3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "inline_invlist.c"
35 #include "charclass_invlists.h"
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
41 =head1 Unicode Support
42 These are various utility functions for manipulating UTF8-encoded
43 strings. For the uninitiated, this is a method of representing arbitrary
44 Unicode characters as a variable number of bytes, in such a way that
45 characters in the ASCII range are unmodified, and a zero byte never appears
46 within non-zero characters.
52 =for apidoc is_invariant_string
54 Returns true iff the first C<len> bytes of the string C<s> are the same
55 regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
56 EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
57 machines, all the ASCII characters and only the ASCII characters fit this
58 definition. On EBCDIC machines, the ASCII-range characters are invariant, but
59 so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
62 If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
63 use this option, that C<s> can't have embedded C<NUL> characters and has to
64 have a terminating C<NUL> byte).
66 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
72 Perl_is_invariant_string(const U8 *s, STRLEN len)
74 const U8* const send = s + (len ? len : strlen((const char *)s));
77 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
79 for (; x < send; ++x) {
80 if (!UTF8_IS_INVARIANT(*x))
88 =for apidoc uvoffuni_to_utf8_flags
90 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
91 Instead, B<Almost all code should use L</uvchr_to_utf8> or
92 L</uvchr_to_utf8_flags>>.
94 This function is like them, but the input is a strict Unicode
95 (as opposed to native) code point. Only in very rare circumstances should code
96 not be using the native code point.
98 For details, see the description for L</uvchr_to_utf8_flags>>.
104 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
106 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
108 if (UNI_IS_INVARIANT(uv)) {
109 *d++ = (U8) LATIN1_TO_NATIVE(uv);
114 /* Not representable in UTF-EBCDIC */
115 flags |= UNICODE_DISALLOW_FE_FF;
118 /* The first problematic code point is the first surrogate */
119 if (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);
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 illegal 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<UNISKIP(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<UNISKIP(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 UNICODE_WARN_SURROGATE is set,
292 the function will raise a warning, provided UTF8 warnings are enabled. If instead
293 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 UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
297 affect how the function handles a Unicode non-character. And likewise, the
298 UNICODE_WARN_SUPER and 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 UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
305 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
306 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
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_char_buf
324 This is identical to the macro L</isUTF8_CHAR>.
329 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
332 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
334 return isUTF8_CHAR(buf, buf_end);
338 =for apidoc is_utf8_string
340 Returns true if the first C<len> bytes of string C<s> form a valid
341 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
342 using C<strlen(s)> (which means if you use this option, that C<s> can't have
343 embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
344 that all characters being ASCII constitute 'a valid UTF-8 string'.
346 See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
352 Perl_is_utf8_string(const U8 *s, STRLEN len)
354 const U8* const send = s + (len ? len : strlen((const char *)s));
357 PERL_ARGS_ASSERT_IS_UTF8_STRING;
360 STRLEN len = isUTF8_CHAR(x, send);
361 if (UNLIKELY(! len)) {
371 Implemented as a macro in utf8.h
373 =for apidoc is_utf8_string_loc
375 Like L</is_utf8_string> but stores the location of the failure (in the
376 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
377 "utf8ness success") in the C<ep>.
379 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
381 =for apidoc is_utf8_string_loclen
383 Like L</is_utf8_string>() but stores the location of the failure (in the
384 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
385 "utf8ness success") in the C<ep>, and the number of UTF-8
386 encoded characters in the C<el>.
388 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
394 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
396 const U8* const send = s + (len ? len : strlen((const char *)s));
400 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
403 STRLEN len = isUTF8_CHAR(x, send);
404 if (UNLIKELY(! len)) {
422 =for apidoc utf8n_to_uvchr
424 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
425 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
427 Bottom level UTF-8 decode routine.
428 Returns the native code point value of the first character in the string C<s>,
429 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
430 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
431 the length, in bytes, of that character.
433 The value of C<flags> determines the behavior when C<s> does not point to a
434 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
435 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
436 next possible position in C<s> that could begin a non-malformed character.
437 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
439 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
440 individual types of malformations, such as the sequence being overlong (that
441 is, when there is a shorter sequence that can express the same code point;
442 overlong sequences are expressly forbidden in the UTF-8 standard due to
443 potential security issues). Another malformation example is the first byte of
444 a character not being a legal first byte. See F<utf8.h> for the list of such
445 flags. For allowed 0 length strings, this function returns 0; for allowed
446 overlong sequences, the computed code point is returned; for all other allowed
447 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
448 determinable reasonable value.
450 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
451 flags) malformation is found. If this flag is set, the routine assumes that
452 the caller will raise a warning, and this function will silently just set
453 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
455 Note that this API requires disambiguation between successful decoding a C<NUL>
456 character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
457 in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
458 first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
461 Certain code points are considered problematic. These are Unicode surrogates,
462 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
463 By default these are considered regular code points, but certain situations
464 warrant special handling for them. If C<flags> contains
465 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
466 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
467 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
468 maximum) can be set to disallow these categories individually.
470 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
471 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
472 for their respective categories, but otherwise the code points are considered
473 valid (not malformations). To get a category to both be treated as a
474 malformation and raise a warning, specify both the WARN and DISALLOW flags.
475 (But note that warnings are not raised if lexically disabled nor if
476 UTF8_CHECK_ONLY is also specified.)
478 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
479 the others that are above the Unicode legal maximum. There are several
480 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
481 not representable at all on EBCDIC platforms, and the original UTF-8
482 specification never went above this number (the current 0x10FFFF limit was
483 imposed later). (The smaller ones, those that fit into 32 bits, are
484 representable by a UV on ASCII platforms, but not by an IV, which means that
485 the number of operations that can be performed on them is quite restricted.)
486 The UTF-8 encoding on ASCII platforms for these large code points begins with a
487 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
488 be treated as malformations, while allowing smaller above-Unicode code points.
489 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
490 including these, as malformations.)
491 Similarly, UTF8_WARN_FE_FF acts just like
492 the other WARN flags, but applies just to these code points.
494 All other code points corresponding to Unicode characters, including private
495 use and those yet to be assigned, are never considered malformed and never
502 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
504 const U8 * const s0 = s;
505 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
510 UV outlier_ret = 0; /* return value when input is in error or problematic
512 UV pack_warn = 0; /* Save result of packWARN() for later */
513 bool unexpected_non_continuation = FALSE;
514 bool overflowed = FALSE;
515 bool do_overlong_test = TRUE; /* May have to skip this test */
517 const char* const malformed_text = "Malformed UTF-8 character";
519 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
521 /* The order of malformation tests here is important. We should consume as
522 * few bytes as possible in order to not skip any valid character. This is
523 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
524 * http://unicode.org/reports/tr36 for more discussion as to why. For
525 * example, once we've done a UTF8SKIP, we can tell the expected number of
526 * bytes, and could fail right off the bat if the input parameters indicate
527 * that there are too few available. But it could be that just that first
528 * byte is garbled, and the intended character occupies fewer bytes. If we
529 * blindly assumed that the first byte is correct, and skipped based on
530 * that number, we could skip over a valid input character. So instead, we
531 * always examine the sequence byte-by-byte.
533 * We also should not consume too few bytes, otherwise someone could inject
534 * things. For example, an input could be deliberately designed to
535 * overflow, and if this code bailed out immediately upon discovering that,
536 * returning to the caller C<*retlen> pointing to the very next byte (one
537 * which is actually part of of the overflowing sequence), that could look
538 * legitimate to the caller, which could discard the initial partial
539 * sequence and process the rest, inappropriately */
541 /* Zero length strings, if allowed, of necessity are zero */
542 if (UNLIKELY(curlen == 0)) {
547 if (flags & UTF8_ALLOW_EMPTY) {
550 if (! (flags & UTF8_CHECK_ONLY)) {
551 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
556 expectlen = UTF8SKIP(s);
558 /* A well-formed UTF-8 character, as the vast majority of calls to this
559 * function will be for, has this expected length. For efficiency, set
560 * things up here to return it. It will be overriden only in those rare
561 * cases where a malformation is found */
566 /* An invariant is trivially well-formed */
567 if (UTF8_IS_INVARIANT(uv)) {
571 /* A continuation character can't start a valid sequence */
572 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
573 if (flags & UTF8_ALLOW_CONTINUATION) {
577 return UNICODE_REPLACEMENT;
580 if (! (flags & UTF8_CHECK_ONLY)) {
581 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
587 /* Here is not a continuation byte, nor an invariant. The only thing left
588 * is a start byte (possibly for an overlong) */
591 uv = NATIVE_UTF8_TO_I8(uv);
594 /* Remove the leading bits that indicate the number of bytes in the
595 * character's whole UTF-8 sequence, leaving just the bits that are part of
597 uv &= UTF_START_MASK(expectlen);
599 /* Now, loop through the remaining bytes in the character's sequence,
600 * accumulating each into the working value as we go. Be sure to not look
601 * past the end of the input string */
602 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
604 for (s = s0 + 1; s < send; s++) {
605 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
606 #ifndef EBCDIC /* Can't overflow in EBCDIC */
607 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
609 /* The original implementors viewed this malformation as more
610 * serious than the others (though I, khw, don't understand
611 * why, since other malformations also give very very wrong
612 * results), so there is no way to turn off checking for it.
613 * Set a flag, but keep going in the loop, so that we absorb
614 * the rest of the bytes that comprise the character. */
616 overflow_byte = *s; /* Save for warning message's use */
619 uv = UTF8_ACCUMULATE(uv, *s);
622 /* Here, found a non-continuation before processing all expected
623 * bytes. This byte begins a new character, so quit, even if
624 * allowing this malformation. */
625 unexpected_non_continuation = TRUE;
628 } /* End of loop through the character's bytes */
630 /* Save how many bytes were actually in the character */
633 /* The loop above finds two types of malformations: non-continuation and/or
634 * overflow. The non-continuation malformation is really a too-short
635 * malformation, as it means that the current character ended before it was
636 * expected to (being terminated prematurely by the beginning of the next
637 * character, whereas in the too-short malformation there just are too few
638 * bytes available to hold the character. In both cases, the check below
639 * that we have found the expected number of bytes would fail if executed.)
640 * Thus the non-continuation malformation is really unnecessary, being a
641 * subset of the too-short malformation. But there may be existing
642 * applications that are expecting the non-continuation type, so we retain
643 * it, and return it in preference to the too-short malformation. (If this
644 * code were being written from scratch, the two types might be collapsed
645 * into one.) I, khw, am also giving priority to returning the
646 * non-continuation and too-short malformations over overflow when multiple
647 * ones are present. I don't know of any real reason to prefer one over
648 * the other, except that it seems to me that multiple-byte errors trumps
649 * errors from a single byte */
650 if (UNLIKELY(unexpected_non_continuation)) {
651 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
652 if (! (flags & UTF8_CHECK_ONLY)) {
654 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
657 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));
662 uv = UNICODE_REPLACEMENT;
664 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
665 * as what the original expectations were. */
666 do_overlong_test = FALSE;
671 else if (UNLIKELY(curlen < expectlen)) {
672 if (! (flags & UTF8_ALLOW_SHORT)) {
673 if (! (flags & UTF8_CHECK_ONLY)) {
674 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));
678 uv = UNICODE_REPLACEMENT;
679 do_overlong_test = FALSE;
685 #ifndef EBCDIC /* EBCDIC can't overflow */
686 if (UNLIKELY(overflowed)) {
687 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
693 && expectlen > (STRLEN) OFFUNISKIP(uv)
694 && ! (flags & UTF8_ALLOW_LONG))
696 /* The overlong malformation has lower precedence than the others.
697 * Note that if this malformation is allowed, we return the actual
698 * value, instead of the replacement character. This is because this
699 * value is actually well-defined. */
700 if (! (flags & UTF8_CHECK_ONLY)) {
701 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));
706 /* Here, the input is considered to be well-formed, but it still could be a
707 * problematic code point that is not allowed by the input parameters. */
708 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
709 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
710 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
712 if (UNICODE_IS_SURROGATE(uv)) {
714 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
715 * generation of the sv, since no warnings are raised under CHECK */
716 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
717 && ckWARN_d(WARN_SURROGATE))
719 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
720 pack_warn = packWARN(WARN_SURROGATE);
722 if (flags & UTF8_DISALLOW_SURROGATE) {
726 else if ((uv > PERL_UNICODE_MAX)) {
727 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
728 && ckWARN_d(WARN_NON_UNICODE))
730 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
731 pack_warn = packWARN(WARN_NON_UNICODE);
733 #ifndef EBCDIC /* EBCDIC always allows FE, FF */
735 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
736 * points. We test for these after the regular SUPER ones, and
737 * before possibly bailing out, so that the more dire warning
738 * overrides the regular one, if applicable */
739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
742 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
744 && ckWARN_d(WARN_UTF8))
746 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
747 pack_warn = packWARN(WARN_UTF8);
749 if (flags & UTF8_DISALLOW_FE_FF) {
754 if (flags & UTF8_DISALLOW_SUPER) {
758 else if (UNICODE_IS_NONCHAR(uv)) {
759 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
760 && ckWARN_d(WARN_NONCHAR))
762 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
763 pack_warn = packWARN(WARN_NONCHAR);
765 if (flags & UTF8_DISALLOW_NONCHAR) {
771 outlier_ret = uv; /* Note we don't bother to convert to native,
772 as all the outlier code points are the same
773 in both ASCII and EBCDIC */
777 /* Here, this is not considered a malformed character, so drop through
781 return UNI_TO_NATIVE(uv);
783 /* There are three cases which get to beyond this point. In all 3 cases:
784 * <sv> if not null points to a string to print as a warning.
785 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
787 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
788 * This is done by initializing it to 0, and changing it only
791 * 1) The input is valid but problematic, and to be warned about. The
792 * return value is the resultant code point; <*retlen> is set to
793 * <curlen>, the number of bytes that comprise the code point.
794 * <pack_warn> contains the result of packWARN() for the warning
795 * types. The entry point for this case is the label <do_warn>;
796 * 2) The input is a valid code point but disallowed by the parameters to
797 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
798 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
799 * comprise the code point. <pack_warn> contains the result of
800 * packWARN() for the warning types. The entry point for this case is
801 * the label <disallowed>.
802 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
803 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
804 * bytes that comprise the malformation. All such malformations are
805 * assumed to be warning type <utf8>. The entry point for this case
806 * is the label <malformed>.
811 if (sv && ckWARN_d(WARN_UTF8)) {
812 pack_warn = packWARN(WARN_UTF8);
817 if (flags & UTF8_CHECK_ONLY) {
819 *retlen = ((STRLEN) -1);
825 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
826 if warnings are to be raised. */
827 const char * const string = SvPVX_const(sv);
830 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
832 Perl_warner(aTHX_ pack_warn, "%s", string);
843 =for apidoc utf8_to_uvchr_buf
845 Returns the native code point of the first character in the string C<s> which
846 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
847 C<*retlen> will be set to the length, in bytes, of that character.
849 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
850 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
851 NULL) to -1. If those warnings are off, the computed value, if well-defined
852 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
853 C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
854 the next possible position in C<s> that could begin a non-malformed character.
855 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
863 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
867 return utf8n_to_uvchr(s, send - s, retlen,
868 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
871 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
872 * there are no malformations in the input UTF-8 string C<s>. surrogates,
873 * non-character code points, and non-Unicode code points are allowed. */
876 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
878 UV expectlen = UTF8SKIP(s);
879 const U8* send = s + expectlen;
882 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
889 /* An invariant is trivially returned */
890 if (expectlen == 1) {
895 uv = NATIVE_UTF8_TO_I8(uv);
898 /* Remove the leading bits that indicate the number of bytes, leaving just
899 * the bits that are part of the value */
900 uv &= UTF_START_MASK(expectlen);
902 /* Now, loop through the remaining bytes, accumulating each into the
903 * working total as we go. (I khw tried unrolling the loop for up to 4
904 * bytes, but there was no performance improvement) */
905 for (++s; s < send; s++) {
906 uv = UTF8_ACCUMULATE(uv, *s);
909 return UNI_TO_NATIVE(uv);
914 =for apidoc utf8_to_uvuni_buf
916 Only in very rare circumstances should code need to be dealing in Unicode
917 (as opposed to native) code points. In those few cases, use
918 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
920 Returns the Unicode (not-native) code point of the first character in the
922 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
923 C<retlen> will be set to the length, in bytes, of that character.
925 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
926 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
927 NULL) to -1. If those warnings are off, the computed value if well-defined (or
928 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
929 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
930 next possible position in C<s> that could begin a non-malformed character.
931 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
937 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
939 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
943 /* Call the low level routine asking for checks */
944 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
945 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
949 =for apidoc utf8_length
951 Return the length of the UTF-8 char encoded string C<s> in characters.
952 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
953 up past C<e>, croaks.
959 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
963 PERL_ARGS_ASSERT_UTF8_LENGTH;
965 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
966 * the bitops (especially ~) can create illegal UTF-8.
967 * In other words: in Perl UTF-8 is not just for Unicode. */
970 goto warn_and_return;
980 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
981 "%s in %s", unees, OP_DESC(PL_op));
983 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
990 =for apidoc utf8_distance
992 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
995 WARNING: use only if you *know* that the pointers point inside the
1002 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1004 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1006 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1010 =for apidoc utf8_hop
1012 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1013 forward or backward.
1015 WARNING: do not use the following unless you *know* C<off> is within
1016 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1017 on the first byte of character or just after the last byte of a character.
1023 Perl_utf8_hop(const U8 *s, I32 off)
1025 PERL_ARGS_ASSERT_UTF8_HOP;
1027 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1028 * the bitops (especially ~) can create illegal UTF-8.
1029 * In other words: in Perl UTF-8 is not just for Unicode. */
1038 while (UTF8_IS_CONTINUATION(*s))
1046 =for apidoc bytes_cmp_utf8
1048 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1049 sequence of characters (stored as UTF-8)
1050 in C<u>, C<ulen>. Returns 0 if they are
1051 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1052 if the first string is greater than the second string.
1054 -1 or +1 is returned if the shorter string was identical to the start of the
1055 longer string. -2 or +2 is returned if
1056 there was a difference between characters
1063 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1065 const U8 *const bend = b + blen;
1066 const U8 *const uend = u + ulen;
1068 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1070 while (b < bend && u < uend) {
1072 if (!UTF8_IS_INVARIANT(c)) {
1073 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1076 if (UTF8_IS_CONTINUATION(c1)) {
1077 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
1079 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1080 "Malformed UTF-8 character "
1081 "(unexpected non-continuation byte 0x%02x"
1082 ", immediately after start byte 0x%02x)"
1083 /* Dear diag.t, it's in the pod. */
1085 PL_op ? " in " : "",
1086 PL_op ? OP_DESC(PL_op) : "");
1091 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1092 "%s in %s", unees, OP_DESC(PL_op));
1094 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1095 return -2; /* Really want to return undef :-) */
1102 return *b < c ? -2 : +2;
1107 if (b == bend && u == uend)
1110 return b < bend ? +1 : -1;
1114 =for apidoc utf8_to_bytes
1116 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1117 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1118 updates C<len> to contain the new length.
1119 Returns zero on failure, setting C<len> to -1.
1121 If you need a copy of the string, see L</bytes_from_utf8>.
1127 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1129 U8 * const save = s;
1130 U8 * const send = s + *len;
1133 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1134 PERL_UNUSED_CONTEXT;
1136 /* ensure valid UTF-8 and chars < 256 before updating string */
1138 if (! UTF8_IS_INVARIANT(*s)) {
1139 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1140 *len = ((STRLEN) -1);
1151 if (! UTF8_IS_INVARIANT(c)) {
1152 /* Then it is two-byte encoded */
1153 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1164 =for apidoc bytes_from_utf8
1166 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1167 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1168 the newly-created string, and updates C<len> to contain the new
1169 length. Returns the original string if no conversion occurs, C<len>
1170 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1171 0 if C<s> is converted or consisted entirely of characters that are invariant
1172 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1178 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1181 const U8 *start = s;
1185 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1186 PERL_UNUSED_CONTEXT;
1190 /* ensure valid UTF-8 and chars < 256 before converting string */
1191 for (send = s + *len; s < send;) {
1192 if (! UTF8_IS_INVARIANT(*s)) {
1193 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1204 Newx(d, (*len) - count + 1, U8);
1205 s = start; start = d;
1208 if (! UTF8_IS_INVARIANT(c)) {
1209 /* Then it is two-byte encoded */
1210 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1221 =for apidoc bytes_to_utf8
1223 Converts a string C<s> of length C<len> bytes from the native encoding into
1225 Returns a pointer to the newly-created string, and sets C<len> to
1226 reflect the new length in bytes.
1228 A C<NUL> character will be written after the end of the string.
1230 If you want to convert to UTF-8 from encodings other than
1231 the native (Latin1 or EBCDIC),
1232 see L</sv_recode_to_utf8>().
1237 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1238 likewise need duplication. */
1241 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1243 const U8 * const send = s + (*len);
1247 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1248 PERL_UNUSED_CONTEXT;
1250 Newx(d, (*len) * 2 + 1, U8);
1254 append_utf8_from_native_byte(*s, &d);
1263 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1265 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1266 * We optimize for native, for obvious reasons. */
1269 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1274 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1277 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1282 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1284 if (UNI_IS_INVARIANT(uv)) {
1285 *d++ = LATIN1_TO_NATIVE((U8) uv);
1288 if (uv <= MAX_UTF8_TWO_BYTE) {
1289 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1290 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1293 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1294 #define LAST_HIGH_SURROGATE 0xDBFF
1295 #define FIRST_LOW_SURROGATE 0xDC00
1296 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1297 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
1299 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1301 UV low = (p[0] << 8) + p[1];
1303 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
1304 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1305 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1306 + (low - FIRST_LOW_SURROGATE) + 0x10000;
1308 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
1309 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1312 d = uvoffuni_to_utf8_flags(d, uv, 0);
1315 *d++ = (U8)(( uv >> 12) | 0xe0);
1316 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1317 *d++ = (U8)(( uv & 0x3f) | 0x80);
1321 *d++ = (U8)(( uv >> 18) | 0xf0);
1322 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1323 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1324 *d++ = (U8)(( uv & 0x3f) | 0x80);
1329 *newlen = d - dstart;
1333 /* Note: this one is slightly destructive of the source. */
1336 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1339 U8* const send = s + bytelen;
1341 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1344 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1348 const U8 tmp = s[0];
1353 return utf16_to_utf8(p, d, bytelen, newlen);
1357 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1359 U8 tmpbuf[UTF8_MAXBYTES+1];
1360 uvchr_to_utf8(tmpbuf, c);
1361 return _is_utf8_FOO(classnum, tmpbuf);
1364 /* Internal function so we can deprecate the external one, and call
1365 this one from other deprecated functions in this file */
1368 Perl__is_utf8_idstart(pTHX_ const U8 *p)
1370 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
1374 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
1378 Perl__is_uni_perl_idcont(pTHX_ UV c)
1380 U8 tmpbuf[UTF8_MAXBYTES+1];
1381 uvchr_to_utf8(tmpbuf, c);
1382 return _is_utf8_perl_idcont(tmpbuf);
1386 Perl__is_uni_perl_idstart(pTHX_ UV c)
1388 U8 tmpbuf[UTF8_MAXBYTES+1];
1389 uvchr_to_utf8(tmpbuf, c);
1390 return _is_utf8_perl_idstart(tmpbuf);
1394 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1396 /* We have the latin1-range values compiled into the core, so just use
1397 * those, converting the result to utf8. The only difference between upper
1398 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1399 * either "SS" or "Ss". Which one to use is passed into the routine in
1400 * 'S_or_s' to avoid a test */
1402 UV converted = toUPPER_LATIN1_MOD(c);
1404 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1406 assert(S_or_s == 'S' || S_or_s == 's');
1408 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
1409 characters in this range */
1410 *p = (U8) converted;
1415 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1416 * which it maps to one of them, so as to only have to have one check for
1417 * it in the main case */
1418 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1420 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1421 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1424 converted = GREEK_CAPITAL_LETTER_MU;
1426 case LATIN_SMALL_LETTER_SHARP_S:
1432 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1433 NOT_REACHED; /* NOTREACHED */
1437 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1438 *p = UTF8_TWO_BYTE_LO(converted);
1444 /* Call the function to convert a UTF-8 encoded character to the specified case.
1445 * Note that there may be more than one character in the result.
1446 * INP is a pointer to the first byte of the input character
1447 * OUTP will be set to the first byte of the string of changed characters. It
1448 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1449 * LENP will be set to the length in bytes of the string of changed characters
1451 * The functions return the ordinal of the first character in the string of OUTP */
1452 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1453 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1454 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
1456 /* This additionally has the input parameter SPECIALS, which if non-zero will
1457 * cause this to use the SPECIALS hash for folding (meaning get full case
1458 * folding); otherwise, when zero, this implies a simple case fold */
1459 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
1462 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1464 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1465 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1466 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1467 * the changed version may be longer than the original character.
1469 * The ordinal of the first character of the changed version is returned
1470 * (but note, as explained above, that there may be more.) */
1472 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1475 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1478 uvchr_to_utf8(p, c);
1479 return CALL_UPPER_CASE(p, p, lenp);
1483 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1485 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1488 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1491 uvchr_to_utf8(p, c);
1492 return CALL_TITLE_CASE(p, p, lenp);
1496 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
1498 /* We have the latin1-range values compiled into the core, so just use
1499 * those, converting the result to utf8. Since the result is always just
1500 * one character, we allow <p> to be NULL */
1502 U8 converted = toLOWER_LATIN1(c);
1505 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
1510 /* Result is known to always be < 256, so can use the EIGHT_BIT
1512 *p = UTF8_EIGHT_BIT_HI(converted);
1513 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
1521 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1523 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1526 return to_lower_latin1((U8) c, p, lenp);
1529 uvchr_to_utf8(p, c);
1530 return CALL_LOWER_CASE(p, p, lenp);
1534 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
1536 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1537 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1538 * FOLD_FLAGS_FULL iff full folding is to be used;
1540 * Not to be used for locale folds
1545 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1546 PERL_UNUSED_CONTEXT;
1548 assert (! (flags & FOLD_FLAGS_LOCALE));
1550 if (c == MICRO_SIGN) {
1551 converted = GREEK_SMALL_LETTER_MU;
1553 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1555 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1556 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1557 * under those circumstances. */
1558 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1559 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1560 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1562 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 /* Tread a UTF-8 locale as not being in locale at all */
1604 if (IN_UTF8_CTYPE_LOCALE) {
1605 flags &= ~FOLD_FLAGS_LOCALE;
1609 UV result = _to_fold_latin1((U8) c, p, lenp,
1610 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
1611 /* It is illegal for the fold to cross the 255/256 boundary under
1612 * locale; in this case return the original */
1613 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1618 /* If no special needs, just use the macro */
1619 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1620 uvchr_to_utf8(p, c);
1621 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1623 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1624 the special flags. */
1625 U8 utf8_c[UTF8_MAXBYTES + 1];
1626 uvchr_to_utf8(utf8_c, c);
1627 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
1631 PERL_STATIC_INLINE bool
1632 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1633 const char *const swashname, SV* const invlist)
1635 /* returns a boolean giving whether or not the UTF8-encoded character that
1636 * starts at <p> is in the swash indicated by <swashname>. <swash>
1637 * contains a pointer to where the swash indicated by <swashname>
1638 * is to be stored; which this routine will do, so that future calls will
1639 * look at <*swash> and only generate a swash if it is not null. <invlist>
1640 * is NULL or an inversion list that defines the swash. If not null, it
1641 * saves time during initialization of the swash.
1643 * Note that it is assumed that the buffer length of <p> is enough to
1644 * contain all the bytes that comprise the character. Thus, <*p> should
1645 * have been checked before this call for mal-formedness enough to assure
1648 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1650 /* The API should have included a length for the UTF-8 character in <p>,
1651 * but it doesn't. We therefore assume that p has been validated at least
1652 * as far as there being enough bytes available in it to accommodate the
1653 * character without reading beyond the end, and pass that number on to the
1654 * validating routine */
1655 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
1656 if (ckWARN_d(WARN_UTF8)) {
1657 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
1658 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
1659 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1660 what the malformation is */
1661 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1667 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1668 *swash = _core_swash_init("utf8",
1670 /* Only use the name if there is no inversion
1671 * list; otherwise will go out to disk */
1672 (invlist) ? "" : swashname,
1674 &PL_sv_undef, 1, 0, invlist, &flags);
1677 return swash_fetch(*swash, p, TRUE) != 0;
1681 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1683 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1685 assert(classnum < _FIRST_NON_SWASH_CC);
1687 return is_utf8_common(p,
1688 &PL_utf8_swash_ptrs[classnum],
1689 swash_property_names[classnum],
1690 PL_XPosix_ptrs[classnum]);
1694 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
1698 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
1700 if (! PL_utf8_perl_idstart) {
1701 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1703 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
1707 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
1709 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
1713 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
1717 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1721 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1723 if (! PL_utf8_perl_idcont) {
1724 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1726 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
1730 Perl__is_utf8_idcont(pTHX_ const U8 *p)
1732 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
1734 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
1738 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
1740 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
1742 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
1746 Perl__is_utf8_mark(pTHX_ const U8 *p)
1748 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1750 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
1754 =for apidoc to_utf8_case
1756 C<p> contains the pointer to the UTF-8 string encoding
1757 the character that is being converted. This routine assumes that the character
1758 at C<p> is well-formed.
1760 C<ustrp> is a pointer to the character buffer to put the
1761 conversion result to. C<lenp> is a pointer to the length
1764 C<swashp> is a pointer to the swash to use.
1766 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
1767 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
1768 but not always, a multicharacter mapping), is tried first.
1770 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1771 any special mappings; C<""> means to use the special mappings. Values other
1772 than these two are treated as the name of the hash containing the special
1773 mappings, like C<"utf8::ToSpecLower">.
1775 C<normal> is a string like "ToLower" which means the swash
1781 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1782 SV **swashp, const char *normal, const char *special)
1785 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
1787 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1789 /* Note that swash_fetch() doesn't output warnings for these because it
1790 * assumes we will */
1791 if (uv1 >= UNICODE_SURROGATE_FIRST) {
1792 if (uv1 <= UNICODE_SURROGATE_LAST) {
1793 if (ckWARN_d(WARN_SURROGATE)) {
1794 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1795 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1796 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1799 else if (UNICODE_IS_SUPER(uv1)) {
1800 if (ckWARN_d(WARN_NON_UNICODE)) {
1801 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1802 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1803 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1807 /* Note that non-characters are perfectly legal, so no warning should
1811 if (!*swashp) /* load on-demand */
1812 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
1815 /* It might be "special" (sometimes, but not always,
1816 * a multicharacter mapping) */
1820 /* If passed in the specials name, use that; otherwise use any
1821 * given in the swash */
1822 if (*special != '\0') {
1823 hv = get_hv(special, 0);
1826 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1828 hv = MUTABLE_HV(SvRV(*svp));
1833 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1838 s = SvPV_const(*svp, len);
1841 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
1843 Copy(s, ustrp, len, U8);
1848 if (!len && *swashp) {
1849 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
1852 /* It was "normal" (a single character mapping). */
1853 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
1861 return valid_utf8_to_uvchr(ustrp, 0);
1864 /* Here, there was no mapping defined, which means that the code point maps
1865 * to itself. Return the inputs */
1867 if (p != ustrp) { /* Don't copy onto itself */
1868 Copy(p, ustrp, len, U8);
1879 S_check_locale_boundary_crossing(pTHX_ const char * const func_name, const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
1881 /* This is called when changing the case of a utf8-encoded character above
1882 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1883 * result contains a character that crosses the 255/256 boundary, disallow
1884 * the change, and return the original code point. See L<perlfunc/lc> for
1887 * p points to the original string whose case was changed; assumed
1888 * by this routine to be well-formed
1889 * result the code point of the first character in the changed-case string
1890 * ustrp points to the changed-case string (<result> represents its first char)
1891 * lenp points to the length of <ustrp> */
1893 UV original; /* To store the first code point of <p> */
1895 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1897 assert(UTF8_IS_ABOVE_LATIN1(*p));
1899 /* We know immediately if the first character in the string crosses the
1900 * boundary, so can skip */
1903 /* Look at every character in the result; if any cross the
1904 * boundary, the whole thing is disallowed */
1905 U8* s = ustrp + UTF8SKIP(ustrp);
1906 U8* e = ustrp + *lenp;
1908 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
1914 /* Here, no characters crossed, result is ok as-is */
1920 /* Failed, have to return the original */
1921 original = valid_utf8_to_uvchr(p, lenp);
1923 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1924 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1925 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1926 "resolved to \"\\x{%"UVXf"}\".",
1930 Copy(p, ustrp, *lenp, char);
1935 =for apidoc to_utf8_upper
1937 Instead use L</toUPPER_utf8>.
1941 /* Not currently externally documented, and subject to change:
1942 * <flags> is set iff iff the rules from the current underlying locale are to
1946 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
1950 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
1952 if (flags && IN_UTF8_CTYPE_LOCALE) {
1956 if (UTF8_IS_INVARIANT(*p)) {
1958 result = toUPPER_LC(*p);
1961 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
1964 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
1966 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
1967 result = toUPPER_LC(c);
1970 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
1974 else { /* utf8, ord above 255 */
1975 result = CALL_UPPER_CASE(p, ustrp, lenp);
1978 result = check_locale_boundary_crossing("uc", p, result, ustrp, lenp);
1983 /* Here, used locale rules. Convert back to utf8 */
1984 if (UTF8_IS_INVARIANT(result)) {
1985 *ustrp = (U8) result;
1989 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1990 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
1998 =for apidoc to_utf8_title
2000 Instead use L</toTITLE_utf8>.
2004 /* Not currently externally documented, and subject to change:
2005 * <flags> is set iff the rules from the current underlying locale are to be
2006 * used. Since titlecase is not defined in POSIX, for other than a
2007 * UTF-8 locale, uppercase is used instead for code points < 256.
2011 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2015 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2017 if (flags && IN_UTF8_CTYPE_LOCALE) {
2021 if (UTF8_IS_INVARIANT(*p)) {
2023 result = toUPPER_LC(*p);
2026 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2029 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2031 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2032 result = toUPPER_LC(c);
2035 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2039 else { /* utf8, ord above 255 */
2040 result = CALL_TITLE_CASE(p, ustrp, lenp);
2043 result = check_locale_boundary_crossing("ucfirst", p, result, ustrp, lenp);
2048 /* Here, used locale rules. Convert back to utf8 */
2049 if (UTF8_IS_INVARIANT(result)) {
2050 *ustrp = (U8) result;
2054 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2055 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2063 =for apidoc to_utf8_lower
2065 Instead use L</toLOWER_utf8>.
2069 /* Not currently externally documented, and subject to change:
2070 * <flags> is set iff iff the rules from the current underlying locale are to
2075 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2079 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2081 if (flags && IN_UTF8_CTYPE_LOCALE) {
2085 if (UTF8_IS_INVARIANT(*p)) {
2087 result = toLOWER_LC(*p);
2090 return to_lower_latin1(*p, ustrp, lenp);
2093 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2095 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2096 result = toLOWER_LC(c);
2099 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2103 else { /* utf8, ord above 255 */
2104 result = CALL_LOWER_CASE(p, ustrp, lenp);
2107 result = check_locale_boundary_crossing("lc", p, result, ustrp, lenp);
2113 /* Here, used locale rules. Convert back to utf8 */
2114 if (UTF8_IS_INVARIANT(result)) {
2115 *ustrp = (U8) result;
2119 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2120 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2128 =for apidoc to_utf8_fold
2130 Instead use L</toFOLD_utf8>.
2134 /* Not currently externally documented, and subject to change,
2136 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2137 * locale are to be used.
2138 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2139 * otherwise simple folds
2140 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2145 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2149 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2151 /* These are mutually exclusive */
2152 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2154 assert(p != ustrp); /* Otherwise overwrites */
2156 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2157 flags &= ~FOLD_FLAGS_LOCALE;
2160 if (UTF8_IS_INVARIANT(*p)) {
2161 if (flags & FOLD_FLAGS_LOCALE) {
2162 result = toFOLD_LC(*p);
2165 return _to_fold_latin1(*p, ustrp, lenp,
2166 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2169 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2170 if (flags & FOLD_FLAGS_LOCALE) {
2171 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
2172 result = toFOLD_LC(c);
2175 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
2177 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2180 else { /* utf8, ord above 255 */
2181 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2183 if (flags & FOLD_FLAGS_LOCALE) {
2185 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2186 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2188 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2189 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2191 /* Special case these two characters, as what normally gets
2192 * returned under locale doesn't work */
2193 if (UTF8SKIP(p) == cap_sharp_s_len
2194 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
2196 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2197 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2198 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2199 "resolved to \"\\x{17F}\\x{17F}\".");
2202 else if (UTF8SKIP(p) == long_s_t_len
2203 && memEQ((char *) p, LONG_S_T, long_s_t_len))
2205 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2206 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2207 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2208 "resolved to \"\\x{FB06}\".");
2209 goto return_ligature_st;
2211 return check_locale_boundary_crossing("fc", p, result, ustrp, lenp);
2213 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2217 /* This is called when changing the case of a utf8-encoded
2218 * character above the ASCII range, and the result should not
2219 * contain an ASCII character. */
2221 UV original; /* To store the first code point of <p> */
2223 /* Look at every character in the result; if any cross the
2224 * boundary, the whole thing is disallowed */
2226 U8* e = ustrp + *lenp;
2229 /* Crossed, have to return the original */
2230 original = valid_utf8_to_uvchr(p, lenp);
2232 /* But in these instances, there is an alternative we can
2233 * return that is valid */
2234 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2235 || original == LATIN_SMALL_LETTER_SHARP_S)
2239 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2240 goto return_ligature_st;
2242 Copy(p, ustrp, *lenp, char);
2248 /* Here, no characters crossed, result is ok as-is */
2253 /* Here, used locale rules. Convert back to utf8 */
2254 if (UTF8_IS_INVARIANT(result)) {
2255 *ustrp = (U8) result;
2259 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2260 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2267 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2268 * folds to a string of two of these characters. By returning this
2269 * instead, then, e.g.,
2270 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2273 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2274 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2276 return LATIN_SMALL_LETTER_LONG_S;
2279 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2280 * have the other one fold to it */
2282 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2283 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2284 return LATIN_SMALL_LIGATURE_ST;
2288 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2289 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2290 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2294 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2296 PERL_ARGS_ASSERT_SWASH_INIT;
2298 /* Returns a copy of a swash initiated by the called function. This is the
2299 * public interface, and returning a copy prevents others from doing
2300 * mischief on the original */
2302 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2306 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2309 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2310 * use the following define */
2312 #define CORE_SWASH_INIT_RETURN(x) \
2313 PL_curpm= old_PL_curpm; \
2316 /* Initialize and return a swash, creating it if necessary. It does this
2317 * by calling utf8_heavy.pl in the general case. The returned value may be
2318 * the swash's inversion list instead if the input parameters allow it.
2319 * Which is returned should be immaterial to callers, as the only
2320 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2321 * and swash_to_invlist() handle both these transparently.
2323 * This interface should only be used by functions that won't destroy or
2324 * adversely change the swash, as doing so affects all other uses of the
2325 * swash in the program; the general public should use 'Perl_swash_init'
2328 * pkg is the name of the package that <name> should be in.
2329 * name is the name of the swash to find. Typically it is a Unicode
2330 * property name, including user-defined ones
2331 * listsv is a string to initialize the swash with. It must be of the form
2332 * documented as the subroutine return value in
2333 * L<perlunicode/User-Defined Character Properties>
2334 * minbits is the number of bits required to represent each data element.
2335 * It is '1' for binary properties.
2336 * none I (khw) do not understand this one, but it is used only in tr///.
2337 * invlist is an inversion list to initialize the swash with (or NULL)
2338 * flags_p if non-NULL is the address of various input and output flag bits
2339 * to the routine, as follows: ('I' means is input to the routine;
2340 * 'O' means output from the routine. Only flags marked O are
2341 * meaningful on return.)
2342 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2343 * came from a user-defined property. (I O)
2344 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2345 * when the swash cannot be located, to simply return NULL. (I)
2346 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2347 * return of an inversion list instead of a swash hash if this routine
2348 * thinks that would result in faster execution of swash_fetch() later
2351 * Thus there are three possible inputs to find the swash: <name>,
2352 * <listsv>, and <invlist>. At least one must be specified. The result
2353 * will be the union of the specified ones, although <listsv>'s various
2354 * actions can intersect, etc. what <name> gives. To avoid going out to
2355 * disk at all, <invlist> should specify completely what the swash should
2356 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
2358 * <invlist> is only valid for binary properties */
2360 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2362 SV* retval = &PL_sv_undef;
2363 HV* swash_hv = NULL;
2364 const int invlist_swash_boundary =
2365 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2366 ? 512 /* Based on some benchmarking, but not extensive, see commit
2368 : -1; /* Never return just an inversion list */
2370 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2371 assert(! invlist || minbits == 1);
2373 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2374 that triggered the swash init and the swash init perl logic itself.
2377 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2379 if (listsv != &PL_sv_undef || strNE(name, "")) {
2381 const size_t pkg_len = strlen(pkg);
2382 const size_t name_len = strlen(name);
2383 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2387 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2389 PUSHSTACKi(PERLSI_MAGIC);
2392 /* We might get here via a subroutine signature which uses a utf8
2393 * parameter name, at which point PL_subname will have been set
2394 * but not yet used. */
2395 save_item(PL_subname);
2396 if (PL_parser && PL_parser->error_count)
2397 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2398 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2399 if (!method) { /* demand load utf8 */
2401 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2402 GvSV(PL_errgv) = NULL;
2403 #ifndef NO_TAINT_SUPPORT
2404 /* It is assumed that callers of this routine are not passing in
2405 * any user derived data. */
2406 SAVEBOOL(TAINT_get);
2409 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2412 /* Not ERRSV, as there is no need to vivify a scalar we are
2413 about to discard. */
2414 SV * const errsv = GvSV(PL_errgv);
2415 if (!SvTRUE(errsv)) {
2416 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2417 SvREFCNT_dec(errsv);
2425 mPUSHp(pkg, pkg_len);
2426 mPUSHp(name, name_len);
2431 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2432 GvSV(PL_errgv) = NULL;
2433 /* If we already have a pointer to the method, no need to use
2434 * call_method() to repeat the lookup. */
2436 ? call_sv(MUTABLE_SV(method), G_SCALAR)
2437 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2439 retval = *PL_stack_sp--;
2440 SvREFCNT_inc(retval);
2443 /* Not ERRSV. See above. */
2444 SV * const errsv = GvSV(PL_errgv);
2445 if (!SvTRUE(errsv)) {
2446 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2447 SvREFCNT_dec(errsv);
2452 if (IN_PERL_COMPILETIME) {
2453 CopHINTS_set(PL_curcop, PL_hints);
2455 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2458 /* If caller wants to handle missing properties, let them */
2459 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2460 CORE_SWASH_INIT_RETURN(NULL);
2463 "Can't find Unicode property definition \"%"SVf"\"",
2465 NOT_REACHED; /* NOTREACHED */
2467 } /* End of calling the module to find the swash */
2469 /* If this operation fetched a swash, and we will need it later, get it */
2470 if (retval != &PL_sv_undef
2471 && (minbits == 1 || (flags_p
2473 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2475 swash_hv = MUTABLE_HV(SvRV(retval));
2477 /* If we don't already know that there is a user-defined component to
2478 * this swash, and the user has indicated they wish to know if there is
2479 * one (by passing <flags_p>), find out */
2480 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2481 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2482 if (user_defined && SvUV(*user_defined)) {
2483 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2488 /* Make sure there is an inversion list for binary properties */
2490 SV** swash_invlistsvp = NULL;
2491 SV* swash_invlist = NULL;
2492 bool invlist_in_swash_is_valid = FALSE;
2493 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2494 an unclaimed reference count */
2496 /* If this operation fetched a swash, get its already existing
2497 * inversion list, or create one for it */
2500 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2501 if (swash_invlistsvp) {
2502 swash_invlist = *swash_invlistsvp;
2503 invlist_in_swash_is_valid = TRUE;
2506 swash_invlist = _swash_to_invlist(retval);
2507 swash_invlist_unclaimed = TRUE;
2511 /* If an inversion list was passed in, have to include it */
2514 /* Any fetched swash will by now have an inversion list in it;
2515 * otherwise <swash_invlist> will be NULL, indicating that we
2516 * didn't fetch a swash */
2517 if (swash_invlist) {
2519 /* Add the passed-in inversion list, which invalidates the one
2520 * already stored in the swash */
2521 invlist_in_swash_is_valid = FALSE;
2522 _invlist_union(invlist, swash_invlist, &swash_invlist);
2526 /* Here, there is no swash already. Set up a minimal one, if
2527 * we are going to return a swash */
2528 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2530 retval = newRV_noinc(MUTABLE_SV(swash_hv));
2532 swash_invlist = invlist;
2536 /* Here, we have computed the union of all the passed-in data. It may
2537 * be that there was an inversion list in the swash which didn't get
2538 * touched; otherwise save the computed one */
2539 if (! invlist_in_swash_is_valid
2540 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2542 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
2544 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2546 /* We just stole a reference count. */
2547 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2548 else SvREFCNT_inc_simple_void_NN(swash_invlist);
2551 SvREADONLY_on(swash_invlist);
2553 /* Use the inversion list stand-alone if small enough */
2554 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2555 SvREFCNT_dec(retval);
2556 if (!swash_invlist_unclaimed)
2557 SvREFCNT_inc_simple_void_NN(swash_invlist);
2558 retval = newRV_noinc(swash_invlist);
2562 CORE_SWASH_INIT_RETURN(retval);
2563 #undef CORE_SWASH_INIT_RETURN
2567 /* This API is wrong for special case conversions since we may need to
2568 * return several Unicode characters for a single Unicode character
2569 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2570 * the lower-level routine, and it is similarly broken for returning
2571 * multiple values. --jhi
2572 * For those, you should use to_utf8_case() instead */
2573 /* Now SWASHGET is recasted into S_swatch_get in this file. */
2576 * Returns the value of property/mapping C<swash> for the first character
2577 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2578 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2579 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2581 * A "swash" is a hash which contains initially the keys/values set up by
2582 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2583 * property for all possible code points. Things are stored in a compact form
2584 * (see utf8_heavy.pl) so that calculation is required to find the actual
2585 * property value for a given code point. As code points are looked up, new
2586 * key/value pairs are added to the hash, so that the calculation doesn't have
2587 * to ever be re-done. Further, each calculation is done, not just for the
2588 * desired one, but for a whole block of code points adjacent to that one.
2589 * For binary properties on ASCII machines, the block is usually for 64 code
2590 * points, starting with a code point evenly divisible by 64. Thus if the
2591 * property value for code point 257 is requested, the code goes out and
2592 * calculates the property values for all 64 code points between 256 and 319,
2593 * and stores these as a single 64-bit long bit vector, called a "swatch",
2594 * under the key for code point 256. The key is the UTF-8 encoding for code
2595 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2596 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2597 * for code point 258 is then requested, this code realizes that it would be
2598 * stored under the key for 256, and would find that value and extract the
2599 * relevant bit, offset from 256.
2601 * Non-binary properties are stored in as many bits as necessary to represent
2602 * their values (32 currently, though the code is more general than that), not
2603 * as single bits, but the principal is the same: the value for each key is a
2604 * vector that encompasses the property values for all code points whose UTF-8
2605 * representations are represented by the key. That is, for all code points
2606 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2610 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2612 HV *const hv = MUTABLE_HV(SvRV(swash));
2617 const U8 *tmps = NULL;
2621 PERL_ARGS_ASSERT_SWASH_FETCH;
2623 /* If it really isn't a hash, it isn't really swash; must be an inversion
2625 if (SvTYPE(hv) != SVt_PVHV) {
2626 return _invlist_contains_cp((SV*)hv,
2628 ? valid_utf8_to_uvchr(ptr, NULL)
2632 /* We store the values in a "swatch" which is a vec() value in a swash
2633 * hash. Code points 0-255 are a single vec() stored with key length
2634 * (klen) 0. All other code points have a UTF-8 representation
2635 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2636 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2637 * length for them is the length of the encoded char - 1. ptr[klen] is the
2638 * final byte in the sequence representing the character */
2639 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2644 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2647 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
2650 klen = UTF8SKIP(ptr) - 1;
2652 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2653 * the vec is the final byte in the sequence. (In EBCDIC this is
2654 * converted to I8 to get consecutive values.) To help you visualize
2656 * Straight 1047 After final byte
2657 * UTF-8 UTF-EBCDIC I8 transform
2658 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2659 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2661 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2662 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2664 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2665 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2667 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2668 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2670 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2671 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2673 * (There are no discontinuities in the elided (...) entries.)
2674 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2675 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2676 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2677 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2678 * index into the vec() swatch (after subtracting 0x80, which we
2679 * actually do with an '&').
2680 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2681 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2682 * dicontinuities which go away by transforming it into I8, and we
2683 * effectively subtract 0xA0 to get the index. */
2684 needents = (1 << UTF_ACCUMULATION_SHIFT);
2685 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
2689 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2690 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2691 * it's nothing to sniff at.) Pity we usually come through at least
2692 * two function calls to get here...
2694 * NB: this code assumes that swatches are never modified, once generated!
2697 if (hv == PL_last_swash_hv &&
2698 klen == PL_last_swash_klen &&
2699 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2701 tmps = PL_last_swash_tmps;
2702 slen = PL_last_swash_slen;
2705 /* Try our second-level swatch cache, kept in a hash. */
2706 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2708 /* If not cached, generate it via swatch_get */
2709 if (!svp || !SvPOK(*svp)
2710 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2713 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2714 swatch = swatch_get(swash,
2715 code_point & ~((UV)needents - 1),
2718 else { /* For the first 256 code points, the swatch has a key of
2720 swatch = swatch_get(swash, 0, needents);
2723 if (IN_PERL_COMPILETIME)
2724 CopHINTS_set(PL_curcop, PL_hints);
2726 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2728 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2729 || (slen << 3) < needents)
2730 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2731 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2732 svp, tmps, (UV)slen, (UV)needents);
2735 PL_last_swash_hv = hv;
2736 assert(klen <= sizeof(PL_last_swash_key));
2737 PL_last_swash_klen = (U8)klen;
2738 /* FIXME change interpvar.h? */
2739 PL_last_swash_tmps = (U8 *) tmps;
2740 PL_last_swash_slen = slen;
2742 Copy(ptr, PL_last_swash_key, klen, U8);
2745 switch ((int)((slen << 3) / needents)) {
2747 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
2749 return ((UV) tmps[off]);
2753 ((UV) tmps[off ] << 8) +
2754 ((UV) tmps[off + 1]);
2758 ((UV) tmps[off ] << 24) +
2759 ((UV) tmps[off + 1] << 16) +
2760 ((UV) tmps[off + 2] << 8) +
2761 ((UV) tmps[off + 3]);
2763 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2764 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
2765 NORETURN_FUNCTION_END;
2768 /* Read a single line of the main body of the swash input text. These are of
2771 * where each number is hex. The first two numbers form the minimum and
2772 * maximum of a range, and the third is the value associated with the range.
2773 * Not all swashes should have a third number
2775 * On input: l points to the beginning of the line to be examined; it points
2776 * to somewhere in the string of the whole input text, and is
2777 * terminated by a \n or the null string terminator.
2778 * lend points to the null terminator of that string
2779 * wants_value is non-zero if the swash expects a third number
2780 * typestr is the name of the swash's mapping, like 'ToLower'
2781 * On output: *min, *max, and *val are set to the values read from the line.
2782 * returns a pointer just beyond the line examined. If there was no
2783 * valid min number on the line, returns lend+1
2787 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2788 const bool wants_value, const U8* const typestr)
2790 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2791 STRLEN numlen; /* Length of the number */
2792 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2793 | PERL_SCAN_DISALLOW_PREFIX
2794 | PERL_SCAN_SILENT_NON_PORTABLE;
2796 /* nl points to the next \n in the scan */
2797 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2799 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2801 /* Get the first number on the line: the range minimum */
2803 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2804 *max = *min; /* So can never return without setting max */
2805 if (numlen) /* If found a hex number, position past it */
2807 else if (nl) { /* Else, go handle next line, if any */
2808 return nl + 1; /* 1 is length of "\n" */
2810 else { /* Else, no next line */
2811 return lend + 1; /* to LIST's end at which \n is not found */
2814 /* The max range value follows, separated by a BLANK */
2817 flags = PERL_SCAN_SILENT_ILLDIGIT
2818 | PERL_SCAN_DISALLOW_PREFIX
2819 | PERL_SCAN_SILENT_NON_PORTABLE;
2821 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2824 else /* If no value here, it is a single element range */
2827 /* Non-binary tables have a third entry: what the first element of the
2828 * range maps to. The map for those currently read here is in hex */
2832 flags = PERL_SCAN_SILENT_ILLDIGIT
2833 | PERL_SCAN_DISALLOW_PREFIX
2834 | PERL_SCAN_SILENT_NON_PORTABLE;
2836 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2845 /* diag_listed_as: To%s: illegal mapping '%s' */
2846 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2852 *val = 0; /* bits == 1, then any val should be ignored */
2854 else { /* Nothing following range min, should be single element with no
2859 /* diag_listed_as: To%s: illegal mapping '%s' */
2860 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2864 *val = 0; /* bits == 1, then val should be ignored */
2867 /* Position to next line if any, or EOF */
2877 * Returns a swatch (a bit vector string) for a code point sequence
2878 * that starts from the value C<start> and comprises the number C<span>.
2879 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2880 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2883 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
2886 U8 *l, *lend, *x, *xend, *s, *send;
2887 STRLEN lcur, xcur, scur;
2888 HV *const hv = MUTABLE_HV(SvRV(swash));
2889 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
2891 SV** listsvp = NULL; /* The string containing the main body of the table */
2892 SV** extssvp = NULL;
2893 SV** invert_it_svp = NULL;
2896 STRLEN octets; /* if bits == 1, then octets == 0 */
2898 UV end = start + span;
2900 if (invlistsvp == NULL) {
2901 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2902 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2903 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2904 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2905 listsvp = hv_fetchs(hv, "LIST", FALSE);
2906 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2908 bits = SvUV(*bitssvp);
2909 none = SvUV(*nonesvp);
2910 typestr = (U8*)SvPV_nolen(*typesvp);
2916 octets = bits >> 3; /* if bits == 1, then octets == 0 */
2918 PERL_ARGS_ASSERT_SWATCH_GET;
2920 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
2921 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
2925 /* If overflowed, use the max possible */
2931 /* create and initialize $swatch */
2932 scur = octets ? (span * octets) : (span + 7) / 8;
2933 swatch = newSV(scur);
2935 s = (U8*)SvPVX(swatch);
2936 if (octets && none) {
2937 const U8* const e = s + scur;
2940 *s++ = (U8)(none & 0xff);
2941 else if (bits == 16) {
2942 *s++ = (U8)((none >> 8) & 0xff);
2943 *s++ = (U8)( none & 0xff);
2945 else if (bits == 32) {
2946 *s++ = (U8)((none >> 24) & 0xff);
2947 *s++ = (U8)((none >> 16) & 0xff);
2948 *s++ = (U8)((none >> 8) & 0xff);
2949 *s++ = (U8)( none & 0xff);
2955 (void)memzero((U8*)s, scur + 1);
2957 SvCUR_set(swatch, scur);
2958 s = (U8*)SvPVX(swatch);
2960 if (invlistsvp) { /* If has an inversion list set up use that */
2961 _invlist_populate_swatch(*invlistsvp, start, end, s);
2965 /* read $swash->{LIST} */
2966 l = (U8*)SvPV(*listsvp, lcur);
2969 UV min, max, val, upper;
2970 l = swash_scan_list_line(l, lend, &min, &max, &val,
2971 cBOOL(octets), typestr);
2976 /* If looking for something beyond this range, go try the next one */
2980 /* <end> is generally 1 beyond where we want to set things, but at the
2981 * platform's infinity, where we can't go any higher, we want to
2982 * include the code point at <end> */
2985 : (max != UV_MAX || end != UV_MAX)
2992 if (!none || val < none) {
2997 for (key = min; key <= upper; key++) {
2999 /* offset must be non-negative (start <= min <= key < end) */
3000 offset = octets * (key - start);
3002 s[offset] = (U8)(val & 0xff);
3003 else if (bits == 16) {
3004 s[offset ] = (U8)((val >> 8) & 0xff);
3005 s[offset + 1] = (U8)( val & 0xff);
3007 else if (bits == 32) {
3008 s[offset ] = (U8)((val >> 24) & 0xff);
3009 s[offset + 1] = (U8)((val >> 16) & 0xff);
3010 s[offset + 2] = (U8)((val >> 8) & 0xff);
3011 s[offset + 3] = (U8)( val & 0xff);
3014 if (!none || val < none)
3018 else { /* bits == 1, then val should be ignored */
3023 for (key = min; key <= upper; key++) {
3024 const STRLEN offset = (STRLEN)(key - start);
3025 s[offset >> 3] |= 1 << (offset & 7);
3030 /* Invert if the data says it should be. Assumes that bits == 1 */
3031 if (invert_it_svp && SvUV(*invert_it_svp)) {
3033 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3034 * be 0, and their inversion should also be 0, as we don't succeed any
3035 * Unicode property matches for non-Unicode code points */
3036 if (start <= PERL_UNICODE_MAX) {
3038 /* The code below assumes that we never cross the
3039 * Unicode/above-Unicode boundary in a range, as otherwise we would
3040 * have to figure out where to stop flipping the bits. Since this
3041 * boundary is divisible by a large power of 2, and swatches comes
3042 * in small powers of 2, this should be a valid assumption */
3043 assert(start + span - 1 <= PERL_UNICODE_MAX);
3053 /* read $swash->{EXTRAS}
3054 * This code also copied to swash_to_invlist() below */
3055 x = (U8*)SvPV(*extssvp, xcur);
3063 SV **otherbitssvp, *other;
3067 const U8 opc = *x++;
3071 nl = (U8*)memchr(x, '\n', xend - x);
3073 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3075 x = nl + 1; /* 1 is length of "\n" */
3079 x = xend; /* to EXTRAS' end at which \n is not found */
3086 namelen = nl - namestr;
3090 namelen = xend - namestr;
3094 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3095 otherhv = MUTABLE_HV(SvRV(*othersvp));
3096 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3097 otherbits = (STRLEN)SvUV(*otherbitssvp);
3098 if (bits < otherbits)
3099 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3100 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3102 /* The "other" swatch must be destroyed after. */
3103 other = swatch_get(*othersvp, start, span);
3104 o = (U8*)SvPV(other, olen);
3107 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3109 s = (U8*)SvPV(swatch, slen);
3110 if (bits == 1 && otherbits == 1) {
3112 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3113 "mismatch, slen=%"UVuf", olen=%"UVuf,
3114 (UV)slen, (UV)olen);
3138 STRLEN otheroctets = otherbits >> 3;
3140 U8* const send = s + slen;
3145 if (otherbits == 1) {
3146 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3150 STRLEN vlen = otheroctets;
3158 if (opc == '+' && otherval)
3159 NOOP; /* replace with otherval */
3160 else if (opc == '!' && !otherval)
3162 else if (opc == '-' && otherval)
3164 else if (opc == '&' && !otherval)
3167 s += octets; /* no replacement */
3172 *s++ = (U8)( otherval & 0xff);
3173 else if (bits == 16) {
3174 *s++ = (U8)((otherval >> 8) & 0xff);
3175 *s++ = (U8)( otherval & 0xff);
3177 else if (bits == 32) {
3178 *s++ = (U8)((otherval >> 24) & 0xff);
3179 *s++ = (U8)((otherval >> 16) & 0xff);
3180 *s++ = (U8)((otherval >> 8) & 0xff);
3181 *s++ = (U8)( otherval & 0xff);
3185 sv_free(other); /* through with it! */
3191 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3194 /* Subject to change or removal. For use only in regcomp.c and regexec.c
3195 * Can't be used on a property that is subject to user override, as it
3196 * relies on the value of SPECIALS in the swash which would be set by
3197 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3198 * for overridden properties
3200 * Returns a hash which is the inversion and closure of a swash mapping.
3201 * For example, consider the input lines:
3206 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3207 * 006C. The value for each key is an array. For 006C, the array would
3208 * have two elements, the utf8 for itself, and for 004C. For 006B, there
3209 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3211 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3212 * keys are only code points that are folded-to, so it isn't a full closure.
3214 * Essentially, for any code point, it gives all the code points that map to
3215 * it, or the list of 'froms' for that point.
3217 * Currently it ignores any additions or deletions from other swashes,
3218 * looking at just the main body of the swash, and if there are SPECIALS
3219 * in the swash, at that hash
3221 * The specials hash can be extra code points, and most likely consists of
3222 * maps from single code points to multiple ones (each expressed as a string
3223 * of utf8 characters). This function currently returns only 1-1 mappings.
3224 * However consider this possible input in the specials hash:
3225 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3226 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3228 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3229 * currently handle. But it also means that FB05 and FB06 are equivalent in
3230 * a 1-1 mapping which we should handle, and this relationship may not be in
3231 * the main table. Therefore this function examines all the multi-char
3232 * sequences and adds the 1-1 mappings that come out of that. */
3236 HV *const hv = MUTABLE_HV(SvRV(swash));
3238 /* The string containing the main body of the table. This will have its
3239 * assertion fail if the swash has been converted to its inversion list */
3240 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3242 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3243 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3244 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3245 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3246 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3247 const STRLEN bits = SvUV(*bitssvp);
3248 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3249 const UV none = SvUV(*nonesvp);
3250 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3254 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3256 /* Must have at least 8 bits to get the mappings */
3257 if (bits != 8 && bits != 16 && bits != 32) {
3258 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3262 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3263 mapping to more than one character */
3265 /* Construct an inverse mapping hash for the specials */
3266 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3267 HV * specials_inverse = newHV();
3268 char *char_from; /* the lhs of the map */
3269 I32 from_len; /* its byte length */
3270 char *char_to; /* the rhs of the map */
3271 I32 to_len; /* its byte length */
3272 SV *sv_to; /* and in a sv */
3273 AV* from_list; /* list of things that map to each 'to' */
3275 hv_iterinit(specials_hv);
3277 /* The keys are the characters (in utf8) that map to the corresponding
3278 * utf8 string value. Iterate through the list creating the inverse
3280 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3282 if (! SvPOK(sv_to)) {
3283 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3284 "unexpectedly is not a string, flags=%lu",
3285 (unsigned long)SvFLAGS(sv_to));
3287 /*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)));*/
3289 /* Each key in the inverse list is a mapped-to value, and the key's
3290 * hash value is a list of the strings (each in utf8) that map to
3291 * it. Those strings are all one character long */
3292 if ((listp = hv_fetch(specials_inverse,
3296 from_list = (AV*) *listp;
3298 else { /* No entry yet for it: create one */
3299 from_list = newAV();
3300 if (! hv_store(specials_inverse,
3303 (SV*) from_list, 0))
3305 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3309 /* Here have the list associated with this 'to' (perhaps newly
3310 * created and empty). Just add to it. Note that we ASSUME that
3311 * the input is guaranteed to not have duplications, so we don't
3312 * check for that. Duplications just slow down execution time. */
3313 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3316 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3317 * it looking for cases like the FB05/FB06 examples above. There would
3318 * be an entry in the hash like
3319 * 'st' => [ FB05, FB06 ]
3320 * In this example we will create two lists that get stored in the
3321 * returned hash, 'ret':
3322 * FB05 => [ FB05, FB06 ]
3323 * FB06 => [ FB05, FB06 ]
3325 * Note that there is nothing to do if the array only has one element.
3326 * (In the normal 1-1 case handled below, we don't have to worry about
3327 * two lists, as everything gets tied to the single list that is
3328 * generated for the single character 'to'. But here, we are omitting
3329 * that list, ('st' in the example), so must have multiple lists.) */
3330 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3331 &char_to, &to_len)))
3333 if (av_tindex(from_list) > 0) {
3336 /* We iterate over all combinations of i,j to place each code
3337 * point on each list */
3338 for (i = 0; i <= av_tindex(from_list); i++) {
3340 AV* i_list = newAV();
3341 SV** entryp = av_fetch(from_list, i, FALSE);
3342 if (entryp == NULL) {
3343 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3345 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3346 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3348 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3349 (SV*) i_list, FALSE))
3351 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3354 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3355 for (j = 0; j <= av_tindex(from_list); j++) {
3356 entryp = av_fetch(from_list, j, FALSE);
3357 if (entryp == NULL) {
3358 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3361 /* When i==j this adds itself to the list */
3362 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3363 (U8*) SvPVX(*entryp),
3364 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3366 /*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));*/
3371 SvREFCNT_dec(specials_inverse); /* done with it */
3372 } /* End of specials */
3374 /* read $swash->{LIST} */
3375 l = (U8*)SvPV(*listsvp, lcur);
3378 /* Go through each input line */
3382 l = swash_scan_list_line(l, lend, &min, &max, &val,
3383 cBOOL(octets), typestr);
3388 /* Each element in the range is to be inverted */
3389 for (inverse = min; inverse <= max; inverse++) {
3393 bool found_key = FALSE;
3394 bool found_inverse = FALSE;
3396 /* The key is the inverse mapping */
3397 char key[UTF8_MAXBYTES+1];
3398 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
3399 STRLEN key_len = key_end - key;
3401 /* Get the list for the map */
3402 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3403 list = (AV*) *listp;
3405 else { /* No entry yet for it: create one */
3407 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3408 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3412 /* Look through list to see if this inverse mapping already is
3413 * listed, or if there is a mapping to itself already */
3414 for (i = 0; i <= av_tindex(list); i++) {
3415 SV** entryp = av_fetch(list, i, FALSE);
3418 if (entryp == NULL) {
3419 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3423 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3427 if (uv == inverse) {
3428 found_inverse = TRUE;
3431 /* No need to continue searching if found everything we are
3433 if (found_key && found_inverse) {
3438 /* Make sure there is a mapping to itself on the list */
3440 av_push(list, newSVuv(val));
3441 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3445 /* Simply add the value to the list */
3446 if (! found_inverse) {
3447 av_push(list, newSVuv(inverse));
3448 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3451 /* swatch_get() increments the value of val for each element in the
3452 * range. That makes more compact tables possible. You can
3453 * express the capitalization, for example, of all consecutive
3454 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3455 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3456 * and it's not documented; it appears to be used only in
3457 * implementing tr//; I copied the semantics from swatch_get(), just
3459 if (!none || val < none) {
3469 Perl__swash_to_invlist(pTHX_ SV* const swash)
3472 /* Subject to change or removal. For use only in one place in regcomp.c.
3473 * Ownership is given to one reference count in the returned SV* */
3478 HV *const hv = MUTABLE_HV(SvRV(swash));
3479 UV elements = 0; /* Number of elements in the inversion list */
3489 STRLEN octets; /* if bits == 1, then octets == 0 */
3495 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3497 /* If not a hash, it must be the swash's inversion list instead */
3498 if (SvTYPE(hv) != SVt_PVHV) {
3499 return SvREFCNT_inc_simple_NN((SV*) hv);
3502 /* The string containing the main body of the table */
3503 listsvp = hv_fetchs(hv, "LIST", FALSE);
3504 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3505 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3506 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3507 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3509 typestr = (U8*)SvPV_nolen(*typesvp);
3510 bits = SvUV(*bitssvp);
3511 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3513 /* read $swash->{LIST} */
3514 if (SvPOK(*listsvp)) {
3515 l = (U8*)SvPV(*listsvp, lcur);
3518 /* LIST legitimately doesn't contain a string during compilation phases
3519 * of Perl itself, before the Unicode tables are generated. In this
3520 * case, just fake things up by creating an empty list */
3527 if (*l == 'V') { /* Inversion list format */
3528 const char *after_atou = (char *) lend;
3530 UV* other_elements_ptr;
3532 /* The first number is a count of the rest */
3534 elements = grok_atou((const char *)l, &after_atou);
3535 if (elements == 0) {
3536 invlist = _new_invlist(0);
3539 while (isSPACE(*l)) l++;
3540 l = (U8 *) after_atou;
3542 /* Get the 0th element, which is needed to setup the inversion list */
3543 while (isSPACE(*l)) l++;
3544 element0 = (UV) grok_atou((const char *)l, &after_atou);
3545 l = (U8 *) after_atou;
3546 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3549 /* Then just populate the rest of the input */
3550 while (elements-- > 0) {
3552 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3554 while (isSPACE(*l)) l++;
3555 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3556 l = (U8 *) after_atou;
3562 /* Scan the input to count the number of lines to preallocate array
3563 * size based on worst possible case, which is each line in the input
3564 * creates 2 elements in the inversion list: 1) the beginning of a
3565 * range in the list; 2) the beginning of a range not in the list. */
3566 while ((loc = (strchr(loc, '\n'))) != NULL) {
3571 /* If the ending is somehow corrupt and isn't a new line, add another
3572 * element for the final range that isn't in the inversion list */
3573 if (! (*lend == '\n'
3574 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3579 invlist = _new_invlist(elements);
3581 /* Now go through the input again, adding each range to the list */
3584 UV val; /* Not used by this function */
3586 l = swash_scan_list_line(l, lend, &start, &end, &val,
3587 cBOOL(octets), typestr);
3593 invlist = _add_range_to_invlist(invlist, start, end);
3597 /* Invert if the data says it should be */
3598 if (invert_it_svp && SvUV(*invert_it_svp)) {
3599 _invlist_invert(invlist);
3602 /* This code is copied from swatch_get()
3603 * read $swash->{EXTRAS} */
3604 x = (U8*)SvPV(*extssvp, xcur);
3612 SV **otherbitssvp, *other;
3615 const U8 opc = *x++;
3619 nl = (U8*)memchr(x, '\n', xend - x);
3621 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3623 x = nl + 1; /* 1 is length of "\n" */
3627 x = xend; /* to EXTRAS' end at which \n is not found */
3634 namelen = nl - namestr;
3638 namelen = xend - namestr;
3642 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3643 otherhv = MUTABLE_HV(SvRV(*othersvp));
3644 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3645 otherbits = (STRLEN)SvUV(*otherbitssvp);
3647 if (bits != otherbits || bits != 1) {
3648 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3649 "properties, bits=%"UVuf", otherbits=%"UVuf,
3650 (UV)bits, (UV)otherbits);
3653 /* The "other" swatch must be destroyed after. */
3654 other = _swash_to_invlist((SV *)*othersvp);
3656 /* End of code copied from swatch_get() */
3659 _invlist_union(invlist, other, &invlist);
3662 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
3665 _invlist_subtract(invlist, other, &invlist);
3668 _invlist_intersection(invlist, other, &invlist);
3673 sv_free(other); /* through with it! */
3676 SvREADONLY_on(invlist);
3681 Perl__get_swash_invlist(pTHX_ SV* const swash)
3685 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3687 if (! SvROK(swash)) {
3691 /* If it really isn't a hash, it isn't really swash; must be an inversion
3693 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
3697 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
3706 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3708 /* May change: warns if surrogates, non-character code points, or
3709 * non-Unicode code points are in s which has length len bytes. Returns
3710 * TRUE if none found; FALSE otherwise. The only other validity check is
3711 * to make sure that this won't exceed the string's length */
3713 const U8* const e = s + len;
3716 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3719 if (UTF8SKIP(s) > len) {
3720 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3721 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3724 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3726 if (UTF8_IS_SUPER(s)) {
3727 if (ckWARN_d(WARN_NON_UNICODE)) {
3728 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3729 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3730 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3734 else if (UTF8_IS_SURROGATE(s)) {
3735 if (ckWARN_d(WARN_SURROGATE)) {
3736 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3737 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3738 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3743 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3744 && (ckWARN_d(WARN_NONCHAR)))
3746 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
3747 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3748 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3759 =for apidoc pv_uni_display
3761 Build to the scalar C<dsv> a displayable version of the string C<spv>,
3762 length C<len>, the displayable version being at most C<pvlim> bytes long
3763 (if longer, the rest is truncated and "..." will be appended).
3765 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
3766 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3767 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3768 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3769 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3770 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3772 The pointer to the PV of the C<dsv> is returned.
3776 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3781 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3785 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3787 /* This serves double duty as a flag and a character to print after
3788 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3792 if (pvlim && SvCUR(dsv) >= pvlim) {
3796 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
3798 const unsigned char c = (unsigned char)u & 0xFF;
3799 if (flags & UNI_DISPLAY_BACKSLASH) {
3816 const char string = ok;
3817 sv_catpvs(dsv, "\\");
3818 sv_catpvn(dsv, &string, 1);
3821 /* isPRINT() is the locale-blind version. */
3822 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3823 const char string = c;
3824 sv_catpvn(dsv, &string, 1);
3829 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
3832 sv_catpvs(dsv, "...");
3838 =for apidoc sv_uni_display
3840 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
3841 the displayable version being at most C<pvlim> bytes long
3842 (if longer, the rest is truncated and "..." will be appended).
3844 The C<flags> argument is as in L</pv_uni_display>().
3846 The pointer to the PV of the C<dsv> is returned.
3851 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
3853 const char * const ptr =
3854 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
3856 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
3858 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
3859 SvCUR(ssv), pvlim, flags);
3863 =for apidoc foldEQ_utf8
3865 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
3866 of which may be in UTF-8) are the same case-insensitively; false otherwise.
3867 How far into the strings to compare is determined by other input parameters.
3869 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
3870 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
3871 with respect to C<s2>.
3873 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
3874 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
3875 scan will not be considered to be a match unless the goal is reached, and
3876 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
3879 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
3880 considered an end pointer to the position 1 byte past the maximum point
3881 in C<s1> beyond which scanning will not continue under any circumstances.
3882 (This routine assumes that UTF-8 encoded input strings are not malformed;
3883 malformed input can cause it to read past C<pe1>).
3884 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
3885 is less than C<s1>+C<l1>, the match will never be successful because it can
3887 get as far as its goal (and in fact is asserted against). Correspondingly for
3888 C<pe2> with respect to C<s2>.
3890 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
3891 C<l2> must be non-zero), and if both do, both have to be
3892 reached for a successful match. Also, if the fold of a character is multiple
3893 characters, all of them must be matched (see tr21 reference below for
3896 Upon a successful match, if C<pe1> is non-NULL,
3897 it will be set to point to the beginning of the I<next> character of C<s1>
3898 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
3900 For case-insensitiveness, the "casefolding" of Unicode is used
3901 instead of upper/lowercasing both the characters, see
3902 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
3906 /* A flags parameter has been added which may change, and hence isn't
3907 * externally documented. Currently it is:
3908 * 0 for as-documented above
3909 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
3910 ASCII one, to not match
3911 * FOLDEQ_LOCALE is set iff the rules from the current underlying
3912 * locale are to be used.
3913 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
3914 * routine. This allows that step to be skipped.
3915 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
3918 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)
3920 const U8 *p1 = (const U8*)s1; /* Point to current char */
3921 const U8 *p2 = (const U8*)s2;
3922 const U8 *g1 = NULL; /* goal for s1 */
3923 const U8 *g2 = NULL;
3924 const U8 *e1 = NULL; /* Don't scan s1 past this */
3925 U8 *f1 = NULL; /* Point to current folded */
3926 const U8 *e2 = NULL;
3928 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
3929 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
3930 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
3932 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
3934 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
3935 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
3936 /* The algorithm is to trial the folds without regard to the flags on
3937 * the first line of the above assert(), and then see if the result
3938 * violates them. This means that the inputs can't be pre-folded to a
3939 * violating result, hence the assert. This could be changed, with the
3940 * addition of extra tests here for the already-folded case, which would
3941 * slow it down. That cost is more than any possible gain for when these
3942 * flags are specified, as the flags indicate /il or /iaa matching which
3943 * is less common than /iu, and I (khw) also believe that real-world /il
3944 * and /iaa matches are most likely to involve code points 0-255, and this
3945 * function only under rare conditions gets called for 0-255. */
3947 if (IN_UTF8_CTYPE_LOCALE) {
3948 flags &= ~FOLDEQ_LOCALE;
3956 g1 = (const U8*)s1 + l1;
3964 g2 = (const U8*)s2 + l2;
3967 /* Must have at least one goal */
3972 /* Will never match if goal is out-of-bounds */
3973 assert(! e1 || e1 >= g1);
3975 /* Here, there isn't an end pointer, or it is beyond the goal. We
3976 * only go as far as the goal */
3980 assert(e1); /* Must have an end for looking at s1 */
3983 /* Same for goal for s2 */
3985 assert(! e2 || e2 >= g2);
3992 /* If both operands are already folded, we could just do a memEQ on the
3993 * whole strings at once, but it would be better if the caller realized
3994 * this and didn't even call us */
3996 /* Look through both strings, a character at a time */
3997 while (p1 < e1 && p2 < e2) {
3999 /* If at the beginning of a new character in s1, get its fold to use
4000 * and the length of the fold. (exception: locale rules just get the
4001 * character to a single byte) */
4003 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4008 /* If in locale matching, we use two sets of rules, depending
4009 * on if the code point is above or below 255. Here, we test
4010 * for and handle locale rules */
4011 if ((flags & FOLDEQ_LOCALE)
4012 && (! u1 || ! UTF8_IS_ABOVE_LATIN1(*p1)))
4014 /* There is no mixing of code points above and below 255. */
4015 if (u2 && UTF8_IS_ABOVE_LATIN1(*p2)) {
4019 /* We handle locale rules by converting, if necessary, the
4020 * code point to a single byte. */
4021 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4025 *foldbuf1 = TWO_BYTE_UTF8_TO_NATIVE(*p1, *(p1 + 1));
4029 else if (isASCII(*p1)) { /* Note, that here won't be both
4030 ASCII and using locale rules */
4032 /* If trying to mix non- with ASCII, and not supposed to,
4034 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4038 *foldbuf1 = toFOLD(*p1);
4041 to_utf8_fold(p1, foldbuf1, &n1);
4043 else { /* Not utf8, get utf8 fold */
4044 to_uni_fold(*p1, foldbuf1, &n1);
4050 if (n2 == 0) { /* Same for s2 */
4051 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4056 if ((flags & FOLDEQ_LOCALE)
4057 && (! u2 || ! UTF8_IS_ABOVE_LATIN1(*p2)))
4059 /* Here, the next char in s2 is < 256. We've already
4060 * worked on s1, and if it isn't also < 256, can't match */
4061 if (u1 && UTF8_IS_ABOVE_LATIN1(*p1)) {
4064 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4068 *foldbuf2 = TWO_BYTE_UTF8_TO_NATIVE(*p2, *(p2 + 1));
4071 /* Use another function to handle locale rules. We've made
4072 * sure that both characters to compare are single bytes */
4073 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4078 else if (isASCII(*p2)) {
4079 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4083 *foldbuf2 = toFOLD(*p2);
4086 to_utf8_fold(p2, foldbuf2, &n2);
4089 to_uni_fold(*p2, foldbuf2, &n2);
4095 /* Here f1 and f2 point to the beginning of the strings to compare.
4096 * These strings are the folds of the next character from each input
4097 * string, stored in utf8. */
4099 /* While there is more to look for in both folds, see if they
4100 * continue to match */
4102 U8 fold_length = UTF8SKIP(f1);
4103 if (fold_length != UTF8SKIP(f2)
4104 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4105 function call for single
4107 || memNE((char*)f1, (char*)f2, fold_length))
4109 return 0; /* mismatch */
4112 /* Here, they matched, advance past them */
4119 /* When reach the end of any fold, advance the input past it */
4121 p1 += u1 ? UTF8SKIP(p1) : 1;
4124 p2 += u2 ? UTF8SKIP(p2) : 1;
4126 } /* End of loop through both strings */
4128 /* A match is defined by each scan that specified an explicit length
4129 * reaching its final goal, and the other not having matched a partial
4130 * character (which can happen when the fold of a character is more than one
4132 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4136 /* Successful match. Set output pointers */
4146 /* XXX The next two functions should likely be moved to mathoms.c once all
4147 * occurrences of them are removed from the core; some cpan-upstream modules
4151 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4153 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4155 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
4159 =for apidoc utf8n_to_uvuni
4161 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
4163 This function was useful for code that wanted to handle both EBCDIC and
4164 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4165 distinctions between the platforms have mostly been made invisible to most
4166 code, so this function is quite unlikely to be what you want. If you do need
4167 this precise functionality, use instead
4168 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
4169 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
4175 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4177 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4179 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4183 =for apidoc uvuni_to_utf8_flags
4185 Instead you almost certainly want to use L</uvchr_to_utf8> or
4186 L</uvchr_to_utf8_flags>>.
4188 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4189 which itself, while not deprecated, should be used only in isolated
4190 circumstances. These functions were useful for code that wanted to handle
4191 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4192 v5.20, the distinctions between the platforms have mostly been made invisible
4193 to most code, so this function is quite unlikely to be what you want.
4199 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4201 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4203 return uvoffuni_to_utf8_flags(d, uv, flags);
4208 * c-indentation-style: bsd
4210 * indent-tabs-mode: nil
4213 * ex: set ts=8 sts=4 sw=4 et: