_title
2010 Instead use L</toTITLE_utf8>.
2014 /* Not currently externally documented, and subject to change:
2015 * <flags> is set iff the rules from the current underlying locale are to be
2016 * used. Since titlecase is not defined in POSIX, for other than a
2017 * UTF-8 locale, uppercase is used instead for code points < 256.
2021 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2025 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2028 /* Treat a UTF-8 locale as not being in locale at all */
2029 if (IN_UTF8_CTYPE_LOCALE) {
2033 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2037 if (UTF8_IS_INVARIANT(*p)) {
2039 result = toUPPER_LC(*p);
2042 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2045 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2047 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2048 result = toUPPER_LC(c);
2051 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2055 else { /* UTF-8, ord above 255 */
2056 result = CALL_TITLE_CASE(p, ustrp, lenp);
2059 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2064 /* Here, used locale rules. Convert back to UTF-8 */
2065 if (UTF8_IS_INVARIANT(result)) {
2066 *ustrp = (U8) result;
2070 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2071 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2079 =for apidoc to_utf8_lower
2081 Instead use L</toLOWER_utf8>.
2085 /* Not currently externally documented, and subject to change:
2086 * <flags> is set iff iff the rules from the current underlying locale are to
2091 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2095 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2098 /* Treat a UTF-8 locale as not being in locale at all */
2099 if (IN_UTF8_CTYPE_LOCALE) {
2103 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2107 if (UTF8_IS_INVARIANT(*p)) {
2109 result = toLOWER_LC(*p);
2112 return to_lower_latin1(*p, ustrp, lenp);
2115 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2117 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2118 result = toLOWER_LC(c);
2121 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2125 else { /* UTF-8, ord above 255 */
2126 result = CALL_LOWER_CASE(p, ustrp, lenp);
2129 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2135 /* Here, used locale rules. Convert back to UTF-8 */
2136 if (UTF8_IS_INVARIANT(result)) {
2137 *ustrp = (U8) result;
2141 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2142 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2150 =for apidoc to_utf8_fold
2152 Instead use L</toFOLD_utf8>.
2156 /* Not currently externally documented, and subject to change,
2158 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2159 * locale are to be used.
2160 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2161 * otherwise simple folds
2162 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2167 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2171 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2173 /* These are mutually exclusive */
2174 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2176 assert(p != ustrp); /* Otherwise overwrites */
2178 if (flags & FOLD_FLAGS_LOCALE) {
2179 /* Treat a UTF-8 locale as not being in locale at all */
2180 if (IN_UTF8_CTYPE_LOCALE) {
2181 flags &= ~FOLD_FLAGS_LOCALE;
2184 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2188 if (UTF8_IS_INVARIANT(*p)) {
2189 if (flags & FOLD_FLAGS_LOCALE) {
2190 result = toFOLD_LC(*p);
2193 return _to_fold_latin1(*p, ustrp, lenp,
2194 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2197 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2198 if (flags & FOLD_FLAGS_LOCALE) {
2199 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2200 result = toFOLD_LC(c);
2203 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2205 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2208 else { /* UTF-8, ord above 255 */
2209 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2211 if (flags & FOLD_FLAGS_LOCALE) {
2213 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2214 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2216 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2217 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2219 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2221 /* Special case these two characters, as what normally gets
2222 * returned under locale doesn't work */
2223 if (UTF8SKIP(p) == cap_sharp_s_len
2224 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
2226 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2227 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2228 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2229 "resolved to \"\\x{17F}\\x{17F}\".");
2234 if (UTF8SKIP(p) == long_s_t_len
2235 && memEQ((char *) p, LONG_S_T, long_s_t_len))
2237 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2238 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2239 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2240 "resolved to \"\\x{FB06}\".");
2241 goto return_ligature_st;
2244 #if UNICODE_MAJOR_VERSION == 3 \
2245 && UNICODE_DOT_VERSION == 0 \
2246 && UNICODE_DOT_DOT_VERSION == 1
2247 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2249 /* And special case this on this Unicode version only, for the same
2250 * reaons the other two are special cased. They would cross the
2251 * 255/256 boundary which is forbidden under /l, and so the code
2252 * wouldn't catch that they are equivalent (which they are only in
2254 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2255 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2257 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2258 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2259 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2260 "resolved to \"\\x{0131}\".");
2261 goto return_dotless_i;
2265 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2267 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2271 /* This is called when changing the case of a UTF-8-encoded
2272 * character above the ASCII range, and the result should not
2273 * contain an ASCII character. */
2275 UV original; /* To store the first code point of <p> */
2277 /* Look at every character in the result; if any cross the
2278 * boundary, the whole thing is disallowed */
2280 U8* e = ustrp + *lenp;
2283 /* Crossed, have to return the original */
2284 original = valid_utf8_to_uvchr(p, lenp);
2286 /* But in these instances, there is an alternative we can
2287 * return that is valid */
2288 if (original == LATIN_SMALL_LETTER_SHARP_S
2289 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2290 || original == LATIN_CAPITAL_LETTER_SHARP_S
2295 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2296 goto return_ligature_st;
2298 #if UNICODE_MAJOR_VERSION == 3 \
2299 && UNICODE_DOT_VERSION == 0 \
2300 && UNICODE_DOT_DOT_VERSION == 1
2302 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2303 goto return_dotless_i;
2306 Copy(p, ustrp, *lenp, char);
2312 /* Here, no characters crossed, result is ok as-is */
2317 /* Here, used locale rules. Convert back to UTF-8 */
2318 if (UTF8_IS_INVARIANT(result)) {
2319 *ustrp = (U8) result;
2323 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2324 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2331 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2332 * folds to a string of two of these characters. By returning this
2333 * instead, then, e.g.,
2334 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2337 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2338 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2340 return LATIN_SMALL_LETTER_LONG_S;
2343 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2344 * have the other one fold to it */
2346 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2347 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2348 return LATIN_SMALL_LIGATURE_ST;
2350 #if UNICODE_MAJOR_VERSION == 3 \
2351 && UNICODE_DOT_VERSION == 0 \
2352 && UNICODE_DOT_DOT_VERSION == 1
2355 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2356 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2357 return LATIN_SMALL_LETTER_DOTLESS_I;
2364 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2365 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2366 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2370 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2372 PERL_ARGS_ASSERT_SWASH_INIT;
2374 /* Returns a copy of a swash initiated by the called function. This is the
2375 * public interface, and returning a copy prevents others from doing
2376 * mischief on the original */
2378 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2382 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2385 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2386 * use the following define */
2388 #define CORE_SWASH_INIT_RETURN(x) \
2389 PL_curpm= old_PL_curpm; \
2392 /* Initialize and return a swash, creating it if necessary. It does this
2393 * by calling utf8_heavy.pl in the general case. The returned value may be
2394 * the swash's inversion list instead if the input parameters allow it.
2395 * Which is returned should be immaterial to callers, as the only
2396 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2397 * and swash_to_invlist() handle both these transparently.
2399 * This interface should only be used by functions that won't destroy or
2400 * adversely change the swash, as doing so affects all other uses of the
2401 * swash in the program; the general public should use 'Perl_swash_init'
2404 * pkg is the name of the package that <name> should be in.
2405 * name is the name of the swash to find. Typically it is a Unicode
2406 * property name, including user-defined ones
2407 * listsv is a string to initialize the swash with. It must be of the form
2408 * documented as the subroutine return value in
2409 * L<perlunicode/User-Defined Character Properties>
2410 * minbits is the number of bits required to represent each data element.
2411 * It is '1' for binary properties.
2412 * none I (khw) do not understand this one, but it is used only in tr///.
2413 * invlist is an inversion list to initialize the swash with (or NULL)
2414 * flags_p if non-NULL is the address of various input and output flag bits
2415 * to the routine, as follows: ('I' means is input to the routine;
2416 * 'O' means output from the routine. Only flags marked O are
2417 * meaningful on return.)
2418 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2419 * came from a user-defined property. (I O)
2420 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2421 * when the swash cannot be located, to simply return NULL. (I)
2422 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2423 * return of an inversion list instead of a swash hash if this routine
2424 * thinks that would result in faster execution of swash_fetch() later
2427 * Thus there are three possible inputs to find the swash: <name>,
2428 * <listsv>, and <invlist>. At least one must be specified. The result
2429 * will be the union of the specified ones, although <listsv>'s various
2430 * actions can intersect, etc. what <name> gives. To avoid going out to
2431 * disk at all, <invlist> should specify completely what the swash should
2432 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2434 * <invlist> is only valid for binary properties */
2436 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2438 SV* retval = &PL_sv_undef;
2439 HV* swash_hv = NULL;
2440 const int invlist_swash_boundary =
2441 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2442 ? 512 /* Based on some benchmarking, but not extensive, see commit
2444 : -1; /* Never return just an inversion list */
2446 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2447 assert(! invlist || minbits == 1);
2449 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2450 that triggered the swash init and the swash init perl logic itself.
2453 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2455 if (listsv != &PL_sv_undef || strNE(name, "")) {
2457 const size_t pkg_len = strlen(pkg);
2458 const size_t name_len = strlen(name);
2459 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2463 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2465 PUSHSTACKi(PERLSI_MAGIC);
2469 /* We might get here via a subroutine signature which uses a utf8
2470 * parameter name, at which point PL_subname will have been set
2471 * but not yet used. */
2472 save_item(PL_subname);
2473 if (PL_parser && PL_parser->error_count)
2474 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2475 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2476 if (!method) { /* demand load UTF-8 */
2478 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2479 GvSV(PL_errgv) = NULL;
2480 #ifndef NO_TAINT_SUPPORT
2481 /* It is assumed that callers of this routine are not passing in
2482 * any user derived data. */
2483 /* Need to do this after save_re_context() as it will set
2484 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2485 * in Perl_magic_get). Even line to create errsv_save can turn on
2487 SAVEBOOL(TAINT_get);
2490 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2493 /* Not ERRSV, as there is no need to vivify a scalar we are
2494 about to discard. */
2495 SV * const errsv = GvSV(PL_errgv);
2496 if (!SvTRUE(errsv)) {
2497 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2498 SvREFCNT_dec(errsv);
2506 mPUSHp(pkg, pkg_len);
2507 mPUSHp(name, name_len);
2512 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2513 GvSV(PL_errgv) = NULL;
2514 /* If we already have a pointer to the method, no need to use
2515 * call_method() to repeat the lookup. */
2517 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2518 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2520 retval = *PL_stack_sp--;
2521 SvREFCNT_inc(retval);
2524 /* Not ERRSV. See above. */
2525 SV * const errsv = GvSV(PL_errgv);
2526 if (!SvTRUE(errsv)) {
2527 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2528 SvREFCNT_dec(errsv);
2533 if (IN_PERL_COMPILETIME) {
2534 CopHINTS_set(PL_curcop, PL_hints);
2536 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2539 /* If caller wants to handle missing properties, let them */
2540 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2541 CORE_SWASH_INIT_RETURN(NULL);
2544 "Can't find Unicode property definition \"%"SVf"\"",
2546 NOT_REACHED; /* NOTREACHED */
2548 } /* End of calling the module to find the swash */
2550 /* If this operation fetched a swash, and we will need it later, get it */
2551 if (retval != &PL_sv_undef
2552 && (minbits == 1 || (flags_p
2554 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2556 swash_hv = MUTABLE_HV(SvRV(retval));
2558 /* If we don't already know that there is a user-defined component to
2559 * this swash, and the user has indicated they wish to know if there is
2560 * one (by passing <flags_p>), find out */
2561 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2562 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2563 if (user_defined && SvUV(*user_defined)) {
2564 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2569 /* Make sure there is an inversion list for binary properties */
2571 SV** swash_invlistsvp = NULL;
2572 SV* swash_invlist = NULL;
2573 bool invlist_in_swash_is_valid = FALSE;
2574 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2575 an unclaimed reference count */
2577 /* If this operation fetched a swash, get its already existing
2578 * inversion list, or create one for it */
2581 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2582 if (swash_invlistsvp) {
2583 swash_invlist = *swash_invlistsvp;
2584 invlist_in_swash_is_valid = TRUE;
2587 swash_invlist = _swash_to_invlist(retval);
2588 swash_invlist_unclaimed = TRUE;
2592 /* If an inversion list was passed in, have to include it */
2595 /* Any fetched swash will by now have an inversion list in it;
2596 * otherwise <swash_invlist> will be NULL, indicating that we
2597 * didn't fetch a swash */
2598 if (swash_invlist) {
2600 /* Add the passed-in inversion list, which invalidates the one
2601 * already stored in the swash */
2602 invlist_in_swash_is_valid = FALSE;
2603 _invlist_union(invlist, swash_invlist, &swash_invlist);
2607 /* Here, there is no swash already. Set up a minimal one, if
2608 * we are going to return a swash */
2609 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2611 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2613 swash_invlist = invlist;
2617 /* Here, we have computed the union of all the passed-in data. It may
2618 * be that there was an inversion list in the swash which didn't get
2619 * touched; otherwise save the computed one */
2620 if (! invlist_in_swash_is_valid
2621 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2623 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2625 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2627 /* We just stole a reference count. */
2628 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2629 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2632 SvREADONLY_on(swash_invlist);
2634 /* Use the inversion list stand-alone if small enough */
2635 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2636 SvREFCNT_dec(retval);
2637 if (!swash_invlist_unclaimed)
2638 SvREFCNT_inc_simple_void_NN(swash_invlist);
2639 retval = newRV_noinc(swash_invlist);
2643 CORE_SWASH_INIT_RETURN(retval);
2644 #undef CORE_SWASH_INIT_RETURN
2648 /* This API is wrong for special case conversions since we may need to
2649 * return several Unicode characters for a single Unicode character
2650 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2651 * the lower-level routine, and it is similarly broken for returning
2652 * multiple values. --jhi
2653 * For those, you should use to_utf8_case() instead */
2654 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2657 * Returns the value of property/mapping C<swash> for the first character
2658 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2659 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
2660 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2662 * A "swash" is a hash which contains initially the keys/values set up by
2663 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2664 * property for all possible code points. Things are stored in a compact form
2665 * (see utf8_heavy.pl) so that calculation is required to find the actual
2666 * property value for a given code point. As code points are looked up, new
2667 * key/value pairs are added to the hash, so that the calculation doesn't have
2668 * to ever be re-done. Further, each calculation is done, not just for the
2669 * desired one, but for a whole block of code points adjacent to that one.
2670 * For binary properties on ASCII machines, the block is usually for 64 code
2671 * points, starting with a code point evenly divisible by 64. Thus if the
2672 * property value for code point 257 is requested, the code goes out and
2673 * calculates the property values for all 64 code points between 256 and 319,
2674 * and stores these as a single 64-bit long bit vector, called a "swatch",
2675 * under the key for code point 256. The key is the UTF-8 encoding for code
2676 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2677 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2678 * for code point 258 is then requested, this code realizes that it would be
2679 * stored under the key for 256, and would find that value and extract the
2680 * relevant bit, offset from 256.
2682 * Non-binary properties are stored in as many bits as necessary to represent
2683 * their values (32 currently, though the code is more general than that), not
2684 * as single bits, but the principal is the same: the value for each key is a
2685 * vector that encompasses the property values for all code points whose UTF-8
2686 * representations are represented by the key. That is, for all code points
2687 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2691 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2693 HV *const hv = MUTABLE_HV(SvRV(swash));
2698 const U8 *tmps = NULL;
2702 PERL_ARGS_ASSERT_SWASH_FETCH;
2704 /* If it really isn't a hash, it isn't really swash; must be an inversion
2706 if (SvTYPE(hv) != SVt_PVHV) {
2707 return _invlist_contains_cp((SV*)hv,
2709 ? valid_utf8_to_uvchr(ptr, NULL)
2713 /* We store the values in a "swatch" which is a vec() value in a swash
2714 * hash. Code points 0-255 are a single vec() stored with key length
2715 * (klen) 0. All other code points have a UTF-8 representation
2716 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2717 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2718 * length for them is the length of the encoded char - 1. ptr[klen] is the
2719 * final byte in the sequence representing the character */
2720 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2725 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2728 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
2731 klen = UTF8SKIP(ptr) - 1;
2733 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2734 * the vec is the final byte in the sequence. (In EBCDIC this is
2735 * converted to I8 to get consecutive values.) To help you visualize
2737 * Straight 1047 After final byte
2738 * UTF-8 UTF-EBCDIC I8 transform
2739 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2740 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2742 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2743 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2745 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2746 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2748 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2749 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2751 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2752 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2754 * (There are no discontinuities in the elided (...) entries.)
2755 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2756 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2757 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2758 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2759 * index into the vec() swatch (after subtracting 0x80, which we
2760 * actually do with an '&').
2761 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2762 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2763 * dicontinuities which go away by transforming it into I8, and we
2764 * effectively subtract 0xA0 to get the index. */
2765 needents = (1 << UTF_ACCUMULATION_SHIFT);
2766 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2770 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
2771 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2772 * it's nothing to sniff at.) Pity we usually come through at least
2773 * two function calls to get here...
2775 * NB: this code assumes that swatches are never modified, once generated!
2778 if (hv == PL_last_swash_hv &&
2779 klen == PL_last_swash_klen &&
2780 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2782 tmps = PL_last_swash_tmps;
2783 slen = PL_last_swash_slen;
2786 /* Try our second-level swatch cache, kept in a hash. */
2787 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2789 /* If not cached, generate it via swatch_get */
2790 if (!svp || !SvPOK(*svp)
2791 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2794 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2795 swatch = swatch_get(swash,
2796 code_point & ~((UV)needents - 1),
2799 else { /* For the first 256 code points, the swatch has a key of
2801 swatch = swatch_get(swash, 0, needents);
2804 if (IN_PERL_COMPILETIME)
2805 CopHINTS_set(PL_curcop, PL_hints);
2807 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2809 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2810 || (slen << 3) < needents)
2811 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2812 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2813 svp, tmps, (UV)slen, (UV)needents);
2816 PL_last_swash_hv = hv;
2817 assert(klen <= sizeof(PL_last_swash_key));
2818 PL_last_swash_klen = (U8)klen;
2819 /* FIXME change interpvar.h? */
2820 PL_last_swash_tmps = (U8 *) tmps;
2821 PL_last_swash_slen = slen;
2823 Copy(ptr, PL_last_swash_key, klen, U8);
2826 switch ((int)((slen << 3) / needents)) {
2828 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2830 return ((UV) tmps[off]);
2834 ((UV) tmps[off ] << 8) +
2835 ((UV) tmps[off + 1]);
2839 ((UV) tmps[off ] << 24) +
2840 ((UV) tmps[off + 1] << 16) +
2841 ((UV) tmps[off + 2] << 8) +
2842 ((UV) tmps[off + 3]);
2844 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2845 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2846 NORETURN_FUNCTION_END;
2849 /* Read a single line of the main body of the swash input text. These are of
2852 * where each number is hex. The first two numbers form the minimum and
2853 * maximum of a range, and the third is the value associated with the range.
2854 * Not all swashes should have a third number
2856 * On input: l points to the beginning of the line to be examined; it points
2857 * to somewhere in the string of the whole input text, and is
2858 * terminated by a \n or the null string terminator.
2859 * lend points to the null terminator of that string
2860 * wants_value is non-zero if the swash expects a third number
2861 * typestr is the name of the swash's mapping, like 'ToLower'
2862 * On output: *min, *max, and *val are set to the values read from the line.
2863 * returns a pointer just beyond the line examined. If there was no
2864 * valid min number on the line, returns lend+1
2868 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2869 const bool wants_value, const U8* const typestr)
2871 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2872 STRLEN numlen; /* Length of the number */
2873 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2874 | PERL_SCAN_DISALLOW_PREFIX
2875 | PERL_SCAN_SILENT_NON_PORTABLE;
2877 /* nl points to the next \n in the scan */
2878 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2880 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2882 /* Get the first number on the line: the range minimum */
2884 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2885 *max = *min; /* So can never return without setting max */
2886 if (numlen) /* If found a hex number, position past it */
2888 else if (nl) { /* Else, go handle next line, if any */
2889 return nl + 1; /* 1 is length of "\n" */
2891 else { /* Else, no next line */
2892 return lend + 1; /* to LIST's end at which \n is not found */
2895 /* The max range value follows, separated by a BLANK */
2898 flags = PERL_SCAN_SILENT_ILLDIGIT
2899 | PERL_SCAN_DISALLOW_PREFIX
2900 | PERL_SCAN_SILENT_NON_PORTABLE;
2902 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2905 else /* If no value here, it is a single element range */
2908 /* Non-binary tables have a third entry: what the first element of the
2909 * range maps to. The map for those currently read here is in hex */
2913 flags = PERL_SCAN_SILENT_ILLDIGIT
2914 | PERL_SCAN_DISALLOW_PREFIX
2915 | PERL_SCAN_SILENT_NON_PORTABLE;
2917 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2926 /* diag_listed_as: To%s: illegal mapping '%s' */
2927 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2933 *val = 0; /* bits == 1, then any val should be ignored */
2935 else { /* Nothing following range min, should be single element with no
2940 /* diag_listed_as: To%s: illegal mapping '%s' */
2941 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2945 *val = 0; /* bits == 1, then val should be ignored */
2948 /* Position to next line if any, or EOF */
2958 * Returns a swatch (a bit vector string) for a code point sequence
2959 * that starts from the value C<start> and comprises the number C<span>.
2960 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2961 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2964 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
2967 U8 *l, *lend, *x, *xend, *s, *send;
2968 STRLEN lcur, xcur, scur;
2969 HV *const hv = MUTABLE_HV(SvRV(swash));
2970 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
2972 SV** listsvp = NULL; /* The string containing the main body of the table */
2973 SV** extssvp = NULL;
2974 SV** invert_it_svp = NULL;
2977 STRLEN octets; /* if bits == 1, then octets == 0 */
2979 UV end = start + span;
2981 if (invlistsvp == NULL) {
2982 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2983 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2984 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2985 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2986 listsvp = hv_fetchs(hv, "LIST", FALSE);
2987 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2989 bits = SvUV(*bitssvp);
2990 none = SvUV(*nonesvp);
2991 typestr = (U8*)SvPV_nolen(*typesvp);
2997 octets = bits >> 3; /* if bits == 1, then octets == 0 */
2999 PERL_ARGS_ASSERT_SWATCH_GET;
3001 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3002 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3006 /* If overflowed, use the max possible */
3012 /* create and initialize $swatch */
3013 scur = octets ? (span * octets) : (span + 7) / 8;
3014 swatch = newSV(scur);
3016 s = (U8*)SvPVX(swatch);
3017 if (octets && none) {
3018 const U8* const e = s + scur;
3021 *s++ = (U8)(none & 0xff);
3022 else if (bits == 16) {
3023 *s++ = (U8)((none >> 8) & 0xff);
3024 *s++ = (U8)( none & 0xff);
3026 else if (bits == 32) {
3027 *s++ = (U8)((none >> 24) & 0xff);
3028 *s++ = (U8)((none >> 16) & 0xff);
3029 *s++ = (U8)((none >> 8) & 0xff);
3030 *s++ = (U8)( none & 0xff);
3036 (void)memzero((U8*)s, scur + 1);
3038 SvCUR_set(swatch, scur);
3039 s = (U8*)SvPVX(swatch);
3041 if (invlistsvp) { /* If has an inversion list set up use that */
3042 _invlist_populate_swatch(*invlistsvp, start, end, s);
3046 /* read $swash->{LIST} */
3047 l = (U8*)SvPV(*listsvp, lcur);
3050 UV min, max, val, upper;
3051 l = swash_scan_list_line(l, lend, &min, &max, &val,
3052 cBOOL(octets), typestr);
3057 /* If looking for something beyond this range, go try the next one */
3061 /* <end> is generally 1 beyond where we want to set things, but at the
3062 * platform's infinity, where we can't go any higher, we want to
3063 * include the code point at <end> */
3066 : (max != UV_MAX || end != UV_MAX)
3073 if (!none || val < none) {
3078 for (key = min; key <= upper; key++) {
3080 /* offset must be non-negative (start <= min <= key < end) */
3081 offset = octets * (key - start);
3083 s[offset] = (U8)(val & 0xff);
3084 else if (bits == 16) {
3085 s[offset ] = (U8)((val >> 8) & 0xff);
3086 s[offset + 1] = (U8)( val & 0xff);
3088 else if (bits == 32) {
3089 s[offset ] = (U8)((val >> 24) & 0xff);
3090 s[offset + 1] = (U8)((val >> 16) & 0xff);
3091 s[offset + 2] = (U8)((val >> 8) & 0xff);
3092 s[offset + 3] = (U8)( val & 0xff);
3095 if (!none || val < none)
3099 else { /* bits == 1, then val should be ignored */
3104 for (key = min; key <= upper; key++) {
3105 const STRLEN offset = (STRLEN)(key - start);
3106 s[offset >> 3] |= 1 << (offset & 7);
3111 /* Invert if the data says it should be. Assumes that bits == 1 */
3112 if (invert_it_svp && SvUV(*invert_it_svp)) {
3114 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3115 * be 0, and their inversion should also be 0, as we don't succeed any
3116 * Unicode property matches for non-Unicode code points */
3117 if (start <= PERL_UNICODE_MAX) {
3119 /* The code below assumes that we never cross the
3120 * Unicode/above-Unicode boundary in a range, as otherwise we would
3121 * have to figure out where to stop flipping the bits. Since this
3122 * boundary is divisible by a large power of 2, and swatches comes
3123 * in small powers of 2, this should be a valid assumption */
3124 assert(start + span - 1 <= PERL_UNICODE_MAX);
3134 /* read $swash->{EXTRAS}
3135 * This code also copied to swash_to_invlist() below */
3136 x = (U8*)SvPV(*extssvp, xcur);
3144 SV **otherbitssvp, *other;
3148 const U8 opc = *x++;
3152 nl = (U8*)memchr(x, '\n', xend - x);
3154 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3156 x = nl + 1; /* 1 is length of "\n" */
3160 x = xend; /* to EXTRAS' end at which \n is not found */
3167 namelen = nl - namestr;
3171 namelen = xend - namestr;
3175 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3176 otherhv = MUTABLE_HV(SvRV(*othersvp));
3177 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3178 otherbits = (STRLEN)SvUV(*otherbitssvp);
3179 if (bits < otherbits)
3180 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3181 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3183 /* The "other" swatch must be destroyed after. */
3184 other = swatch_get(*othersvp, start, span);
3185 o = (U8*)SvPV(other, olen);
3188 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3190 s = (U8*)SvPV(swatch, slen);
3191 if (bits == 1 && otherbits == 1) {
3193 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3194 "mismatch, slen=%"UVuf", olen=%"UVuf,
3195 (UV)slen, (UV)olen);
3219 STRLEN otheroctets = otherbits >> 3;
3221 U8* const send = s + slen;
3226 if (otherbits == 1) {
3227 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3231 STRLEN vlen = otheroctets;
3239 if (opc == '+' && otherval)
3240 NOOP; /* replace with otherval */
3241 else if (opc == '!' && !otherval)
3243 else if (opc == '-' && otherval)
3245 else if (opc == '&' && !otherval)
3248 s += octets; /* no replacement */
3253 *s++ = (U8)( otherval & 0xff);
3254 else if (bits == 16) {
3255 *s++ = (U8)((otherval >> 8) & 0xff);
3256 *s++ = (U8)( otherval & 0xff);
3258 else if (bits == 32) {
3259 *s++ = (U8)((otherval >> 24) & 0xff);
3260 *s++ = (U8)((otherval >> 16) & 0xff);
3261 *s++ = (U8)((otherval >> 8) & 0xff);
3262 *s++ = (U8)( otherval & 0xff);
3266 sv_free(other); /* through with it! */
3272 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3275 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3276 * Can't be used on a property that is subject to user override, as it
3277 * relies on the value of SPECIALS in the swash which would be set by
3278 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3279 * for overridden properties
3281 * Returns a hash which is the inversion and closure of a swash mapping.
3282 * For example, consider the input lines:
3287 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
3288 * 006C. The value for each key is an array. For 006C, the array would
3289 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3290 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
3292 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3293 * keys are only code points that are folded-to, so it isn't a full closure.
3295 * Essentially, for any code point, it gives all the code points that map to
3296 * it, or the list of 'froms' for that point.
3298 * Currently it ignores any additions or deletions from other swashes,
3299 * looking at just the main body of the swash, and if there are SPECIALS
3300 * in the swash, at that hash
3302 * The specials hash can be extra code points, and most likely consists of
3303 * maps from single code points to multiple ones (each expressed as a string
3304 * of UTF-8 characters). This function currently returns only 1-1 mappings.
3305 * However consider this possible input in the specials hash:
3306 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3307 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3309 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3310 * currently handle. But it also means that FB05 and FB06 are equivalent in
3311 * a 1-1 mapping which we should handle, and this relationship may not be in
3312 * the main table. Therefore this function examines all the multi-char
3313 * sequences and adds the 1-1 mappings that come out of that.
3315 * XXX This function was originally intended to be multipurpose, but its
3316 * only use is quite likely to remain for constructing the inversion of
3317 * the CaseFolding (//i) property. If it were more general purpose for
3318 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3319 * because certain folds are prohibited under /iaa and /il. As an example,
3320 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3321 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3322 * prohibited, so we would not figure out that they fold to each other.
3323 * Code could be written to automatically figure this out, similar to the
3324 * code that does this for multi-character folds, but this is the only case
3325 * where something like this is ever likely to happen, as all the single
3326 * char folds to The 0-255 range are now quite settled. Instead there is a
3327 * little special code that is compiled only for this Unicode version. This
3328 * is smaller and didn't require much coding time to do. But this makes
3329 * this routine strongly tied to being used just for CaseFolding. If ever
3330 * it should be generalized, this would have to be fixed */
3334 HV *const hv = MUTABLE_HV(SvRV(swash));
3336 /* The string containing the main body of the table. This will have its
3337 * assertion fail if the swash has been converted to its inversion list */
3338 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3340 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3341 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3342 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3343 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3344 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3345 const STRLEN bits = SvUV(*bitssvp);
3346 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3347 const UV none = SvUV(*nonesvp);
3348 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3352 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3354 /* Must have at least 8 bits to get the mappings */
3355 if (bits != 8 && bits != 16 && bits != 32) {
3356 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3360 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3361 mapping to more than one character */
3363 /* Construct an inverse mapping hash for the specials */
3364 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3365 HV * specials_inverse = newHV();
3366 char *char_from; /* the lhs of the map */
3367 I32 from_len; /* its byte length */
3368 char *char_to; /* the rhs of the map */
3369 I32 to_len; /* its byte length */
3370 SV *sv_to; /* and in a sv */
3371 AV* from_list; /* list of things that map to each 'to' */
3373 hv_iterinit(specials_hv);
3375 /* The keys are the characters (in UTF-8) that map to the corresponding
3376 * UTF-8 string value. Iterate through the list creating the inverse
3378 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3380 if (! SvPOK(sv_to)) {
3381 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3382 "unexpectedly is not a string, flags=%lu",
3383 (unsigned long)SvFLAGS(sv_to));
3385 /*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)));*/
3387 /* Each key in the inverse list is a mapped-to value, and the key's
3388 * hash value is a list of the strings (each in UTF-8) that map to
3389 * it. Those strings are all one character long */
3390 if ((listp = hv_fetch(specials_inverse,
3394 from_list = (AV*) *listp;
3396 else { /* No entry yet for it: create one */
3397 from_list = newAV();
3398 if (! hv_store(specials_inverse,
3401 (SV*) from_list, 0))
3403 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3407 /* Here have the list associated with this 'to' (perhaps newly
3408 * created and empty). Just add to it. Note that we ASSUME that
3409 * the input is guaranteed to not have duplications, so we don't
3410 * check for that. Duplications just slow down execution time. */
3411 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3414 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3415 * it looking for cases like the FB05/FB06 examples above. There would
3416 * be an entry in the hash like
3417 * 'st' => [ FB05, FB06 ]
3418 * In this example we will create two lists that get stored in the
3419 * returned hash, 'ret':
3420 * FB05 => [ FB05, FB06 ]
3421 * FB06 => [ FB05, FB06 ]
3423 * Note that there is nothing to do if the array only has one element.
3424 * (In the normal 1-1 case handled below, we don't have to worry about
3425 * two lists, as everything gets tied to the single list that is
3426 * generated for the single character 'to'. But here, we are omitting
3427 * that list, ('st' in the example), so must have multiple lists.) */
3428 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3429 &char_to, &to_len)))
3431 if (av_tindex(from_list) > 0) {
3434 /* We iterate over all combinations of i,j to place each code
3435 * point on each list */
3436 for (i = 0; i <= av_tindex(from_list); i++) {
3438 AV* i_list = newAV();
3439 SV** entryp = av_fetch(from_list, i, FALSE);
3440 if (entryp == NULL) {
3441 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3443 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3444 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3446 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3447 (SV*) i_list, FALSE))
3449 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3452 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3453 for (j = 0; j <= av_tindex(from_list); j++) {
3454 entryp = av_fetch(from_list, j, FALSE);
3455 if (entryp == NULL) {
3456 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3459 /* When i==j this adds itself to the list */
3460 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3461 (U8*) SvPVX(*entryp),
3462 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3464 /*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));*/
3469 SvREFCNT_dec(specials_inverse); /* done with it */
3470 } /* End of specials */
3472 /* read $swash->{LIST} */
3474 #if UNICODE_MAJOR_VERSION == 3 \
3475 && UNICODE_DOT_VERSION == 0 \
3476 && UNICODE_DOT_DOT_VERSION == 1
3478 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3479 * rule so that things work under /iaa and /il */
3481 SV * mod_listsv = sv_mortalcopy(*listsvp);
3482 sv_catpv(mod_listsv, "130\t130\t131\n");
3483 l = (U8*)SvPV(mod_listsv, lcur);
3487 l = (U8*)SvPV(*listsvp, lcur);
3493 /* Go through each input line */
3497 l = swash_scan_list_line(l, lend, &min, &max, &val,
3498 cBOOL(octets), typestr);
3503 /* Each element in the range is to be inverted */
3504 for (inverse = min; inverse <= max; inverse++) {
3508 bool found_key = FALSE;
3509 bool found_inverse = FALSE;
3511 /* The key is the inverse mapping */
3512 char key[UTF8_MAXBYTES+1];
3513 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3514 STRLEN key_len = key_end - key;
3516 /* Get the list for the map */
3517 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3518 list = (AV*) *listp;
3520 else { /* No entry yet for it: create one */
3522 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3523 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3527 /* Look through list to see if this inverse mapping already is
3528 * listed, or if there is a mapping to itself already */
3529 for (i = 0; i <= av_tindex(list); i++) {
3530 SV** entryp = av_fetch(list, i, FALSE);
3533 if (entryp == NULL) {
3534 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3538 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3542 if (uv == inverse) {
3543 found_inverse = TRUE;
3546 /* No need to continue searching if found everything we are
3548 if (found_key && found_inverse) {
3553 /* Make sure there is a mapping to itself on the list */
3555 av_push(list, newSVuv(val));
3556 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3560 /* Simply add the value to the list */
3561 if (! found_inverse) {
3562 av_push(list, newSVuv(inverse));
3563 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3566 /* swatch_get() increments the value of val for each element in the
3567 * range. That makes more compact tables possible. You can
3568 * express the capitalization, for example, of all consecutive
3569 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3570 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3571 * and it's not documented; it appears to be used only in
3572 * implementing tr//; I copied the semantics from swatch_get(), just
3574 if (!none || val < none) {
3584 Perl__swash_to_invlist(pTHX_ SV* const swash)
3587 /* Subject to change or removal. For use only in one place in regcomp.c.
3588 * Ownership is given to one reference count in the returned SV* */
3593 HV *const hv = MUTABLE_HV(SvRV(swash));
3594 UV elements = 0; /* Number of elements in the inversion list */
3604 STRLEN octets; /* if bits == 1, then octets == 0 */
3610 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3612 /* If not a hash, it must be the swash's inversion list instead */
3613 if (SvTYPE(hv) != SVt_PVHV) {
3614 return SvREFCNT_inc_simple_NN((SV*) hv);
3617 /* The string containing the main body of the table */
3618 listsvp = hv_fetchs(hv, "LIST", FALSE);
3619 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3620 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3621 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3622 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3624 typestr = (U8*)SvPV_nolen(*typesvp);
3625 bits = SvUV(*bitssvp);
3626 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3628 /* read $swash->{LIST} */
3629 if (SvPOK(*listsvp)) {
3630 l = (U8*)SvPV(*listsvp, lcur);
3633 /* LIST legitimately doesn't contain a string during compilation phases
3634 * of Perl itself, before the Unicode tables are generated. In this
3635 * case, just fake things up by creating an empty list */
3642 if (*l == 'V') { /* Inversion list format */
3643 const char *after_atou = (char *) lend;
3645 UV* other_elements_ptr;
3647 /* The first number is a count of the rest */
3649 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
3650 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
3652 if (elements == 0) {
3653 invlist = _new_invlist(0);
3656 while (isSPACE(*l)) l++;
3657 l = (U8 *) after_atou;
3659 /* Get the 0th element, which is needed to setup the inversion list */
3660 while (isSPACE(*l)) l++;
3661 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
3662 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
3664 l = (U8 *) after_atou;
3665 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3668 /* Then just populate the rest of the input */
3669 while (elements-- > 0) {
3671 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3673 while (isSPACE(*l)) l++;
3674 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
3675 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
3677 l = (U8 *) after_atou;
3683 /* Scan the input to count the number of lines to preallocate array
3684 * size based on worst possible case, which is each line in the input
3685 * creates 2 elements in the inversion list: 1) the beginning of a
3686 * range in the list; 2) the beginning of a range not in the list. */
3687 while ((loc = (strchr(loc, '\n'))) != NULL) {
3692 /* If the ending is somehow corrupt and isn't a new line, add another
3693 * element for the final range that isn't in the inversion list */
3694 if (! (*lend == '\n'
3695 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3700 invlist = _new_invlist(elements);
3702 /* Now go through the input again, adding each range to the list */
3705 UV val; /* Not used by this function */
3707 l = swash_scan_list_line(l, lend, &start, &end, &val,
3708 cBOOL(octets), typestr);
3714 invlist = _add_range_to_invlist(invlist, start, end);
3718 /* Invert if the data says it should be */
3719 if (invert_it_svp && SvUV(*invert_it_svp)) {
3720 _invlist_invert(invlist);
3723 /* This code is copied from swatch_get()
3724 * read $swash->{EXTRAS} */
3725 x = (U8*)SvPV(*extssvp, xcur);
3733 SV **otherbitssvp, *other;
3736 const U8 opc = *x++;
3740 nl = (U8*)memchr(x, '\n', xend - x);
3742 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3744 x = nl + 1; /* 1 is length of "\n" */
3748 x = xend; /* to EXTRAS' end at which \n is not found */
3755 namelen = nl - namestr;
3759 namelen = xend - namestr;
3763 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3764 otherhv = MUTABLE_HV(SvRV(*othersvp));
3765 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3766 otherbits = (STRLEN)SvUV(*otherbitssvp);
3768 if (bits != otherbits || bits != 1) {
3769 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3770 "properties, bits=%"UVuf", otherbits=%"UVuf,
3771 (UV)bits, (UV)otherbits);
3774 /* The "other" swatch must be destroyed after. */
3775 other = _swash_to_invlist((SV *)*othersvp);
3777 /* End of code copied from swatch_get() */
3780 _invlist_union(invlist, other, &invlist);
3783 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3786 _invlist_subtract(invlist, other, &invlist);
3789 _invlist_intersection(invlist, other, &invlist);
3794 sv_free(other); /* through with it! */
3797 SvREADONLY_on(invlist);
3802 Perl__get_swash_invlist(pTHX_ SV* const swash)
3806 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3808 if (! SvROK(swash)) {
3812 /* If it really isn't a hash, it isn't really swash; must be an inversion
3814 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3818 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3827 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3829 /* May change: warns if surrogates, non-character code points, or
3830 * non-Unicode code points are in s which has length len bytes. Returns
3831 * TRUE if none found; FALSE otherwise. The only other validity check is
3832 * to make sure that this won't exceed the string's length */
3834 const U8* const e = s + len;
3837 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3840 if (UTF8SKIP(s) > len) {
3841 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3842 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3845 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
3847 if (UTF8_IS_SUPER(s, e)) {
3848 if (ckWARN_d(WARN_NON_UNICODE)) {
3849 /* A side effect of this function will be to warn */
3850 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_SUPER);
3854 else if (UTF8_IS_SURROGATE(s, e)) {
3855 if (ckWARN_d(WARN_SURROGATE)) {
3856 /* This has a different warning than the one the called
3857 * function would output, so can't just call it, unlike we
3858 * do for the non-chars and above-unicodes */
3859 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3860 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3861 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3865 else if ((UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
3866 /* A side effect of this function will be to warn */
3867 (void) utf8n_to_uvchr(s, e - s, &char_len, UTF8_WARN_NONCHAR);
3878 =for apidoc pv_uni_display
3880 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3881 length C<len>, the displayable version being at most C<pvlim> bytes long
3882 (if longer, the rest is truncated and C<"..."> will be appended).
3884 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
3885 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
3886 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
3887 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
3888 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
3889 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
3891 The pointer to the PV of the C<dsv> is returned.
3893 See also L</sv_uni_display>.
3897 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3902 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3906 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3908 /* This serves double duty as a flag and a character to print after
3909 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3913 if (pvlim && SvCUR(dsv) >= pvlim) {
3917 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3919 const unsigned char c = (unsigned char)u & 0xFF;
3920 if (flags & UNI_DISPLAY_BACKSLASH) {
3937 const char string = ok;
3938 sv_catpvs(dsv, "\\");
3939 sv_catpvn(dsv, &string, 1);
3942 /* isPRINT() is the locale-blind version. */
3943 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3944 const char string = c;
3945 sv_catpvn(dsv, &string, 1);
3950 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
3953 sv_catpvs(dsv, "...");
3959 =for apidoc sv_uni_display
3961 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
3962 the displayable version being at most C<pvlim> bytes long
3963 (if longer, the rest is truncated and "..." will be appended).
3965 The C<flags> argument is as in L</pv_uni_display>().
3967 The pointer to the PV of the C<dsv> is returned.
3972 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
3974 const char * const ptr =
3975 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
3977 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
3979 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
3980 SvCUR(ssv), pvlim, flags);
3984 =for apidoc foldEQ_utf8
3986 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
3987 of which may be in UTF-8) are the same case-insensitively; false otherwise.
3988 How far into the strings to compare is determined by other input parameters.
3990 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
3991 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
3992 with respect to C<s2>.
3994 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
3995 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
3996 scan will not be considered to be a match unless the goal is reached, and
3997 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4000 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
4001 considered an end pointer to the position 1 byte past the maximum point
4002 in C<s1> beyond which scanning will not continue under any circumstances.
4003 (This routine assumes that UTF-8 encoded input strings are not malformed;
4004 malformed input can cause it to read past C<pe1>).
4005 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4006 is less than C<s1>+C<l1>, the match will never be successful because it can
4008 get as far as its goal (and in fact is asserted against). Correspondingly for
4009 C<pe2> with respect to C<s2>.
4011 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4012 C<l2> must be non-zero), and if both do, both have to be
4013 reached for a successful match. Also, if the fold of a character is multiple
4014 characters, all of them must be matched (see tr21 reference below for
4017 Upon a successful match, if C<pe1> is non-C<NULL>,
4018 it will be set to point to the beginning of the I<next> character of C<s1>
4019 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4021 For case-insensitiveness, the "casefolding" of Unicode is used
4022 instead of upper/lowercasing both the characters, see
4023 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4027 /* A flags parameter has been added which may change, and hence isn't
4028 * externally documented. Currently it is:
4029 * 0 for as-documented above
4030 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4031 ASCII one, to not match
4032 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4033 * locale are to be used.
4034 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4035 * routine. This allows that step to be skipped.
4036 * Currently, this requires s1 to be encoded as UTF-8
4037 * (u1 must be true), which is asserted for.
4038 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4039 * cross certain boundaries. Hence, the caller should
4040 * let this function do the folding instead of
4041 * pre-folding. This code contains an assertion to
4042 * that effect. However, if the caller knows what
4043 * it's doing, it can pass this flag to indicate that,
4044 * and the assertion is skipped.
4045 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4046 * FOLDEQ_S2_FOLDS_SANE
4049 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)
4051 const U8 *p1 = (const U8*)s1; /* Point to current char */
4052 const U8 *p2 = (const U8*)s2;
4053 const U8 *g1 = NULL; /* goal for s1 */
4054 const U8 *g2 = NULL;
4055 const U8 *e1 = NULL; /* Don't scan s1 past this */
4056 U8 *f1 = NULL; /* Point to current folded */
4057 const U8 *e2 = NULL;
4059 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4060 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4061 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4062 U8 flags_for_folder = FOLD_FLAGS_FULL;
4064 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4066 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4067 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
4068 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4069 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
4070 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4071 /* The algorithm is to trial the folds without regard to the flags on
4072 * the first line of the above assert(), and then see if the result
4073 * violates them. This means that the inputs can't be pre-folded to a
4074 * violating result, hence the assert. This could be changed, with the
4075 * addition of extra tests here for the already-folded case, which would
4076 * slow it down. That cost is more than any possible gain for when these
4077 * flags are specified, as the flags indicate /il or /iaa matching which
4078 * is less common than /iu, and I (khw) also believe that real-world /il
4079 * and /iaa matches are most likely to involve code points 0-255, and this
4080 * function only under rare conditions gets called for 0-255. */
4082 if (flags & FOLDEQ_LOCALE) {
4083 if (IN_UTF8_CTYPE_LOCALE) {
4084 flags &= ~FOLDEQ_LOCALE;
4087 flags_for_folder |= FOLD_FLAGS_LOCALE;
4096 g1 = (const U8*)s1 + l1;
4104 g2 = (const U8*)s2 + l2;
4107 /* Must have at least one goal */
4112 /* Will never match if goal is out-of-bounds */
4113 assert(! e1 || e1 >= g1);
4115 /* Here, there isn't an end pointer, or it is beyond the goal. We
4116 * only go as far as the goal */
4120 assert(e1); /* Must have an end for looking at s1 */
4123 /* Same for goal for s2 */
4125 assert(! e2 || e2 >= g2);
4132 /* If both operands are already folded, we could just do a memEQ on the
4133 * whole strings at once, but it would be better if the caller realized
4134 * this and didn't even call us */
4136 /* Look through both strings, a character at a time */
4137 while (p1 < e1 && p2 < e2) {
4139 /* If at the beginning of a new character in s1, get its fold to use
4140 * and the length of the fold. */
4142 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4148 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4150 /* We have to forbid mixing ASCII with non-ASCII if the
4151 * flags so indicate. And, we can short circuit having to
4152 * call the general functions for this common ASCII case,
4153 * all of whose non-locale folds are also ASCII, and hence
4154 * UTF-8 invariants, so the UTF8ness of the strings is not
4156 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4160 *foldbuf1 = toFOLD(*p1);
4163 _to_utf8_fold_flags(p1, foldbuf1, &n1, flags_for_folder);
4165 else { /* Not UTF-8, get UTF-8 fold */
4166 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4172 if (n2 == 0) { /* Same for s2 */
4173 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4179 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4180 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4184 *foldbuf2 = toFOLD(*p2);
4187 _to_utf8_fold_flags(p2, foldbuf2, &n2, flags_for_folder);
4190 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4196 /* Here f1 and f2 point to the beginning of the strings to compare.
4197 * These strings are the folds of the next character from each input
4198 * string, stored in UTF-8. */
4200 /* While there is more to look for in both folds, see if they
4201 * continue to match */
4203 U8 fold_length = UTF8SKIP(f1);
4204 if (fold_length != UTF8SKIP(f2)
4205 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4206 function call for single
4208 || memNE((char*)f1, (char*)f2, fold_length))
4210 return 0; /* mismatch */
4213 /* Here, they matched, advance past them */
4220 /* When reach the end of any fold, advance the input past it */
4222 p1 += u1 ? UTF8SKIP(p1) : 1;
4225 p2 += u2 ? UTF8SKIP(p2) : 1;
4227 } /* End of loop through both strings */
4229 /* A match is defined by each scan that specified an explicit length
4230 * reaching its final goal, and the other not having matched a partial
4231 * character (which can happen when the fold of a character is more than one
4233 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4237 /* Successful match. Set output pointers */
4247 /* XXX The next two functions should likely be moved to mathoms.c once all
4248 * occurrences of them are removed from the core; some cpan-upstream modules
4252 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4254 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4256 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4260 =for apidoc utf8n_to_uvuni
4262 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4264 This function was useful for code that wanted to handle both EBCDIC and
4265 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4266 distinctions between the platforms have mostly been made invisible to most
4267 code, so this function is quite unlikely to be what you want. If you do need
4268 this precise functionality, use instead
4269 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4270 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4276 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4278 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4280 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4284 =for apidoc uvuni_to_utf8_flags
4286 Instead you almost certainly want to use L</uvchr_to_utf8> or
4287 L</uvchr_to_utf8_flags>.
4289 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4290 which itself, while not deprecated, should be used only in isolated
4291 circumstances. These functions were useful for code that wanted to handle
4292 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4293 v5.20, the distinctions between the platforms have mostly been made invisible
4294 to most code, so this function is quite unlikely to be what you want.
4300 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4302 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4304 return uvoffuni_to_utf8_flags(d, uv, flags);
4308 * ex: set ts=8 sts=4 sw=4 et: