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"
37 /* Separate prototypes needed because in ASCII systems these are
38 * usually macros but they still are compiled as code, too. */
39 PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
40 PERL_CALLCONV UV Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen);
41 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
44 static const char unees[] =
45 "Malformed UTF-8 character (unexpected end of string)";
48 =head1 Unicode Support
50 This file contains various utility functions for manipulating UTF8-encoded
51 strings. For the uninitiated, this is a method of representing arbitrary
52 Unicode characters as a variable number of bytes, in such a way that
53 characters in the ASCII range are unmodified, and a zero byte never appears
54 within non-zero characters.
60 =for apidoc is_ascii_string
62 Returns true if the first C<len> bytes of the string C<s> are the same whether
63 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
64 is, if they are invariant. On ASCII-ish machines, only ASCII characters
65 fit this definition, hence the function's name.
67 If C<len> is 0, it will be calculated using C<strlen(s)>.
69 See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
75 Perl_is_ascii_string(const U8 *s, STRLEN len)
77 const U8* const send = s + (len ? len : strlen((const char *)s));
80 PERL_ARGS_ASSERT_IS_ASCII_STRING;
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
91 =for apidoc uvuni_to_utf8_flags
93 Adds the UTF-8 representation of the code point C<uv> to the end
94 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
95 bytes available. The return value is the pointer to the byte after the
96 end of the new character. In other words,
98 d = uvuni_to_utf8_flags(d, uv, flags);
102 d = uvuni_to_utf8(d, uv);
104 (which is equivalent to)
106 d = uvuni_to_utf8_flags(d, uv, 0);
108 This is the recommended Unicode-aware way of saying
112 This function will convert to UTF-8 (and not warn) even code points that aren't
113 legal Unicode or are problematic, unless C<flags> contains one or more of the
116 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
117 the function will raise a warning, provided UTF8 warnings are enabled. If instead
118 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
119 If both flags are set, the function will both warn and return NULL.
121 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
122 affect how the function handles a Unicode non-character. And, likewise for the
123 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
124 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
125 even less portable) can be warned and/or disallowed even if other above-Unicode
126 code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
129 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
130 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
138 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
140 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
142 /* The first problematic code point is the first surrogate */
143 if (uv >= UNICODE_SURROGATE_FIRST
144 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
146 if (UNICODE_IS_SURROGATE(uv)) {
147 if (flags & UNICODE_WARN_SURROGATE) {
148 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
149 "UTF-16 surrogate U+%04"UVXf, uv);
151 if (flags & UNICODE_DISALLOW_SURROGATE) {
155 else if (UNICODE_IS_SUPER(uv)) {
156 if (flags & UNICODE_WARN_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
159 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
160 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
162 if (flags & UNICODE_DISALLOW_SUPER
163 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
168 else if (UNICODE_IS_NONCHAR(uv)) {
169 if (flags & UNICODE_WARN_NONCHAR) {
170 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
171 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
174 if (flags & UNICODE_DISALLOW_NONCHAR) {
179 if (UNI_IS_INVARIANT(uv)) {
180 *d++ = (U8)UTF_TO_NATIVE(uv);
185 STRLEN len = UNISKIP(uv);
188 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
189 uv >>= UTF_ACCUMULATION_SHIFT;
191 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
194 #else /* Non loop style */
196 *d++ = (U8)(( uv >> 6) | 0xc0);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
201 *d++ = (U8)(( uv >> 12) | 0xe0);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
207 *d++ = (U8)(( uv >> 18) | 0xf0);
208 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
209 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
210 *d++ = (U8)(( uv & 0x3f) | 0x80);
213 if (uv < 0x4000000) {
214 *d++ = (U8)(( uv >> 24) | 0xf8);
215 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
218 *d++ = (U8)(( uv & 0x3f) | 0x80);
221 if (uv < 0x80000000) {
222 *d++ = (U8)(( uv >> 30) | 0xfc);
223 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
227 *d++ = (U8)(( uv & 0x3f) | 0x80);
231 if (uv < UTF8_QUAD_MAX)
234 *d++ = 0xfe; /* Can't match U+FEFF! */
235 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
239 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
240 *d++ = (U8)(( uv & 0x3f) | 0x80);
245 *d++ = 0xff; /* Can't match U+FFFE! */
246 *d++ = 0x80; /* 6 Reserved bits */
247 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
248 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
257 *d++ = (U8)(( uv & 0x3f) | 0x80);
261 #endif /* Loop style */
266 Tests if the first C<len> bytes of string C<s> form a valid UTF-8
267 character. Note that an INVARIANT (i.e. ASCII) character is a valid
268 UTF-8 character. The number of bytes in the UTF-8 character
269 will be returned if it is valid, otherwise 0.
271 This is the "slow" version as opposed to the "fast" version which is
272 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
273 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
274 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
275 you should use the _slow(). In practice this means that the _slow()
276 will be used very rarely, since the maximum Unicode code point (as of
277 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
278 the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
283 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
285 dTHX; /* The function called below requires thread context */
289 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
291 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
293 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
297 =for apidoc is_utf8_char_buf
299 Returns the number of bytes that comprise the first UTF-8 encoded character in
300 buffer C<buf>. C<buf_end> should point to one position beyond the end of the
301 buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
304 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
305 machines) is a valid UTF-8 character.
310 Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
315 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
317 if (buf_end <= buf) {
322 if (len > UTF8SKIP(buf)) {
327 if (IS_UTF8_CHAR_FAST(len))
328 return IS_UTF8_CHAR(buf, len) ? len : 0;
329 #endif /* #ifdef IS_UTF8_CHAR */
330 return is_utf8_char_slow(buf, len);
334 =for apidoc is_utf8_char
338 Tests if some arbitrary number of bytes begins in a valid UTF-8
339 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
340 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
341 character will be returned if it is valid, otherwise 0.
343 This function is deprecated due to the possibility that malformed input could
344 cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
350 Perl_is_utf8_char(const U8 *s)
352 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
354 /* Assumes we have enough space, which is why this is deprecated */
355 return is_utf8_char_buf(s, s + UTF8SKIP(s));
360 =for apidoc is_utf8_string
362 Returns true if the first C<len> bytes of string C<s> form a valid
363 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
364 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
365 terminating NUL byte). Note that all characters being ASCII constitute 'a
368 See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
374 Perl_is_utf8_string(const U8 *s, STRLEN len)
376 const U8* const send = s + (len ? len : strlen((const char *)s));
379 PERL_ARGS_ASSERT_IS_UTF8_STRING;
382 /* Inline the easy bits of is_utf8_char() here for speed... */
383 if (UTF8_IS_INVARIANT(*x)) {
386 else if (!UTF8_IS_START(*x))
389 /* ... and call is_utf8_char() only if really needed. */
390 const STRLEN c = UTF8SKIP(x);
391 const U8* const next_char_ptr = x + c;
393 if (next_char_ptr > send) {
397 if (IS_UTF8_CHAR_FAST(c)) {
398 if (!IS_UTF8_CHAR(x, c))
401 else if (! is_utf8_char_slow(x, c)) {
412 Implemented as a macro in utf8.h
414 =for apidoc is_utf8_string_loc
416 Like L</is_utf8_string> but stores the location of the failure (in the
417 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
418 "utf8ness success") in the C<ep>.
420 See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
422 =for apidoc is_utf8_string_loclen
424 Like L</is_utf8_string>() but stores the location of the failure (in the
425 case of "utf8ness failure") or the location C<s>+C<len> (in the case of
426 "utf8ness success") in the C<ep>, and the number of UTF-8
427 encoded characters in the C<el>.
429 See also L</is_utf8_string_loc>() and L</is_utf8_string>().
435 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
437 const U8* const send = s + (len ? len : strlen((const char *)s));
442 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
445 const U8* next_char_ptr;
447 /* Inline the easy bits of is_utf8_char() here for speed... */
448 if (UTF8_IS_INVARIANT(*x))
449 next_char_ptr = x + 1;
450 else if (!UTF8_IS_START(*x))
453 /* ... and call is_utf8_char() only if really needed. */
455 next_char_ptr = c + x;
456 if (next_char_ptr > send) {
459 if (IS_UTF8_CHAR_FAST(c)) {
460 if (!IS_UTF8_CHAR(x, c))
463 c = is_utf8_char_slow(x, c);
482 =for apidoc utf8n_to_uvuni
484 Bottom level UTF-8 decode routine.
485 Returns the code point value of the first character in the string C<s>,
486 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
487 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
488 the length, in bytes, of that character.
490 The value of C<flags> determines the behavior when C<s> does not point to a
491 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
492 zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
493 next possible position in C<s> that could begin a non-malformed character.
494 Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
496 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
497 individual types of malformations, such as the sequence being overlong (that
498 is, when there is a shorter sequence that can express the same code point;
499 overlong sequences are expressly forbidden in the UTF-8 standard due to
500 potential security issues). Another malformation example is the first byte of
501 a character not being a legal first byte. See F<utf8.h> for the list of such
502 flags. For allowed 0 length strings, this function returns 0; for allowed
503 overlong sequences, the computed code point is returned; for all other allowed
504 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
505 determinable reasonable value.
507 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
508 flags) malformation is found. If this flag is set, the routine assumes that
509 the caller will raise a warning, and this function will silently just set
510 C<retlen> to C<-1> and return zero.
512 Certain code points are considered problematic. These are Unicode surrogates,
513 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
514 By default these are considered regular code points, but certain situations
515 warrant special handling for them. If C<flags> contains
516 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
517 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
518 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
519 maximum) can be set to disallow these categories individually.
521 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
522 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
523 for their respective categories, but otherwise the code points are considered
524 valid (not malformations). To get a category to both be treated as a
525 malformation and raise a warning, specify both the WARN and DISALLOW flags.
526 (But note that warnings are not raised if lexically disabled nor if
527 UTF8_CHECK_ONLY is also specified.)
529 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
530 the others that are above the Unicode legal maximum. There are several
531 reasons: they requre at least 32 bits to represent them on ASCII platforms, are
532 not representable at all on EBCDIC platforms, and the original UTF-8
533 specification never went above this number (the current 0x10FFFF limit was
534 imposed later). (The smaller ones, those that fit into 32 bits, are
535 representable by a UV on ASCII platforms, but not by an IV, which means that
536 the number of operations that can be performed on them is quite restricted.)
537 The UTF-8 encoding on ASCII platforms for these large code points begins with a
538 byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
539 be treated as malformations, while allowing smaller above-Unicode code points.
540 (Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
541 including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
542 the other WARN flags, but applies just to these code points.
544 All other code points corresponding to Unicode characters, including private
545 use and those yet to be assigned, are never considered malformed and never
548 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
554 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
557 const U8 * const s0 = s;
558 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
563 UV outlier_ret = 0; /* return value when input is in error or problematic
565 UV pack_warn = 0; /* Save result of packWARN() for later */
566 bool unexpected_non_continuation = FALSE;
567 bool overflowed = FALSE;
568 bool do_overlong_test = TRUE; /* May have to skip this test */
570 const char* const malformed_text = "Malformed UTF-8 character";
572 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
574 /* The order of malformation tests here is important. We should consume as
575 * few bytes as possible in order to not skip any valid character. This is
576 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
577 * http://unicode.org/reports/tr36 for more discussion as to why. For
578 * example, once we've done a UTF8SKIP, we can tell the expected number of
579 * bytes, and could fail right off the bat if the input parameters indicate
580 * that there are too few available. But it could be that just that first
581 * byte is garbled, and the intended character occupies fewer bytes. If we
582 * blindly assumed that the first byte is correct, and skipped based on
583 * that number, we could skip over a valid input character. So instead, we
584 * always examine the sequence byte-by-byte.
586 * We also should not consume too few bytes, otherwise someone could inject
587 * things. For example, an input could be deliberately designed to
588 * overflow, and if this code bailed out immediately upon discovering that,
589 * returning to the caller *retlen pointing to the very next byte (one
590 * which is actually part of of the overflowing sequence), that could look
591 * legitimate to the caller, which could discard the initial partial
592 * sequence and process the rest, inappropriately */
594 /* Zero length strings, if allowed, of necessity are zero */
595 if (UNLIKELY(curlen == 0)) {
600 if (flags & UTF8_ALLOW_EMPTY) {
603 if (! (flags & UTF8_CHECK_ONLY)) {
604 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
609 expectlen = UTF8SKIP(s);
611 /* A well-formed UTF-8 character, as the vast majority of calls to this
612 * function will be for, has this expected length. For efficiency, set
613 * things up here to return it. It will be overriden only in those rare
614 * cases where a malformation is found */
619 /* An invariant is trivially well-formed */
620 if (UTF8_IS_INVARIANT(uv)) {
621 return (UV) (NATIVE_TO_UTF(*s));
624 /* A continuation character can't start a valid sequence */
625 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
626 if (flags & UTF8_ALLOW_CONTINUATION) {
630 return UNICODE_REPLACEMENT;
633 if (! (flags & UTF8_CHECK_ONLY)) {
634 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
641 uv = NATIVE_TO_UTF(uv);
644 /* Here is not a continuation byte, nor an invariant. The only thing left
645 * is a start byte (possibly for an overlong) */
647 /* Remove the leading bits that indicate the number of bytes in the
648 * character's whole UTF-8 sequence, leaving just the bits that are part of
650 uv &= UTF_START_MASK(expectlen);
652 /* Now, loop through the remaining bytes in the character's sequence,
653 * accumulating each into the working value as we go. Be sure to not look
654 * past the end of the input string */
655 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
657 for (s = s0 + 1; s < send; s++) {
658 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
659 #ifndef EBCDIC /* Can't overflow in EBCDIC */
660 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
662 /* The original implementors viewed this malformation as more
663 * serious than the others (though I, khw, don't understand
664 * why, since other malformations also give very very wrong
665 * results), so there is no way to turn off checking for it.
666 * Set a flag, but keep going in the loop, so that we absorb
667 * the rest of the bytes that comprise the character. */
669 overflow_byte = *s; /* Save for warning message's use */
672 uv = UTF8_ACCUMULATE(uv, *s);
675 /* Here, found a non-continuation before processing all expected
676 * bytes. This byte begins a new character, so quit, even if
677 * allowing this malformation. */
678 unexpected_non_continuation = TRUE;
681 } /* End of loop through the character's bytes */
683 /* Save how many bytes were actually in the character */
686 /* The loop above finds two types of malformations: non-continuation and/or
687 * overflow. The non-continuation malformation is really a too-short
688 * malformation, as it means that the current character ended before it was
689 * expected to (being terminated prematurely by the beginning of the next
690 * character, whereas in the too-short malformation there just are too few
691 * bytes available to hold the character. In both cases, the check below
692 * that we have found the expected number of bytes would fail if executed.)
693 * Thus the non-continuation malformation is really unnecessary, being a
694 * subset of the too-short malformation. But there may be existing
695 * applications that are expecting the non-continuation type, so we retain
696 * it, and return it in preference to the too-short malformation. (If this
697 * code were being written from scratch, the two types might be collapsed
698 * into one.) I, khw, am also giving priority to returning the
699 * non-continuation and too-short malformations over overflow when multiple
700 * ones are present. I don't know of any real reason to prefer one over
701 * the other, except that it seems to me that multiple-byte errors trumps
702 * errors from a single byte */
703 if (UNLIKELY(unexpected_non_continuation)) {
704 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
705 if (! (flags & UTF8_CHECK_ONLY)) {
707 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
710 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));
715 uv = UNICODE_REPLACEMENT;
717 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
718 * as what the original expectations were. */
719 do_overlong_test = FALSE;
724 else if (UNLIKELY(curlen < expectlen)) {
725 if (! (flags & UTF8_ALLOW_SHORT)) {
726 if (! (flags & UTF8_CHECK_ONLY)) {
727 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));
731 uv = UNICODE_REPLACEMENT;
732 do_overlong_test = FALSE;
738 #ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
742 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
743 * generation of the sv, since no warnings are raised under CHECK */
744 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
745 && ckWARN_d(WARN_UTF8))
747 /* This message is deliberately not of the same syntax as the other
748 * messages for malformations, for backwards compatibility in the
749 * unlikely event that code is relying on its precise earlier text
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
752 pack_warn = packWARN(WARN_UTF8);
754 if (flags & UTF8_DISALLOW_FE_FF) {
758 if (UNLIKELY(overflowed)) {
760 /* If the first byte is FF, it will overflow a 32-bit word. If the
761 * first byte is FE, it will overflow a signed 32-bit word. The
762 * above preserves backward compatibility, since its message was used
763 * in earlier versions of this code in preference to overflow */
764 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
770 && expectlen > (STRLEN)UNISKIP(uv)
771 && ! (flags & UTF8_ALLOW_LONG))
773 /* The overlong malformation has lower precedence than the others.
774 * Note that if this malformation is allowed, we return the actual
775 * value, instead of the replacement character. This is because this
776 * value is actually well-defined. */
777 if (! (flags & UTF8_CHECK_ONLY)) {
778 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), *s0));
783 /* Here, the input is considered to be well-formed , but could be a
784 * problematic code point that is not allowed by the input parameters. */
785 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
786 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
787 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
789 if (UNICODE_IS_SURROGATE(uv)) {
790 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
791 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
793 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
794 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
796 if (flags & UTF8_DISALLOW_SURROGATE) {
800 else if ((uv > PERL_UNICODE_MAX)) {
801 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
802 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
804 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
805 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
807 if (flags & UTF8_DISALLOW_SUPER) {
811 else if (UNICODE_IS_NONCHAR(uv)) {
812 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
813 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
815 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
816 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
818 if (flags & UTF8_DISALLOW_NONCHAR) {
828 /* Here, this is not considered a malformed character, so drop through
834 /* There are three cases which get to beyond this point. In all 3 cases:
835 * <sv> if not null points to a string to print as a warning.
836 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
838 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
839 * This is done by initializing it to 0, and changing it only
842 * 1) The input is valid but problematic, and to be warned about. The
843 * return value is the resultant code point; <*retlen> is set to
844 * <curlen>, the number of bytes that comprise the code point.
845 * <pack_warn> contains the result of packWARN() for the warning
846 * types. The entry point for this case is the label <do_warn>;
847 * 2) The input is a valid code point but disallowed by the parameters to
848 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
849 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
850 * comprise the code point. <pack_warn> contains the result of
851 * packWARN() for the warning types. The entry point for this case is
852 * the label <disallowed>.
853 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
854 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
855 * bytes that comprise the malformation. All such malformations are
856 * assumed to be warning type <utf8>. The entry point for this case
857 * is the label <malformed>.
862 if (sv && ckWARN_d(WARN_UTF8)) {
863 pack_warn = packWARN(WARN_UTF8);
868 if (flags & UTF8_CHECK_ONLY) {
870 *retlen = ((STRLEN) -1);
876 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
877 if warnings are to be raised. */
878 const char * const string = SvPVX_const(sv);
881 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
883 Perl_warner(aTHX_ pack_warn, "%s", string);
894 =for apidoc utf8_to_uvchr_buf
896 Returns the native code point of the first character in the string C<s> which
897 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
898 C<*retlen> will be set to the length, in bytes, of that character.
900 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
901 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
902 NULL) to -1. If those warnings are off, the computed value if well-defined (or
903 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
904 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
905 next possible position in C<s> that could begin a non-malformed character.
906 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
913 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
915 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
919 return utf8n_to_uvchr(s, send - s, retlen,
920 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
923 /* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
924 * there are no malformations in the input UTF-8 string C<s>. surrogates,
925 * non-character code points, and non-Unicode code points are allowed. A macro
926 * in utf8.h is used to normally avoid this function wrapper */
929 Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
931 const UV uv = valid_utf8_to_uvuni(s, retlen);
933 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
935 return UNI_TO_NATIVE(uv);
939 =for apidoc utf8_to_uvchr
943 Returns the native code point of the first character in the string C<s>
944 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
945 length, in bytes, of that character.
947 Some, but not all, UTF-8 malformations are detected, and in fact, some
948 malformed input could cause reading beyond the end of the input buffer, which
949 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
951 If C<s> points to one of the detected malformations, and UTF8 warnings are
952 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
953 NULL) to -1. If those warnings are off, the computed value if well-defined (or
954 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
955 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
956 next possible position in C<s> that could begin a non-malformed character.
957 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
963 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
965 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
967 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
971 =for apidoc utf8_to_uvuni_buf
973 Returns the Unicode code point of the first character in the string C<s> which
974 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
975 C<retlen> will be set to the length, in bytes, of that character.
977 This function should only be used when the returned UV is considered
978 an index into the Unicode semantic tables (e.g. swashes).
980 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
981 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
982 NULL) to -1. If those warnings are off, the computed value if well-defined (or
983 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
984 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
985 next possible position in C<s> that could begin a non-malformed character.
986 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
992 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
994 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
998 /* Call the low level routine asking for checks */
999 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1000 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1003 /* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
1004 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
1005 * non-character code points, and non-Unicode code points are allowed */
1008 Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1010 UV expectlen = UTF8SKIP(s);
1011 const U8* send = s + expectlen;
1012 UV uv = NATIVE_TO_UTF(*s);
1014 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1017 *retlen = expectlen;
1020 /* An invariant is trivially returned */
1021 if (expectlen == 1) {
1025 /* Remove the leading bits that indicate the number of bytes, leaving just
1026 * the bits that are part of the value */
1027 uv &= UTF_START_MASK(expectlen);
1029 /* Now, loop through the remaining bytes, accumulating each into the
1030 * working total as we go. (I khw tried unrolling the loop for up to 4
1031 * bytes, but there was no performance improvement) */
1032 for (++s; s < send; s++) {
1033 uv = UTF8_ACCUMULATE(uv, *s);
1040 =for apidoc utf8_to_uvuni
1044 Returns the Unicode code point of the first character in the string C<s>
1045 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
1046 length, in bytes, of that character.
1048 This function should only be used when the returned UV is considered
1049 an index into the Unicode semantic tables (e.g. swashes).
1051 Some, but not all, UTF-8 malformations are detected, and in fact, some
1052 malformed input could cause reading beyond the end of the input buffer, which
1053 is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
1055 If C<s> points to one of the detected malformations, and UTF8 warnings are
1056 enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1057 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1058 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1059 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1060 next possible position in C<s> that could begin a non-malformed character.
1061 See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
1067 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1069 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1071 return valid_utf8_to_uvuni(s, retlen);
1075 =for apidoc utf8_length
1077 Return the length of the UTF-8 char encoded string C<s> in characters.
1078 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1079 up past C<e>, croaks.
1085 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1090 PERL_ARGS_ASSERT_UTF8_LENGTH;
1092 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1093 * the bitops (especially ~) can create illegal UTF-8.
1094 * In other words: in Perl UTF-8 is not just for Unicode. */
1097 goto warn_and_return;
1099 if (!UTF8_IS_INVARIANT(*s))
1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1111 "%s in %s", unees, OP_DESC(PL_op));
1113 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1120 =for apidoc utf8_distance
1122 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
1125 WARNING: use only if you *know* that the pointers point inside the
1132 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
1134 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1136 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
1140 =for apidoc utf8_hop
1142 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1143 forward or backward.
1145 WARNING: do not use the following unless you *know* C<off> is within
1146 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1147 on the first byte of character or just after the last byte of a character.
1153 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
1155 PERL_ARGS_ASSERT_UTF8_HOP;
1157 PERL_UNUSED_CONTEXT;
1158 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1159 * the bitops (especially ~) can create illegal UTF-8.
1160 * In other words: in Perl UTF-8 is not just for Unicode. */
1169 while (UTF8_IS_CONTINUATION(*s))
1177 =for apidoc bytes_cmp_utf8
1179 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1180 sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
1181 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1182 if the first string is greater than the second string.
1184 -1 or +1 is returned if the shorter string was identical to the start of the
1185 longer string. -2 or +2 is returned if the was a difference between characters
1192 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1194 const U8 *const bend = b + blen;
1195 const U8 *const uend = u + ulen;
1197 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1199 PERL_UNUSED_CONTEXT;
1201 while (b < bend && u < uend) {
1203 if (!UTF8_IS_INVARIANT(c)) {
1204 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1207 if (UTF8_IS_CONTINUATION(c1)) {
1208 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
1210 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1211 "Malformed UTF-8 character "
1212 "(unexpected non-continuation byte 0x%02x"
1213 ", immediately after start byte 0x%02x)"
1214 /* Dear diag.t, it's in the pod. */
1216 PL_op ? " in " : "",
1217 PL_op ? OP_DESC(PL_op) : "");
1222 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1223 "%s in %s", unees, OP_DESC(PL_op));
1225 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1226 return -2; /* Really want to return undef :-) */
1233 return *b < c ? -2 : +2;
1238 if (b == bend && u == uend)
1241 return b < bend ? +1 : -1;
1245 =for apidoc utf8_to_bytes
1247 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1248 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1249 updates C<len> to contain the new length.
1250 Returns zero on failure, setting C<len> to -1.
1252 If you need a copy of the string, see L</bytes_from_utf8>.
1258 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1260 U8 * const save = s;
1261 U8 * const send = s + *len;
1264 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1266 /* ensure valid UTF-8 and chars < 256 before updating string */
1270 if (!UTF8_IS_INVARIANT(c) &&
1271 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1272 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
1273 *len = ((STRLEN) -1);
1281 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
1290 =for apidoc bytes_from_utf8
1292 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1293 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1294 the newly-created string, and updates C<len> to contain the new
1295 length. Returns the original string if no conversion occurs, C<len>
1296 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1297 0 if C<s> is converted or consisted entirely of characters that are invariant
1298 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1304 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1307 const U8 *start = s;
1311 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1313 PERL_UNUSED_CONTEXT;
1317 /* ensure valid UTF-8 and chars < 256 before converting string */
1318 for (send = s + *len; s < send;) {
1320 if (!UTF8_IS_INVARIANT(c)) {
1321 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1322 (c = *s++) && UTF8_IS_CONTINUATION(c))
1331 Newx(d, (*len) - count + 1, U8);
1332 s = start; start = d;
1335 if (!UTF8_IS_INVARIANT(c)) {
1336 /* Then it is two-byte encoded */
1337 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
1347 =for apidoc bytes_to_utf8
1349 Converts a string C<s> of length C<len> bytes from the native encoding into
1351 Returns a pointer to the newly-created string, and sets C<len> to
1352 reflect the new length in bytes.
1354 A NUL character will be written after the end of the string.
1356 If you want to convert to UTF-8 from encodings other than
1357 the native (Latin1 or EBCDIC),
1358 see L</sv_recode_to_utf8>().
1363 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1364 likewise need duplication. */
1367 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1369 const U8 * const send = s + (*len);
1373 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1374 PERL_UNUSED_CONTEXT;
1376 Newx(d, (*len) * 2 + 1, U8);
1380 const UV uv = NATIVE_TO_ASCII(*s++);
1381 if (UNI_IS_INVARIANT(uv))
1382 *d++ = (U8)UTF_TO_NATIVE(uv);
1384 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1385 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
1394 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1396 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1397 * We optimize for native, for obvious reasons. */
1400 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1405 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1408 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1413 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1417 *d++ = UNI_TO_NATIVE(uv);
1424 *d++ = (U8)(( uv >> 6) | 0xc0);
1425 *d++ = (U8)(( uv & 0x3f) | 0x80);
1428 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
1430 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1432 UV low = (p[0] << 8) + p[1];
1434 if (low < 0xdc00 || low > 0xdfff)
1435 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1436 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1438 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1439 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1442 *d++ = (U8)(( uv >> 12) | 0xe0);
1443 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1444 *d++ = (U8)(( uv & 0x3f) | 0x80);
1448 *d++ = (U8)(( uv >> 18) | 0xf0);
1449 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1450 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1451 *d++ = (U8)(( uv & 0x3f) | 0x80);
1455 *newlen = d - dstart;
1459 /* Note: this one is slightly destructive of the source. */
1462 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1465 U8* const send = s + bytelen;
1467 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1470 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1474 const U8 tmp = s[0];
1479 return utf16_to_utf8(p, d, bytelen, newlen);
1482 /* for now these are all defined (inefficiently) in terms of the utf8 versions.
1483 * Note that the macros in handy.h that call these short-circuit calling them
1484 * for Latin-1 range inputs */
1487 Perl_is_uni_alnum(pTHX_ UV c)
1489 U8 tmpbuf[UTF8_MAXBYTES+1];
1490 uvchr_to_utf8(tmpbuf, c);
1491 return is_utf8_alnum(tmpbuf);
1495 Perl_is_uni_idfirst(pTHX_ UV c)
1497 U8 tmpbuf[UTF8_MAXBYTES+1];
1498 uvchr_to_utf8(tmpbuf, c);
1499 return is_utf8_idfirst(tmpbuf);
1503 Perl_is_uni_alpha(pTHX_ UV c)
1505 U8 tmpbuf[UTF8_MAXBYTES+1];
1506 uvchr_to_utf8(tmpbuf, c);
1507 return is_utf8_alpha(tmpbuf);
1511 Perl_is_uni_ascii(pTHX_ UV c)
1517 Perl_is_uni_blank(pTHX_ UV c)
1519 U8 tmpbuf[UTF8_MAXBYTES+1];
1520 uvchr_to_utf8(tmpbuf, c);
1521 return is_utf8_blank(tmpbuf);
1525 Perl_is_uni_space(pTHX_ UV c)
1527 U8 tmpbuf[UTF8_MAXBYTES+1];
1528 uvchr_to_utf8(tmpbuf, c);
1529 return is_utf8_space(tmpbuf);
1533 Perl_is_uni_digit(pTHX_ UV c)
1535 U8 tmpbuf[UTF8_MAXBYTES+1];
1536 uvchr_to_utf8(tmpbuf, c);
1537 return is_utf8_digit(tmpbuf);
1541 Perl_is_uni_upper(pTHX_ UV c)
1543 U8 tmpbuf[UTF8_MAXBYTES+1];
1544 uvchr_to_utf8(tmpbuf, c);
1545 return is_utf8_upper(tmpbuf);
1549 Perl_is_uni_lower(pTHX_ UV c)
1551 U8 tmpbuf[UTF8_MAXBYTES+1];
1552 uvchr_to_utf8(tmpbuf, c);
1553 return is_utf8_lower(tmpbuf);
1557 Perl_is_uni_cntrl(pTHX_ UV c)
1559 return isCNTRL_L1(c);
1563 Perl_is_uni_graph(pTHX_ UV c)
1565 U8 tmpbuf[UTF8_MAXBYTES+1];
1566 uvchr_to_utf8(tmpbuf, c);
1567 return is_utf8_graph(tmpbuf);
1571 Perl_is_uni_print(pTHX_ UV c)
1573 U8 tmpbuf[UTF8_MAXBYTES+1];
1574 uvchr_to_utf8(tmpbuf, c);
1575 return is_utf8_print(tmpbuf);
1579 Perl_is_uni_punct(pTHX_ UV c)
1581 U8 tmpbuf[UTF8_MAXBYTES+1];
1582 uvchr_to_utf8(tmpbuf, c);
1583 return is_utf8_punct(tmpbuf);
1587 Perl_is_uni_xdigit(pTHX_ UV c)
1589 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1590 uvchr_to_utf8(tmpbuf, c);
1591 return is_utf8_xdigit(tmpbuf);
1595 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1597 /* We have the latin1-range values compiled into the core, so just use
1598 * those, converting the result to utf8. The only difference between upper
1599 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1600 * either "SS" or "Ss". Which one to use is passed into the routine in
1601 * 'S_or_s' to avoid a test */
1603 UV converted = toUPPER_LATIN1_MOD(c);
1605 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1607 assert(S_or_s == 'S' || S_or_s == 's');
1609 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1610 characters in this range */
1611 *p = (U8) converted;
1616 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1617 * which it maps to one of them, so as to only have to have one check for
1618 * it in the main case */
1619 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1621 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1622 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1625 converted = GREEK_CAPITAL_LETTER_MU;
1627 case LATIN_SMALL_LETTER_SHARP_S:
1633 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1634 assert(0); /* NOTREACHED */
1638 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1639 *p = UTF8_TWO_BYTE_LO(converted);
1645 /* Call the function to convert a UTF-8 encoded character to the specified case.
1646 * Note that there may be more than one character in the result.
1647 * INP is a pointer to the first byte of the input character
1648 * OUTP will be set to the first byte of the string of changed characters. It
1649 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1650 * LENP will be set to the length in bytes of the string of changed characters
1652 * The functions return the ordinal of the first character in the string of OUTP */
1653 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1654 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1655 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
1657 /* This additionally has the input parameter SPECIALS, which if non-zero will
1658 * cause this to use the SPECIALS hash for folding (meaning get full case
1659 * folding); otherwise, when zero, this implies a simple case fold */
1660 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
1663 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1667 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1668 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1669 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1670 * the changed version may be longer than the original character.
1672 * The ordinal of the first character of the changed version is returned
1673 * (but note, as explained above, that there may be more.) */
1675 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1678 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1681 uvchr_to_utf8(p, c);
1682 return CALL_UPPER_CASE(p, p, lenp);
1686 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1690 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1693 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1696 uvchr_to_utf8(p, c);
1697 return CALL_TITLE_CASE(p, p, lenp);
1701 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1703 /* We have the latin1-range values compiled into the core, so just use
1704 * those, converting the result to utf8. Since the result is always just
1705 * one character, we allow <p> to be NULL */
1707 U8 converted = toLOWER_LATIN1(c);
1710 if (UNI_IS_INVARIANT(converted)) {
1715 *p = UTF8_TWO_BYTE_HI(converted);
1716 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1724 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1728 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1731 return to_lower_latin1((U8) c, p, lenp);
1734 uvchr_to_utf8(p, c);
1735 return CALL_LOWER_CASE(p, p, lenp);
1739 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
1741 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
1746 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1748 if (c == MICRO_SIGN) {
1749 converted = GREEK_SMALL_LETTER_MU;
1751 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1757 else { /* In this range the fold of all other characters is their lower
1759 converted = toLOWER_LATIN1(c);
1762 if (UNI_IS_INVARIANT(converted)) {
1763 *p = (U8) converted;
1767 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1768 *p = UTF8_TWO_BYTE_LO(converted);
1776 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
1779 /* Not currently externally documented, and subject to change
1780 * <flags> bits meanings:
1781 * FOLD_FLAGS_FULL iff full folding is to be used;
1782 * FOLD_FLAGS_LOCALE iff in locale
1783 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1786 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1789 UV result = _to_fold_latin1((U8) c, p, lenp,
1790 cBOOL(((flags & FOLD_FLAGS_FULL)
1791 /* If ASCII-safe, don't allow full folding,
1792 * as that could include SHARP S => ss;
1793 * otherwise there is no crossing of
1794 * ascii/non-ascii in the latin1 range */
1795 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1796 /* It is illegal for the fold to cross the 255/256 boundary under
1797 * locale; in this case return the original */
1798 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1803 /* If no special needs, just use the macro */
1804 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1805 uvchr_to_utf8(p, c);
1806 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1808 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1809 the special flags. */
1810 U8 utf8_c[UTF8_MAXBYTES + 1];
1811 uvchr_to_utf8(utf8_c, c);
1812 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1816 /* for now these all assume no locale info available for Unicode > 255; and
1817 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1818 * called instead, so that these don't get called for < 255 */
1821 Perl_is_uni_alnum_lc(pTHX_ UV c)
1823 return is_uni_alnum(c); /* XXX no locale support yet */
1827 Perl_is_uni_idfirst_lc(pTHX_ UV c)
1829 return is_uni_idfirst(c); /* XXX no locale support yet */
1833 Perl_is_uni_alpha_lc(pTHX_ UV c)
1835 return is_uni_alpha(c); /* XXX no locale support yet */
1839 Perl_is_uni_ascii_lc(pTHX_ UV c)
1841 return is_uni_ascii(c); /* XXX no locale support yet */
1845 Perl_is_uni_blank_lc(pTHX_ UV c)
1847 return is_uni_blank(c); /* XXX no locale support yet */
1851 Perl_is_uni_space_lc(pTHX_ UV c)
1853 return is_uni_space(c); /* XXX no locale support yet */
1857 Perl_is_uni_digit_lc(pTHX_ UV c)
1859 return is_uni_digit(c); /* XXX no locale support yet */
1863 Perl_is_uni_upper_lc(pTHX_ UV c)
1865 return is_uni_upper(c); /* XXX no locale support yet */
1869 Perl_is_uni_lower_lc(pTHX_ UV c)
1871 return is_uni_lower(c); /* XXX no locale support yet */
1875 Perl_is_uni_cntrl_lc(pTHX_ UV c)
1877 return is_uni_cntrl(c); /* XXX no locale support yet */
1881 Perl_is_uni_graph_lc(pTHX_ UV c)
1883 return is_uni_graph(c); /* XXX no locale support yet */
1887 Perl_is_uni_print_lc(pTHX_ UV c)
1889 return is_uni_print(c); /* XXX no locale support yet */
1893 Perl_is_uni_punct_lc(pTHX_ UV c)
1895 return is_uni_punct(c); /* XXX no locale support yet */
1899 Perl_is_uni_xdigit_lc(pTHX_ UV c)
1901 return is_uni_xdigit(c); /* XXX no locale support yet */
1905 Perl_to_uni_upper_lc(pTHX_ U32 c)
1907 /* XXX returns only the first character -- do not use XXX */
1908 /* XXX no locale support yet */
1910 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1911 return (U32)to_uni_upper(c, tmpbuf, &len);
1915 Perl_to_uni_title_lc(pTHX_ U32 c)
1917 /* XXX returns only the first character XXX -- do not use XXX */
1918 /* XXX no locale support yet */
1920 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1921 return (U32)to_uni_title(c, tmpbuf, &len);
1925 Perl_to_uni_lower_lc(pTHX_ U32 c)
1927 /* XXX returns only the first character -- do not use XXX */
1928 /* XXX no locale support yet */
1930 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1931 return (U32)to_uni_lower(c, tmpbuf, &len);
1935 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1936 const char *const swashname)
1938 /* returns a boolean giving whether or not the UTF8-encoded character that
1939 * starts at <p> is in the swash indicated by <swashname>. <swash>
1940 * contains a pointer to where the swash indicated by <swashname>
1941 * is to be stored; which this routine will do, so that future calls will
1942 * look at <*swash> and only generate a swash if it is not null
1944 * Note that it is assumed that the buffer length of <p> is enough to
1945 * contain all the bytes that comprise the character. Thus, <*p> should
1946 * have been checked before this call for mal-formedness enough to assure
1951 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1953 /* The API should have included a length for the UTF-8 character in <p>,
1954 * but it doesn't. We therefor assume that p has been validated at least
1955 * as far as there being enough bytes available in it to accommodate the
1956 * character without reading beyond the end, and pass that number on to the
1957 * validating routine */
1958 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
1961 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1962 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
1964 return swash_fetch(*swash, p, TRUE) != 0;
1968 Perl_is_utf8_alnum(pTHX_ const U8 *p)
1972 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1974 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1975 * descendant of isalnum(3), in other words, it doesn't
1976 * contain the '_'. --jhi */
1977 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
1981 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
1985 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1989 /* is_utf8_idstart would be more logical. */
1990 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1994 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1998 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
2002 /* is_utf8_idstart would be more logical. */
2003 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2007 Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2011 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2013 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2017 Perl_is_utf8_idcont(pTHX_ const U8 *p)
2021 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2023 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
2027 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2031 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2033 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2037 Perl_is_utf8_alpha(pTHX_ const U8 *p)
2041 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2043 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
2047 Perl_is_utf8_ascii(pTHX_ const U8 *p)
2051 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2053 /* ASCII characters are the same whether in utf8 or not. So the macro
2054 * works on both utf8 and non-utf8 representations. */
2059 Perl_is_utf8_blank(pTHX_ const U8 *p)
2063 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2065 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2069 Perl_is_utf8_space(pTHX_ const U8 *p)
2073 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2075 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
2079 Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2083 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2085 /* Only true if is an ASCII space-like character, and ASCII is invariant
2086 * under utf8, so can just use the macro */
2087 return isSPACE_A(*p);
2091 Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2095 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2097 /* Only true if is an ASCII word character, and ASCII is invariant
2098 * under utf8, so can just use the macro */
2099 return isWORDCHAR_A(*p);
2103 Perl_is_utf8_digit(pTHX_ const U8 *p)
2107 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2109 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
2113 Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2117 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2119 /* Only true if is an ASCII digit character, and ASCII is invariant
2120 * under utf8, so can just use the macro */
2121 return isDIGIT_A(*p);
2125 Perl_is_utf8_upper(pTHX_ const U8 *p)
2129 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2131 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
2135 Perl_is_utf8_lower(pTHX_ const U8 *p)
2139 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2141 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
2145 Perl_is_utf8_cntrl(pTHX_ const U8 *p)
2149 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2152 return isCNTRL_A(*p);
2155 /* All controls are in Latin1 */
2156 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2159 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2163 Perl_is_utf8_graph(pTHX_ const U8 *p)
2167 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2169 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
2173 Perl_is_utf8_print(pTHX_ const U8 *p)
2177 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2179 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
2183 Perl_is_utf8_punct(pTHX_ const U8 *p)
2187 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2189 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
2193 Perl_is_utf8_xdigit(pTHX_ const U8 *p)
2197 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2199 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
2203 Perl_is_utf8_mark(pTHX_ const U8 *p)
2207 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2209 return is_utf8_common(p, &PL_utf8_mark, "IsM");
2213 Perl_is_utf8_X_regular_begin(pTHX_ const U8 *p)
2217 PERL_ARGS_ASSERT_IS_UTF8_X_REGULAR_BEGIN;
2219 return is_utf8_common(p, &PL_utf8_X_regular_begin, "_X_Regular_Begin");
2223 Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2227 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2229 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2233 Perl__is_utf8_quotemeta(pTHX_ const U8 *p)
2235 /* For exclusive use of pp_quotemeta() */
2239 PERL_ARGS_ASSERT__IS_UTF8_QUOTEMETA;
2241 return is_utf8_common(p, &PL_utf8_quotemeta, "_Perl_Quotemeta");
2244 =for apidoc to_utf8_case
2246 The C<p> contains the pointer to the UTF-8 string encoding
2247 the character that is being converted. This routine assumes that the character
2248 at C<p> is well-formed.
2250 The C<ustrp> is a pointer to the character buffer to put the
2251 conversion result to. The C<lenp> is a pointer to the length
2254 The C<swashp> is a pointer to the swash to use.
2256 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2257 and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
2258 but not always, a multicharacter mapping), is tried first.
2260 The C<special> is a string like "utf8::ToSpecLower", which means the
2261 hash %utf8::ToSpecLower. The access to the hash is through
2262 Perl_to_utf8_case().
2264 The C<normal> is a string like "ToLower" which means the swash
2270 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2271 SV **swashp, const char *normal, const char *special)
2274 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2276 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
2277 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2278 * are necessary in EBCDIC, they are redundant no-ops
2279 * in ASCII-ish platforms, and hopefully optimized away. */
2280 const UV uv1 = NATIVE_TO_UNI(uv0);
2282 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2284 /* Note that swash_fetch() doesn't output warnings for these because it
2285 * assumes we will */
2286 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2287 if (uv1 <= UNICODE_SURROGATE_LAST) {
2288 if (ckWARN_d(WARN_SURROGATE)) {
2289 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2290 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2291 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2294 else if (UNICODE_IS_SUPER(uv1)) {
2295 if (ckWARN_d(WARN_NON_UNICODE)) {
2296 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2297 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2298 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2302 /* Note that non-characters are perfectly legal, so no warning should
2306 uvuni_to_utf8(tmpbuf, uv1);
2308 if (!*swashp) /* load on-demand */
2309 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2312 /* It might be "special" (sometimes, but not always,
2313 * a multicharacter mapping) */
2314 HV * const hv = get_hv(special, 0);
2318 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2322 s = SvPV_const(*svp, len);
2324 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2327 /* If we have EBCDIC we need to remap the characters
2328 * since any characters in the low 256 are Unicode
2329 * code points, not EBCDIC. */
2330 U8 *t = (U8*)s, *tend = t + len, *d;
2337 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2339 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2348 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2353 Copy(tmpbuf, ustrp, len, U8);
2355 Copy(s, ustrp, len, U8);
2361 if (!len && *swashp) {
2362 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
2365 /* It was "normal" (a single character mapping). */
2366 const UV uv3 = UNI_TO_NATIVE(uv2);
2367 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2375 return valid_utf8_to_uvchr(ustrp, 0);
2378 /* Here, there was no mapping defined, which means that the code point maps
2379 * to itself. Return the inputs */
2381 Copy(p, ustrp, len, U8);
2391 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2393 /* This is called when changing the case of a utf8-encoded character above
2394 * the Latin1 range, and the operation is in locale. If the result
2395 * contains a character that crosses the 255/256 boundary, disallow the
2396 * change, and return the original code point. See L<perlfunc/lc> for why;
2398 * p points to the original string whose case was changed; assumed
2399 * by this routine to be well-formed
2400 * result the code point of the first character in the changed-case string
2401 * ustrp points to the changed-case string (<result> represents its first char)
2402 * lenp points to the length of <ustrp> */
2404 UV original; /* To store the first code point of <p> */
2406 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2408 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2410 /* We know immediately if the first character in the string crosses the
2411 * boundary, so can skip */
2414 /* Look at every character in the result; if any cross the
2415 * boundary, the whole thing is disallowed */
2416 U8* s = ustrp + UTF8SKIP(ustrp);
2417 U8* e = ustrp + *lenp;
2419 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2426 /* Here, no characters crossed, result is ok as-is */
2432 /* Failed, have to return the original */
2433 original = valid_utf8_to_uvchr(p, lenp);
2434 Copy(p, ustrp, *lenp, char);
2439 =for apidoc to_utf8_upper
2441 Convert the UTF-8 encoded character at C<p> to its uppercase version and
2442 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2443 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2444 the uppercase version may be longer than the original character.
2446 The first character of the uppercased version is returned
2447 (but note, as explained above, that there may be more.)
2449 The character at C<p> is assumed by this routine to be well-formed.
2453 /* Not currently externally documented, and subject to change:
2454 * <flags> is set iff locale semantics are to be used for code points < 256
2455 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2456 * were used in the calculation; otherwise unchanged. */
2459 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2465 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2467 if (UTF8_IS_INVARIANT(*p)) {
2469 result = toUPPER_LC(*p);
2472 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2475 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2477 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2480 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2484 else { /* utf8, ord above 255 */
2485 result = CALL_UPPER_CASE(p, ustrp, lenp);
2488 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2493 /* Here, used locale rules. Convert back to utf8 */
2494 if (UTF8_IS_INVARIANT(result)) {
2495 *ustrp = (U8) result;
2499 *ustrp = UTF8_EIGHT_BIT_HI(result);
2500 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2505 *tainted_ptr = TRUE;
2511 =for apidoc to_utf8_title
2513 Convert the UTF-8 encoded character at C<p> to its titlecase version and
2514 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2515 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2516 titlecase version may be longer than the original character.
2518 The first character of the titlecased version is returned
2519 (but note, as explained above, that there may be more.)
2521 The character at C<p> is assumed by this routine to be well-formed.
2525 /* Not currently externally documented, and subject to change:
2526 * <flags> is set iff locale semantics are to be used for code points < 256
2527 * Since titlecase is not defined in POSIX, uppercase is used instead
2529 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2530 * were used in the calculation; otherwise unchanged. */
2533 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2539 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2541 if (UTF8_IS_INVARIANT(*p)) {
2543 result = toUPPER_LC(*p);
2546 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2549 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2551 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2554 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2558 else { /* utf8, ord above 255 */
2559 result = CALL_TITLE_CASE(p, ustrp, lenp);
2562 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2567 /* Here, used locale rules. Convert back to utf8 */
2568 if (UTF8_IS_INVARIANT(result)) {
2569 *ustrp = (U8) result;
2573 *ustrp = UTF8_EIGHT_BIT_HI(result);
2574 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2579 *tainted_ptr = TRUE;
2585 =for apidoc to_utf8_lower
2587 Convert the UTF-8 encoded character at C<p> to its lowercase version and
2588 store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2589 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2590 lowercase version may be longer than the original character.
2592 The first character of the lowercased version is returned
2593 (but note, as explained above, that there may be more.)
2595 The character at C<p> is assumed by this routine to be well-formed.
2599 /* Not currently externally documented, and subject to change:
2600 * <flags> is set iff locale semantics are to be used for code points < 256
2601 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2602 * were used in the calculation; otherwise unchanged. */
2605 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2611 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2613 if (UTF8_IS_INVARIANT(*p)) {
2615 result = toLOWER_LC(*p);
2618 return to_lower_latin1(*p, ustrp, lenp);
2621 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2623 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2626 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2630 else { /* utf8, ord above 255 */
2631 result = CALL_LOWER_CASE(p, ustrp, lenp);
2634 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2640 /* Here, used locale rules. Convert back to utf8 */
2641 if (UTF8_IS_INVARIANT(result)) {
2642 *ustrp = (U8) result;
2646 *ustrp = UTF8_EIGHT_BIT_HI(result);
2647 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2652 *tainted_ptr = TRUE;
2658 =for apidoc to_utf8_fold
2660 Convert the UTF-8 encoded character at C<p> to its foldcase version and
2661 store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2662 that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2663 foldcase version may be longer than the original character (up to
2666 The first character of the foldcased version is returned
2667 (but note, as explained above, that there may be more.)
2669 The character at C<p> is assumed by this routine to be well-formed.
2673 /* Not currently externally documented, and subject to change,
2675 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2676 * points < 256. Since foldcase is not defined in
2677 * POSIX, lowercase is used instead
2678 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2679 * otherwise simple folds
2680 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2682 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2683 * were used in the calculation; otherwise unchanged. */
2686 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2692 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2694 /* These are mutually exclusive */
2695 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2697 assert(p != ustrp); /* Otherwise overwrites */
2699 if (UTF8_IS_INVARIANT(*p)) {
2700 if (flags & FOLD_FLAGS_LOCALE) {
2701 result = toLOWER_LC(*p);
2704 return _to_fold_latin1(*p, ustrp, lenp,
2705 cBOOL(flags & FOLD_FLAGS_FULL));
2708 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2709 if (flags & FOLD_FLAGS_LOCALE) {
2710 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2713 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2715 cBOOL((flags & FOLD_FLAGS_FULL
2716 /* If ASCII safe, don't allow full
2717 * folding, as that could include SHARP
2718 * S => ss; otherwise there is no
2719 * crossing of ascii/non-ascii in the
2721 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
2724 else { /* utf8, ord above 255 */
2725 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
2727 if ((flags & FOLD_FLAGS_LOCALE)) {
2728 return check_locale_boundary_crossing(p, result, ustrp, lenp);
2730 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2734 /* This is called when changing the case of a utf8-encoded
2735 * character above the Latin1 range, and the result should not
2736 * contain an ASCII character. */
2738 UV original; /* To store the first code point of <p> */
2740 /* Look at every character in the result; if any cross the
2741 * boundary, the whole thing is disallowed */
2743 U8* e = ustrp + *lenp;
2746 /* Crossed, have to return the original */
2747 original = valid_utf8_to_uvchr(p, lenp);
2748 Copy(p, ustrp, *lenp, char);
2754 /* Here, no characters crossed, result is ok as-is */
2759 /* Here, used locale rules. Convert back to utf8 */
2760 if (UTF8_IS_INVARIANT(result)) {
2761 *ustrp = (U8) result;
2765 *ustrp = UTF8_EIGHT_BIT_HI(result);
2766 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2771 *tainted_ptr = TRUE;
2777 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
2778 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2779 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2783 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2785 PERL_ARGS_ASSERT_SWASH_INIT;
2787 /* Returns a copy of a swash initiated by the called function. This is the
2788 * public interface, and returning a copy prevents others from doing
2789 * mischief on the original */
2791 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
2795 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
2797 /* Initialize and return a swash, creating it if necessary. It does this
2798 * by calling utf8_heavy.pl in the general case. The returned value may be
2799 * the swash's inversion list instead if the input parameters allow it.
2800 * Which is returned should be immaterial to callers, as the only
2801 * operations permitted on a swash, swash_fetch() and
2802 * _get_swash_invlist(), handle both these transparently.
2804 * This interface should only be used by functions that won't destroy or
2805 * adversely change the swash, as doing so affects all other uses of the
2806 * swash in the program; the general public should use 'Perl_swash_init'
2809 * pkg is the name of the package that <name> should be in.
2810 * name is the name of the swash to find. Typically it is a Unicode
2811 * property name, including user-defined ones
2812 * listsv is a string to initialize the swash with. It must be of the form
2813 * documented as the subroutine return value in
2814 * L<perlunicode/User-Defined Character Properties>
2815 * minbits is the number of bits required to represent each data element.
2816 * It is '1' for binary properties.
2817 * none I (khw) do not understand this one, but it is used only in tr///.
2818 * invlist is an inversion list to initialize the swash with (or NULL)
2819 * flags_p if non-NULL is the address of various input and output flag bits
2820 * to the routine, as follows: ('I' means is input to the routine;
2821 * 'O' means output from the routine. Only flags marked O are
2822 * meaningful on return.)
2823 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2824 * came from a user-defined property. (I O)
2825 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2826 * when the swash cannot be located, to simply return NULL. (I)
2827 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2828 * return of an inversion list instead of a swash hash if this routine
2829 * thinks that would result in faster execution of swash_fetch() later
2832 * Thus there are three possible inputs to find the swash: <name>,
2833 * <listsv>, and <invlist>. At least one must be specified. The result
2834 * will be the union of the specified ones, although <listsv>'s various
2835 * actions can intersect, etc. what <name> gives.
2837 * <invlist> is only valid for binary properties */
2840 SV* retval = &PL_sv_undef;
2841 HV* swash_hv = NULL;
2842 const int invlist_swash_boundary =
2843 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2844 ? 512 /* Based on some benchmarking, but not extensive, see commit
2846 : -1; /* Never return just an inversion list */
2848 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2849 assert(! invlist || minbits == 1);
2851 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2853 if (listsv != &PL_sv_undef || strNE(name, "")) {
2855 const size_t pkg_len = strlen(pkg);
2856 const size_t name_len = strlen(name);
2857 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2861 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2863 PUSHSTACKi(PERLSI_MAGIC);
2867 /* We might get here via a subroutine signature which uses a utf8
2868 * parameter name, at which point PL_subname will have been set
2869 * but not yet used. */
2870 save_item(PL_subname);
2871 if (PL_parser && PL_parser->error_count)
2872 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2873 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2874 if (!method) { /* demand load utf8 */
2876 errsv_save = newSVsv(ERRSV);
2877 /* It is assumed that callers of this routine are not passing in
2878 * any user derived data. */
2879 /* Need to do this after save_re_context() as it will set
2880 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2881 * in Perl_magic_get). Even line to create errsv_save can turn on
2883 SAVEBOOL(PL_tainted);
2885 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2888 sv_setsv(ERRSV, errsv_save);
2889 SvREFCNT_dec(errsv_save);
2895 mPUSHp(pkg, pkg_len);
2896 mPUSHp(name, name_len);
2901 errsv_save = newSVsv(ERRSV);
2902 /* If we already have a pointer to the method, no need to use
2903 * call_method() to repeat the lookup. */
2904 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
2905 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2907 retval = *PL_stack_sp--;
2908 SvREFCNT_inc(retval);
2911 sv_setsv(ERRSV, errsv_save);
2912 SvREFCNT_dec(errsv_save);
2915 if (IN_PERL_COMPILETIME) {
2916 CopHINTS_set(PL_curcop, PL_hints);
2918 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2921 /* If caller wants to handle missing properties, let them */
2922 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2926 "Can't find Unicode property definition \"%"SVf"\"",
2928 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2930 } /* End of calling the module to find the swash */
2932 /* If this operation fetched a swash, and we will need it later, get it */
2933 if (retval != &PL_sv_undef
2934 && (minbits == 1 || (flags_p
2936 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2938 swash_hv = MUTABLE_HV(SvRV(retval));
2940 /* If we don't already know that there is a user-defined component to
2941 * this swash, and the user has indicated they wish to know if there is
2942 * one (by passing <flags_p>), find out */
2943 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2944 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2945 if (user_defined && SvUV(*user_defined)) {
2946 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2951 /* Make sure there is an inversion list for binary properties */
2953 SV** swash_invlistsvp = NULL;
2954 SV* swash_invlist = NULL;
2955 bool invlist_in_swash_is_valid = FALSE;
2957 /* If this operation fetched a swash, get its already existing
2958 * inversion list, or create one for it */
2961 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
2962 if (swash_invlistsvp) {
2963 swash_invlist = *swash_invlistsvp;
2964 invlist_in_swash_is_valid = TRUE;
2967 swash_invlist = _swash_to_invlist(retval);
2971 /* If an inversion list was passed in, have to include it */
2974 /* Any fetched swash will by now have an inversion list in it;
2975 * otherwise <swash_invlist> will be NULL, indicating that we
2976 * didn't fetch a swash */
2977 if (swash_invlist) {
2979 /* Add the passed-in inversion list, which invalidates the one
2980 * already stored in the swash */
2981 invlist_in_swash_is_valid = FALSE;
2982 _invlist_union(invlist, swash_invlist, &swash_invlist);
2986 /* Here, there is no swash already. Set up a minimal one, if
2987 * we are going to return a swash */
2988 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
2990 retval = newRV_inc(MUTABLE_SV(swash_hv));
2992 swash_invlist = invlist;
2996 /* Here, we have computed the union of all the passed-in data. It may
2997 * be that there was an inversion list in the swash which didn't get
2998 * touched; otherwise save the one computed one */
2999 if (! invlist_in_swash_is_valid
3000 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3002 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3004 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3008 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3009 SvREFCNT_dec(retval);
3010 retval = newRV_inc(swash_invlist);
3018 /* This API is wrong for special case conversions since we may need to
3019 * return several Unicode characters for a single Unicode character
3020 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3021 * the lower-level routine, and it is similarly broken for returning
3022 * multiple values. --jhi
3023 * For those, you should use to_utf8_case() instead */
3024 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3027 * Returns the value of property/mapping C<swash> for the first character
3028 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3029 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3030 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3032 * A "swash" is a hash which contains initially the keys/values set up by
3033 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3034 * property for all possible code points. Things are stored in a compact form
3035 * (see utf8_heavy.pl) so that calculation is required to find the actual
3036 * property value for a given code point. As code points are looked up, new
3037 * key/value pairs are added to the hash, so that the calculation doesn't have
3038 * to ever be re-done. Further, each calculation is done, not just for the
3039 * desired one, but for a whole block of code points adjacent to that one.
3040 * For binary properties on ASCII machines, the block is usually for 64 code
3041 * points, starting with a code point evenly divisible by 64. Thus if the
3042 * property value for code point 257 is requested, the code goes out and
3043 * calculates the property values for all 64 code points between 256 and 319,
3044 * and stores these as a single 64-bit long bit vector, called a "swatch",
3045 * under the key for code point 256. The key is the UTF-8 encoding for code
3046 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3047 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3048 * for code point 258 is then requested, this code realizes that it would be
3049 * stored under the key for 256, and would find that value and extract the
3050 * relevant bit, offset from 256.
3052 * Non-binary properties are stored in as many bits as necessary to represent
3053 * their values (32 currently, though the code is more general than that), not
3054 * as single bits, but the principal is the same: the value for each key is a
3055 * vector that encompasses the property values for all code points whose UTF-8
3056 * representations are represented by the key. That is, for all code points
3057 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3061 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3064 HV *const hv = MUTABLE_HV(SvRV(swash));
3069 const U8 *tmps = NULL;
3073 const UV c = NATIVE_TO_ASCII(*ptr);
3075 PERL_ARGS_ASSERT_SWASH_FETCH;
3077 /* If it really isn't a hash, it isn't really swash; must be an inversion
3079 if (SvTYPE(hv) != SVt_PVHV) {
3080 return _invlist_contains_cp((SV*)hv,
3082 ? valid_utf8_to_uvchr(ptr, NULL)
3086 /* Convert to utf8 if not already */
3087 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
3088 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3089 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3092 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
3093 * then the "swatch" is a vec() for all the chars which start
3095 * So the key in the hash (klen) is length of encoded char -1
3097 klen = UTF8SKIP(ptr) - 1;
3101 /* If char is invariant then swatch is for all the invariant chars
3102 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
3104 needents = UTF_CONTINUATION_MARK;
3105 off = NATIVE_TO_UTF(ptr[klen]);
3108 /* If char is encoded then swatch is for the prefix */
3109 needents = (1 << UTF_ACCUMULATION_SHIFT);
3110 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3114 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3115 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3116 * it's nothing to sniff at.) Pity we usually come through at least
3117 * two function calls to get here...
3119 * NB: this code assumes that swatches are never modified, once generated!
3122 if (hv == PL_last_swash_hv &&
3123 klen == PL_last_swash_klen &&
3124 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3126 tmps = PL_last_swash_tmps;
3127 slen = PL_last_swash_slen;
3130 /* Try our second-level swatch cache, kept in a hash. */
3131 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3133 /* If not cached, generate it via swatch_get */
3134 if (!svp || !SvPOK(*svp)
3135 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
3136 /* We use utf8n_to_uvuni() as we want an index into
3137 Unicode tables, not a native character number.
3139 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
3141 0 : UTF8_ALLOW_ANY);
3142 swatch = swatch_get(swash,
3143 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
3144 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
3147 if (IN_PERL_COMPILETIME)
3148 CopHINTS_set(PL_curcop, PL_hints);
3150 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3152 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3153 || (slen << 3) < needents)
3154 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3155 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3156 svp, tmps, (UV)slen, (UV)needents);
3159 PL_last_swash_hv = hv;
3160 assert(klen <= sizeof(PL_last_swash_key));
3161 PL_last_swash_klen = (U8)klen;
3162 /* FIXME change interpvar.h? */
3163 PL_last_swash_tmps = (U8 *) tmps;
3164 PL_last_swash_slen = slen;
3166 Copy(ptr, PL_last_swash_key, klen, U8);
3169 switch ((int)((slen << 3) / needents)) {
3171 bit = 1 << (off & 7);
3173 return (tmps[off] & bit) != 0;
3178 return (tmps[off] << 8) + tmps[off + 1] ;
3181 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3183 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3184 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
3185 NORETURN_FUNCTION_END;
3188 /* Read a single line of the main body of the swash input text. These are of
3191 * where each number is hex. The first two numbers form the minimum and
3192 * maximum of a range, and the third is the value associated with the range.
3193 * Not all swashes should have a third number
3195 * On input: l points to the beginning of the line to be examined; it points
3196 * to somewhere in the string of the whole input text, and is
3197 * terminated by a \n or the null string terminator.
3198 * lend points to the null terminator of that string
3199 * wants_value is non-zero if the swash expects a third number
3200 * typestr is the name of the swash's mapping, like 'ToLower'
3201 * On output: *min, *max, and *val are set to the values read from the line.
3202 * returns a pointer just beyond the line examined. If there was no
3203 * valid min number on the line, returns lend+1
3207 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3208 const bool wants_value, const U8* const typestr)
3210 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3211 STRLEN numlen; /* Length of the number */
3212 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3213 | PERL_SCAN_DISALLOW_PREFIX
3214 | PERL_SCAN_SILENT_NON_PORTABLE;
3216 /* nl points to the next \n in the scan */
3217 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3219 /* Get the first number on the line: the range minimum */
3221 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3222 if (numlen) /* If found a hex number, position past it */
3224 else if (nl) { /* Else, go handle next line, if any */
3225 return nl + 1; /* 1 is length of "\n" */
3227 else { /* Else, no next line */
3228 return lend + 1; /* to LIST's end at which \n is not found */
3231 /* The max range value follows, separated by a BLANK */
3234 flags = PERL_SCAN_SILENT_ILLDIGIT
3235 | PERL_SCAN_DISALLOW_PREFIX
3236 | PERL_SCAN_SILENT_NON_PORTABLE;
3238 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3241 else /* If no value here, it is a single element range */
3244 /* Non-binary tables have a third entry: what the first element of the
3250 /* The ToLc, etc table mappings are not in hex, and must be
3251 * corrected by adding the code point to them */
3253 char *after_strtol = (char *) lend;
3254 *val = Strtol((char *)l, &after_strtol, 10);
3255 l = (U8 *) after_strtol;
3257 else { /* Other tables are in hex, and are the correct result
3259 flags = PERL_SCAN_SILENT_ILLDIGIT
3260 | PERL_SCAN_DISALLOW_PREFIX
3261 | PERL_SCAN_SILENT_NON_PORTABLE;
3263 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3273 /* diag_listed_as: To%s: illegal mapping '%s' */
3274 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3280 *val = 0; /* bits == 1, then any val should be ignored */
3282 else { /* Nothing following range min, should be single element with no
3288 /* diag_listed_as: To%s: illegal mapping '%s' */
3289 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3293 *val = 0; /* bits == 1, then val should be ignored */
3296 /* Position to next line if any, or EOF */
3306 * Returns a swatch (a bit vector string) for a code point sequence
3307 * that starts from the value C<start> and comprises the number C<span>.
3308 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3309 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3312 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3315 U8 *l, *lend, *x, *xend, *s, *send;
3316 STRLEN lcur, xcur, scur;
3317 HV *const hv = MUTABLE_HV(SvRV(swash));
3318 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3320 SV** listsvp = NULL; /* The string containing the main body of the table */
3321 SV** extssvp = NULL;
3322 SV** invert_it_svp = NULL;
3325 STRLEN octets; /* if bits == 1, then octets == 0 */
3327 UV end = start + span;
3329 if (invlistsvp == NULL) {
3330 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3331 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3332 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3333 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3334 listsvp = hv_fetchs(hv, "LIST", FALSE);
3335 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3337 bits = SvUV(*bitssvp);
3338 none = SvUV(*nonesvp);
3339 typestr = (U8*)SvPV_nolen(*typesvp);
3345 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3347 PERL_ARGS_ASSERT_SWATCH_GET;
3349 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3350 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
3354 /* If overflowed, use the max possible */
3360 /* create and initialize $swatch */
3361 scur = octets ? (span * octets) : (span + 7) / 8;
3362 swatch = newSV(scur);
3364 s = (U8*)SvPVX(swatch);
3365 if (octets && none) {
3366 const U8* const e = s + scur;
3369 *s++ = (U8)(none & 0xff);
3370 else if (bits == 16) {
3371 *s++ = (U8)((none >> 8) & 0xff);
3372 *s++ = (U8)( none & 0xff);
3374 else if (bits == 32) {
3375 *s++ = (U8)((none >> 24) & 0xff);
3376 *s++ = (U8)((none >> 16) & 0xff);
3377 *s++ = (U8)((none >> 8) & 0xff);
3378 *s++ = (U8)( none & 0xff);
3384 (void)memzero((U8*)s, scur + 1);
3386 SvCUR_set(swatch, scur);
3387 s = (U8*)SvPVX(swatch);
3389 if (invlistsvp) { /* If has an inversion list set up use that */
3390 _invlist_populate_swatch(*invlistsvp, start, end, s);
3394 /* read $swash->{LIST} */
3395 l = (U8*)SvPV(*listsvp, lcur);
3398 UV min, max, val, upper;
3399 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3400 cBOOL(octets), typestr);
3405 /* If looking for something beyond this range, go try the next one */
3409 /* <end> is generally 1 beyond where we want to set things, but at the
3410 * platform's infinity, where we can't go any higher, we want to
3411 * include the code point at <end> */
3414 : (max != UV_MAX || end != UV_MAX)
3421 if (!none || val < none) {
3426 for (key = min; key <= upper; key++) {
3428 /* offset must be non-negative (start <= min <= key < end) */
3429 offset = octets * (key - start);
3431 s[offset] = (U8)(val & 0xff);
3432 else if (bits == 16) {
3433 s[offset ] = (U8)((val >> 8) & 0xff);
3434 s[offset + 1] = (U8)( val & 0xff);
3436 else if (bits == 32) {
3437 s[offset ] = (U8)((val >> 24) & 0xff);
3438 s[offset + 1] = (U8)((val >> 16) & 0xff);
3439 s[offset + 2] = (U8)((val >> 8) & 0xff);
3440 s[offset + 3] = (U8)( val & 0xff);
3443 if (!none || val < none)
3447 else { /* bits == 1, then val should be ignored */
3452 for (key = min; key <= upper; key++) {
3453 const STRLEN offset = (STRLEN)(key - start);
3454 s[offset >> 3] |= 1 << (offset & 7);
3459 /* Invert if the data says it should be. Assumes that bits == 1 */
3460 if (invert_it_svp && SvUV(*invert_it_svp)) {
3462 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3463 * be 0, and their inversion should also be 0, as we don't succeed any
3464 * Unicode property matches for non-Unicode code points */
3465 if (start <= PERL_UNICODE_MAX) {
3467 /* The code below assumes that we never cross the
3468 * Unicode/above-Unicode boundary in a range, as otherwise we would
3469 * have to figure out where to stop flipping the bits. Since this
3470 * boundary is divisible by a large power of 2, and swatches comes
3471 * in small powers of 2, this should be a valid assumption */
3472 assert(start + span - 1 <= PERL_UNICODE_MAX);
3482 /* read $swash->{EXTRAS}
3483 * This code also copied to swash_to_invlist() below */
3484 x = (U8*)SvPV(*extssvp, xcur);
3492 SV **otherbitssvp, *other;
3496 const U8 opc = *x++;
3500 nl = (U8*)memchr(x, '\n', xend - x);
3502 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3504 x = nl + 1; /* 1 is length of "\n" */
3508 x = xend; /* to EXTRAS' end at which \n is not found */
3515 namelen = nl - namestr;
3519 namelen = xend - namestr;
3523 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3524 otherhv = MUTABLE_HV(SvRV(*othersvp));
3525 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3526 otherbits = (STRLEN)SvUV(*otherbitssvp);
3527 if (bits < otherbits)
3528 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3529 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
3531 /* The "other" swatch must be destroyed after. */
3532 other = swatch_get(*othersvp, start, span);
3533 o = (U8*)SvPV(other, olen);
3536 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
3538 s = (U8*)SvPV(swatch, slen);
3539 if (bits == 1 && otherbits == 1) {
3541 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3542 "mismatch, slen=%"UVuf", olen=%"UVuf,
3543 (UV)slen, (UV)olen);
3567 STRLEN otheroctets = otherbits >> 3;
3569 U8* const send = s + slen;
3574 if (otherbits == 1) {
3575 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3579 STRLEN vlen = otheroctets;
3587 if (opc == '+' && otherval)
3588 NOOP; /* replace with otherval */
3589 else if (opc == '!' && !otherval)
3591 else if (opc == '-' && otherval)
3593 else if (opc == '&' && !otherval)
3596 s += octets; /* no replacement */
3601 *s++ = (U8)( otherval & 0xff);
3602 else if (bits == 16) {
3603 *s++ = (U8)((otherval >> 8) & 0xff);
3604 *s++ = (U8)( otherval & 0xff);
3606 else if (bits == 32) {
3607 *s++ = (U8)((otherval >> 24) & 0xff);
3608 *s++ = (U8)((otherval >> 16) & 0xff);
3609 *s++ = (U8)((otherval >> 8) & 0xff);
3610 *s++ = (U8)( otherval & 0xff);
3614 sv_free(other); /* through with it! */
3620 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3623 /* Subject to change or removal. For use only in one place in regcomp.c.
3624 * Can't be used on a property that is subject to user override, as it
3625 * relies on the value of SPECIALS in the swash which would be set by
3626 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3627 * for overridden properties
3629 * Returns a hash which is the inversion and closure of a swash mapping.
3630 * For example, consider the input lines:
3635 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3636 * 006C. The value for each key is an array. For 006C, the array would
3637 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3638 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3640 * Essentially, for any code point, it gives all the code points that map to
3641 * it, or the list of 'froms' for that point.
3643 * Currently it ignores any additions or deletions from other swashes,
3644 * looking at just the main body of the swash, and if there are SPECIALS
3645 * in the swash, at that hash
3647 * The specials hash can be extra code points, and most likely consists of
3648 * maps from single code points to multiple ones (each expressed as a string
3649 * of utf8 characters). This function currently returns only 1-1 mappings.
3650 * However consider this possible input in the specials hash:
3651 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3652 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3654 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3655 * currently handle. But it also means that FB05 and FB06 are equivalent in
3656 * a 1-1 mapping which we should handle, and this relationship may not be in
3657 * the main table. Therefore this function examines all the multi-char
3658 * sequences and adds the 1-1 mappings that come out of that. */
3662 HV *const hv = MUTABLE_HV(SvRV(swash));
3664 /* The string containing the main body of the table */
3665 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3667 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3668 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3669 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3670 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3671 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3672 const STRLEN bits = SvUV(*bitssvp);
3673 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3674 const UV none = SvUV(*nonesvp);
3675 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3679 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3681 /* Must have at least 8 bits to get the mappings */
3682 if (bits != 8 && bits != 16 && bits != 32) {
3683 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3687 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3688 mapping to more than one character */
3690 /* Construct an inverse mapping hash for the specials */
3691 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3692 HV * specials_inverse = newHV();
3693 char *char_from; /* the lhs of the map */
3694 I32 from_len; /* its byte length */
3695 char *char_to; /* the rhs of the map */
3696 I32 to_len; /* its byte length */
3697 SV *sv_to; /* and in a sv */
3698 AV* from_list; /* list of things that map to each 'to' */
3700 hv_iterinit(specials_hv);
3702 /* The keys are the characters (in utf8) that map to the corresponding
3703 * utf8 string value. Iterate through the list creating the inverse
3705 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3707 if (! SvPOK(sv_to)) {
3708 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3709 "unexpectedly is not a string, flags=%lu",
3710 (unsigned long)SvFLAGS(sv_to));
3712 /*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)));*/
3714 /* Each key in the inverse list is a mapped-to value, and the key's
3715 * hash value is a list of the strings (each in utf8) that map to
3716 * it. Those strings are all one character long */
3717 if ((listp = hv_fetch(specials_inverse,
3721 from_list = (AV*) *listp;
3723 else { /* No entry yet for it: create one */
3724 from_list = newAV();
3725 if (! hv_store(specials_inverse,
3728 (SV*) from_list, 0))
3730 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3734 /* Here have the list associated with this 'to' (perhaps newly
3735 * created and empty). Just add to it. Note that we ASSUME that
3736 * the input is guaranteed to not have duplications, so we don't
3737 * check for that. Duplications just slow down execution time. */
3738 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3741 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3742 * it looking for cases like the FB05/FB06 examples above. There would
3743 * be an entry in the hash like
3744 * 'st' => [ FB05, FB06 ]
3745 * In this example we will create two lists that get stored in the
3746 * returned hash, 'ret':
3747 * FB05 => [ FB05, FB06 ]
3748 * FB06 => [ FB05, FB06 ]
3750 * Note that there is nothing to do if the array only has one element.
3751 * (In the normal 1-1 case handled below, we don't have to worry about
3752 * two lists, as everything gets tied to the single list that is
3753 * generated for the single character 'to'. But here, we are omitting
3754 * that list, ('st' in the example), so must have multiple lists.) */
3755 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3756 &char_to, &to_len)))
3758 if (av_len(from_list) > 0) {
3761 /* We iterate over all combinations of i,j to place each code
3762 * point on each list */
3763 for (i = 0; i <= av_len(from_list); i++) {
3765 AV* i_list = newAV();
3766 SV** entryp = av_fetch(from_list, i, FALSE);
3767 if (entryp == NULL) {
3768 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3770 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3771 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3773 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3774 (SV*) i_list, FALSE))
3776 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3779 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3780 for (j = 0; j <= av_len(from_list); j++) {
3781 entryp = av_fetch(from_list, j, FALSE);
3782 if (entryp == NULL) {
3783 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3786 /* When i==j this adds itself to the list */
3787 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3788 (U8*) SvPVX(*entryp),
3789 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3791 /*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));*/
3796 SvREFCNT_dec(specials_inverse); /* done with it */
3797 } /* End of specials */
3799 /* read $swash->{LIST} */
3800 l = (U8*)SvPV(*listsvp, lcur);
3803 /* Go through each input line */
3807 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3808 cBOOL(octets), typestr);
3813 /* Each element in the range is to be inverted */
3814 for (inverse = min; inverse <= max; inverse++) {
3818 bool found_key = FALSE;
3819 bool found_inverse = FALSE;
3821 /* The key is the inverse mapping */
3822 char key[UTF8_MAXBYTES+1];
3823 char* key_end = (char *) uvuni_to_utf8((U8*) key, val);
3824 STRLEN key_len = key_end - key;
3826 /* Get the list for the map */
3827 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3828 list = (AV*) *listp;
3830 else { /* No entry yet for it: create one */
3832 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3833 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3837 /* Look through list to see if this inverse mapping already is
3838 * listed, or if there is a mapping to itself already */
3839 for (i = 0; i <= av_len(list); i++) {
3840 SV** entryp = av_fetch(list, i, FALSE);
3842 if (entryp == NULL) {
3843 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3846 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3847 if (SvUV(entry) == val) {
3850 if (SvUV(entry) == inverse) {
3851 found_inverse = TRUE;
3854 /* No need to continue searching if found everything we are
3856 if (found_key && found_inverse) {
3861 /* Make sure there is a mapping to itself on the list */
3863 av_push(list, newSVuv(val));
3864 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
3868 /* Simply add the value to the list */
3869 if (! found_inverse) {
3870 av_push(list, newSVuv(inverse));
3871 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
3874 /* swatch_get() increments the value of val for each element in the
3875 * range. That makes more compact tables possible. You can
3876 * express the capitalization, for example, of all consecutive
3877 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3878 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3879 * and it's not documented; it appears to be used only in
3880 * implementing tr//; I copied the semantics from swatch_get(), just
3882 if (!none || val < none) {
3892 Perl__swash_to_invlist(pTHX_ SV* const swash)
3895 /* Subject to change or removal. For use only in one place in regcomp.c */
3900 HV *const hv = MUTABLE_HV(SvRV(swash));
3901 UV elements = 0; /* Number of elements in the inversion list */
3904 /* The string containing the main body of the table */
3905 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3906 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3907 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3908 SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3909 SV** const invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3911 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3912 const STRLEN bits = SvUV(*bitssvp);
3913 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3919 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3921 /* read $swash->{LIST} */
3922 if (SvPOK(*listsvp)) {
3923 l = (U8*)SvPV(*listsvp, lcur);
3926 /* LIST legitimately doesn't contain a string during compilation phases
3927 * of Perl itself, before the Unicode tables are generated. In this
3928 * case, just fake things up by creating an empty list */
3935 /* Scan the input to count the number of lines to preallocate array size
3936 * based on worst possible case, which is each line in the input creates 2
3937 * elements in the inversion list: 1) the beginning of a range in the list;
3938 * 2) the beginning of a range not in the list. */
3939 while ((loc = (strchr(loc, '\n'))) != NULL) {
3944 /* If the ending is somehow corrupt and isn't a new line, add another
3945 * element for the final range that isn't in the inversion list */
3946 if (! (*lend == '\n'
3947 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3952 invlist = _new_invlist(elements);
3954 /* Now go through the input again, adding each range to the list */
3957 UV val; /* Not used by this function */
3959 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
3960 cBOOL(octets), typestr);
3966 invlist = _add_range_to_invlist(invlist, start, end);
3969 /* Invert if the data says it should be */
3970 if (invert_it_svp && SvUV(*invert_it_svp)) {
3971 _invlist_invert_prop(invlist);
3974 /* This code is copied from swatch_get()
3975 * read $swash->{EXTRAS} */
3976 x = (U8*)SvPV(*extssvp, xcur);
3984 SV **otherbitssvp, *other;
3987 const U8 opc = *x++;
3991 nl = (U8*)memchr(x, '\n', xend - x);
3993 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3995 x = nl + 1; /* 1 is length of "\n" */
3999 x = xend; /* to EXTRAS' end at which \n is not found */
4006 namelen = nl - namestr;
4010 namelen = xend - namestr;
4014 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4015 otherhv = MUTABLE_HV(SvRV(*othersvp));
4016 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4017 otherbits = (STRLEN)SvUV(*otherbitssvp);
4019 if (bits != otherbits || bits != 1) {
4020 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4021 "properties, bits=%"UVuf", otherbits=%"UVuf,
4022 (UV)bits, (UV)otherbits);
4025 /* The "other" swatch must be destroyed after. */
4026 other = _swash_to_invlist((SV *)*othersvp);
4028 /* End of code copied from swatch_get() */
4031 _invlist_union(invlist, other, &invlist);
4034 _invlist_invert(other);
4035 _invlist_union(invlist, other, &invlist);
4038 _invlist_subtract(invlist, other, &invlist);
4041 _invlist_intersection(invlist, other, &invlist);
4046 sv_free(other); /* through with it! */
4053 Perl__get_swash_invlist(pTHX_ SV* const swash)
4057 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4059 if (! SvROK(swash)) {
4063 /* If it really isn't a hash, it isn't really swash; must be an inversion
4065 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4069 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4078 =for apidoc uvchr_to_utf8
4080 Adds the UTF-8 representation of the Native code point C<uv> to the end
4081 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
4082 bytes available. The return value is the pointer to the byte after the
4083 end of the new character. In other words,
4085 d = uvchr_to_utf8(d, uv);
4087 is the recommended wide native character-aware way of saying
4094 /* On ASCII machines this is normally a macro but we want a
4095 real function in case XS code wants it
4098 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
4100 PERL_ARGS_ASSERT_UVCHR_TO_UTF8;
4102 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), 0);
4106 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4108 PERL_ARGS_ASSERT_UVCHR_TO_UTF8_FLAGS;
4110 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), flags);
4114 =for apidoc utf8n_to_uvchr
4116 Returns the native character value of the first character in the string
4118 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
4119 length, in bytes, of that character.
4121 C<length> and C<flags> are the same as L</utf8n_to_uvuni>().
4125 /* On ASCII machines this is normally a macro but we want
4126 a real function in case XS code wants it
4129 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen,
4132 const UV uv = Perl_utf8n_to_uvuni(aTHX_ s, curlen, retlen, flags);
4134 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
4136 return UNI_TO_NATIVE(uv);
4140 Perl_check_utf8_print(pTHX_ register const U8* s, const STRLEN len)
4142 /* May change: warns if surrogates, non-character code points, or
4143 * non-Unicode code points are in s which has length len bytes. Returns
4144 * TRUE if none found; FALSE otherwise. The only other validity check is
4145 * to make sure that this won't exceed the string's length */
4147 const U8* const e = s + len;
4150 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4153 if (UTF8SKIP(s) > len) {
4154 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4155 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4158 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
4160 if (UTF8_IS_SUPER(s)) {
4161 if (ckWARN_d(WARN_NON_UNICODE)) {
4162 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4163 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
4164 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
4168 else if (UTF8_IS_SURROGATE(s)) {
4169 if (ckWARN_d(WARN_SURROGATE)) {
4170 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4171 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4172 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
4177 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
4178 && (ckWARN_d(WARN_NONCHAR)))
4180 UV uv = utf8_to_uvchr_buf(s, e, &char_len);
4181 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
4182 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
4193 =for apidoc pv_uni_display
4195 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4196 length C<len>, the displayable version being at most C<pvlim> bytes long
4197 (if longer, the rest is truncated and "..." will be appended).
4199 The C<flags> argument can have UNI_DISPLAY_ISPRINT set to display
4200 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
4201 to display the \\[nrfta\\] as the backslashed versions (like '\n')
4202 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
4203 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
4204 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
4206 The pointer to the PV of the C<dsv> is returned.
4210 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4215 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4219 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4221 /* This serves double duty as a flag and a character to print after
4222 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4226 if (pvlim && SvCUR(dsv) >= pvlim) {
4230 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4232 const unsigned char c = (unsigned char)u & 0xFF;
4233 if (flags & UNI_DISPLAY_BACKSLASH) {
4250 const char string = ok;
4251 sv_catpvs(dsv, "\\");
4252 sv_catpvn(dsv, &string, 1);
4255 /* isPRINT() is the locale-blind version. */
4256 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4257 const char string = c;
4258 sv_catpvn(dsv, &string, 1);
4263 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
4266 sv_catpvs(dsv, "...");
4272 =for apidoc sv_uni_display
4274 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4275 the displayable version being at most C<pvlim> bytes long
4276 (if longer, the rest is truncated and "..." will be appended).
4278 The C<flags> argument is as in L</pv_uni_display>().
4280 The pointer to the PV of the C<dsv> is returned.
4285 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4287 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4289 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)SvPVX_const(ssv),
4290 SvCUR(ssv), pvlim, flags);
4294 =for apidoc foldEQ_utf8
4296 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4297 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4298 How far into the strings to compare is determined by other input parameters.
4300 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4301 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4302 with respect to C<s2>.
4304 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4305 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4306 scan will not be considered to be a match unless the goal is reached, and
4307 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4310 If C<pe1> is non-NULL and the pointer it points to is not NULL, that pointer is
4311 considered an end pointer beyond which scanning of C<s1> will not continue under
4312 any circumstances. This means that if both C<l1> and C<pe1> are specified, and
4314 is less than C<s1>+C<l1>, the match will never be successful because it can
4316 get as far as its goal (and in fact is asserted against). Correspondingly for
4317 C<pe2> with respect to C<s2>.
4319 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4320 C<l2> must be non-zero), and if both do, both have to be
4321 reached for a successful match. Also, if the fold of a character is multiple
4322 characters, all of them must be matched (see tr21 reference below for
4325 Upon a successful match, if C<pe1> is non-NULL,
4326 it will be set to point to the beginning of the I<next> character of C<s1>
4327 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4329 For case-insensitiveness, the "casefolding" of Unicode is used
4330 instead of upper/lowercasing both the characters, see
4331 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4335 /* A flags parameter has been added which may change, and hence isn't
4336 * externally documented. Currently it is:
4337 * 0 for as-documented above
4338 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4339 ASCII one, to not match
4340 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
4341 * points below 256; unicode rules for above 255; and
4342 * folds that cross those boundaries are disallowed,
4343 * like the NOMIX_ASCII option
4344 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4345 * routine. This allows that step to be skipped.
4346 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4349 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, register UV l1, bool u1, const char *s2, char **pe2, register UV l2, bool u2, U32 flags)
4352 const U8 *p1 = (const U8*)s1; /* Point to current char */
4353 const U8 *p2 = (const U8*)s2;
4354 const U8 *g1 = NULL; /* goal for s1 */
4355 const U8 *g2 = NULL;
4356 const U8 *e1 = NULL; /* Don't scan s1 past this */
4357 U8 *f1 = NULL; /* Point to current folded */
4358 const U8 *e2 = NULL;
4360 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4361 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4362 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4364 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4366 /* The algorithm requires that input with the flags on the first line of
4367 * the assert not be pre-folded. */
4368 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
4369 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
4376 g1 = (const U8*)s1 + l1;
4384 g2 = (const U8*)s2 + l2;
4387 /* Must have at least one goal */
4392 /* Will never match if goal is out-of-bounds */
4393 assert(! e1 || e1 >= g1);
4395 /* Here, there isn't an end pointer, or it is beyond the goal. We
4396 * only go as far as the goal */
4400 assert(e1); /* Must have an end for looking at s1 */
4403 /* Same for goal for s2 */
4405 assert(! e2 || e2 >= g2);
4412 /* If both operands are already folded, we could just do a memEQ on the
4413 * whole strings at once, but it would be better if the caller realized
4414 * this and didn't even call us */
4416 /* Look through both strings, a character at a time */
4417 while (p1 < e1 && p2 < e2) {
4419 /* If at the beginning of a new character in s1, get its fold to use
4420 * and the length of the fold. (exception: locale rules just get the
4421 * character to a single byte) */
4423 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4429 /* If in locale matching, we use two sets of rules, depending
4430 * on if the code point is above or below 255. Here, we test
4431 * for and handle locale rules */
4432 if ((flags & FOLDEQ_UTF8_LOCALE)
4433 && (! u1 || UTF8_IS_INVARIANT(*p1)
4434 || UTF8_IS_DOWNGRADEABLE_START(*p1)))
4436 /* There is no mixing of code points above and below 255. */
4437 if (u2 && (! UTF8_IS_INVARIANT(*p2)
4438 && ! UTF8_IS_DOWNGRADEABLE_START(*p2)))
4443 /* We handle locale rules by converting, if necessary, the
4444 * code point to a single byte. */
4445 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
4449 *foldbuf1 = TWO_BYTE_UTF8_TO_UNI(*p1, *(p1 + 1));
4453 else if (isASCII(*p1)) { /* Note, that here won't be both
4454 ASCII and using locale rules */
4456 /* If trying to mix non- with ASCII, and not supposed to,
4458 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4462 *foldbuf1 = toLOWER(*p1); /* Folds in the ASCII range are
4466 to_utf8_fold(p1, foldbuf1, &n1);
4468 else { /* Not utf8, get utf8 fold */
4469 to_uni_fold(NATIVE_TO_UNI(*p1), foldbuf1, &n1);
4475 if (n2 == 0) { /* Same for s2 */
4476 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4481 if ((flags & FOLDEQ_UTF8_LOCALE)
4482 && (! u2 || UTF8_IS_INVARIANT(*p2) || UTF8_IS_DOWNGRADEABLE_START(*p2)))
4484 /* Here, the next char in s2 is < 256. We've already
4485 * worked on s1, and if it isn't also < 256, can't match */
4486 if (u1 && (! UTF8_IS_INVARIANT(*p1)
4487 && ! UTF8_IS_DOWNGRADEABLE_START(*p1)))
4491 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
4495 *foldbuf2 = TWO_BYTE_UTF8_TO_UNI(*p2, *(p2 + 1));
4498 /* Use another function to handle locale rules. We've made
4499 * sure that both characters to compare are single bytes */
4500 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
4505 else if (isASCII(*p2)) {
4506 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4510 *foldbuf2 = toLOWER(*p2);
4513 to_utf8_fold(p2, foldbuf2, &n2);
4516 to_uni_fold(NATIVE_TO_UNI(*p2), foldbuf2, &n2);
4522 /* Here f1 and f2 point to the beginning of the strings to compare.
4523 * These strings are the folds of the next character from each input
4524 * string, stored in utf8. */
4526 /* While there is more to look for in both folds, see if they
4527 * continue to match */
4529 U8 fold_length = UTF8SKIP(f1);
4530 if (fold_length != UTF8SKIP(f2)
4531 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4532 function call for single
4534 || memNE((char*)f1, (char*)f2, fold_length))
4536 return 0; /* mismatch */
4539 /* Here, they matched, advance past them */
4546 /* When reach the end of any fold, advance the input past it */
4548 p1 += u1 ? UTF8SKIP(p1) : 1;
4551 p2 += u2 ? UTF8SKIP(p2) : 1;
4553 } /* End of loop through both strings */
4555 /* A match is defined by each scan that specified an explicit length
4556 * reaching its final goal, and the other not having matched a partial
4557 * character (which can happen when the fold of a character is more than one
4559 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4563 /* Successful match. Set output pointers */
4575 * c-indentation-style: bsd
4577 * indent-tabs-mode: nil
4580 * ex: set ts=8 sts=4 sw=4 et: