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
36 /* Separate prototypes needed because in ASCII systems these are
37 * usually macros but they still are compiled as code, too. */
38 PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
39 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
42 static const char unees[] =
43 "Malformed UTF-8 character (unexpected end of string)";
46 =head1 Unicode Support
48 This file contains various utility functions for manipulating UTF8-encoded
49 strings. For the uninitiated, this is a method of representing arbitrary
50 Unicode characters as a variable number of bytes, in such a way that
51 characters in the ASCII range are unmodified, and a zero byte never appears
52 within non-zero characters.
58 =for apidoc is_ascii_string
60 Returns true if the first C<len> bytes of the given string are the same whether
61 or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
62 is, if they are invariant. On ASCII-ish machines, only ASCII characters
63 fit this definition, hence the function's name.
65 If C<len> is 0, it will be calculated using C<strlen(s)>.
67 See also is_utf8_string(), is_utf8_string_loclen(), and is_utf8_string_loc().
73 Perl_is_ascii_string(const U8 *s, STRLEN len)
75 const U8* const send = s + (len ? len : strlen((const char *)s));
78 PERL_ARGS_ASSERT_IS_ASCII_STRING;
80 for (; x < send; ++x) {
81 if (!UTF8_IS_INVARIANT(*x))
89 =for apidoc uvuni_to_utf8_flags
91 Adds the UTF-8 representation of the code point C<uv> to the end
92 of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
93 bytes available. The return value is the pointer to the byte after the
94 end of the new character. In other words,
96 d = uvuni_to_utf8_flags(d, uv, flags);
100 d = uvuni_to_utf8(d, uv);
102 (which is equivalent to)
104 d = uvuni_to_utf8_flags(d, uv, 0);
106 This is the recommended Unicode-aware way of saying
110 This function will convert to UTF-8 (and not warn) even code points that aren't
111 legal Unicode or are problematic, unless C<flags> contains one or more of the
113 If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
114 the function will raise a warning, provided UTF8 warnings are enabled. If instead
115 UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
116 If both flags are set, the function will both warn and return NULL.
118 The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
119 affect how the function handles a Unicode non-character. And, likewise for the
120 UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
121 above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
122 even less portable) can be warned and/or disallowed even if other above-Unicode
123 code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
126 And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
127 above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
135 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
137 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
139 if (ckWARN_d(WARN_UTF8)) {
140 if (UNICODE_IS_SURROGATE(uv)) {
141 if (flags & UNICODE_WARN_SURROGATE) {
142 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
143 "UTF-16 surrogate U+%04"UVXf, uv);
145 if (flags & UNICODE_DISALLOW_SURROGATE) {
149 else if (UNICODE_IS_SUPER(uv)) {
150 if (flags & UNICODE_WARN_SUPER
151 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
153 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
154 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
156 if (flags & UNICODE_DISALLOW_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
162 else if (UNICODE_IS_NONCHAR(uv)) {
163 if (flags & UNICODE_WARN_NONCHAR) {
164 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
165 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
168 if (flags & UNICODE_DISALLOW_NONCHAR) {
173 if (UNI_IS_INVARIANT(uv)) {
174 *d++ = (U8)UTF_TO_NATIVE(uv);
179 STRLEN len = UNISKIP(uv);
182 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
183 uv >>= UTF_ACCUMULATION_SHIFT;
185 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
188 #else /* Non loop style */
190 *d++ = (U8)(( uv >> 6) | 0xc0);
191 *d++ = (U8)(( uv & 0x3f) | 0x80);
195 *d++ = (U8)(( uv >> 12) | 0xe0);
196 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
201 *d++ = (U8)(( uv >> 18) | 0xf0);
202 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
204 *d++ = (U8)(( uv & 0x3f) | 0x80);
207 if (uv < 0x4000000) {
208 *d++ = (U8)(( uv >> 24) | 0xf8);
209 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
210 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
211 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
212 *d++ = (U8)(( uv & 0x3f) | 0x80);
215 if (uv < 0x80000000) {
216 *d++ = (U8)(( uv >> 30) | 0xfc);
217 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
221 *d++ = (U8)(( uv & 0x3f) | 0x80);
225 if (uv < UTF8_QUAD_MAX)
228 *d++ = 0xfe; /* Can't match U+FEFF! */
229 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
233 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
234 *d++ = (U8)(( uv & 0x3f) | 0x80);
239 *d++ = 0xff; /* Can't match U+FFFE! */
240 *d++ = 0x80; /* 6 Reserved bits */
241 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
242 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
243 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
244 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
245 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
246 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
247 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
248 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
251 *d++ = (U8)(( uv & 0x3f) | 0x80);
255 #endif /* Loop style */
260 Tests if some arbitrary number of bytes begins in a valid UTF-8
261 character. Note that an INVARIANT (i.e. ASCII) character is a valid
262 UTF-8 character. The actual number of bytes in the UTF-8 character
263 will be returned if it is valid, otherwise 0.
265 This is the "slow" version as opposed to the "fast" version which is
266 the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
267 difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
268 or less you should use the IS_UTF8_CHAR(), for lengths of five or more
269 you should use the _slow(). In practice this means that the _slow()
270 will be used very rarely, since the maximum Unicode code point (as of
271 Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
272 the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
277 S_is_utf8_char_slow(const U8 *s, const STRLEN len)
283 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
285 if (UTF8_IS_INVARIANT(u))
288 if (!UTF8_IS_START(u))
291 if (len < 2 || !UTF8_IS_CONTINUATION(s[1]))
297 u = NATIVE_TO_UTF(u);
299 u &= UTF_START_MASK(len);
303 if (!UTF8_IS_CONTINUATION(*s))
305 uv = UTF8_ACCUMULATE(uv, *s);
312 if ((STRLEN)UNISKIP(uv) < len)
319 =for apidoc is_utf8_char
321 Tests if some arbitrary number of bytes begins in a valid UTF-8
322 character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
323 character is a valid UTF-8 character. The actual number of bytes in the UTF-8
324 character will be returned if it is valid, otherwise 0.
326 WARNING: use only if you *know* that C<s> has at least either UTF8_MAXBYTES or
331 Perl_is_utf8_char(const U8 *s)
333 const STRLEN len = UTF8SKIP(s);
335 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
337 if (IS_UTF8_CHAR_FAST(len))
338 return IS_UTF8_CHAR(s, len) ? len : 0;
339 #endif /* #ifdef IS_UTF8_CHAR */
340 return is_utf8_char_slow(s, len);
345 =for apidoc is_utf8_string
347 Returns true if first C<len> bytes of the given string form a valid
348 UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
349 using C<strlen(s)> (which means if you use this option, that C<s> has to have a
350 terminating NUL byte). Note that all characters being ASCII constitute 'a
353 See also is_ascii_string(), is_utf8_string_loclen(), and is_utf8_string_loc().
359 Perl_is_utf8_string(const U8 *s, STRLEN len)
361 const U8* const send = s + (len ? len : strlen((const char *)s));
364 PERL_ARGS_ASSERT_IS_UTF8_STRING;
367 /* Inline the easy bits of is_utf8_char() here for speed... */
368 if (UTF8_IS_INVARIANT(*x)) {
371 else if (!UTF8_IS_START(*x))
374 /* ... and call is_utf8_char() only if really needed. */
375 const STRLEN c = UTF8SKIP(x);
376 const U8* const next_char_ptr = x + c;
378 if (next_char_ptr > send) {
382 if (IS_UTF8_CHAR_FAST(c)) {
383 if (!IS_UTF8_CHAR(x, c))
386 else if (! is_utf8_char_slow(x, c)) {
397 Implemented as a macro in utf8.h
399 =for apidoc is_utf8_string_loc
401 Like is_utf8_string() but stores the location of the failure (in the
402 case of "utf8ness failure") or the location s+len (in the case of
403 "utf8ness success") in the C<ep>.
405 See also is_utf8_string_loclen() and is_utf8_string().
407 =for apidoc is_utf8_string_loclen
409 Like is_utf8_string() but stores the location of the failure (in the
410 case of "utf8ness failure") or the location s+len (in the case of
411 "utf8ness success") in the C<ep>, and the number of UTF-8
412 encoded characters in the C<el>.
414 See also is_utf8_string_loc() and is_utf8_string().
420 Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
422 const U8* const send = s + (len ? len : strlen((const char *)s));
427 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
430 const U8* next_char_ptr;
432 /* Inline the easy bits of is_utf8_char() here for speed... */
433 if (UTF8_IS_INVARIANT(*x))
434 next_char_ptr = x + 1;
435 else if (!UTF8_IS_START(*x))
438 /* ... and call is_utf8_char() only if really needed. */
440 next_char_ptr = c + x;
441 if (next_char_ptr > send) {
444 if (IS_UTF8_CHAR_FAST(c)) {
445 if (!IS_UTF8_CHAR(x, c))
448 c = is_utf8_char_slow(x, c);
467 =for apidoc utf8n_to_uvuni
469 Bottom level UTF-8 decode routine.
470 Returns the code point value of the first character in the string C<s>
471 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding and no longer than
472 C<curlen> bytes; C<retlen> will be set to the length, in bytes, of that
475 The value of C<flags> determines the behavior when C<s> does not point to a
476 well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
477 C<retlen> is set to the expected length of the UTF-8 character in bytes, zero
478 is returned, and if UTF-8 warnings haven't been lexically disabled, a warning
481 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
482 individual types of malformations, such as the sequence being overlong (that
483 is, when there is a shorter sequence that can express the same code point;
484 overlong sequences are expressly forbidden in the UTF-8 standard due to
485 potential security issues). Another malformation example is the first byte of
486 a character not being a legal first byte. See F<utf8.h> for the list of such
487 flags. Of course, the value returned by this function under such conditions is
490 The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
491 flags) malformation is found. If this flag is set, the routine assumes that
492 the caller will raise a warning, and this function will silently just set
493 C<retlen> to C<-1> and return zero.
495 Certain code points are considered problematic. These are Unicode surrogates,
496 Unicode non-characters, and code points above the Unicode maximum of 0x10FFF.
497 By default these are considered regular code points, but certain situations
498 warrant special handling for them. If C<flags> contains
499 UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
500 malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
501 UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
502 maximum) can be set to disallow these categories individually.
504 The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
505 UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
506 for their respective categories, but otherwise the code points are considered
507 valid (not malformations). To get a category to both be treated as a
508 malformation and raise a warning, specify both the WARN and DISALLOW flags.
509 (But note that warnings are not raised if lexically disabled nor if
510 UTF8_CHECK_ONLY is also specified.)
512 Very large code points (above 0x7FFF_FFFF) are considered more problematic than
513 the others that are above the Unicode legal maximum. There are several
514 reasons, one of which is that the original UTF-8 specification never went above
515 this number (the current 0x10FFF limit was imposed later). The UTF-8 encoding
516 on ASCII platforms for these large code points begins with a byte containing
517 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to be treated as
518 malformations, while allowing smaller above-Unicode code points. (Of course
519 UTF8_DISALLOW_SUPER will treat all above-Unicode code points, including these,
520 as malformations.) Similarly, UTF8_WARN_FE_FF acts just like the other WARN
521 flags, but applies just to these code points.
523 All other code points corresponding to Unicode characters, including private
524 use and those yet to be assigned, are never considered malformed and never
527 Most code should use utf8_to_uvchr() rather than call this directly.
533 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
536 const U8 * const s0 = s;
539 bool dowarn = ckWARN_d(WARN_UTF8);
540 const UV startbyte = *s;
541 STRLEN expectlen = 0;
545 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
547 /* This list is a superset of the UTF8_ALLOW_XXX. */
549 #define UTF8_WARN_EMPTY 1
550 #define UTF8_WARN_CONTINUATION 2
551 #define UTF8_WARN_NON_CONTINUATION 3
552 #define UTF8_WARN_SHORT 4
553 #define UTF8_WARN_OVERFLOW 5
554 #define UTF8_WARN_LONG 6
557 !(flags & UTF8_ALLOW_EMPTY)) {
558 warning = UTF8_WARN_EMPTY;
562 if (UTF8_IS_INVARIANT(uv)) {
565 return (UV) (NATIVE_TO_UTF(*s));
568 if (UTF8_IS_CONTINUATION(uv) &&
569 !(flags & UTF8_ALLOW_CONTINUATION)) {
570 warning = UTF8_WARN_CONTINUATION;
574 if (UTF8_IS_START(uv) && curlen > 1 && !UTF8_IS_CONTINUATION(s[1]) &&
575 !(flags & UTF8_ALLOW_NON_CONTINUATION)) {
576 warning = UTF8_WARN_NON_CONTINUATION;
581 uv = NATIVE_TO_UTF(uv);
583 if (uv == 0xfe || uv == 0xff) {
584 if (flags & (UTF8_WARN_SUPER|UTF8_WARN_FE_FF)) {
585 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point beginning with byte 0x%02"UVXf" is not Unicode, and not portable", uv));
586 flags &= ~UTF8_WARN_SUPER; /* Only warn once on this problem */
588 if (flags & (UTF8_DISALLOW_SUPER|UTF8_DISALLOW_FE_FF)) {
594 if (!(uv & 0x20)) { len = 2; uv &= 0x1f; }
595 else if (!(uv & 0x10)) { len = 3; uv &= 0x0f; }
596 else if (!(uv & 0x08)) { len = 4; uv &= 0x07; }
597 else if (!(uv & 0x04)) { len = 5; uv &= 0x03; }
599 else if (!(uv & 0x02)) { len = 6; uv &= 0x01; }
600 else { len = 7; uv &= 0x01; }
602 else if (!(uv & 0x02)) { len = 6; uv &= 0x01; }
603 else if (!(uv & 0x01)) { len = 7; uv = 0; }
604 else { len = 13; uv = 0; } /* whoa! */
612 if ((curlen < expectlen) &&
613 !(flags & UTF8_ALLOW_SHORT)) {
614 warning = UTF8_WARN_SHORT;
620 ouv = uv; /* ouv is the value from the previous iteration */
623 if (!UTF8_IS_CONTINUATION(*s) &&
624 !(flags & UTF8_ALLOW_NON_CONTINUATION)) {
626 warning = UTF8_WARN_NON_CONTINUATION;
630 uv = UTF8_ACCUMULATE(uv, *s);
631 if (!(uv > ouv)) { /* If the value didn't grow from the previous
632 iteration, something is horribly wrong */
633 /* These cannot be allowed. */
635 if (expectlen != 13 && !(flags & UTF8_ALLOW_LONG)) {
636 warning = UTF8_WARN_LONG;
640 else { /* uv < ouv */
641 /* This cannot be allowed. */
642 warning = UTF8_WARN_OVERFLOW;
650 if ((expectlen > (STRLEN)UNISKIP(uv)) && !(flags & UTF8_ALLOW_LONG)) {
651 warning = UTF8_WARN_LONG;
653 } else if (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE|UTF8_WARN_ILLEGAL_INTERCHANGE)) {
654 if (UNICODE_IS_SURROGATE(uv)) {
655 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE) {
656 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
658 if (flags & UTF8_DISALLOW_SURROGATE) {
662 else if (UNICODE_IS_NONCHAR(uv)) {
663 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR ) {
664 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
666 if (flags & UTF8_DISALLOW_NONCHAR) {
670 else if ((uv > PERL_UNICODE_MAX)) {
671 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER) {
672 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
674 if (flags & UTF8_DISALLOW_SUPER) {
679 /* Here, this is not considered a malformed character, so drop through
685 disallowed: /* Is disallowed, but otherwise not malformed. 'sv' will have been
686 set if there is to be a warning. */
693 if (flags & UTF8_CHECK_ONLY) {
695 *retlen = ((STRLEN) -1);
701 sv = newSVpvs_flags("Malformed UTF-8 character ", SVs_TEMP);
705 case 0: /* Intentionally empty. */ break;
706 case UTF8_WARN_EMPTY:
707 sv_catpvs(sv, "(empty string)");
709 case UTF8_WARN_CONTINUATION:
710 Perl_sv_catpvf(aTHX_ sv, "(unexpected continuation byte 0x%02"UVxf", with no preceding start byte)", uv);
712 case UTF8_WARN_NON_CONTINUATION:
714 Perl_sv_catpvf(aTHX_ sv, "(unexpected non-continuation byte 0x%02"UVxf", immediately after start byte 0x%02"UVxf")",
715 (UV)s[1], startbyte);
717 const int len = (int)(s-s0);
718 Perl_sv_catpvf(aTHX_ sv, "(unexpected non-continuation byte 0x%02"UVxf", %d byte%s after start byte 0x%02"UVxf", expected %d bytes)",
719 (UV)s[1], len, len > 1 ? "s" : "", startbyte, (int)expectlen);
723 case UTF8_WARN_SHORT:
724 Perl_sv_catpvf(aTHX_ sv, "(%d byte%s, need %d, after start byte 0x%02"UVxf")",
725 (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, startbyte);
726 expectlen = curlen; /* distance for caller to skip */
728 case UTF8_WARN_OVERFLOW:
729 Perl_sv_catpvf(aTHX_ sv, "(overflow at 0x%"UVxf", byte 0x%02x, after start byte 0x%02"UVxf")",
733 Perl_sv_catpvf(aTHX_ sv, "(%d byte%s, need %d, after start byte 0x%02"UVxf")",
734 (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), startbyte);
737 sv_catpvs(sv, "(unknown reason)");
742 const char * const s = SvPVX_const(sv);
745 Perl_warner(aTHX_ packWARN(WARN_UTF8),
746 "%s in %s", s, OP_DESC(PL_op));
748 Perl_warner(aTHX_ packWARN(WARN_UTF8), "%s", s);
753 *retlen = expectlen ? expectlen : len;
759 =for apidoc utf8_to_uvchr
761 Returns the native code point of the first character in the string C<s>
762 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
763 length, in bytes, of that character.
765 If C<s> does not point to a well-formed UTF-8 character, zero is
766 returned and retlen is set, if possible, to -1.
773 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
775 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
777 return utf8n_to_uvchr(s, UTF8_MAXBYTES, retlen,
778 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
782 =for apidoc utf8_to_uvuni
784 Returns the Unicode code point of the first character in the string C<s>
785 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
786 length, in bytes, of that character.
788 This function should only be used when the returned UV is considered
789 an index into the Unicode semantic tables (e.g. swashes).
791 If C<s> does not point to a well-formed UTF-8 character, zero is
792 returned and retlen is set, if possible, to -1.
798 Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
800 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
802 /* Call the low level routine asking for checks */
803 return Perl_utf8n_to_uvuni(aTHX_ s, UTF8_MAXBYTES, retlen,
804 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
808 =for apidoc utf8_length
810 Return the length of the UTF-8 char encoded string C<s> in characters.
811 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
812 up past C<e>, croaks.
818 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
823 PERL_ARGS_ASSERT_UTF8_LENGTH;
825 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
826 * the bitops (especially ~) can create illegal UTF-8.
827 * In other words: in Perl UTF-8 is not just for Unicode. */
830 goto warn_and_return;
832 if (!UTF8_IS_INVARIANT(*s))
843 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
844 "%s in %s", unees, OP_DESC(PL_op));
846 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
853 =for apidoc utf8_distance
855 Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
858 WARNING: use only if you *know* that the pointers point inside the
865 Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
867 PERL_ARGS_ASSERT_UTF8_DISTANCE;
869 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
875 Return the UTF-8 pointer C<s> displaced by C<off> characters, either
878 WARNING: do not use the following unless you *know* C<off> is within
879 the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
880 on the first byte of character or just after the last byte of a character.
886 Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
888 PERL_ARGS_ASSERT_UTF8_HOP;
891 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
892 * the bitops (especially ~) can create illegal UTF-8.
893 * In other words: in Perl UTF-8 is not just for Unicode. */
902 while (UTF8_IS_CONTINUATION(*s))
910 =for apidoc bytes_cmp_utf8
912 Compares the sequence of characters (stored as octets) in b, blen with the
913 sequence of characters (stored as UTF-8) in u, ulen. Returns 0 if they are
914 equal, -1 or -2 if the first string is less than the second string, +1 or +2
915 if the first string is greater than the second string.
917 -1 or +1 is returned if the shorter string was identical to the start of the
918 longer string. -2 or +2 is returned if the was a difference between characters
925 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
927 const U8 *const bend = b + blen;
928 const U8 *const uend = u + ulen;
930 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
934 while (b < bend && u < uend) {
936 if (!UTF8_IS_INVARIANT(c)) {
937 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
940 if (UTF8_IS_CONTINUATION(c1)) {
941 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
943 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
944 "Malformed UTF-8 character "
945 "(unexpected non-continuation byte 0x%02x"
946 ", immediately after start byte 0x%02x)"
947 /* Dear diag.t, it's in the pod. */
950 PL_op ? OP_DESC(PL_op) : "");
955 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
956 "%s in %s", unees, OP_DESC(PL_op));
958 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
959 return -2; /* Really want to return undef :-) */
966 return *b < c ? -2 : +2;
971 if (b == bend && u == uend)
974 return b < bend ? +1 : -1;
978 =for apidoc utf8_to_bytes
980 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
981 Unlike C<bytes_to_utf8>, this over-writes the original string, and
982 updates len to contain the new length.
983 Returns zero on failure, setting C<len> to -1.
985 If you need a copy of the string, see C<bytes_from_utf8>.
991 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
994 U8 * const send = s + *len;
997 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
999 /* ensure valid UTF-8 and chars < 256 before updating string */
1003 if (!UTF8_IS_INVARIANT(c) &&
1004 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1005 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
1006 *len = ((STRLEN) -1);
1014 *d++ = (U8)utf8_to_uvchr(s, &ulen);
1023 =for apidoc bytes_from_utf8
1025 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1026 Unlike C<utf8_to_bytes> but like C<bytes_to_utf8>, returns a pointer to
1027 the newly-created string, and updates C<len> to contain the new
1028 length. Returns the original string if no conversion occurs, C<len>
1029 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1030 0 if C<s> is converted or consisted entirely of characters that are invariant
1031 in utf8 (i.e., US-ASCII on non-EBCDIC machines).
1037 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1040 const U8 *start = s;
1044 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1046 PERL_UNUSED_CONTEXT;
1050 /* ensure valid UTF-8 and chars < 256 before converting string */
1051 for (send = s + *len; s < send;) {
1053 if (!UTF8_IS_INVARIANT(c)) {
1054 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1055 (c = *s++) && UTF8_IS_CONTINUATION(c))
1064 Newx(d, (*len) - count + 1, U8);
1065 s = start; start = d;
1068 if (!UTF8_IS_INVARIANT(c)) {
1069 /* Then it is two-byte encoded */
1070 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
1080 =for apidoc bytes_to_utf8
1082 Converts a string C<s> of length C<len> bytes from the native encoding into
1084 Returns a pointer to the newly-created string, and sets C<len> to
1085 reflect the new length in bytes.
1087 A NUL character will be written after the end of the string.
1089 If you want to convert to UTF-8 from encodings other than
1090 the native (Latin1 or EBCDIC),
1091 see sv_recode_to_utf8().
1096 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1097 likewise need duplication. */
1100 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1102 const U8 * const send = s + (*len);
1106 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1107 PERL_UNUSED_CONTEXT;
1109 Newx(d, (*len) * 2 + 1, U8);
1113 const UV uv = NATIVE_TO_ASCII(*s++);
1114 if (UNI_IS_INVARIANT(uv))
1115 *d++ = (U8)UTF_TO_NATIVE(uv);
1117 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1118 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
1127 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1129 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1130 * We optimize for native, for obvious reasons. */
1133 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1138 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1141 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
1146 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1150 *d++ = UNI_TO_NATIVE(uv);
1157 *d++ = (U8)(( uv >> 6) | 0xc0);
1158 *d++ = (U8)(( uv & 0x3f) | 0x80);
1161 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
1163 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1165 UV low = (p[0] << 8) + p[1];
1167 if (low < 0xdc00 || low > 0xdfff)
1168 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1169 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1171 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1172 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1175 *d++ = (U8)(( uv >> 12) | 0xe0);
1176 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1177 *d++ = (U8)(( uv & 0x3f) | 0x80);
1181 *d++ = (U8)(( uv >> 18) | 0xf0);
1182 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1184 *d++ = (U8)(( uv & 0x3f) | 0x80);
1188 *newlen = d - dstart;
1192 /* Note: this one is slightly destructive of the source. */
1195 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1198 U8* const send = s + bytelen;
1200 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1203 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1207 const U8 tmp = s[0];
1212 return utf16_to_utf8(p, d, bytelen, newlen);
1215 /* for now these are all defined (inefficiently) in terms of the utf8 versions.
1216 * Note that the macros in handy.h that call these short-circuit calling them
1217 * for Latin-1 range inputs */
1220 Perl_is_uni_alnum(pTHX_ UV c)
1222 U8 tmpbuf[UTF8_MAXBYTES+1];
1223 uvchr_to_utf8(tmpbuf, c);
1224 return is_utf8_alnum(tmpbuf);
1228 Perl_is_uni_idfirst(pTHX_ UV c)
1230 U8 tmpbuf[UTF8_MAXBYTES+1];
1231 uvchr_to_utf8(tmpbuf, c);
1232 return is_utf8_idfirst(tmpbuf);
1236 Perl_is_uni_alpha(pTHX_ UV c)
1238 U8 tmpbuf[UTF8_MAXBYTES+1];
1239 uvchr_to_utf8(tmpbuf, c);
1240 return is_utf8_alpha(tmpbuf);
1244 Perl_is_uni_ascii(pTHX_ UV c)
1250 Perl_is_uni_space(pTHX_ UV c)
1252 U8 tmpbuf[UTF8_MAXBYTES+1];
1253 uvchr_to_utf8(tmpbuf, c);
1254 return is_utf8_space(tmpbuf);
1258 Perl_is_uni_digit(pTHX_ UV c)
1260 U8 tmpbuf[UTF8_MAXBYTES+1];
1261 uvchr_to_utf8(tmpbuf, c);
1262 return is_utf8_digit(tmpbuf);
1266 Perl_is_uni_upper(pTHX_ UV c)
1268 U8 tmpbuf[UTF8_MAXBYTES+1];
1269 uvchr_to_utf8(tmpbuf, c);
1270 return is_utf8_upper(tmpbuf);
1274 Perl_is_uni_lower(pTHX_ UV c)
1276 U8 tmpbuf[UTF8_MAXBYTES+1];
1277 uvchr_to_utf8(tmpbuf, c);
1278 return is_utf8_lower(tmpbuf);
1282 Perl_is_uni_cntrl(pTHX_ UV c)
1284 return isCNTRL_L1(c);
1288 Perl_is_uni_graph(pTHX_ UV c)
1290 U8 tmpbuf[UTF8_MAXBYTES+1];
1291 uvchr_to_utf8(tmpbuf, c);
1292 return is_utf8_graph(tmpbuf);
1296 Perl_is_uni_print(pTHX_ UV c)
1298 U8 tmpbuf[UTF8_MAXBYTES+1];
1299 uvchr_to_utf8(tmpbuf, c);
1300 return is_utf8_print(tmpbuf);
1304 Perl_is_uni_punct(pTHX_ UV c)
1306 U8 tmpbuf[UTF8_MAXBYTES+1];
1307 uvchr_to_utf8(tmpbuf, c);
1308 return is_utf8_punct(tmpbuf);
1312 Perl_is_uni_xdigit(pTHX_ UV c)
1314 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1315 uvchr_to_utf8(tmpbuf, c);
1316 return is_utf8_xdigit(tmpbuf);
1320 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1322 /* We have the latin1-range values compiled into the core, so just use
1323 * those, converting the result to utf8. The only difference between upper
1324 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1325 * either "SS" or "Ss". Which one to use is passed into the routine in
1326 * 'S_or_s' to avoid a test */
1328 UV converted = toUPPER_LATIN1_MOD(c);
1330 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1332 assert(S_or_s == 'S' || S_or_s == 's');
1334 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1335 characters in this range */
1336 *p = (U8) converted;
1341 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1342 * which it maps to one of them, so as to only have to have one check for
1343 * it in the main case */
1344 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1346 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1347 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1350 converted = GREEK_CAPITAL_LETTER_MU;
1352 case LATIN_SMALL_LETTER_SHARP_S:
1358 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1363 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1364 *p = UTF8_TWO_BYTE_LO(converted);
1370 /* Call the function to convert a UTF-8 encoded character to the specified case.
1371 * Note that there may be more than one character in the result.
1372 * INP is a pointer to the first byte of the input character
1373 * OUTP will be set to the first byte of the string of changed characters. It
1374 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1375 * LENP will be set to the length in bytes of the string of changed characters
1377 * The functions return the ordinal of the first character in the string of OUTP */
1378 #define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUpper")
1379 #define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTitle")
1380 #define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLower")
1382 /* This additionally has the input parameter SPECIALS, which if non-zero will
1383 * cause this to use the SPECIALS hash for folding (meaning get full case
1384 * folding); otherwise, when zero, this implies a simple case fold */
1385 #define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecFold" : NULL)
1388 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
1392 /* Convert the Unicode character whose ordinal is c to its uppercase
1393 * version and store that in UTF-8 in p and its length in bytes in lenp.
1394 * Note that the p needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
1395 * the changed version may be longer than the original character.
1397 * The ordinal of the first character of the changed version is returned
1398 * (but note, as explained above, that there may be more.) */
1400 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1403 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1406 uvchr_to_utf8(p, c);
1407 return CALL_UPPER_CASE(p, p, lenp);
1411 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
1415 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1418 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1421 uvchr_to_utf8(p, c);
1422 return CALL_TITLE_CASE(p, p, lenp);
1426 S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1428 /* We have the latin1-range values compiled into the core, so just use
1429 * those, converting the result to utf8. Since the result is always just
1430 * one character, we allow p to be NULL */
1432 U8 converted = toLOWER_LATIN1(c);
1435 if (UNI_IS_INVARIANT(converted)) {
1440 *p = UTF8_TWO_BYTE_HI(converted);
1441 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1449 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
1453 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1456 return to_lower_latin1((U8) c, p, lenp);
1459 uvchr_to_utf8(p, c);
1460 return CALL_LOWER_CASE(p, p, lenp);
1464 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
1466 /* Corresponds to to_lower_latin1(), flags is TRUE if to use full case
1471 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1473 if (c == MICRO_SIGN) {
1474 converted = GREEK_SMALL_LETTER_MU;
1476 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1482 else { /* In this range the fold of all other characters is their lower
1484 converted = toLOWER_LATIN1(c);
1487 if (UNI_IS_INVARIANT(converted)) {
1488 *p = (U8) converted;
1492 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1493 *p = UTF8_TWO_BYTE_LO(converted);
1501 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const bool flags)
1504 /* Not currently externally documented, and subject to change, <flags> is
1505 * TRUE iff full folding is to be used */
1507 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
1510 return _to_fold_latin1((U8) c, p, lenp, flags);
1513 uvchr_to_utf8(p, c);
1514 return CALL_FOLD_CASE(p, p, lenp, flags);
1517 /* for now these all assume no locale info available for Unicode > 255 */
1520 Perl_is_uni_alnum_lc(pTHX_ UV c)
1522 return is_uni_alnum(c); /* XXX no locale support yet */
1526 Perl_is_uni_idfirst_lc(pTHX_ UV c)
1528 return is_uni_idfirst(c); /* XXX no locale support yet */
1532 Perl_is_uni_alpha_lc(pTHX_ UV c)
1534 return is_uni_alpha(c); /* XXX no locale support yet */
1538 Perl_is_uni_ascii_lc(pTHX_ UV c)
1540 return is_uni_ascii(c); /* XXX no locale support yet */
1544 Perl_is_uni_space_lc(pTHX_ UV c)
1546 return is_uni_space(c); /* XXX no locale support yet */
1550 Perl_is_uni_digit_lc(pTHX_ UV c)
1552 return is_uni_digit(c); /* XXX no locale support yet */
1556 Perl_is_uni_upper_lc(pTHX_ UV c)
1558 return is_uni_upper(c); /* XXX no locale support yet */
1562 Perl_is_uni_lower_lc(pTHX_ UV c)
1564 return is_uni_lower(c); /* XXX no locale support yet */
1568 Perl_is_uni_cntrl_lc(pTHX_ UV c)
1570 return is_uni_cntrl(c); /* XXX no locale support yet */
1574 Perl_is_uni_graph_lc(pTHX_ UV c)
1576 return is_uni_graph(c); /* XXX no locale support yet */
1580 Perl_is_uni_print_lc(pTHX_ UV c)
1582 return is_uni_print(c); /* XXX no locale support yet */
1586 Perl_is_uni_punct_lc(pTHX_ UV c)
1588 return is_uni_punct(c); /* XXX no locale support yet */
1592 Perl_is_uni_xdigit_lc(pTHX_ UV c)
1594 return is_uni_xdigit(c); /* XXX no locale support yet */
1598 Perl_to_uni_upper_lc(pTHX_ U32 c)
1600 /* XXX returns only the first character -- do not use XXX */
1601 /* XXX no locale support yet */
1603 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1604 return (U32)to_uni_upper(c, tmpbuf, &len);
1608 Perl_to_uni_title_lc(pTHX_ U32 c)
1610 /* XXX returns only the first character XXX -- do not use XXX */
1611 /* XXX no locale support yet */
1613 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1614 return (U32)to_uni_title(c, tmpbuf, &len);
1618 Perl_to_uni_lower_lc(pTHX_ U32 c)
1620 /* XXX returns only the first character -- do not use XXX */
1621 /* XXX no locale support yet */
1623 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
1624 return (U32)to_uni_lower(c, tmpbuf, &len);
1628 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
1629 const char *const swashname)
1633 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1635 if (!is_utf8_char(p))
1638 *swash = swash_init("utf8", swashname, &PL_sv_undef, 1, 0);
1639 return swash_fetch(*swash, p, TRUE) != 0;
1643 Perl_is_utf8_alnum(pTHX_ const U8 *p)
1647 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1649 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1650 * descendant of isalnum(3), in other words, it doesn't
1651 * contain the '_'. --jhi */
1652 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
1656 Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
1660 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1664 /* is_utf8_idstart would be more logical. */
1665 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1669 Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1673 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1677 /* is_utf8_idstart would be more logical. */
1678 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
1682 Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
1686 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
1688 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
1692 Perl_is_utf8_idcont(pTHX_ const U8 *p)
1696 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
1698 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
1702 Perl_is_utf8_xidcont(pTHX_ const U8 *p)
1706 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
1708 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
1712 Perl_is_utf8_alpha(pTHX_ const U8 *p)
1716 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
1718 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
1722 Perl_is_utf8_ascii(pTHX_ const U8 *p)
1726 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
1728 /* ASCII characters are the same whether in utf8 or not. So the macro
1729 * works on both utf8 and non-utf8 representations. */
1734 Perl_is_utf8_space(pTHX_ const U8 *p)
1738 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
1740 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
1744 Perl_is_utf8_perl_space(pTHX_ const U8 *p)
1748 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
1750 /* Only true if is an ASCII space-like character, and ASCII is invariant
1751 * under utf8, so can just use the macro */
1752 return isSPACE_A(*p);
1756 Perl_is_utf8_perl_word(pTHX_ const U8 *p)
1760 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
1762 /* Only true if is an ASCII word character, and ASCII is invariant
1763 * under utf8, so can just use the macro */
1764 return isWORDCHAR_A(*p);
1768 Perl_is_utf8_digit(pTHX_ const U8 *p)
1772 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
1774 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
1778 Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
1782 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
1784 /* Only true if is an ASCII digit character, and ASCII is invariant
1785 * under utf8, so can just use the macro */
1786 return isDIGIT_A(*p);
1790 Perl_is_utf8_upper(pTHX_ const U8 *p)
1794 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
1796 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
1800 Perl_is_utf8_lower(pTHX_ const U8 *p)
1804 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
1806 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
1810 Perl_is_utf8_cntrl(pTHX_ const U8 *p)
1814 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
1817 return isCNTRL_A(*p);
1820 /* All controls are in Latin1 */
1821 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
1824 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
1828 Perl_is_utf8_graph(pTHX_ const U8 *p)
1832 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
1834 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
1838 Perl_is_utf8_print(pTHX_ const U8 *p)
1842 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
1844 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
1848 Perl_is_utf8_punct(pTHX_ const U8 *p)
1852 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
1854 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
1858 Perl_is_utf8_xdigit(pTHX_ const U8 *p)
1862 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
1864 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
1868 Perl_is_utf8_mark(pTHX_ const U8 *p)
1872 PERL_ARGS_ASSERT_IS_UTF8_MARK;
1874 return is_utf8_common(p, &PL_utf8_mark, "IsM");
1878 Perl_is_utf8_X_begin(pTHX_ const U8 *p)
1882 PERL_ARGS_ASSERT_IS_UTF8_X_BEGIN;
1884 return is_utf8_common(p, &PL_utf8_X_begin, "_X_Begin");
1888 Perl_is_utf8_X_extend(pTHX_ const U8 *p)
1892 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
1894 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
1898 Perl_is_utf8_X_prepend(pTHX_ const U8 *p)
1902 PERL_ARGS_ASSERT_IS_UTF8_X_PREPEND;
1904 return is_utf8_common(p, &PL_utf8_X_prepend, "GCB=Prepend");
1908 Perl_is_utf8_X_non_hangul(pTHX_ const U8 *p)
1912 PERL_ARGS_ASSERT_IS_UTF8_X_NON_HANGUL;
1914 return is_utf8_common(p, &PL_utf8_X_non_hangul, "HST=Not_Applicable");
1918 Perl_is_utf8_X_L(pTHX_ const U8 *p)
1922 PERL_ARGS_ASSERT_IS_UTF8_X_L;
1924 return is_utf8_common(p, &PL_utf8_X_L, "GCB=L");
1928 Perl_is_utf8_X_LV(pTHX_ const U8 *p)
1932 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
1934 return is_utf8_common(p, &PL_utf8_X_LV, "GCB=LV");
1938 Perl_is_utf8_X_LVT(pTHX_ const U8 *p)
1942 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
1944 return is_utf8_common(p, &PL_utf8_X_LVT, "GCB=LVT");
1948 Perl_is_utf8_X_T(pTHX_ const U8 *p)
1952 PERL_ARGS_ASSERT_IS_UTF8_X_T;
1954 return is_utf8_common(p, &PL_utf8_X_T, "GCB=T");
1958 Perl_is_utf8_X_V(pTHX_ const U8 *p)
1962 PERL_ARGS_ASSERT_IS_UTF8_X_V;
1964 return is_utf8_common(p, &PL_utf8_X_V, "GCB=V");
1968 Perl_is_utf8_X_LV_LVT_V(pTHX_ const U8 *p)
1972 PERL_ARGS_ASSERT_IS_UTF8_X_LV_LVT_V;
1974 return is_utf8_common(p, &PL_utf8_X_LV_LVT_V, "_X_LV_LVT_V");
1978 =for apidoc to_utf8_case
1980 The "p" contains the pointer to the UTF-8 string encoding
1981 the character that is being converted.
1983 The "ustrp" is a pointer to the character buffer to put the
1984 conversion result to. The "lenp" is a pointer to the length
1987 The "swashp" is a pointer to the swash to use.
1989 Both the special and normal mappings are stored in lib/unicore/To/Foo.pl,
1990 and loaded by SWASHNEW, using lib/utf8_heavy.pl. The special (usually,
1991 but not always, a multicharacter mapping), is tried first.
1993 The "special" is a string like "utf8::ToSpecLower", which means the
1994 hash %utf8::ToSpecLower. The access to the hash is through
1995 Perl_to_utf8_case().
1997 The "normal" is a string like "ToLower" which means the swash
2003 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2004 SV **swashp, const char *normal, const char *special)
2007 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
2009 const UV uv0 = utf8_to_uvchr(p, NULL);
2010 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2011 * are necessary in EBCDIC, they are redundant no-ops
2012 * in ASCII-ish platforms, and hopefully optimized away. */
2013 const UV uv1 = NATIVE_TO_UNI(uv0);
2015 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2017 /* Note that swash_fetch() doesn't output warnings for these because it
2018 * assumes we will */
2019 if (uv1 >= UNICODE_SURROGATE_FIRST) {
2020 if (uv1 <= UNICODE_SURROGATE_LAST) {
2021 if (ckWARN_d(WARN_SURROGATE)) {
2022 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2023 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2024 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2027 else if (UNICODE_IS_SUPER(uv1)) {
2028 if (ckWARN_d(WARN_NON_UNICODE)) {
2029 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2030 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2031 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2035 /* Note that non-characters are perfectly legal, so no warning should
2039 uvuni_to_utf8(tmpbuf, uv1);
2041 if (!*swashp) /* load on-demand */
2042 *swashp = swash_init("utf8", normal, &PL_sv_undef, 4, 0);
2045 /* It might be "special" (sometimes, but not always,
2046 * a multicharacter mapping) */
2047 HV * const hv = get_hv(special, 0);
2051 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2055 s = SvPV_const(*svp, len);
2057 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2060 /* If we have EBCDIC we need to remap the characters
2061 * since any characters in the low 256 are Unicode
2062 * code points, not EBCDIC. */
2063 U8 *t = (U8*)s, *tend = t + len, *d;
2070 const UV c = utf8_to_uvchr(t, &tlen);
2072 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2081 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2086 Copy(tmpbuf, ustrp, len, U8);
2088 Copy(s, ustrp, len, U8);
2094 if (!len && *swashp) {
2095 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE);
2098 /* It was "normal" (a single character mapping). */
2099 const UV uv3 = UNI_TO_NATIVE(uv2);
2100 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2104 if (!len) /* Neither: just copy. In other words, there was no mapping
2105 defined, which means that the code point maps to itself */
2106 len = uvchr_to_utf8(ustrp, uv0) - ustrp;
2111 return len ? utf8_to_uvchr(ustrp, 0) : 0;
2115 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2117 /* This is called when changing the case of a utf8-encoded character above
2118 * the Latin1 range, and the operation is in locale. If the result
2119 * contains a character that crosses the 255/256 boundary, disallow the
2120 * change, and return the original code point. See L<perlfunc/lc> for why;
2122 * p points to the original string whose case was changed
2123 * result the code point of the first character in the changed-case string
2124 * ustrp points to the changed-case string (<result> represents its first char)
2125 * lenp points to the length of <ustrp> */
2127 UV original; /* To store the first code point of <p> */
2129 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2131 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2133 /* We know immediately if the first character in the string crosses the
2134 * boundary, so can skip */
2137 /* Look at every character in the result; if any cross the
2138 * boundary, the whole thing is disallowed */
2139 U8* s = ustrp + UTF8SKIP(ustrp);
2140 U8* e = ustrp + *lenp;
2142 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2149 /* Here, no characters crossed, result is ok as-is */
2155 /* Failed, have to return the original */
2156 original = utf8_to_uvchr(p, lenp);
2157 Copy(p, ustrp, *lenp, char);
2162 =for apidoc to_utf8_upper
2164 Convert the UTF-8 encoded character at p to its uppercase version and
2165 store that in UTF-8 in ustrp and its length in bytes in lenp. Note
2166 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2167 the uppercase version may be longer than the original character.
2169 The first character of the uppercased version is returned
2170 (but note, as explained above, that there may be more.)
2174 /* Not currently externally documented, and subject to change:
2175 * <flags> is set iff locale semantics are to be used for code points < 256
2176 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2177 * were used in the calculation; otherwise unchanged. */
2180 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2186 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2188 if (UTF8_IS_INVARIANT(*p)) {
2190 result = toUPPER_LC(*p);
2193 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2196 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2198 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2201 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2205 else { /* utf8, ord above 255 */
2206 result = CALL_UPPER_CASE(p, ustrp, lenp);
2209 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2214 /* Here, used locale rules. Convert back to utf8 */
2215 if (UTF8_IS_INVARIANT(result)) {
2216 *ustrp = (U8) result;
2220 *ustrp = UTF8_EIGHT_BIT_HI(result);
2221 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2226 *tainted_ptr = TRUE;
2232 =for apidoc to_utf8_title
2234 Convert the UTF-8 encoded character at p to its titlecase version and
2235 store that in UTF-8 in ustrp and its length in bytes in lenp. Note
2236 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2237 titlecase version may be longer than the original character.
2239 The first character of the titlecased version is returned
2240 (but note, as explained above, that there may be more.)
2244 /* Not currently externally documented, and subject to change:
2245 * <flags> is set iff locale semantics are to be used for code points < 256
2246 * Since titlecase is not defined in POSIX, uppercase is used instead
2248 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2249 * were used in the calculation; otherwise unchanged. */
2252 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2258 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2260 if (UTF8_IS_INVARIANT(*p)) {
2262 result = toUPPER_LC(*p);
2265 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2268 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2270 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2273 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2277 else { /* utf8, ord above 255 */
2278 result = CALL_TITLE_CASE(p, ustrp, lenp);
2281 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2286 /* Here, used locale rules. Convert back to utf8 */
2287 if (UTF8_IS_INVARIANT(result)) {
2288 *ustrp = (U8) result;
2292 *ustrp = UTF8_EIGHT_BIT_HI(result);
2293 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2298 *tainted_ptr = TRUE;
2304 =for apidoc to_utf8_lower
2306 Convert the UTF-8 encoded character at p to its lowercase version and
2307 store that in UTF-8 in ustrp and its length in bytes in lenp. Note
2308 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2309 lowercase version may be longer than the original character.
2311 The first character of the lowercased version is returned
2312 (but note, as explained above, that there may be more.)
2316 /* Not currently externally documented, and subject to change:
2317 * <flags> is set iff locale semantics are to be used for code points < 256
2318 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2319 * were used in the calculation; otherwise unchanged. */
2322 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
2328 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2330 if (UTF8_IS_INVARIANT(*p)) {
2332 result = toLOWER_LC(*p);
2335 return to_lower_latin1(*p, ustrp, lenp);
2338 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2340 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2343 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2347 else { /* utf8, ord above 255 */
2348 result = CALL_LOWER_CASE(p, ustrp, lenp);
2351 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2357 /* Here, used locale rules. Convert back to utf8 */
2358 if (UTF8_IS_INVARIANT(result)) {
2359 *ustrp = (U8) result;
2363 *ustrp = UTF8_EIGHT_BIT_HI(result);
2364 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2369 *tainted_ptr = TRUE;
2375 =for apidoc to_utf8_fold
2377 Convert the UTF-8 encoded character at p to its foldcase version and
2378 store that in UTF-8 in ustrp and its length in bytes in lenp. Note
2379 that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
2380 foldcase version may be longer than the original character (up to
2383 The first character of the foldcased version is returned
2384 (but note, as explained above, that there may be more.)
2388 /* Not currently externally documented, and subject to change,
2390 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2391 * points < 256. Since foldcase is not defined in
2392 * POSIX, lowercase is used instead
2393 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2394 * otherwise simple folds
2395 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2396 * were used in the calculation; otherwise unchanged. */
2399 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
2405 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2407 if (UTF8_IS_INVARIANT(*p)) {
2408 if (flags & FOLD_FLAGS_LOCALE) {
2409 result = toLOWER_LC(*p);
2412 return _to_fold_latin1(*p, ustrp, lenp,
2413 cBOOL(flags & FOLD_FLAGS_FULL));
2416 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2417 if (flags & FOLD_FLAGS_LOCALE) {
2418 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2421 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2422 ustrp, lenp, cBOOL(flags & FOLD_FLAGS_FULL));
2425 else { /* utf8, ord above 255 */
2426 result = CALL_FOLD_CASE(p, ustrp, lenp, flags);
2428 if ((flags & FOLD_FLAGS_LOCALE)) {
2429 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2435 /* Here, used locale rules. Convert back to utf8 */
2436 if (UTF8_IS_INVARIANT(result)) {
2437 *ustrp = (U8) result;
2441 *ustrp = UTF8_EIGHT_BIT_HI(result);
2442 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2447 *tainted_ptr = TRUE;
2453 * Returns a "swash" which is a hash described in utf8.c:S_swash_fetch().
2454 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2455 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2458 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
2463 const size_t pkg_len = strlen(pkg);
2464 const size_t name_len = strlen(name);
2465 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2469 PERL_ARGS_ASSERT_SWASH_INIT;
2471 PUSHSTACKi(PERLSI_MAGIC);
2475 if (PL_parser && PL_parser->error_count)
2476 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2477 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2478 if (!method) { /* demand load utf8 */
2480 errsv_save = newSVsv(ERRSV);
2481 /* It is assumed that callers of this routine are not passing in any
2482 user derived data. */
2483 /* Need to do this after save_re_context() as it will set PL_tainted to
2484 1 while saving $1 etc (see the code after getrx: in Perl_magic_get).
2485 Even line to create errsv_save can turn on PL_tainted. */
2486 SAVEBOOL(PL_tainted);
2488 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2491 sv_setsv(ERRSV, errsv_save);
2492 SvREFCNT_dec(errsv_save);
2498 mPUSHp(pkg, pkg_len);
2499 mPUSHp(name, name_len);
2504 errsv_save = newSVsv(ERRSV);
2505 /* If we already have a pointer to the method, no need to use call_method()
2506 to repeat the lookup. */
2507 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
2508 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2509 retval = newSVsv(*PL_stack_sp--);
2511 retval = &PL_sv_undef;
2513 sv_setsv(ERRSV, errsv_save);
2514 SvREFCNT_dec(errsv_save);
2517 if (IN_PERL_COMPILETIME) {
2518 CopHINTS_set(PL_curcop, PL_hints);
2520 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2522 Perl_croak(aTHX_ "Can't find Unicode property definition \"%"SVf"\"",
2524 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2530 /* This API is wrong for special case conversions since we may need to
2531 * return several Unicode characters for a single Unicode character
2532 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2533 * the lower-level routine, and it is similarly broken for returning
2534 * multiple values. --jhi
2535 * For those, you should use to_utf8_case() instead */
2536 /* Now SWASHGET is recasted into S_swash_get in this file. */
2539 * Returns the value of property/mapping C<swash> for the first character
2540 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2541 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
2542 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
2544 * A "swash" is a hash which contains initially the keys/values set up by
2545 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2546 * property for all possible code points. Things are stored in a compact form
2547 * (see utf8_heavy.pl) so that calculation is required to find the actual
2548 * property value for a given code point. As code points are looked up, new
2549 * key/value pairs are added to the hash, so that the calculation doesn't have
2550 * to ever be re-done. Further, each calculation is done, not just for the
2551 * desired one, but for a whole block of code points adjacent to that one.
2552 * For binary properties on ASCII machines, the block is usually for 64 code
2553 * points, starting with a code point evenly divisible by 64. Thus if the
2554 * property value for code point 257 is requested, the code goes out and
2555 * calculates the property values for all 64 code points between 256 and 319,
2556 * and stores these as a single 64-bit long bit vector, called a "swatch",
2557 * under the key for code point 256. The key is the UTF-8 encoding for code
2558 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2559 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2560 * for code point 258 is then requested, this code realizes that it would be
2561 * stored under the key for 256, and would find that value and extract the
2562 * relevant bit, offset from 256.
2564 * Non-binary properties are stored in as many bits as necessary to represent
2565 * their values (32 currently, though the code is more general than that), not
2566 * as single bits, but the principal is the same: the value for each key is a
2567 * vector that encompasses the property values for all code points whose UTF-8
2568 * representations are represented by the key. That is, for all code points
2569 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2573 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
2576 HV *const hv = MUTABLE_HV(SvRV(swash));
2581 const U8 *tmps = NULL;
2585 const UV c = NATIVE_TO_ASCII(*ptr);
2587 PERL_ARGS_ASSERT_SWASH_FETCH;
2589 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
2590 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
2591 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
2594 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
2595 * then the "swatch" is a vec() for all the chars which start
2597 * So the key in the hash (klen) is length of encoded char -1
2599 klen = UTF8SKIP(ptr) - 1;
2603 /* If char is invariant then swatch is for all the invariant chars
2604 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
2606 needents = UTF_CONTINUATION_MARK;
2607 off = NATIVE_TO_UTF(ptr[klen]);
2610 /* If char is encoded then swatch is for the prefix */
2611 needents = (1 << UTF_ACCUMULATION_SHIFT);
2612 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
2616 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2617 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2618 * it's nothing to sniff at.) Pity we usually come through at least
2619 * two function calls to get here...
2621 * NB: this code assumes that swatches are never modified, once generated!
2624 if (hv == PL_last_swash_hv &&
2625 klen == PL_last_swash_klen &&
2626 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
2628 tmps = PL_last_swash_tmps;
2629 slen = PL_last_swash_slen;
2632 /* Try our second-level swatch cache, kept in a hash. */
2633 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
2635 /* If not cached, generate it via swash_get */
2636 if (!svp || !SvPOK(*svp)
2637 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2638 /* We use utf8n_to_uvuni() as we want an index into
2639 Unicode tables, not a native character number.
2641 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
2643 0 : UTF8_ALLOW_ANY);
2644 swatch = swash_get(swash,
2645 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
2646 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
2649 if (IN_PERL_COMPILETIME)
2650 CopHINTS_set(PL_curcop, PL_hints);
2652 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
2654 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2655 || (slen << 3) < needents)
2656 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch");
2659 PL_last_swash_hv = hv;
2660 assert(klen <= sizeof(PL_last_swash_key));
2661 PL_last_swash_klen = (U8)klen;
2662 /* FIXME change interpvar.h? */
2663 PL_last_swash_tmps = (U8 *) tmps;
2664 PL_last_swash_slen = slen;
2666 Copy(ptr, PL_last_swash_key, klen, U8);
2669 if (UTF8_IS_SUPER(ptr) && ckWARN_d(WARN_NON_UNICODE)) {
2670 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2672 /* This outputs warnings for binary properties only, assuming that
2673 * to_utf8_case() will output any for non-binary. Also, surrogates
2674 * aren't checked for, as that would warn on things like /\p{Gc=Cs}/ */
2676 if (SvUV(*bitssvp) == 1) {
2677 /* User-defined properties can silently match above-Unicode */
2678 SV** const user_defined_svp = hv_fetchs(hv, "USER_DEFINED", FALSE);
2679 if (! user_defined_svp || ! SvUV(*user_defined_svp)) {
2680 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0, 0);
2681 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2682 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", code_point);
2687 switch ((int)((slen << 3) / needents)) {
2689 bit = 1 << (off & 7);
2691 return (tmps[off] & bit) != 0;
2696 return (tmps[off] << 8) + tmps[off + 1] ;
2699 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
2701 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width");
2702 NORETURN_FUNCTION_END;
2705 /* Read a single line of the main body of the swash input text. These are of
2708 * where each number is hex. The first two numbers form the minimum and
2709 * maximum of a range, and the third is the value associated with the range.
2710 * Not all swashes should have a third number
2712 * On input: l points to the beginning of the line to be examined; it points
2713 * to somewhere in the string of the whole input text, and is
2714 * terminated by a \n or the null string terminator.
2715 * lend points to the null terminator of that string
2716 * wants_value is non-zero if the swash expects a third number
2717 * typestr is the name of the swash's mapping, like 'ToLower'
2718 * On output: *min, *max, and *val are set to the values read from the line.
2719 * returns a pointer just beyond the line examined. If there was no
2720 * valid min number on the line, returns lend+1
2724 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2725 const bool wants_value, const U8* const typestr)
2727 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2728 STRLEN numlen; /* Length of the number */
2729 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2730 | PERL_SCAN_DISALLOW_PREFIX
2731 | PERL_SCAN_SILENT_NON_PORTABLE;
2733 /* nl points to the next \n in the scan */
2734 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2736 /* Get the first number on the line: the range minimum */
2738 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2739 if (numlen) /* If found a hex number, position past it */
2741 else if (nl) { /* Else, go handle next line, if any */
2742 return nl + 1; /* 1 is length of "\n" */
2744 else { /* Else, no next line */
2745 return lend + 1; /* to LIST's end at which \n is not found */
2748 /* The max range value follows, separated by a BLANK */
2751 flags = PERL_SCAN_SILENT_ILLDIGIT
2752 | PERL_SCAN_DISALLOW_PREFIX
2753 | PERL_SCAN_SILENT_NON_PORTABLE;
2755 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2758 else /* If no value here, it is a single element range */
2761 /* Non-binary tables have a third entry: what the first element of the
2766 flags = PERL_SCAN_SILENT_ILLDIGIT
2767 | PERL_SCAN_DISALLOW_PREFIX
2768 | PERL_SCAN_SILENT_NON_PORTABLE;
2770 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2779 /* diag_listed_as: To%s: illegal mapping '%s' */
2780 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2786 *val = 0; /* bits == 1, then any val should be ignored */
2788 else { /* Nothing following range min, should be single element with no
2794 /* diag_listed_as: To%s: illegal mapping '%s' */
2795 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2799 *val = 0; /* bits == 1, then val should be ignored */
2802 /* Position to next line if any, or EOF */
2812 * Returns a swatch (a bit vector string) for a code point sequence
2813 * that starts from the value C<start> and comprises the number C<span>.
2814 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2815 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2818 S_swash_get(pTHX_ SV* swash, UV start, UV span)
2821 U8 *l, *lend, *x, *xend, *s, *send;
2822 STRLEN lcur, xcur, scur;
2823 HV *const hv = MUTABLE_HV(SvRV(swash));
2825 /* The string containing the main body of the table */
2826 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
2828 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2829 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2830 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2831 SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2832 SV** const invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2833 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
2834 const STRLEN bits = SvUV(*bitssvp);
2835 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
2836 const UV none = SvUV(*nonesvp);
2837 UV end = start + span;
2839 PERL_ARGS_ASSERT_SWASH_GET;
2841 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
2842 Perl_croak(aTHX_ "panic: swash_get doesn't expect bits %"UVuf,
2846 /* If overflowed, use the max possible */
2852 /* create and initialize $swatch */
2853 scur = octets ? (span * octets) : (span + 7) / 8;
2854 swatch = newSV(scur);
2856 s = (U8*)SvPVX(swatch);
2857 if (octets && none) {
2858 const U8* const e = s + scur;
2861 *s++ = (U8)(none & 0xff);
2862 else if (bits == 16) {
2863 *s++ = (U8)((none >> 8) & 0xff);
2864 *s++ = (U8)( none & 0xff);
2866 else if (bits == 32) {
2867 *s++ = (U8)((none >> 24) & 0xff);
2868 *s++ = (U8)((none >> 16) & 0xff);
2869 *s++ = (U8)((none >> 8) & 0xff);
2870 *s++ = (U8)( none & 0xff);
2876 (void)memzero((U8*)s, scur + 1);
2878 SvCUR_set(swatch, scur);
2879 s = (U8*)SvPVX(swatch);
2881 /* read $swash->{LIST}. XXX Note that this is a linear scan through a
2882 * sorted list. A binary search would be much more efficient */
2883 l = (U8*)SvPV(*listsvp, lcur);
2887 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
2888 cBOOL(octets), typestr);
2893 /* If looking for something beyond this range, go try the next one */
2900 if (!none || val < none) {
2905 for (key = min; key <= max; key++) {
2909 /* XXX If it should ever happen (very unlikely) that we would
2910 * want a non-binary result for the code point at UV_MAX,
2911 * special handling would need to be inserted here, as is done
2912 * below for the binary case */
2913 /* offset must be non-negative (start <= min <= key < end) */
2914 offset = octets * (key - start);
2916 s[offset] = (U8)(val & 0xff);
2917 else if (bits == 16) {
2918 s[offset ] = (U8)((val >> 8) & 0xff);
2919 s[offset + 1] = (U8)( val & 0xff);
2921 else if (bits == 32) {
2922 s[offset ] = (U8)((val >> 24) & 0xff);
2923 s[offset + 1] = (U8)((val >> 16) & 0xff);
2924 s[offset + 2] = (U8)((val >> 8) & 0xff);
2925 s[offset + 3] = (U8)( val & 0xff);
2928 if (!none || val < none)
2932 else { /* bits == 1, then val should be ignored */
2937 /* Special case when the upper-end is the highest possible code
2938 * point representable on the platform. Otherwise, the code below
2939 * exits before setting this bit. Done here to avoid testing for
2940 * this extremely unlikely possibility in the loop */
2941 if (UNLIKELY(end == UV_MAX && max == UV_MAX)) {
2942 const STRLEN offset = (STRLEN)(max - start);
2943 s[offset >> 3] |= 1 << (offset & 7);
2945 for (key = min; key <= max; key++) {
2946 const STRLEN offset = (STRLEN)(key - start);
2949 s[offset >> 3] |= 1 << (offset & 7);
2955 /* Invert if the data says it should be. Assumes that bits == 1 */
2956 if (invert_it_svp && SvUV(*invert_it_svp)) {
2958 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
2959 * be 0, and their inversion should also be 0, as we don't succeed any
2960 * Unicode property matches for non-Unicode code points */
2961 if (start <= PERL_UNICODE_MAX) {
2963 /* The code below assumes that we never cross the
2964 * Unicode/above-Unicode boundary in a range, as otherwise we would
2965 * have to figure out where to stop flipping the bits. Since this
2966 * boundary is divisible by a large power of 2, and swatches comes
2967 * in small powers of 2, this should be a valid assumption */
2968 assert(start + span - 1 <= PERL_UNICODE_MAX);
2978 /* read $swash->{EXTRAS}
2979 * This code also copied to swash_to_invlist() below */
2980 x = (U8*)SvPV(*extssvp, xcur);
2988 SV **otherbitssvp, *other;
2992 const U8 opc = *x++;
2996 nl = (U8*)memchr(x, '\n', xend - x);
2998 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3000 x = nl + 1; /* 1 is length of "\n" */
3004 x = xend; /* to EXTRAS' end at which \n is not found */
3011 namelen = nl - namestr;
3015 namelen = xend - namestr;
3019 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3020 otherhv = MUTABLE_HV(SvRV(*othersvp));
3021 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3022 otherbits = (STRLEN)SvUV(*otherbitssvp);
3023 if (bits < otherbits)
3024 Perl_croak(aTHX_ "panic: swash_get found swatch size mismatch");
3026 /* The "other" swatch must be destroyed after. */
3027 other = swash_get(*othersvp, start, span);
3028 o = (U8*)SvPV(other, olen);
3031 Perl_croak(aTHX_ "panic: swash_get got improper swatch");
3033 s = (U8*)SvPV(swatch, slen);
3034 if (bits == 1 && otherbits == 1) {
3036 Perl_croak(aTHX_ "panic: swash_get found swatch length mismatch");
3060 STRLEN otheroctets = otherbits >> 3;
3062 U8* const send = s + slen;
3067 if (otherbits == 1) {
3068 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3072 STRLEN vlen = otheroctets;
3080 if (opc == '+' && otherval)
3081 NOOP; /* replace with otherval */
3082 else if (opc == '!' && !otherval)
3084 else if (opc == '-' && otherval)
3086 else if (opc == '&' && !otherval)
3089 s += octets; /* no replacement */
3094 *s++ = (U8)( otherval & 0xff);
3095 else if (bits == 16) {
3096 *s++ = (U8)((otherval >> 8) & 0xff);
3097 *s++ = (U8)( otherval & 0xff);
3099 else if (bits == 32) {
3100 *s++ = (U8)((otherval >> 24) & 0xff);
3101 *s++ = (U8)((otherval >> 16) & 0xff);
3102 *s++ = (U8)((otherval >> 8) & 0xff);
3103 *s++ = (U8)( otherval & 0xff);
3107 sv_free(other); /* through with it! */
3113 Perl__swash_inversion_hash(pTHX_ SV* const swash)
3116 /* Subject to change or removal. For use only in one place in regcomp.c.
3117 * Can't be used on a property that is subject to user override, as it
3118 * relies on the value of SPECIALS in the swash which would be set by
3119 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3120 * for overridden properties
3122 * Returns a hash which is the inversion and closure of a swash mapping.
3123 * For example, consider the input lines:
3128 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3129 * 006C. The value for each key is an array. For 006C, the array would
3130 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3131 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3133 * Essentially, for any code point, it gives all the code points that map to
3134 * it, or the list of 'froms' for that point.
3136 * Currently it ignores any additions or deletions from other swashes,
3137 * looking at just the main body of the swash, and if there are SPECIALS
3138 * in the swash, at that hash
3140 * The specials hash can be extra code points, and most likely consists of
3141 * maps from single code points to multiple ones (each expressed as a string
3142 * of utf8 characters). This function currently returns only 1-1 mappings.
3143 * However consider this possible input in the specials hash:
3144 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3145 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3147 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3148 * currently handle. But it also means that FB05 and FB06 are equivalent in
3149 * a 1-1 mapping which we should handle, and this relationship may not be in
3150 * the main table. Therefore this function examines all the multi-char
3151 * sequences and adds the 1-1 mappings that come out of that. */
3155 HV *const hv = MUTABLE_HV(SvRV(swash));
3157 /* The string containing the main body of the table */
3158 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3160 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3161 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3162 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3163 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3164 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3165 const STRLEN bits = SvUV(*bitssvp);
3166 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3167 const UV none = SvUV(*nonesvp);
3168 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
3172 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3174 /* Must have at least 8 bits to get the mappings */
3175 if (bits != 8 && bits != 16 && bits != 32) {
3176 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3180 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3181 mapping to more than one character */
3183 /* Construct an inverse mapping hash for the specials */
3184 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3185 HV * specials_inverse = newHV();
3186 char *char_from; /* the lhs of the map */
3187 I32 from_len; /* its byte length */
3188 char *char_to; /* the rhs of the map */
3189 I32 to_len; /* its byte length */
3190 SV *sv_to; /* and in a sv */
3191 AV* from_list; /* list of things that map to each 'to' */
3193 hv_iterinit(specials_hv);
3195 /* The keys are the characters (in utf8) that map to the corresponding
3196 * utf8 string value. Iterate through the list creating the inverse
3198 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3200 if (! SvPOK(sv_to)) {
3201 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() unexpectedly is not a string");
3203 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %"UVXf", First char of to is %"UVXf"\n", utf8_to_uvchr((U8*) char_from, 0), utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
3205 /* Each key in the inverse list is a mapped-to value, and the key's
3206 * hash value is a list of the strings (each in utf8) that map to
3207 * it. Those strings are all one character long */
3208 if ((listp = hv_fetch(specials_inverse,
3212 from_list = (AV*) *listp;
3214 else { /* No entry yet for it: create one */
3215 from_list = newAV();
3216 if (! hv_store(specials_inverse,
3219 (SV*) from_list, 0))
3221 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3225 /* Here have the list associated with this 'to' (perhaps newly
3226 * created and empty). Just add to it. Note that we ASSUME that
3227 * the input is guaranteed to not have duplications, so we don't
3228 * check for that. Duplications just slow down execution time. */
3229 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3232 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3233 * it looking for cases like the FB05/FB06 examples above. There would
3234 * be an entry in the hash like
3235 * 'st' => [ FB05, FB06 ]
3236 * In this example we will create two lists that get stored in the
3237 * returned hash, 'ret':
3238 * FB05 => [ FB05, FB06 ]
3239 * FB06 => [ FB05, FB06 ]
3241 * Note that there is nothing to do if the array only has one element.
3242 * (In the normal 1-1 case handled below, we don't have to worry about
3243 * two lists, as everything gets tied to the single list that is
3244 * generated for the single character 'to'. But here, we are omitting
3245 * that list, ('st' in the example), so must have multiple lists.) */
3246 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3247 &char_to, &to_len)))
3249 if (av_len(from_list) > 0) {
3252 /* We iterate over all combinations of i,j to place each code
3253 * point on each list */
3254 for (i = 0; i <= av_len(from_list); i++) {
3256 AV* i_list = newAV();
3257 SV** entryp = av_fetch(from_list, i, FALSE);
3258 if (entryp == NULL) {
3259 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3261 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3262 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3264 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3265 (SV*) i_list, FALSE))
3267 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3270 /* For debugging: UV u = utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
3271 for (j = 0; j <= av_len(from_list); j++) {
3272 entryp = av_fetch(from_list, j, FALSE);
3273 if (entryp == NULL) {
3274 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3277 /* When i==j this adds itself to the list */
3278 av_push(i_list, newSVuv(utf8_to_uvchr(
3279 (U8*) SvPVX(*entryp), 0)));
3280 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Adding %"UVXf" to list for %"UVXf"\n", utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
3285 SvREFCNT_dec(specials_inverse); /* done with it */
3286 } /* End of specials */
3288 /* read $swash->{LIST} */
3289 l = (U8*)SvPV(*listsvp, lcur);
3292 /* Go through each input line */
3296 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3297 cBOOL(octets), typestr);
3302 /* Each element in the range is to be inverted */
3303 for (inverse = min; inverse <= max; inverse++) {
3307 bool found_key = FALSE;
3308 bool found_inverse = FALSE;
3310 /* The key is the inverse mapping */
3311 char key[UTF8_MAXBYTES+1];
3312 char* key_end = (char *) uvuni_to_utf8((U8*) key, val);
3313 STRLEN key_len = key_end - key;
3315 /* Get the list for the map */
3316 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3317 list = (AV*) *listp;
3319 else { /* No entry yet for it: create one */
3321 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3322 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3326 /* Look through list to see if this inverse mapping already is
3327 * listed, or if there is a mapping to itself already */
3328 for (i = 0; i <= av_len(list); i++) {
3329 SV** entryp = av_fetch(list, i, FALSE);
3331 if (entryp == NULL) {
3332 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3335 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
3336 if (SvUV(entry) == val) {
3339 if (SvUV(entry) == inverse) {
3340 found_inverse = TRUE;
3343 /* No need to continue searching if found everything we are
3345 if (found_key && found_inverse) {
3350 /* Make sure there is a mapping to itself on the list */
3352 av_push(list, newSVuv(val));
3353 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Adding %"UVXf" to list for %"UVXf"\n", val, val));*/
3357 /* Simply add the value to the list */
3358 if (! found_inverse) {
3359 av_push(list, newSVuv(inverse));
3360 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Adding %"UVXf" to list for %"UVXf"\n", inverse, val));*/
3363 /* swash_get() increments the value of val for each element in the
3364 * range. That makes more compact tables possible. You can
3365 * express the capitalization, for example, of all consecutive
3366 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3367 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
3368 * and it's not documented; it appears to be used only in
3369 * implementing tr//; I copied the semantics from swash_get(), just
3371 if (!none || val < none) {
3381 Perl__swash_to_invlist(pTHX_ SV* const swash)
3384 /* Subject to change or removal. For use only in one place in regcomp.c */
3389 HV *const hv = MUTABLE_HV(SvRV(swash));
3390 UV elements = 0; /* Number of elements in the inversion list */
3393 /* The string containing the main body of the table */
3394 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3395 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3396 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3397 SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3398 SV** const invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3400 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3401 const STRLEN bits = SvUV(*bitssvp);
3402 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3408 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3410 /* read $swash->{LIST} */
3411 if (SvPOK(*listsvp)) {
3412 l = (U8*)SvPV(*listsvp, lcur);
3415 /* LIST legitimately doesn't contain a string during compilation phases
3416 * of Perl itself, before the Unicode tables are generated. In this
3417 * case, just fake things up by creating an empty list */
3424 /* Scan the input to count the number of lines to preallocate array size
3425 * based on worst possible case, which is each line in the input creates 2
3426 * elements in the inversion list: 1) the beginning of a range in the list;
3427 * 2) the beginning of a range not in the list. */
3428 while ((loc = (strchr(loc, '\n'))) != NULL) {
3433 /* If the ending is somehow corrupt and isn't a new line, add another
3434 * element for the final range that isn't in the inversion list */
3435 if (! (*lend == '\n' || (*lend == '\0' && *(lend - 1) == '\n'))) {
3439 invlist = _new_invlist(elements);
3441 /* Now go through the input again, adding each range to the list */
3444 UV val; /* Not used by this function */
3446 l = S_swash_scan_list_line(aTHX_ l, lend, &start, &end, &val,
3447 cBOOL(octets), typestr);
3453 _append_range_to_invlist(invlist, start, end);
3456 /* Invert if the data says it should be */
3457 if (invert_it_svp && SvUV(*invert_it_svp)) {
3458 _invlist_invert_prop(invlist);
3461 /* This code is copied from swash_get()
3462 * read $swash->{EXTRAS} */
3463 x = (U8*)SvPV(*extssvp, xcur);
3471 SV **otherbitssvp, *other;
3474 const U8 opc = *x++;
3478 nl = (U8*)memchr(x, '\n', xend - x);
3480 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3482 x = nl + 1; /* 1 is length of "\n" */
3486 x = xend; /* to EXTRAS' end at which \n is not found */
3493 namelen = nl - namestr;
3497 namelen = xend - namestr;
3501 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3502 otherhv = MUTABLE_HV(SvRV(*othersvp));
3503 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3504 otherbits = (STRLEN)SvUV(*otherbitssvp);
3506 if (bits != otherbits || bits != 1) {
3507 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean properties");
3510 /* The "other" swatch must be destroyed after. */
3511 other = _swash_to_invlist((SV *)*othersvp);
3513 /* End of code copied from swash_get() */
3516 _invlist_union(invlist, other, &invlist);
3519 _invlist_invert(other);
3520 _invlist_union(invlist, other, &invlist);
3523 _invlist_subtract(invlist, other, &invlist);
3526 _invlist_intersection(invlist, other, &invlist);
3531 sv_free(other); /* through with it! */
3538 =for apidoc uvchr_to_utf8
3540 Adds the UTF-8 representation of the Native code point C<uv> to the end
3541 of the string C<d>; C<d> should be have at least C<UTF8_MAXBYTES+1> free
3542 bytes available. The return value is the pointer to the byte after the
3543 end of the new character. In other words,
3545 d = uvchr_to_utf8(d, uv);
3547 is the recommended wide native character-aware way of saying
3554 /* On ASCII machines this is normally a macro but we want a
3555 real function in case XS code wants it
3558 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
3560 PERL_ARGS_ASSERT_UVCHR_TO_UTF8;
3562 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), 0);
3566 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
3568 PERL_ARGS_ASSERT_UVCHR_TO_UTF8_FLAGS;
3570 return Perl_uvuni_to_utf8_flags(aTHX_ d, NATIVE_TO_UNI(uv), flags);
3574 =for apidoc utf8n_to_uvchr
3576 Returns the native character value of the first character in the string
3578 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
3579 length, in bytes, of that character.
3581 length and flags are the same as utf8n_to_uvuni().
3585 /* On ASCII machines this is normally a macro but we want
3586 a real function in case XS code wants it
3589 Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen,
3592 const UV uv = Perl_utf8n_to_uvuni(aTHX_ s, curlen, retlen, flags);
3594 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
3596 return UNI_TO_NATIVE(uv);
3600 Perl_check_utf8_print(pTHX_ register const U8* s, const STRLEN len)
3602 /* May change: warns if surrogates, non-character code points, or
3603 * non-Unicode code points are in s which has length len bytes. Returns
3604 * TRUE if none found; FALSE otherwise. The only other validity check is
3605 * to make sure that this won't exceed the string's length */
3607 const U8* const e = s + len;
3610 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3613 if (UTF8SKIP(s) > len) {
3614 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3615 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3618 if (UNLIKELY(*s >= UTF8_FIRST_PROBLEMATIC_CODE_POINT_FIRST_BYTE)) {
3620 if (UTF8_IS_SUPER(s)) {
3621 if (ckWARN_d(WARN_NON_UNICODE)) {
3622 UV uv = utf8_to_uvchr(s, &char_len);
3623 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3624 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
3628 else if (UTF8_IS_SURROGATE(s)) {
3629 if (ckWARN_d(WARN_SURROGATE)) {
3630 UV uv = utf8_to_uvchr(s, &char_len);
3631 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3632 "Unicode surrogate U+%04"UVXf" is illegal in UTF-8", uv);
3637 ((UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s))
3638 && (ckWARN_d(WARN_NONCHAR)))
3640 UV uv = utf8_to_uvchr(s, &char_len);
3641 Perl_warner(aTHX_ packWARN(WARN_NONCHAR),
3642 "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv);
3653 =for apidoc pv_uni_display
3655 Build to the scalar dsv a displayable version of the string spv,
3656 length len, the displayable version being at most pvlim bytes long
3657 (if longer, the rest is truncated and "..." will be appended).
3659 The flags argument can have UNI_DISPLAY_ISPRINT set to display
3660 isPRINT()able characters as themselves, UNI_DISPLAY_BACKSLASH
3661 to display the \\[nrfta\\] as the backslashed versions (like '\n')
3662 (UNI_DISPLAY_BACKSLASH is preferred over UNI_DISPLAY_ISPRINT for \\).
3663 UNI_DISPLAY_QQ (and its alias UNI_DISPLAY_REGEX) have both
3664 UNI_DISPLAY_BACKSLASH and UNI_DISPLAY_ISPRINT turned on.
3666 The pointer to the PV of the dsv is returned.
3670 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
3675 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
3679 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
3681 /* This serves double duty as a flag and a character to print after
3682 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
3686 if (pvlim && SvCUR(dsv) >= pvlim) {
3690 u = utf8_to_uvchr((U8*)s, 0);
3692 const unsigned char c = (unsigned char)u & 0xFF;
3693 if (flags & UNI_DISPLAY_BACKSLASH) {
3710 const char string = ok;
3711 sv_catpvs(dsv, "\\");
3712 sv_catpvn(dsv, &string, 1);
3715 /* isPRINT() is the locale-blind version. */
3716 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
3717 const char string = c;
3718 sv_catpvn(dsv, &string, 1);
3723 Perl_sv_catpvf(aTHX_ dsv, "\\x{%"UVxf"}", u);
3726 sv_catpvs(dsv, "...");
3732 =for apidoc sv_uni_display
3734 Build to the scalar dsv a displayable version of the scalar sv,
3735 the displayable version being at most pvlim bytes long
3736 (if longer, the rest is truncated and "..." will be appended).
3738 The flags argument is as in pv_uni_display().
3740 The pointer to the PV of the dsv is returned.
3745 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
3747 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
3749 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)SvPVX_const(ssv),
3750 SvCUR(ssv), pvlim, flags);
3754 =for apidoc foldEQ_utf8
3756 Returns true if the leading portions of the strings s1 and s2 (either or both
3757 of which may be in UTF-8) are the same case-insensitively; false otherwise.
3758 How far into the strings to compare is determined by other input parameters.
3760 If u1 is true, the string s1 is assumed to be in UTF-8-encoded Unicode;
3761 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for u2
3764 If the byte length l1 is non-zero, it says how far into s1 to check for fold
3765 equality. In other words, s1+l1 will be used as a goal to reach. The
3766 scan will not be considered to be a match unless the goal is reached, and
3767 scanning won't continue past that goal. Correspondingly for l2 with respect to
3770 If pe1 is non-NULL and the pointer it points to is not NULL, that pointer is
3771 considered an end pointer beyond which scanning of s1 will not continue under
3772 any circumstances. This means that if both l1 and pe1 are specified, and pe1
3773 is less than s1+l1, the match will never be successful because it can never
3774 get as far as its goal (and in fact is asserted against). Correspondingly for
3775 pe2 with respect to s2.
3777 At least one of s1 and s2 must have a goal (at least one of l1 and l2 must be
3778 non-zero), and if both do, both have to be
3779 reached for a successful match. Also, if the fold of a character is multiple
3780 characters, all of them must be matched (see tr21 reference below for
3783 Upon a successful match, if pe1 is non-NULL,
3784 it will be set to point to the beginning of the I<next> character of s1 beyond
3785 what was matched. Correspondingly for pe2 and s2.
3787 For case-insensitiveness, the "casefolding" of Unicode is used
3788 instead of upper/lowercasing both the characters, see
3789 http://www.unicode.org/unicode/reports/tr21/ (Case Mappings).
3793 /* A flags parameter has been added which may change, and hence isn't
3794 * externally documented. Currently it is:
3795 * 0 for as-documented above
3796 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
3797 ASCII one, to not match
3798 * FOLDEQ_UTF8_LOCALE meaning that locale rules are to be used for code
3799 * points below 256; unicode rules for above 255; and
3800 * folds that cross those boundaries are disallowed,
3801 * like the NOMIX_ASCII option
3802 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
3803 * routine. This allows that step to be skipped.
3804 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
3807 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)
3810 register const U8 *p1 = (const U8*)s1; /* Point to current char */
3811 register const U8 *p2 = (const U8*)s2;
3812 register const U8 *g1 = NULL; /* goal for s1 */
3813 register const U8 *g2 = NULL;
3814 register const U8 *e1 = NULL; /* Don't scan s1 past this */
3815 register U8 *f1 = NULL; /* Point to current folded */
3816 register const U8 *e2 = NULL;
3817 register U8 *f2 = NULL;
3818 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
3819 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
3820 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
3822 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
3824 /* The algorithm requires that input with the flags on the first line of
3825 * the assert not be pre-folded. */
3826 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_UTF8_LOCALE))
3827 && (flags & (FOLDEQ_S1_ALREADY_FOLDED | FOLDEQ_S2_ALREADY_FOLDED))));
3834 g1 = (const U8*)s1 + l1;
3842 g2 = (const U8*)s2 + l2;
3845 /* Must have at least one goal */
3850 /* Will never match if goal is out-of-bounds */
3851 assert(! e1 || e1 >= g1);
3853 /* Here, there isn't an end pointer, or it is beyond the goal. We
3854 * only go as far as the goal */
3858 assert(e1); /* Must have an end for looking at s1 */
3861 /* Same for goal for s2 */
3863 assert(! e2 || e2 >= g2);
3870 /* If both operands are already folded, we could just do a memEQ on the
3871 * whole strings at once, but it would be better if the caller realized
3872 * this and didn't even call us */
3874 /* Look through both strings, a character at a time */
3875 while (p1 < e1 && p2 < e2) {
3877 /* If at the beginning of a new character in s1, get its fold to use
3878 * and the length of the fold. (exception: locale rules just get the
3879 * character to a single byte) */
3881 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
3887 /* If in locale matching, we use two sets of rules, depending
3888 * on if the code point is above or below 255. Here, we test
3889 * for and handle locale rules */
3890 if ((flags & FOLDEQ_UTF8_LOCALE)
3891 && (! u1 || UTF8_IS_INVARIANT(*p1)
3892 || UTF8_IS_DOWNGRADEABLE_START(*p1)))
3894 /* There is no mixing of code points above and below 255. */
3895 if (u2 && (! UTF8_IS_INVARIANT(*p2)
3896 && ! UTF8_IS_DOWNGRADEABLE_START(*p2)))
3901 /* We handle locale rules by converting, if necessary, the
3902 * code point to a single byte. */
3903 if (! u1 || UTF8_IS_INVARIANT(*p1)) {
3907 *foldbuf1 = TWO_BYTE_UTF8_TO_UNI(*p1, *(p1 + 1));
3911 else if (isASCII(*p1)) { /* Note, that here won't be both
3912 ASCII and using locale rules */
3914 /* If trying to mix non- with ASCII, and not supposed to,
3916 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
3920 *foldbuf1 = toLOWER(*p1); /* Folds in the ASCII range are
3924 to_utf8_fold(p1, foldbuf1, &n1);
3926 else { /* Not utf8, get utf8 fold */
3927 to_uni_fold(NATIVE_TO_UNI(*p1), foldbuf1, &n1);
3933 if (n2 == 0) { /* Same for s2 */
3934 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
3939 if ((flags & FOLDEQ_UTF8_LOCALE)
3940 && (! u2 || UTF8_IS_INVARIANT(*p2) || UTF8_IS_DOWNGRADEABLE_START(*p2)))
3942 /* Here, the next char in s2 is < 256. We've already
3943 * worked on s1, and if it isn't also < 256, can't match */
3944 if (u1 && (! UTF8_IS_INVARIANT(*p1)
3945 && ! UTF8_IS_DOWNGRADEABLE_START(*p1)))
3949 if (! u2 || UTF8_IS_INVARIANT(*p2)) {
3953 *foldbuf2 = TWO_BYTE_UTF8_TO_UNI(*p2, *(p2 + 1));
3956 /* Use another function to handle locale rules. We've made
3957 * sure that both characters to compare are single bytes */
3958 if (! foldEQ_locale((char *) f1, (char *) foldbuf2, 1)) {
3963 else if (isASCII(*p2)) {
3964 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
3968 *foldbuf2 = toLOWER(*p2);
3971 to_utf8_fold(p2, foldbuf2, &n2);
3974 to_uni_fold(NATIVE_TO_UNI(*p2), foldbuf2, &n2);
3980 /* Here f1 and f2 point to the beginning of the strings to compare.
3981 * These strings are the folds of the next character from each input
3982 * string, stored in utf8. */
3984 /* While there is more to look for in both folds, see if they
3985 * continue to match */
3987 U8 fold_length = UTF8SKIP(f1);
3988 if (fold_length != UTF8SKIP(f2)
3989 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
3990 function call for single
3992 || memNE((char*)f1, (char*)f2, fold_length))
3994 return 0; /* mismatch */
3997 /* Here, they matched, advance past them */
4004 /* When reach the end of any fold, advance the input past it */
4006 p1 += u1 ? UTF8SKIP(p1) : 1;
4009 p2 += u2 ? UTF8SKIP(p2) : 1;
4011 } /* End of loop through both strings */
4013 /* A match is defined by each scan that specified an explicit length
4014 * reaching its final goal, and the other not having matched a partial
4015 * character (which can happen when the fold of a character is more than one
4017 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4021 /* Successful match. Set output pointers */
4033 * c-indentation-style: bsd
4035 * indent-tabs-mode: t
4038 * ex: set ts=8 sts=4 sw=4 noet: