This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: Reword comment
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
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.
8 *
9 */
10
11/*
4ac71550
TC
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.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
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,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
a0ed51b3 35
a0c21aa1 36#ifndef EBCDIC
970ea3cb 37/* Separate prototypes needed because in ASCII systems these are
a0c21aa1
JH
38 * usually macros but they still are compiled as code, too. */
39PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
6dd9dce9 40PERL_CALLCONV UV Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen);
a0c21aa1
JH
41PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
42#endif
43
27da23d5
JH
44static const char unees[] =
45 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 46
48ef279e 47/*
ccfc67b7 48=head1 Unicode Support
a0ed51b3 49
166f8a29
DM
50This file contains various utility functions for manipulating UTF8-encoded
51strings. For the uninitiated, this is a method of representing arbitrary
61296642 52Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
53characters in the ASCII range are unmodified, and a zero byte never appears
54within non-zero characters.
166f8a29 55
eaf7a4d2
CS
56=cut
57*/
58
59/*
60=for apidoc is_ascii_string
61
a1433954 62Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
63or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
64is, if they are invariant. On ASCII-ish machines, only ASCII characters
65fit this definition, hence the function's name.
eaf7a4d2 66
9f7e3d64
MH
67If C<len> is 0, it will be calculated using C<strlen(s)>.
68
a1433954 69See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
70
71=cut
72*/
73
74bool
668b6d8d 75Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
76{
77 const U8* const send = s + (len ? len : strlen((const char *)s));
78 const U8* x = s;
79
80 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
81
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
84 break;
85 }
86
87 return x == send;
88}
89
90/*
87cea99e 91=for apidoc uvuni_to_utf8_flags
eebe1485 92
6ee84de2
KW
93Adds the UTF-8 representation of the code point C<uv> to the end
94of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
eebe1485 95bytes available. The return value is the pointer to the byte after the
9041c2e3 96end of the new character. In other words,
eebe1485 97
b851fbc1
JH
98 d = uvuni_to_utf8_flags(d, uv, flags);
99
100or, in most cases,
101
9041c2e3 102 d = uvuni_to_utf8(d, uv);
eebe1485 103
b851fbc1
JH
104(which is equivalent to)
105
106 d = uvuni_to_utf8_flags(d, uv, 0);
107
949cf498 108This is the recommended Unicode-aware way of saying
eebe1485
SC
109
110 *(d++) = uv;
111
949cf498
KW
112This function will convert to UTF-8 (and not warn) even code points that aren't
113legal Unicode or are problematic, unless C<flags> contains one or more of the
a1433954
KW
114following flags:
115
949cf498
KW
116If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
117the function will raise a warning, provided UTF8 warnings are enabled. If instead
118UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
119If both flags are set, the function will both warn and return NULL.
120
121The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
122affect how the function handles a Unicode non-character. And, likewise for the
123UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
124above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
125even less portable) can be warned and/or disallowed even if other above-Unicode
126code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
127flags.
128
129And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
130above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
131DISALLOW flags.
132
133
eebe1485
SC
134=cut
135*/
136
dfe13c55 137U8 *
b851fbc1 138Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 139{
7918f24d
NC
140 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
141
979f77b6
KW
142 /* The first problematic code point is the first surrogate */
143 if (uv >= UNICODE_SURROGATE_FIRST
144 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
145 {
949cf498
KW
146 if (UNICODE_IS_SURROGATE(uv)) {
147 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 148 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
149 "UTF-16 surrogate U+%04"UVXf, uv);
150 }
151 if (flags & UNICODE_DISALLOW_SURROGATE) {
152 return NULL;
153 }
154 }
155 else if (UNICODE_IS_SUPER(uv)) {
156 if (flags & UNICODE_WARN_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
158 {
8457b38f 159 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
160 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
161 }
162 if (flags & UNICODE_DISALLOW_SUPER
163 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
164 {
165 return NULL;
166 }
167 }
168 else if (UNICODE_IS_NONCHAR(uv)) {
169 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 170 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
171 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
172 uv);
173 }
174 if (flags & UNICODE_DISALLOW_NONCHAR) {
175 return NULL;
176 }
177 }
507b9800 178 }
c4d5f83a 179 if (UNI_IS_INVARIANT(uv)) {
eb160463 180 *d++ = (U8)UTF_TO_NATIVE(uv);
a0ed51b3
LW
181 return d;
182 }
2d331972 183#if defined(EBCDIC)
1d72bdf6
NIS
184 else {
185 STRLEN len = UNISKIP(uv);
186 U8 *p = d+len-1;
187 while (p > d) {
eb160463 188 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
189 uv >>= UTF_ACCUMULATION_SHIFT;
190 }
eb160463 191 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
192 return d+len;
193 }
194#else /* Non loop style */
a0ed51b3 195 if (uv < 0x800) {
eb160463
GS
196 *d++ = (U8)(( uv >> 6) | 0xc0);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
198 return d;
199 }
200 if (uv < 0x10000) {
eb160463
GS
201 *d++ = (U8)(( uv >> 12) | 0xe0);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
204 return d;
205 }
206 if (uv < 0x200000) {
eb160463
GS
207 *d++ = (U8)(( uv >> 18) | 0xf0);
208 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
209 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
210 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
211 return d;
212 }
213 if (uv < 0x4000000) {
eb160463
GS
214 *d++ = (U8)(( uv >> 24) | 0xf8);
215 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
218 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
219 return d;
220 }
221 if (uv < 0x80000000) {
eb160463
GS
222 *d++ = (U8)(( uv >> 30) | 0xfc);
223 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
227 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
228 return d;
229 }
6b8eaf93 230#ifdef HAS_QUAD
d7578b48 231 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
232#endif
233 {
eb160463
GS
234 *d++ = 0xfe; /* Can't match U+FEFF! */
235 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
239 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
240 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
241 return d;
242 }
6b8eaf93 243#ifdef HAS_QUAD
a0ed51b3 244 {
eb160463
GS
245 *d++ = 0xff; /* Can't match U+FFFE! */
246 *d++ = 0x80; /* 6 Reserved bits */
247 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
248 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
257 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
258 return d;
259 }
260#endif
1d72bdf6 261#endif /* Loop style */
a0ed51b3 262}
9041c2e3 263
646ca15d
JH
264/*
265
f7d739d1 266Tests if the first C<len> bytes of string C<s> form a valid UTF-8
646ca15d 267character. Note that an INVARIANT (i.e. ASCII) character is a valid
f7d739d1 268UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
269will be returned if it is valid, otherwise 0.
270
271This is the "slow" version as opposed to the "fast" version which is
272the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
273difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
274or less you should use the IS_UTF8_CHAR(), for lengths of five or more
275you should use the _slow(). In practice this means that the _slow()
276will be used very rarely, since the maximum Unicode code point (as of
277Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
278the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
279five bytes or more.
280
281=cut */
c053b435 282STATIC STRLEN
5f66b61c 283S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 284{
cd7e6c88 285 dTHX; /* The function called below requires thread context */
646ca15d 286
cd7e6c88 287 STRLEN actual_len;
646ca15d 288
cd7e6c88 289 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 290
cd7e6c88 291 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 292
cd7e6c88 293 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 294}
9041c2e3
NIS
295
296/*
492a624f
KW
297=for apidoc is_utf8_char_buf
298
299Returns the number of bytes that comprise the first UTF-8 encoded character in
300buffer C<buf>. C<buf_end> should point to one position beyond the end of the
301buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
302encoded character.
303
304Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
305machines) is a valid UTF-8 character.
306
307=cut */
308
309STRLEN
310Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
311{
312
313 STRLEN len;
314
315 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
316
317 if (buf_end <= buf) {
318 return 0;
319 }
320
321 len = buf_end - buf;
322 if (len > UTF8SKIP(buf)) {
323 len = UTF8SKIP(buf);
324 }
325
326#ifdef IS_UTF8_CHAR
327 if (IS_UTF8_CHAR_FAST(len))
328 return IS_UTF8_CHAR(buf, len) ? len : 0;
329#endif /* #ifdef IS_UTF8_CHAR */
330 return is_utf8_char_slow(buf, len);
331}
332
333/*
87cea99e 334=for apidoc is_utf8_char
eebe1485 335
76848387
KW
336DEPRECATED!
337
5da9da9e 338Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
339character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
340character is a valid UTF-8 character. The actual number of bytes in the UTF-8
341character will be returned if it is valid, otherwise 0.
9041c2e3 342
76848387 343This function is deprecated due to the possibility that malformed input could
a1433954 344cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 345instead.
e0328548 346
82686b01 347=cut */
76848387 348
067a85ef 349STRLEN
668b6d8d 350Perl_is_utf8_char(const U8 *s)
386d01d6 351{
7918f24d 352 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 353
76848387 354 /* Assumes we have enough space, which is why this is deprecated */
492a624f 355 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
356}
357
eaf7a4d2 358
6662521e 359/*
87cea99e 360=for apidoc is_utf8_string
6662521e 361
a1433954 362Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 363UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
364using C<strlen(s)> (which means if you use this option, that C<s> has to have a
365terminating NUL byte). Note that all characters being ASCII constitute 'a
366valid UTF-8 string'.
6662521e 367
a1433954 368See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 369
6662521e
GS
370=cut
371*/
372
8e84507e 373bool
668b6d8d 374Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 375{
35da51f7 376 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 377 const U8* x = s;
067a85ef 378
7918f24d 379 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 380
6662521e 381 while (x < send) {
1acdb0da 382 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
383 if (UTF8_IS_INVARIANT(*x)) {
384 x++;
385 }
1acdb0da 386 else if (!UTF8_IS_START(*x))
e0328548 387 return FALSE;
1acdb0da
JH
388 else {
389 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
390 const STRLEN c = UTF8SKIP(x);
391 const U8* const next_char_ptr = x + c;
392
393 if (next_char_ptr > send) {
394 return FALSE;
395 }
396
768c67ee
JH
397 if (IS_UTF8_CHAR_FAST(c)) {
398 if (!IS_UTF8_CHAR(x, c))
e0328548 399 return FALSE;
3c614e38 400 }
e0328548
KW
401 else if (! is_utf8_char_slow(x, c)) {
402 return FALSE;
403 }
404 x = next_char_ptr;
1acdb0da 405 }
6662521e 406 }
768c67ee 407
067a85ef 408 return TRUE;
6662521e
GS
409}
410
67e989fb 411/*
814fafa7
NC
412Implemented as a macro in utf8.h
413
87cea99e 414=for apidoc is_utf8_string_loc
814fafa7 415
a1433954
KW
416Like L</is_utf8_string> but stores the location of the failure (in the
417case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
418"utf8ness success") in the C<ep>.
419
a1433954 420See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 421
87cea99e 422=for apidoc is_utf8_string_loclen
81cd54e3 423
a1433954
KW
424Like L</is_utf8_string>() but stores the location of the failure (in the
425case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
426"utf8ness success") in the C<ep>, and the number of UTF-8
427encoded characters in the C<el>.
428
a1433954 429See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
430
431=cut
432*/
433
434bool
668b6d8d 435Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 436{
35da51f7 437 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 438 const U8* x = s;
81cd54e3 439 STRLEN c;
3ebfea28 440 STRLEN outlen = 0;
7918f24d
NC
441
442 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 443
81cd54e3 444 while (x < send) {
e0328548
KW
445 const U8* next_char_ptr;
446
81cd54e3
JH
447 /* Inline the easy bits of is_utf8_char() here for speed... */
448 if (UTF8_IS_INVARIANT(*x))
e0328548 449 next_char_ptr = x + 1;
768c67ee
JH
450 else if (!UTF8_IS_START(*x))
451 goto out;
81cd54e3 452 else {
768c67ee 453 /* ... and call is_utf8_char() only if really needed. */
768c67ee 454 c = UTF8SKIP(x);
e0328548
KW
455 next_char_ptr = c + x;
456 if (next_char_ptr > send) {
457 goto out;
458 }
768c67ee
JH
459 if (IS_UTF8_CHAR_FAST(c)) {
460 if (!IS_UTF8_CHAR(x, c))
461 c = 0;
462 } else
463 c = is_utf8_char_slow(x, c);
768c67ee
JH
464 if (!c)
465 goto out;
81cd54e3 466 }
e0328548 467 x = next_char_ptr;
3ebfea28 468 outlen++;
81cd54e3 469 }
768c67ee
JH
470
471 out:
3ebfea28
AL
472 if (el)
473 *el = outlen;
474
768c67ee
JH
475 if (ep)
476 *ep = x;
3ebfea28 477 return (x == send);
81cd54e3
JH
478}
479
480/*
768c67ee 481
87cea99e 482=for apidoc utf8n_to_uvuni
67e989fb 483
9041c2e3 484Bottom level UTF-8 decode routine.
746afd53
KW
485Returns the code point value of the first character in the string C<s>,
486which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
487C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
488the length, in bytes, of that character.
949cf498
KW
489
490The value of C<flags> determines the behavior when C<s> does not point to a
491well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
492zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
493next possible position in C<s> that could begin a non-malformed character.
494Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
495
496Various ALLOW flags can be set in C<flags> to allow (and not warn on)
497individual types of malformations, such as the sequence being overlong (that
498is, when there is a shorter sequence that can express the same code point;
499overlong sequences are expressly forbidden in the UTF-8 standard due to
500potential security issues). Another malformation example is the first byte of
501a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
502flags. For allowed 0 length strings, this function returns 0; for allowed
503overlong sequences, the computed code point is returned; for all other allowed
504malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
505determinable reasonable value.
949cf498
KW
506
507The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
508flags) malformation is found. If this flag is set, the routine assumes that
509the caller will raise a warning, and this function will silently just set
510C<retlen> to C<-1> and return zero.
511
512Certain code points are considered problematic. These are Unicode surrogates,
746afd53 513Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 514By default these are considered regular code points, but certain situations
5eafe189 515warrant special handling for them. If C<flags> contains
949cf498
KW
516UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
517malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
518UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
519maximum) can be set to disallow these categories individually.
520
521The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
522UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
523for their respective categories, but otherwise the code points are considered
524valid (not malformations). To get a category to both be treated as a
525malformation and raise a warning, specify both the WARN and DISALLOW flags.
526(But note that warnings are not raised if lexically disabled nor if
527UTF8_CHECK_ONLY is also specified.)
528
529Very large code points (above 0x7FFF_FFFF) are considered more problematic than
530the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
531reasons: they requre at least 32 bits to represent them on ASCII platforms, are
532not representable at all on EBCDIC platforms, and the original UTF-8
533specification never went above this number (the current 0x10FFFF limit was
534imposed later). (The smaller ones, those that fit into 32 bits, are
535representable by a UV on ASCII platforms, but not by an IV, which means that
536the number of operations that can be performed on them is quite restricted.)
537The UTF-8 encoding on ASCII platforms for these large code points begins with a
538byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
539be treated as malformations, while allowing smaller above-Unicode code points.
540(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
541including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
542the other WARN flags, but applies just to these code points.
949cf498
KW
543
544All other code points corresponding to Unicode characters, including private
545use and those yet to be assigned, are never considered malformed and never
546warn.
67e989fb 547
ec5f19d0 548Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
9041c2e3 549
37607a96
PK
550=cut
551*/
67e989fb 552
a0ed51b3 553UV
7fc63493 554Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 555{
97aff369 556 dVAR;
d4c19fe8 557 const U8 * const s0 = s;
eb83ed87 558 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 559 U8 * send;
eb83ed87
KW
560 UV uv = *s;
561 STRLEN expectlen;
949cf498 562 SV* sv = NULL;
eb83ed87
KW
563 UV outlier_ret = 0; /* return value when input is in error or problematic
564 */
565 UV pack_warn = 0; /* Save result of packWARN() for later */
566 bool unexpected_non_continuation = FALSE;
567 bool overflowed = FALSE;
2f8f112e 568 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 569
eb83ed87 570 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 571
eb83ed87 572 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
a0dbb045 573
eb83ed87
KW
574 /* The order of malformation tests here is important. We should consume as
575 * few bytes as possible in order to not skip any valid character. This is
576 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
577 * http://unicode.org/reports/tr36 for more discussion as to why. For
578 * example, once we've done a UTF8SKIP, we can tell the expected number of
579 * bytes, and could fail right off the bat if the input parameters indicate
580 * that there are too few available. But it could be that just that first
581 * byte is garbled, and the intended character occupies fewer bytes. If we
582 * blindly assumed that the first byte is correct, and skipped based on
583 * that number, we could skip over a valid input character. So instead, we
584 * always examine the sequence byte-by-byte.
585 *
586 * We also should not consume too few bytes, otherwise someone could inject
587 * things. For example, an input could be deliberately designed to
588 * overflow, and if this code bailed out immediately upon discovering that,
589 * returning to the caller *retlen pointing to the very next byte (one
590 * which is actually part of of the overflowing sequence), that could look
591 * legitimate to the caller, which could discard the initial partial
592 * sequence and process the rest, inappropriately */
593
594 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 595 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
596 if (retlen) {
597 *retlen = 0;
598 }
a0dbb045 599
eb83ed87
KW
600 if (flags & UTF8_ALLOW_EMPTY) {
601 return 0;
602 }
603 if (! (flags & UTF8_CHECK_ONLY)) {
604 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
605 }
0c443dc2
JH
606 goto malformed;
607 }
608
eb83ed87
KW
609 expectlen = UTF8SKIP(s);
610
611 /* A well-formed UTF-8 character, as the vast majority of calls to this
612 * function will be for, has this expected length. For efficiency, set
613 * things up here to return it. It will be overriden only in those rare
614 * cases where a malformation is found */
615 if (retlen) {
616 *retlen = expectlen;
617 }
618
619 /* An invariant is trivially well-formed */
1d72bdf6 620 if (UTF8_IS_INVARIANT(uv)) {
c4d5f83a 621 return (UV) (NATIVE_TO_UTF(*s));
a0ed51b3 622 }
67e989fb 623
eb83ed87 624 /* A continuation character can't start a valid sequence */
b5b9af04 625 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
626 if (flags & UTF8_ALLOW_CONTINUATION) {
627 if (retlen) {
628 *retlen = 1;
629 }
630 return UNICODE_REPLACEMENT;
631 }
ba210ebe 632
eb83ed87
KW
633 if (! (flags & UTF8_CHECK_ONLY)) {
634 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
635 }
636 curlen = 1;
ba210ebe
JH
637 goto malformed;
638 }
9041c2e3 639
1d72bdf6 640#ifdef EBCDIC
75383841 641 uv = NATIVE_TO_UTF(uv);
1d72bdf6
NIS
642#endif
643
eb83ed87
KW
644 /* Here is not a continuation byte, nor an invariant. The only thing left
645 * is a start byte (possibly for an overlong) */
ba210ebe 646
eb83ed87
KW
647 /* Remove the leading bits that indicate the number of bytes in the
648 * character's whole UTF-8 sequence, leaving just the bits that are part of
649 * the value */
650 uv &= UTF_START_MASK(expectlen);
ba210ebe 651
eb83ed87
KW
652 /* Now, loop through the remaining bytes in the character's sequence,
653 * accumulating each into the working value as we go. Be sure to not look
654 * past the end of the input string */
0b8d30e8
KW
655 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
656
eb83ed87 657 for (s = s0 + 1; s < send; s++) {
b5b9af04 658 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
659#ifndef EBCDIC /* Can't overflow in EBCDIC */
660 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
661
662 /* The original implementors viewed this malformation as more
663 * serious than the others (though I, khw, don't understand
664 * why, since other malformations also give very very wrong
665 * results), so there is no way to turn off checking for it.
666 * Set a flag, but keep going in the loop, so that we absorb
667 * the rest of the bytes that comprise the character. */
668 overflowed = TRUE;
669 overflow_byte = *s; /* Save for warning message's use */
670 }
671#endif
8850bf83 672 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
673 }
674 else {
675 /* Here, found a non-continuation before processing all expected
676 * bytes. This byte begins a new character, so quit, even if
677 * allowing this malformation. */
678 unexpected_non_continuation = TRUE;
679 break;
680 }
681 } /* End of loop through the character's bytes */
682
683 /* Save how many bytes were actually in the character */
684 curlen = s - s0;
685
686 /* The loop above finds two types of malformations: non-continuation and/or
687 * overflow. The non-continuation malformation is really a too-short
688 * malformation, as it means that the current character ended before it was
689 * expected to (being terminated prematurely by the beginning of the next
690 * character, whereas in the too-short malformation there just are too few
691 * bytes available to hold the character. In both cases, the check below
692 * that we have found the expected number of bytes would fail if executed.)
693 * Thus the non-continuation malformation is really unnecessary, being a
694 * subset of the too-short malformation. But there may be existing
695 * applications that are expecting the non-continuation type, so we retain
696 * it, and return it in preference to the too-short malformation. (If this
697 * code were being written from scratch, the two types might be collapsed
698 * into one.) I, khw, am also giving priority to returning the
699 * non-continuation and too-short malformations over overflow when multiple
700 * ones are present. I don't know of any real reason to prefer one over
701 * the other, except that it seems to me that multiple-byte errors trumps
702 * errors from a single byte */
b5b9af04 703 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
704 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
705 if (! (flags & UTF8_CHECK_ONLY)) {
706 if (curlen == 1) {
707 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
708 }
709 else {
710 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, %d bytes after start byte 0x%02x, expected %d bytes)", malformed_text, *s, (int) curlen, *s0, (int)expectlen));
a0dbb045
JH
711 }
712 }
eb83ed87
KW
713 goto malformed;
714 }
715 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
716
717 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
718 * as what the original expectations were. */
719 do_overlong_test = FALSE;
eb83ed87
KW
720 if (retlen) {
721 *retlen = curlen;
722 }
723 }
b5b9af04 724 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
725 if (! (flags & UTF8_ALLOW_SHORT)) {
726 if (! (flags & UTF8_CHECK_ONLY)) {
727 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, *s0));
a0dbb045 728 }
eb83ed87
KW
729 goto malformed;
730 }
731 uv = UNICODE_REPLACEMENT;
2f8f112e 732 do_overlong_test = FALSE;
eb83ed87
KW
733 if (retlen) {
734 *retlen = curlen;
735 }
736 }
737
738#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
741 {
742 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
743 * generation of the sv, since no warnings are raised under CHECK */
744 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
745 && ckWARN_d(WARN_UTF8))
746 {
42303544
KW
747 /* This message is deliberately not of the same syntax as the other
748 * messages for malformations, for backwards compatibility in the
749 * unlikely event that code is relying on its precise earlier text
750 */
eb83ed87
KW
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
752 pack_warn = packWARN(WARN_UTF8);
753 }
754 if (flags & UTF8_DISALLOW_FE_FF) {
755 goto malformed;
ba210ebe 756 }
ba210ebe 757 }
b5b9af04 758 if (UNLIKELY(overflowed)) {
ba210ebe 759
eb83ed87
KW
760 /* If the first byte is FF, it will overflow a 32-bit word. If the
761 * first byte is FE, it will overflow a signed 32-bit word. The
762 * above preserves backward compatibility, since its message was used
763 * in earlier versions of this code in preference to overflow */
764 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 765 goto malformed;
eb83ed87
KW
766 }
767#endif
768
2f8f112e
KW
769 if (do_overlong_test
770 && expectlen > (STRLEN)UNISKIP(uv)
771 && ! (flags & UTF8_ALLOW_LONG))
772 {
eb83ed87
KW
773 /* The overlong malformation has lower precedence than the others.
774 * Note that if this malformation is allowed, we return the actual
775 * value, instead of the replacement character. This is because this
776 * value is actually well-defined. */
777 if (! (flags & UTF8_CHECK_ONLY)) {
778 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), *s0));
779 }
780 goto malformed;
781 }
782
783 /* Here, the input is considered to be well-formed , but could be a
784 * problematic code point that is not allowed by the input parameters. */
785 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
786 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
787 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
788 {
949cf498 789 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
790 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
791 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
792 {
111d382d 793 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 794 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
795 }
796 if (flags & UTF8_DISALLOW_SURROGATE) {
797 goto disallowed;
798 }
799 }
949cf498 800 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
801 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
802 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
803 {
111d382d 804 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 805 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
806 }
807 if (flags & UTF8_DISALLOW_SUPER) {
808 goto disallowed;
809 }
810 }
4190d317
KW
811 else if (UNICODE_IS_NONCHAR(uv)) {
812 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
813 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
814 {
815 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
816 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
817 }
818 if (flags & UTF8_DISALLOW_NONCHAR) {
819 goto disallowed;
820 }
821 }
949cf498 822
eb83ed87
KW
823 if (sv) {
824 outlier_ret = uv;
825 goto do_warn;
826 }
827
949cf498
KW
828 /* Here, this is not considered a malformed character, so drop through
829 * to return it */
a0ed51b3 830 }
ba210ebe 831
a0ed51b3 832 return uv;
ba210ebe 833
eb83ed87
KW
834 /* There are three cases which get to beyond this point. In all 3 cases:
835 * <sv> if not null points to a string to print as a warning.
836 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
837 * set.
838 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
839 * This is done by initializing it to 0, and changing it only
840 * for case 1).
841 * The 3 cases are:
842 * 1) The input is valid but problematic, and to be warned about. The
843 * return value is the resultant code point; <*retlen> is set to
844 * <curlen>, the number of bytes that comprise the code point.
845 * <pack_warn> contains the result of packWARN() for the warning
846 * types. The entry point for this case is the label <do_warn>;
847 * 2) The input is a valid code point but disallowed by the parameters to
848 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
849 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
850 * comprise the code point. <pack_warn> contains the result of
851 * packWARN() for the warning types. The entry point for this case is
852 * the label <disallowed>.
853 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
854 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
855 * bytes that comprise the malformation. All such malformations are
856 * assumed to be warning type <utf8>. The entry point for this case
857 * is the label <malformed>.
858 */
949cf498 859
ba210ebe
JH
860malformed:
861
eb83ed87
KW
862 if (sv && ckWARN_d(WARN_UTF8)) {
863 pack_warn = packWARN(WARN_UTF8);
864 }
865
866disallowed:
867
fcc8fcf6 868 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 869 if (retlen)
10edeb5d 870 *retlen = ((STRLEN) -1);
ba210ebe
JH
871 return 0;
872 }
873
eb83ed87 874do_warn:
5b311467 875
eb83ed87
KW
876 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
877 if warnings are to be raised. */
f555bc63 878 const char * const string = SvPVX_const(sv);
a0dbb045 879
f555bc63
KW
880 if (PL_op)
881 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
882 else
883 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
884 }
885
eb83ed87
KW
886 if (retlen) {
887 *retlen = curlen;
888 }
ba210ebe 889
eb83ed87 890 return outlier_ret;
a0ed51b3
LW
891}
892
8e84507e 893/*
ec5f19d0
KW
894=for apidoc utf8_to_uvchr_buf
895
896Returns the native code point of the first character in the string C<s> which
897is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 898C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 899
524080c4
KW
900If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
901enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
902NULL) to -1. If those warnings are off, the computed value if well-defined (or
903the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
904is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
905next possible position in C<s> that could begin a non-malformed character.
906See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
907
908=cut
909*/
910
911
912UV
913Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
914{
915 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
916
917 assert(s < send);
918
919 return utf8n_to_uvchr(s, send - s, retlen,
920 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
921}
922
27d6c58a 923/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 924 * there are no malformations in the input UTF-8 string C<s>. surrogates,
6dd9dce9
KW
925 * non-character code points, and non-Unicode code points are allowed. A macro
926 * in utf8.h is used to normally avoid this function wrapper */
27d6c58a
KW
927
928UV
929Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
930{
3986bb7c
KW
931 const UV uv = valid_utf8_to_uvuni(s, retlen);
932
27d6c58a
KW
933 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
934
3986bb7c 935 return UNI_TO_NATIVE(uv);
27d6c58a
KW
936}
937
ec5f19d0 938/*
87cea99e 939=for apidoc utf8_to_uvchr
9041c2e3 940
977c1d31
KW
941DEPRECATED!
942
6ee84de2 943Returns the native code point of the first character in the string C<s>
1e54db1a 944which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
945length, in bytes, of that character.
946
4b88fb76 947Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
948malformed input could cause reading beyond the end of the input buffer, which
949is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 950
524080c4
KW
951If C<s> points to one of the detected malformations, and UTF8 warnings are
952enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
953NULL) to -1. If those warnings are off, the computed value if well-defined (or
954the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
955is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
956next possible position in C<s> that could begin a non-malformed character.
957See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
958
959=cut
960*/
961
962UV
7fc63493 963Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 964{
7918f24d
NC
965 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
966
2ff6c191 967 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
968}
969
970/*
ec5f19d0
KW
971=for apidoc utf8_to_uvuni_buf
972
973Returns the Unicode code point of the first character in the string C<s> which
974is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
975C<retlen> will be set to the length, in bytes, of that character.
976
977This function should only be used when the returned UV is considered
978an index into the Unicode semantic tables (e.g. swashes).
979
524080c4
KW
980If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
981enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
982NULL) to -1. If those warnings are off, the computed value if well-defined (or
983the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
984is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
985next possible position in C<s> that could begin a non-malformed character.
986See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
987
988=cut
989*/
990
991UV
992Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
993{
994 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
995
996 assert(send > s);
997
998 /* Call the low level routine asking for checks */
999 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1000 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1001}
1002
27d6c58a 1003/* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1004 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1005 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1006
1007UV
1008Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1009{
3986bb7c
KW
1010 UV expectlen = UTF8SKIP(s);
1011 const U8* send = s + expectlen;
1012 UV uv = NATIVE_TO_UTF(*s);
1013
27d6c58a
KW
1014 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1015
3986bb7c
KW
1016 if (retlen) {
1017 *retlen = expectlen;
1018 }
1019
1020 /* An invariant is trivially returned */
1021 if (expectlen == 1) {
1022 return uv;
1023 }
1024
1025 /* Remove the leading bits that indicate the number of bytes, leaving just
1026 * the bits that are part of the value */
1027 uv &= UTF_START_MASK(expectlen);
1028
1029 /* Now, loop through the remaining bytes, accumulating each into the
1030 * working total as we go */
1031 for (++s; s < send; s++) {
1032 uv = UTF8_ACCUMULATE(uv, *s);
1033 }
1034
1035 return uv;
27d6c58a
KW
1036}
1037
ec5f19d0 1038/*
87cea99e 1039=for apidoc utf8_to_uvuni
9041c2e3 1040
977c1d31
KW
1041DEPRECATED!
1042
9041c2e3 1043Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1044which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1045length, in bytes, of that character.
1046
39e518fd
KW
1047This function should only be used when the returned UV is considered
1048an index into the Unicode semantic tables (e.g. swashes).
1049
4b88fb76 1050Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1051malformed input could cause reading beyond the end of the input buffer, which
1052is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
9041c2e3 1053
524080c4
KW
1054If C<s> points to one of the detected malformations, and UTF8 warnings are
1055enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1056NULL) to -1. If those warnings are off, the computed value if well-defined (or
1057the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1058is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1059next possible position in C<s> that could begin a non-malformed character.
1060See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1061
1062=cut
1063*/
1064
1065UV
7fc63493 1066Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1067{
7918f24d
NC
1068 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1069
4b88fb76 1070 return valid_utf8_to_uvuni(s, retlen);
8e84507e
NIS
1071}
1072
b76347f2 1073/*
87cea99e 1074=for apidoc utf8_length
b76347f2
JH
1075
1076Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1077Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1078up past C<e>, croaks.
b76347f2
JH
1079
1080=cut
1081*/
1082
1083STRLEN
35a4481c 1084Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1085{
97aff369 1086 dVAR;
b76347f2
JH
1087 STRLEN len = 0;
1088
7918f24d
NC
1089 PERL_ARGS_ASSERT_UTF8_LENGTH;
1090
8850bf83
JH
1091 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1092 * the bitops (especially ~) can create illegal UTF-8.
1093 * In other words: in Perl UTF-8 is not just for Unicode. */
1094
a3b680e6
AL
1095 if (e < s)
1096 goto warn_and_return;
b76347f2 1097 while (s < e) {
8e91ec7f
AV
1098 if (!UTF8_IS_INVARIANT(*s))
1099 s += UTF8SKIP(s);
1100 else
1101 s++;
1102 len++;
1103 }
1104
1105 if (e != s) {
1106 len--;
1107 warn_and_return:
9b387841
NC
1108 if (PL_op)
1109 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1110 "%s in %s", unees, OP_DESC(PL_op));
1111 else
61a12c31 1112 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1113 }
1114
1115 return len;
1116}
1117
b06226ff 1118/*
87cea99e 1119=for apidoc utf8_distance
b06226ff 1120
1e54db1a 1121Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1122and C<b>.
1123
1124WARNING: use only if you *know* that the pointers point inside the
1125same UTF-8 buffer.
1126
37607a96
PK
1127=cut
1128*/
a0ed51b3 1129
02eb7b47 1130IV
35a4481c 1131Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1132{
7918f24d
NC
1133 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1134
bf1665bc 1135 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1136}
1137
b06226ff 1138/*
87cea99e 1139=for apidoc utf8_hop
b06226ff 1140
8850bf83
JH
1141Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1142forward or backward.
b06226ff
JH
1143
1144WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1145the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1146on the first byte of character or just after the last byte of a character.
b06226ff 1147
37607a96
PK
1148=cut
1149*/
a0ed51b3
LW
1150
1151U8 *
4373e329 1152Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1153{
7918f24d
NC
1154 PERL_ARGS_ASSERT_UTF8_HOP;
1155
96a5add6 1156 PERL_UNUSED_CONTEXT;
8850bf83
JH
1157 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1158 * the bitops (especially ~) can create illegal UTF-8.
1159 * In other words: in Perl UTF-8 is not just for Unicode. */
1160
a0ed51b3
LW
1161 if (off >= 0) {
1162 while (off--)
1163 s += UTF8SKIP(s);
1164 }
1165 else {
1166 while (off++) {
1167 s--;
8850bf83
JH
1168 while (UTF8_IS_CONTINUATION(*s))
1169 s--;
a0ed51b3
LW
1170 }
1171 }
4373e329 1172 return (U8 *)s;
a0ed51b3
LW
1173}
1174
6940069f 1175/*
fed3ba5d
NC
1176=for apidoc bytes_cmp_utf8
1177
a1433954
KW
1178Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1179sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1180equal, -1 or -2 if the first string is less than the second string, +1 or +2
1181if the first string is greater than the second string.
1182
1183-1 or +1 is returned if the shorter string was identical to the start of the
1184longer string. -2 or +2 is returned if the was a difference between characters
1185within the strings.
1186
1187=cut
1188*/
1189
1190int
1191Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1192{
1193 const U8 *const bend = b + blen;
1194 const U8 *const uend = u + ulen;
1195
1196 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1197
1198 PERL_UNUSED_CONTEXT;
1199
1200 while (b < bend && u < uend) {
1201 U8 c = *u++;
1202 if (!UTF8_IS_INVARIANT(c)) {
1203 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1204 if (u < uend) {
1205 U8 c1 = *u++;
1206 if (UTF8_IS_CONTINUATION(c1)) {
356979f4 1207 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
fed3ba5d
NC
1208 } else {
1209 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1210 "Malformed UTF-8 character "
1211 "(unexpected non-continuation byte 0x%02x"
1212 ", immediately after start byte 0x%02x)"
1213 /* Dear diag.t, it's in the pod. */
1214 "%s%s", c1, c,
1215 PL_op ? " in " : "",
1216 PL_op ? OP_DESC(PL_op) : "");
1217 return -2;
1218 }
1219 } else {
1220 if (PL_op)
1221 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1222 "%s in %s", unees, OP_DESC(PL_op));
1223 else
61a12c31 1224 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1225 return -2; /* Really want to return undef :-) */
1226 }
1227 } else {
1228 return -2;
1229 }
1230 }
1231 if (*b != c) {
1232 return *b < c ? -2 : +2;
1233 }
1234 ++b;
1235 }
1236
1237 if (b == bend && u == uend)
1238 return 0;
1239
1240 return b < bend ? +1 : -1;
1241}
1242
1243/*
87cea99e 1244=for apidoc utf8_to_bytes
6940069f 1245
2bbc8d55 1246Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1247Unlike L</bytes_to_utf8>, this over-writes the original string, and
1248updates C<len> to contain the new length.
67e989fb 1249Returns zero on failure, setting C<len> to -1.
6940069f 1250
a1433954 1251If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1252
6940069f
GS
1253=cut
1254*/
1255
1256U8 *
37607a96 1257Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1258{
d4c19fe8
AL
1259 U8 * const save = s;
1260 U8 * const send = s + *len;
6940069f 1261 U8 *d;
246fae53 1262
7918f24d
NC
1263 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1264
1e54db1a 1265 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1266 while (s < send) {
dcad2880
JH
1267 U8 c = *s++;
1268
1d72bdf6
NIS
1269 if (!UTF8_IS_INVARIANT(c) &&
1270 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1271 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
10edeb5d 1272 *len = ((STRLEN) -1);
dcad2880
JH
1273 return 0;
1274 }
246fae53 1275 }
dcad2880
JH
1276
1277 d = s = save;
6940069f 1278 while (s < send) {
ed646e6e 1279 STRLEN ulen;
4b88fb76 1280 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
ed646e6e 1281 s += ulen;
6940069f
GS
1282 }
1283 *d = '\0';
246fae53 1284 *len = d - save;
6940069f
GS
1285 return save;
1286}
1287
1288/*
87cea99e 1289=for apidoc bytes_from_utf8
f9a63242 1290
2bbc8d55 1291Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1292Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1293the newly-created string, and updates C<len> to contain the new
1294length. Returns the original string if no conversion occurs, C<len>
1295is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
12960 if C<s> is converted or consisted entirely of characters that are invariant
1297in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1298
37607a96
PK
1299=cut
1300*/
f9a63242
JH
1301
1302U8 *
e1ec3a88 1303Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1304{
f9a63242 1305 U8 *d;
e1ec3a88
AL
1306 const U8 *start = s;
1307 const U8 *send;
f9a63242
JH
1308 I32 count = 0;
1309
7918f24d
NC
1310 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1311
96a5add6 1312 PERL_UNUSED_CONTEXT;
f9a63242 1313 if (!*is_utf8)
73d840c0 1314 return (U8 *)start;
f9a63242 1315
1e54db1a 1316 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1317 for (send = s + *len; s < send;) {
e1ec3a88 1318 U8 c = *s++;
1d72bdf6 1319 if (!UTF8_IS_INVARIANT(c)) {
db42d148
NIS
1320 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1321 (c = *s++) && UTF8_IS_CONTINUATION(c))
1322 count++;
1323 else
73d840c0 1324 return (U8 *)start;
db42d148 1325 }
f9a63242
JH
1326 }
1327
35da51f7 1328 *is_utf8 = FALSE;
f9a63242 1329
212542aa 1330 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1331 s = start; start = d;
f9a63242
JH
1332 while (s < send) {
1333 U8 c = *s++;
c4d5f83a
NIS
1334 if (!UTF8_IS_INVARIANT(c)) {
1335 /* Then it is two-byte encoded */
356979f4 1336 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
c4d5f83a
NIS
1337 }
1338 *d++ = c;
f9a63242
JH
1339 }
1340 *d = '\0';
1341 *len = d - start;
73d840c0 1342 return (U8 *)start;
f9a63242
JH
1343}
1344
1345/*
87cea99e 1346=for apidoc bytes_to_utf8
6940069f 1347
ff97e5cf
KW
1348Converts a string C<s> of length C<len> bytes from the native encoding into
1349UTF-8.
6662521e 1350Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1351reflect the new length in bytes.
6940069f 1352
2bbc8d55
SP
1353A NUL character will be written after the end of the string.
1354
1355If you want to convert to UTF-8 from encodings other than
1356the native (Latin1 or EBCDIC),
a1433954 1357see L</sv_recode_to_utf8>().
c9ada85f 1358
497711e7 1359=cut
6940069f
GS
1360*/
1361
c682ebef
FC
1362/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1363 likewise need duplication. */
1364
6940069f 1365U8*
35a4481c 1366Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1367{
35a4481c 1368 const U8 * const send = s + (*len);
6940069f
GS
1369 U8 *d;
1370 U8 *dst;
7918f24d
NC
1371
1372 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1373 PERL_UNUSED_CONTEXT;
6940069f 1374
212542aa 1375 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1376 dst = d;
1377
1378 while (s < send) {
35a4481c 1379 const UV uv = NATIVE_TO_ASCII(*s++);
c4d5f83a 1380 if (UNI_IS_INVARIANT(uv))
eb160463 1381 *d++ = (U8)UTF_TO_NATIVE(uv);
6940069f 1382 else {
eb160463
GS
1383 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1384 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
6940069f
GS
1385 }
1386 }
1387 *d = '\0';
6662521e 1388 *len = d-dst;
6940069f
GS
1389 return dst;
1390}
1391
a0ed51b3 1392/*
dea0fc0b 1393 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1394 *
1395 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1396 * We optimize for native, for obvious reasons. */
1397
1398U8*
dea0fc0b 1399Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1400{
dea0fc0b
JH
1401 U8* pend;
1402 U8* dstart = d;
1403
7918f24d
NC
1404 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1405
dea0fc0b 1406 if (bytelen & 1)
f5992bc4 1407 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1408
1409 pend = p + bytelen;
1410
a0ed51b3 1411 while (p < pend) {
dea0fc0b
JH
1412 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1413 p += 2;
a0ed51b3 1414 if (uv < 0x80) {
e294cc5d
JH
1415#ifdef EBCDIC
1416 *d++ = UNI_TO_NATIVE(uv);
1417#else
eb160463 1418 *d++ = (U8)uv;
e294cc5d 1419#endif
a0ed51b3
LW
1420 continue;
1421 }
1422 if (uv < 0x800) {
eb160463
GS
1423 *d++ = (U8)(( uv >> 6) | 0xc0);
1424 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1425 continue;
1426 }
52b9aa85 1427 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
01ea242b 1428 if (p >= pend) {
dea0fc0b 1429 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1430 } else {
1431 UV low = (p[0] << 8) + p[1];
1432 p += 2;
52b9aa85 1433 if (low < 0xdc00 || low > 0xdfff)
01ea242b
NC
1434 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1435 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1436 }
dbde1951
NC
1437 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1438 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3
LW
1439 }
1440 if (uv < 0x10000) {
eb160463
GS
1441 *d++ = (U8)(( uv >> 12) | 0xe0);
1442 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1443 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1444 continue;
1445 }
1446 else {
eb160463
GS
1447 *d++ = (U8)(( uv >> 18) | 0xf0);
1448 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1449 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1450 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1451 continue;
1452 }
1453 }
dea0fc0b 1454 *newlen = d - dstart;
a0ed51b3
LW
1455 return d;
1456}
1457
1458/* Note: this one is slightly destructive of the source. */
1459
1460U8*
dea0fc0b 1461Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1462{
1463 U8* s = (U8*)p;
d4c19fe8 1464 U8* const send = s + bytelen;
7918f24d
NC
1465
1466 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1467
e0ea5e2d
NC
1468 if (bytelen & 1)
1469 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1470 (UV)bytelen);
1471
a0ed51b3 1472 while (s < send) {
d4c19fe8 1473 const U8 tmp = s[0];
a0ed51b3
LW
1474 s[0] = s[1];
1475 s[1] = tmp;
1476 s += 2;
1477 }
dea0fc0b 1478 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1479}
1480
c3fd2246
KW
1481/* for now these are all defined (inefficiently) in terms of the utf8 versions.
1482 * Note that the macros in handy.h that call these short-circuit calling them
1483 * for Latin-1 range inputs */
a0ed51b3
LW
1484
1485bool
84afefe6 1486Perl_is_uni_alnum(pTHX_ UV c)
a0ed51b3 1487{
89ebb4a3 1488 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1489 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1490 return is_utf8_alnum(tmpbuf);
1491}
1492
1493bool
84afefe6 1494Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1495{
89ebb4a3 1496 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1497 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1498 return is_utf8_idfirst(tmpbuf);
1499}
1500
1501bool
84afefe6 1502Perl_is_uni_alpha(pTHX_ UV c)
a0ed51b3 1503{
89ebb4a3 1504 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1505 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1506 return is_utf8_alpha(tmpbuf);
1507}
1508
1509bool
84afefe6 1510Perl_is_uni_ascii(pTHX_ UV c)
4d61ec05 1511{
bc39fe24 1512 return isASCII(c);
4d61ec05
GS
1513}
1514
1515bool
bdd8600f
KW
1516Perl_is_uni_blank(pTHX_ UV c)
1517{
1518 U8 tmpbuf[UTF8_MAXBYTES+1];
1519 uvchr_to_utf8(tmpbuf, c);
1520 return is_utf8_blank(tmpbuf);
1521}
1522
1523bool
84afefe6 1524Perl_is_uni_space(pTHX_ UV c)
a0ed51b3 1525{
89ebb4a3 1526 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1527 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1528 return is_utf8_space(tmpbuf);
1529}
1530
1531bool
84afefe6 1532Perl_is_uni_digit(pTHX_ UV c)
a0ed51b3 1533{
89ebb4a3 1534 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1535 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1536 return is_utf8_digit(tmpbuf);
1537}
1538
1539bool
84afefe6 1540Perl_is_uni_upper(pTHX_ UV c)
a0ed51b3 1541{
89ebb4a3 1542 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1543 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1544 return is_utf8_upper(tmpbuf);
1545}
1546
1547bool
84afefe6 1548Perl_is_uni_lower(pTHX_ UV c)
a0ed51b3 1549{
89ebb4a3 1550 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1551 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1552 return is_utf8_lower(tmpbuf);
1553}
1554
1555bool
84afefe6 1556Perl_is_uni_cntrl(pTHX_ UV c)
b8c5462f 1557{
7b952154 1558 return isCNTRL_L1(c);
b8c5462f
JH
1559}
1560
1561bool
84afefe6 1562Perl_is_uni_graph(pTHX_ UV c)
b8c5462f 1563{
89ebb4a3 1564 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1565 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1566 return is_utf8_graph(tmpbuf);
1567}
1568
1569bool
84afefe6 1570Perl_is_uni_print(pTHX_ UV c)
a0ed51b3 1571{
89ebb4a3 1572 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1573 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1574 return is_utf8_print(tmpbuf);
1575}
1576
b8c5462f 1577bool
84afefe6 1578Perl_is_uni_punct(pTHX_ UV c)
b8c5462f 1579{
89ebb4a3 1580 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1581 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1582 return is_utf8_punct(tmpbuf);
1583}
1584
4d61ec05 1585bool
84afefe6 1586Perl_is_uni_xdigit(pTHX_ UV c)
4d61ec05 1587{
89ebb4a3 1588 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
230880c1 1589 uvchr_to_utf8(tmpbuf, c);
4d61ec05
GS
1590 return is_utf8_xdigit(tmpbuf);
1591}
1592
3a4c58c9
KW
1593UV
1594Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1595{
1596 /* We have the latin1-range values compiled into the core, so just use
1597 * those, converting the result to utf8. The only difference between upper
1598 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1599 * either "SS" or "Ss". Which one to use is passed into the routine in
1600 * 'S_or_s' to avoid a test */
1601
1602 UV converted = toUPPER_LATIN1_MOD(c);
1603
1604 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1605
1606 assert(S_or_s == 'S' || S_or_s == 's');
1607
1608 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1609 characters in this range */
1610 *p = (U8) converted;
1611 *lenp = 1;
1612 return converted;
1613 }
1614
1615 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1616 * which it maps to one of them, so as to only have to have one check for
1617 * it in the main case */
1618 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1619 switch (c) {
1620 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1621 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1622 break;
1623 case MICRO_SIGN:
1624 converted = GREEK_CAPITAL_LETTER_MU;
1625 break;
1626 case LATIN_SMALL_LETTER_SHARP_S:
1627 *(p)++ = 'S';
1628 *p = S_or_s;
1629 *lenp = 2;
1630 return 'S';
1631 default:
1632 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1633 assert(0); /* NOTREACHED */
3a4c58c9
KW
1634 }
1635 }
1636
1637 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1638 *p = UTF8_TWO_BYTE_LO(converted);
1639 *lenp = 2;
1640
1641 return converted;
1642}
1643
50bda2c3
KW
1644/* Call the function to convert a UTF-8 encoded character to the specified case.
1645 * Note that there may be more than one character in the result.
1646 * INP is a pointer to the first byte of the input character
1647 * OUTP will be set to the first byte of the string of changed characters. It
1648 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1649 * LENP will be set to the length in bytes of the string of changed characters
1650 *
1651 * The functions return the ordinal of the first character in the string of OUTP */
f90a9a02
KW
1652#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1653#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1654#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
50bda2c3
KW
1655
1656/* This additionally has the input parameter SPECIALS, which if non-zero will
1657 * cause this to use the SPECIALS hash for folding (meaning get full case
1658 * folding); otherwise, when zero, this implies a simple case fold */
f90a9a02 1659#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
c3fd2246 1660
84afefe6
JH
1661UV
1662Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1663{
3a4c58c9
KW
1664 dVAR;
1665
a1433954
KW
1666 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1667 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1668 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1669 * the changed version may be longer than the original character.
1670 *
1671 * The ordinal of the first character of the changed version is returned
1672 * (but note, as explained above, that there may be more.) */
1673
7918f24d
NC
1674 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1675
3a4c58c9
KW
1676 if (c < 256) {
1677 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1678 }
1679
0ebc6274 1680 uvchr_to_utf8(p, c);
3a4c58c9 1681 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1682}
1683
84afefe6
JH
1684UV
1685Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1686{
3a4c58c9
KW
1687 dVAR;
1688
7918f24d
NC
1689 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1690
3a4c58c9
KW
1691 if (c < 256) {
1692 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1693 }
1694
0ebc6274 1695 uvchr_to_utf8(p, c);
3a4c58c9 1696 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1697}
1698
afc16117
KW
1699STATIC U8
1700S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1701{
1702 /* We have the latin1-range values compiled into the core, so just use
1703 * those, converting the result to utf8. Since the result is always just
a1433954 1704 * one character, we allow <p> to be NULL */
afc16117
KW
1705
1706 U8 converted = toLOWER_LATIN1(c);
1707
1708 if (p != NULL) {
1709 if (UNI_IS_INVARIANT(converted)) {
1710 *p = converted;
1711 *lenp = 1;
1712 }
1713 else {
1714 *p = UTF8_TWO_BYTE_HI(converted);
1715 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1716 *lenp = 2;
1717 }
1718 }
1719 return converted;
1720}
1721
84afefe6
JH
1722UV
1723Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1724{
968c5e6a
KW
1725 dVAR;
1726
7918f24d
NC
1727 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1728
afc16117
KW
1729 if (c < 256) {
1730 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1731 }
1732
afc16117 1733 uvchr_to_utf8(p, c);
968c5e6a 1734 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1735}
1736
84afefe6 1737UV
f673fad4 1738Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
a1dde8de 1739{
a1433954 1740 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
f673fad4
KW
1741 * folding */
1742
a1dde8de
KW
1743 UV converted;
1744
1745 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1746
1747 if (c == MICRO_SIGN) {
1748 converted = GREEK_SMALL_LETTER_MU;
1749 }
1750 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1751 *(p)++ = 's';
1752 *p = 's';
1753 *lenp = 2;
1754 return 's';
1755 }
1756 else { /* In this range the fold of all other characters is their lower
1757 case */
1758 converted = toLOWER_LATIN1(c);
1759 }
1760
1761 if (UNI_IS_INVARIANT(converted)) {
1762 *p = (U8) converted;
1763 *lenp = 1;
1764 }
1765 else {
1766 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1767 *p = UTF8_TWO_BYTE_LO(converted);
1768 *lenp = 2;
1769 }
1770
1771 return converted;
1772}
1773
1774UV
a0270393 1775Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
84afefe6 1776{
4b593389 1777
a0270393
KW
1778 /* Not currently externally documented, and subject to change
1779 * <flags> bits meanings:
1780 * FOLD_FLAGS_FULL iff full folding is to be used;
1781 * FOLD_FLAGS_LOCALE iff in locale
1782 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1783 */
4b593389 1784
36bb2ab6 1785 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1786
a1dde8de 1787 if (c < 256) {
a0270393
KW
1788 UV result = _to_fold_latin1((U8) c, p, lenp,
1789 cBOOL(((flags & FOLD_FLAGS_FULL)
1790 /* If ASCII-safe, don't allow full folding,
1791 * as that could include SHARP S => ss;
1792 * otherwise there is no crossing of
1793 * ascii/non-ascii in the latin1 range */
1794 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1795 /* It is illegal for the fold to cross the 255/256 boundary under
1796 * locale; in this case return the original */
1797 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1798 ? c
1799 : result;
a1dde8de
KW
1800 }
1801
a0270393
KW
1802 /* If no special needs, just use the macro */
1803 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1804 uvchr_to_utf8(p, c);
1805 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1806 }
1807 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1808 the special flags. */
1809 U8 utf8_c[UTF8_MAXBYTES + 1];
1810 uvchr_to_utf8(utf8_c, c);
1811 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1812 }
84afefe6
JH
1813}
1814
ea317ccb
KW
1815/* for now these all assume no locale info available for Unicode > 255; and
1816 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1817 * called instead, so that these don't get called for < 255 */
a0ed51b3
LW
1818
1819bool
84afefe6 1820Perl_is_uni_alnum_lc(pTHX_ UV c)
a0ed51b3
LW
1821{
1822 return is_uni_alnum(c); /* XXX no locale support yet */
1823}
1824
1825bool
84afefe6 1826Perl_is_uni_idfirst_lc(pTHX_ UV c)
a0ed51b3
LW
1827{
1828 return is_uni_idfirst(c); /* XXX no locale support yet */
1829}
1830
1831bool
84afefe6 1832Perl_is_uni_alpha_lc(pTHX_ UV c)
a0ed51b3
LW
1833{
1834 return is_uni_alpha(c); /* XXX no locale support yet */
1835}
1836
1837bool
84afefe6 1838Perl_is_uni_ascii_lc(pTHX_ UV c)
4d61ec05
GS
1839{
1840 return is_uni_ascii(c); /* XXX no locale support yet */
1841}
1842
1843bool
bdd8600f
KW
1844Perl_is_uni_blank_lc(pTHX_ UV c)
1845{
1846 return is_uni_blank(c); /* XXX no locale support yet */
1847}
1848
1849bool
84afefe6 1850Perl_is_uni_space_lc(pTHX_ UV c)
a0ed51b3
LW
1851{
1852 return is_uni_space(c); /* XXX no locale support yet */
1853}
1854
1855bool
84afefe6 1856Perl_is_uni_digit_lc(pTHX_ UV c)
a0ed51b3
LW
1857{
1858 return is_uni_digit(c); /* XXX no locale support yet */
1859}
1860
1861bool
84afefe6 1862Perl_is_uni_upper_lc(pTHX_ UV c)
a0ed51b3
LW
1863{
1864 return is_uni_upper(c); /* XXX no locale support yet */
1865}
1866
1867bool
84afefe6 1868Perl_is_uni_lower_lc(pTHX_ UV c)
a0ed51b3
LW
1869{
1870 return is_uni_lower(c); /* XXX no locale support yet */
1871}
1872
1873bool
84afefe6 1874Perl_is_uni_cntrl_lc(pTHX_ UV c)
b8c5462f
JH
1875{
1876 return is_uni_cntrl(c); /* XXX no locale support yet */
1877}
1878
1879bool
84afefe6 1880Perl_is_uni_graph_lc(pTHX_ UV c)
b8c5462f
JH
1881{
1882 return is_uni_graph(c); /* XXX no locale support yet */
1883}
1884
1885bool
84afefe6 1886Perl_is_uni_print_lc(pTHX_ UV c)
a0ed51b3
LW
1887{
1888 return is_uni_print(c); /* XXX no locale support yet */
1889}
1890
b8c5462f 1891bool
84afefe6 1892Perl_is_uni_punct_lc(pTHX_ UV c)
b8c5462f
JH
1893{
1894 return is_uni_punct(c); /* XXX no locale support yet */
1895}
1896
4d61ec05 1897bool
84afefe6 1898Perl_is_uni_xdigit_lc(pTHX_ UV c)
4d61ec05
GS
1899{
1900 return is_uni_xdigit(c); /* XXX no locale support yet */
1901}
1902
b7ac61fa
JH
1903U32
1904Perl_to_uni_upper_lc(pTHX_ U32 c)
1905{
ee099d14
JH
1906 /* XXX returns only the first character -- do not use XXX */
1907 /* XXX no locale support yet */
1908 STRLEN len;
89ebb4a3 1909 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1910 return (U32)to_uni_upper(c, tmpbuf, &len);
b7ac61fa
JH
1911}
1912
1913U32
1914Perl_to_uni_title_lc(pTHX_ U32 c)
1915{
ee099d14
JH
1916 /* XXX returns only the first character XXX -- do not use XXX */
1917 /* XXX no locale support yet */
1918 STRLEN len;
89ebb4a3 1919 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1920 return (U32)to_uni_title(c, tmpbuf, &len);
b7ac61fa
JH
1921}
1922
1923U32
1924Perl_to_uni_lower_lc(pTHX_ U32 c)
1925{
ee099d14
JH
1926 /* XXX returns only the first character -- do not use XXX */
1927 /* XXX no locale support yet */
1928 STRLEN len;
89ebb4a3 1929 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1930 return (U32)to_uni_lower(c, tmpbuf, &len);
b7ac61fa
JH
1931}
1932
7452cf6a 1933static bool
5141f98e 1934S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
1935 const char *const swashname)
1936{
ea317ccb
KW
1937 /* returns a boolean giving whether or not the UTF8-encoded character that
1938 * starts at <p> is in the swash indicated by <swashname>. <swash>
1939 * contains a pointer to where the swash indicated by <swashname>
1940 * is to be stored; which this routine will do, so that future calls will
1941 * look at <*swash> and only generate a swash if it is not null
1942 *
1943 * Note that it is assumed that the buffer length of <p> is enough to
1944 * contain all the bytes that comprise the character. Thus, <*p> should
1945 * have been checked before this call for mal-formedness enough to assure
1946 * that. */
1947
97aff369 1948 dVAR;
7918f24d
NC
1949
1950 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1951
492a624f
KW
1952 /* The API should have included a length for the UTF-8 character in <p>,
1953 * but it doesn't. We therefor assume that p has been validated at least
1954 * as far as there being enough bytes available in it to accommodate the
1955 * character without reading beyond the end, and pass that number on to the
1956 * validating routine */
1957 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
bde6a22d
NC
1958 return FALSE;
1959 if (!*swash)
711a919c 1960 *swash = swash_init("utf8", swashname, &PL_sv_undef, 1, 0);
bde6a22d
NC
1961 return swash_fetch(*swash, p, TRUE) != 0;
1962}
1963
1964bool
7fc63493 1965Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 1966{
97aff369 1967 dVAR;
7918f24d
NC
1968
1969 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1970
671c33bf
NC
1971 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1972 * descendant of isalnum(3), in other words, it doesn't
1973 * contain the '_'. --jhi */
d4c19fe8 1974 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
a0ed51b3
LW
1975}
1976
1977bool
7fc63493 1978Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1979{
97aff369 1980 dVAR;
7918f24d
NC
1981
1982 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1983
82686b01
JH
1984 if (*p == '_')
1985 return TRUE;
bde6a22d 1986 /* is_utf8_idstart would be more logical. */
d4c19fe8 1987 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
82686b01
JH
1988}
1989
1990bool
c11ff943
KW
1991Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1992{
1993 dVAR;
1994
1995 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1996
1997 if (*p == '_')
1998 return TRUE;
1999 /* is_utf8_idstart would be more logical. */
2000 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2001}
2002
2003bool
b6912c02
KW
2004Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2005{
2006 dVAR;
2007
2008 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2009
2010 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2011}
2012
2013bool
7fc63493 2014Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2015{
97aff369 2016 dVAR;
7918f24d
NC
2017
2018 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2019
d4c19fe8 2020 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
2021}
2022
2023bool
c11ff943
KW
2024Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2025{
2026 dVAR;
2027
2028 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2029
c11ff943
KW
2030 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2031}
2032
2033bool
7fc63493 2034Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 2035{
97aff369 2036 dVAR;
7918f24d
NC
2037
2038 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2039
d4c19fe8 2040 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
a0ed51b3
LW
2041}
2042
2043bool
7fc63493 2044Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2045{
97aff369 2046 dVAR;
7918f24d
NC
2047
2048 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2049
bc39fe24
KW
2050 /* ASCII characters are the same whether in utf8 or not. So the macro
2051 * works on both utf8 and non-utf8 representations. */
2052 return isASCII(*p);
b8c5462f
JH
2053}
2054
2055bool
bdd8600f
KW
2056Perl_is_utf8_blank(pTHX_ const U8 *p)
2057{
2058 dVAR;
2059
2060 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2061
2062 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2063}
2064
2065bool
7fc63493 2066Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2067{
97aff369 2068 dVAR;
7918f24d
NC
2069
2070 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2071
a34094a9 2072 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
a0ed51b3
LW
2073}
2074
2075bool
d1eb3177
YO
2076Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2077{
2078 dVAR;
2079
2080 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2081
c4428693
KW
2082 /* Only true if is an ASCII space-like character, and ASCII is invariant
2083 * under utf8, so can just use the macro */
2084 return isSPACE_A(*p);
d1eb3177
YO
2085}
2086
2087bool
2088Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2089{
2090 dVAR;
2091
2092 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2093
c4428693
KW
2094 /* Only true if is an ASCII word character, and ASCII is invariant
2095 * under utf8, so can just use the macro */
2096 return isWORDCHAR_A(*p);
d1eb3177
YO
2097}
2098
2099bool
7fc63493 2100Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2101{
97aff369 2102 dVAR;
7918f24d
NC
2103
2104 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2105
d4c19fe8 2106 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
a0ed51b3
LW
2107}
2108
2109bool
d1eb3177
YO
2110Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2111{
2112 dVAR;
2113
2114 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2115
c4428693
KW
2116 /* Only true if is an ASCII digit character, and ASCII is invariant
2117 * under utf8, so can just use the macro */
2118 return isDIGIT_A(*p);
d1eb3177
YO
2119}
2120
2121bool
7fc63493 2122Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2123{
97aff369 2124 dVAR;
7918f24d
NC
2125
2126 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2127
d4c19fe8 2128 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
a0ed51b3
LW
2129}
2130
2131bool
7fc63493 2132Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2133{
97aff369 2134 dVAR;
7918f24d
NC
2135
2136 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2137
d4c19fe8 2138 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
a0ed51b3
LW
2139}
2140
2141bool
7fc63493 2142Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2143{
97aff369 2144 dVAR;
7918f24d
NC
2145
2146 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2147
7b952154
KW
2148 if (isASCII(*p)) {
2149 return isCNTRL_A(*p);
2150 }
2151
2152 /* All controls are in Latin1 */
2153 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2154 return 0;
2155 }
2156 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
b8c5462f
JH
2157}
2158
2159bool
7fc63493 2160Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2161{
97aff369 2162 dVAR;
7918f24d
NC
2163
2164 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2165
d4c19fe8 2166 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
b8c5462f
JH
2167}
2168
2169bool
7fc63493 2170Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2171{
97aff369 2172 dVAR;
7918f24d
NC
2173
2174 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2175
d4c19fe8 2176 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
a0ed51b3
LW
2177}
2178
2179bool
7fc63493 2180Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2181{
97aff369 2182 dVAR;
7918f24d
NC
2183
2184 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2185
d4c19fe8 2186 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
b8c5462f
JH
2187}
2188
2189bool
7fc63493 2190Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2191{
97aff369 2192 dVAR;
7918f24d
NC
2193
2194 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2195
d1eb3177 2196 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
b8c5462f
JH
2197}
2198
2199bool
7fc63493 2200Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2201{
97aff369 2202 dVAR;
7918f24d
NC
2203
2204 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2205
d4c19fe8 2206 return is_utf8_common(p, &PL_utf8_mark, "IsM");
a0ed51b3
LW
2207}
2208
37e2e78e
KW
2209bool
2210Perl_is_utf8_X_begin(pTHX_ const U8 *p)
2211{
2212 dVAR;
2213
2214 PERL_ARGS_ASSERT_IS_UTF8_X_BEGIN;
2215
2216 return is_utf8_common(p, &PL_utf8_X_begin, "_X_Begin");
2217}
2218
2219bool
2220Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2221{
2222 dVAR;
2223
2224 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2225
2226 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2227}
2228
2229bool
2230Perl_is_utf8_X_prepend(pTHX_ const U8 *p)
2231{
2232 dVAR;
2233
2234 PERL_ARGS_ASSERT_IS_UTF8_X_PREPEND;
2235
f957fa32 2236 return is_utf8_common(p, &PL_utf8_X_prepend, "_X_GCB_Prepend");
37e2e78e
KW
2237}
2238
2239bool
2240Perl_is_utf8_X_non_hangul(pTHX_ const U8 *p)
2241{
2242 dVAR;
2243
2244 PERL_ARGS_ASSERT_IS_UTF8_X_NON_HANGUL;
2245
f957fa32 2246 return is_utf8_common(p, &PL_utf8_X_non_hangul, "_X_HST_Not_Applicable");
37e2e78e
KW
2247}
2248
2249bool
2250Perl_is_utf8_X_L(pTHX_ const U8 *p)
2251{
2252 dVAR;
2253
2254 PERL_ARGS_ASSERT_IS_UTF8_X_L;
2255
f957fa32 2256 return is_utf8_common(p, &PL_utf8_X_L, "_X_GCB_L");
37e2e78e
KW
2257}
2258
41de4811
KW
2259/* These constants are for finding GCB=LV and GCB=LVT. These are for the
2260 * pre-composed Hangul syllables, which are all in a contiguous block and
2261 * arranged there in such a way so as to facilitate alorithmic determination of
2262 * their characteristics. As such, they don't need a swash, but can be
2263 * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
2264 * is a GCB=LV */
2265#define SBASE 0xAC00 /* Start of block */
2266#define SCount 11172 /* Length of block */
2267#define TCount 28
2268
218282df 2269#if 0 /* This routine is not currently used */
37e2e78e
KW
2270bool
2271Perl_is_utf8_X_LV(pTHX_ const U8 *p)
2272{
41de4811
KW
2273 /* Unlike most other similarly named routines here, this does not create a
2274 * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
2275
37e2e78e
KW
2276 dVAR;
2277
41de4811
KW
2278 UV cp = valid_utf8_to_uvchr(p, NULL);
2279
37e2e78e
KW
2280 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
2281
41de4811
KW
2282 /* The earliest Unicode releases did not have these precomposed Hangul
2283 * syllables. Set to point to undef in that case, so will return false on
2284 * every call */
2285 if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
2286 PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
2287 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
2288 SvREFCNT_dec(PL_utf8_X_LV);
2289 PL_utf8_X_LV = &PL_sv_undef;
2290 }
2291 }
2292
2293 return (PL_utf8_X_LV != &PL_sv_undef
2294 && cp >= SBASE && cp < SBASE + SCount
2295 && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
37e2e78e 2296}
218282df 2297#endif
37e2e78e
KW
2298
2299bool
2300Perl_is_utf8_X_LVT(pTHX_ const U8 *p)
2301{
41de4811
KW
2302 /* Unlike most other similarly named routines here, this does not create a
2303 * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
2304
37e2e78e
KW
2305 dVAR;
2306
41de4811
KW
2307 UV cp = valid_utf8_to_uvchr(p, NULL);
2308
37e2e78e
KW
2309 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
2310
41de4811
KW
2311 /* The earliest Unicode releases did not have these precomposed Hangul
2312 * syllables. Set to point to undef in that case, so will return false on
2313 * every call */
2314 if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
2315 PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
2316 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
2317 SvREFCNT_dec(PL_utf8_X_LVT);
2318 PL_utf8_X_LVT = &PL_sv_undef;
2319 }
2320 }
2321
2322 return (PL_utf8_X_LVT != &PL_sv_undef
2323 && cp >= SBASE && cp < SBASE + SCount
2324 && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
37e2e78e
KW
2325}
2326
2327bool
2328Perl_is_utf8_X_T(pTHX_ const U8 *p)
2329{
2330 dVAR;
2331
2332 PERL_ARGS_ASSERT_IS_UTF8_X_T;
2333
f957fa32 2334 return is_utf8_common(p, &PL_utf8_X_T, "_X_GCB_T");
37e2e78e
KW
2335}
2336
2337bool
2338Perl_is_utf8_X_V(pTHX_ const U8 *p)
2339{
2340 dVAR;
2341
2342 PERL_ARGS_ASSERT_IS_UTF8_X_V;
2343
f957fa32 2344 return is_utf8_common(p, &PL_utf8_X_V, "_X_GCB_V");
37e2e78e
KW
2345}
2346
2347bool
2348Perl_is_utf8_X_LV_LVT_V(pTHX_ const U8 *p)
2349{
2350 dVAR;
2351
2352 PERL_ARGS_ASSERT_IS_UTF8_X_LV_LVT_V;
2353
2354 return is_utf8_common(p, &PL_utf8_X_LV_LVT_V, "_X_LV_LVT_V");
2355}
2356
2e2b2571
KW
2357bool
2358Perl__is_utf8_quotemeta(pTHX_ const U8 *p)
2359{
2360 /* For exclusive use of pp_quotemeta() */
2361
2362 dVAR;
2363
2364 PERL_ARGS_ASSERT__IS_UTF8_QUOTEMETA;
2365
2366 return is_utf8_common(p, &PL_utf8_quotemeta, "_Perl_Quotemeta");
2367}
2368
6b5c0936 2369/*
87cea99e 2370=for apidoc to_utf8_case
6b5c0936 2371
a1433954
KW
2372The C<p> contains the pointer to the UTF-8 string encoding
2373the character that is being converted. This routine assumes that the character
2374at C<p> is well-formed.
6b5c0936 2375
a1433954
KW
2376The C<ustrp> is a pointer to the character buffer to put the
2377conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2378of the result.
2379
a1433954 2380The C<swashp> is a pointer to the swash to use.
6b5c0936 2381
a1433954
KW
2382Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2383and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2384but not always, a multicharacter mapping), is tried first.
6b5c0936 2385
a1433954 2386The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2387hash %utf8::ToSpecLower. The access to the hash is through
2388Perl_to_utf8_case().
6b5c0936 2389
a1433954 2390The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2391%utf8::ToLower.
2392
2393=cut */
6b5c0936 2394
2104c8d9 2395UV
9a957fbc
AL
2396Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2397 SV **swashp, const char *normal, const char *special)
a0ed51b3 2398{
97aff369 2399 dVAR;
89ebb4a3 2400 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
0134edef 2401 STRLEN len = 0;
4b88fb76 2402 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
1feea2c7
JH
2403 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2404 * are necessary in EBCDIC, they are redundant no-ops
2405 * in ASCII-ish platforms, and hopefully optimized away. */
f54cb97a 2406 const UV uv1 = NATIVE_TO_UNI(uv0);
7918f24d
NC
2407
2408 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2409
9ae3ac1a
KW
2410 /* Note that swash_fetch() doesn't output warnings for these because it
2411 * assumes we will */
8457b38f 2412 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2413 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2414 if (ckWARN_d(WARN_SURROGATE)) {
2415 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2416 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2417 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2418 }
9ae3ac1a
KW
2419 }
2420 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2421 if (ckWARN_d(WARN_NON_UNICODE)) {
2422 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2423 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2424 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2425 }
9ae3ac1a
KW
2426 }
2427
2428 /* Note that non-characters are perfectly legal, so no warning should
2429 * be given */
2430 }
2431
1feea2c7 2432 uvuni_to_utf8(tmpbuf, uv1);
0134edef
JH
2433
2434 if (!*swashp) /* load on-demand */
2435 *swashp = swash_init("utf8", normal, &PL_sv_undef, 4, 0);
2436
a6f87d8c 2437 if (special) {
0134edef 2438 /* It might be "special" (sometimes, but not always,
2a37f04d 2439 * a multicharacter mapping) */
6673a63c 2440 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2441 SV **svp;
2442
35da51f7 2443 if (hv &&
b08cf34e
JH
2444 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2445 (*svp)) {
cfd0369c 2446 const char *s;
47654450 2447
cfd0369c 2448 s = SvPV_const(*svp, len);
47654450
JH
2449 if (len == 1)
2450 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2a37f04d 2451 else {
2f9475ad
JH
2452#ifdef EBCDIC
2453 /* If we have EBCDIC we need to remap the characters
2454 * since any characters in the low 256 are Unicode
2455 * code points, not EBCDIC. */
7cda7a3d 2456 U8 *t = (U8*)s, *tend = t + len, *d;
2f9475ad
JH
2457
2458 d = tmpbuf;
b08cf34e 2459 if (SvUTF8(*svp)) {
2f9475ad
JH
2460 STRLEN tlen = 0;
2461
2462 while (t < tend) {
6bd1c396 2463 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2f9475ad
JH
2464 if (tlen > 0) {
2465 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2466 t += tlen;
2467 }
2468 else
2469 break;
2470 }
2471 }
2472 else {
36fec512
JH
2473 while (t < tend) {
2474 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2475 t++;
2476 }
2f9475ad
JH
2477 }
2478 len = d - tmpbuf;
2479 Copy(tmpbuf, ustrp, len, U8);
2480#else
d2dcd0fb 2481 Copy(s, ustrp, len, U8);
2f9475ad 2482#endif
29e98929 2483 }
983ffd37 2484 }
0134edef
JH
2485 }
2486
2487 if (!len && *swashp) {
2114036c 2488 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
d4c19fe8 2489
0134edef
JH
2490 if (uv2) {
2491 /* It was "normal" (a single character mapping). */
d4c19fe8 2492 const UV uv3 = UNI_TO_NATIVE(uv2);
e9101d72 2493 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2a37f04d
JH
2494 }
2495 }
1feea2c7 2496
cbe07460
KW
2497 if (len) {
2498 if (lenp) {
2499 *lenp = len;
2500 }
2501 return valid_utf8_to_uvchr(ustrp, 0);
2502 }
2503
2504 /* Here, there was no mapping defined, which means that the code point maps
2505 * to itself. Return the inputs */
bfdf22ec
KW
2506 len = UTF8SKIP(p);
2507 Copy(p, ustrp, len, U8);
0134edef 2508
2a37f04d
JH
2509 if (lenp)
2510 *lenp = len;
2511
cbe07460
KW
2512 return uv0;
2513
a0ed51b3
LW
2514}
2515
051a06d4
KW
2516STATIC UV
2517S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2518{
2519 /* This is called when changing the case of a utf8-encoded character above
2520 * the Latin1 range, and the operation is in locale. If the result
2521 * contains a character that crosses the 255/256 boundary, disallow the
2522 * change, and return the original code point. See L<perlfunc/lc> for why;
2523 *
a1433954
KW
2524 * p points to the original string whose case was changed; assumed
2525 * by this routine to be well-formed
051a06d4
KW
2526 * result the code point of the first character in the changed-case string
2527 * ustrp points to the changed-case string (<result> represents its first char)
2528 * lenp points to the length of <ustrp> */
2529
2530 UV original; /* To store the first code point of <p> */
2531
2532 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2533
2534 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2535
2536 /* We know immediately if the first character in the string crosses the
2537 * boundary, so can skip */
2538 if (result > 255) {
2539
2540 /* Look at every character in the result; if any cross the
2541 * boundary, the whole thing is disallowed */
2542 U8* s = ustrp + UTF8SKIP(ustrp);
2543 U8* e = ustrp + *lenp;
2544 while (s < e) {
2545 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2546 {
2547 goto bad_crossing;
2548 }
2549 s += UTF8SKIP(s);
2550 }
2551
2552 /* Here, no characters crossed, result is ok as-is */
2553 return result;
2554 }
2555
2556bad_crossing:
2557
2558 /* Failed, have to return the original */
4b88fb76 2559 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2560 Copy(p, ustrp, *lenp, char);
2561 return original;
2562}
2563
d3e79532 2564/*
87cea99e 2565=for apidoc to_utf8_upper
d3e79532 2566
a1433954
KW
2567Convert the UTF-8 encoded character at C<p> to its uppercase version and
2568store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
89ebb4a3
JH
2569that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2570the uppercase version may be longer than the original character.
d3e79532
JH
2571
2572The first character of the uppercased version is returned
2573(but note, as explained above, that there may be more.)
2574
a1433954
KW
2575The character at C<p> is assumed by this routine to be well-formed.
2576
d3e79532
JH
2577=cut */
2578
051a06d4
KW
2579/* Not currently externally documented, and subject to change:
2580 * <flags> is set iff locale semantics are to be used for code points < 256
2581 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2582 * were used in the calculation; otherwise unchanged. */
2583
2104c8d9 2584UV
051a06d4 2585Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2586{
97aff369 2587 dVAR;
7918f24d 2588
051a06d4
KW
2589 UV result;
2590
2591 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2592
3a4c58c9 2593 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2594 if (flags) {
2595 result = toUPPER_LC(*p);
2596 }
2597 else {
81c6c7ce 2598 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2599 }
3a4c58c9
KW
2600 }
2601 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2602 if (flags) {
2603 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2604 }
2605 else {
81c6c7ce
KW
2606 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2607 ustrp, lenp, 'S');
051a06d4
KW
2608 }
2609 }
2610 else { /* utf8, ord above 255 */
2611 result = CALL_UPPER_CASE(p, ustrp, lenp);
2612
2613 if (flags) {
2614 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2615 }
2616 return result;
2617 }
2618
2619 /* Here, used locale rules. Convert back to utf8 */
2620 if (UTF8_IS_INVARIANT(result)) {
2621 *ustrp = (U8) result;
2622 *lenp = 1;
2623 }
2624 else {
2625 *ustrp = UTF8_EIGHT_BIT_HI(result);
2626 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2627 *lenp = 2;
3a4c58c9
KW
2628 }
2629
051a06d4
KW
2630 if (tainted_ptr) {
2631 *tainted_ptr = TRUE;
2632 }
2633 return result;
983ffd37 2634}
a0ed51b3 2635
d3e79532 2636/*
87cea99e 2637=for apidoc to_utf8_title
d3e79532 2638
a1433954
KW
2639Convert the UTF-8 encoded character at C<p> to its titlecase version and
2640store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2641that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2642titlecase version may be longer than the original character.
d3e79532
JH
2643
2644The first character of the titlecased version is returned
2645(but note, as explained above, that there may be more.)
2646
a1433954
KW
2647The character at C<p> is assumed by this routine to be well-formed.
2648
d3e79532
JH
2649=cut */
2650
051a06d4
KW
2651/* Not currently externally documented, and subject to change:
2652 * <flags> is set iff locale semantics are to be used for code points < 256
2653 * Since titlecase is not defined in POSIX, uppercase is used instead
2654 * for these/
2655 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2656 * were used in the calculation; otherwise unchanged. */
2657
983ffd37 2658UV
051a06d4 2659Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2660{
97aff369 2661 dVAR;
7918f24d 2662
051a06d4
KW
2663 UV result;
2664
2665 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2666
3a4c58c9 2667 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2668 if (flags) {
2669 result = toUPPER_LC(*p);
2670 }
2671 else {
81c6c7ce 2672 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2673 }
3a4c58c9
KW
2674 }
2675 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2676 if (flags) {
2677 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2678 }
2679 else {
81c6c7ce
KW
2680 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2681 ustrp, lenp, 's');
051a06d4
KW
2682 }
2683 }
2684 else { /* utf8, ord above 255 */
2685 result = CALL_TITLE_CASE(p, ustrp, lenp);
2686
2687 if (flags) {
2688 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2689 }
2690 return result;
2691 }
2692
2693 /* Here, used locale rules. Convert back to utf8 */
2694 if (UTF8_IS_INVARIANT(result)) {
2695 *ustrp = (U8) result;
2696 *lenp = 1;
2697 }
2698 else {
2699 *ustrp = UTF8_EIGHT_BIT_HI(result);
2700 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2701 *lenp = 2;
3a4c58c9
KW
2702 }
2703
051a06d4
KW
2704 if (tainted_ptr) {
2705 *tainted_ptr = TRUE;
2706 }
2707 return result;
a0ed51b3
LW
2708}
2709
d3e79532 2710/*
87cea99e 2711=for apidoc to_utf8_lower
d3e79532 2712
a1433954
KW
2713Convert the UTF-8 encoded character at C<p> to its lowercase version and
2714store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2715that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2716lowercase version may be longer than the original character.
d3e79532
JH
2717
2718The first character of the lowercased version is returned
2719(but note, as explained above, that there may be more.)
2720
a1433954
KW
2721The character at C<p> is assumed by this routine to be well-formed.
2722
d3e79532
JH
2723=cut */
2724
051a06d4
KW
2725/* Not currently externally documented, and subject to change:
2726 * <flags> is set iff locale semantics are to be used for code points < 256
2727 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2728 * were used in the calculation; otherwise unchanged. */
2729
2104c8d9 2730UV
051a06d4 2731Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2732{
051a06d4
KW
2733 UV result;
2734
97aff369 2735 dVAR;
7918f24d 2736
051a06d4 2737 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2738
968c5e6a 2739 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2740 if (flags) {
2741 result = toLOWER_LC(*p);
2742 }
2743 else {
81c6c7ce 2744 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2745 }
968c5e6a
KW
2746 }
2747 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2748 if (flags) {
2749 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2750 }
2751 else {
81c6c7ce
KW
2752 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2753 ustrp, lenp);
051a06d4 2754 }
968c5e6a 2755 }
051a06d4
KW
2756 else { /* utf8, ord above 255 */
2757 result = CALL_LOWER_CASE(p, ustrp, lenp);
2758
2759 if (flags) {
2760 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2761 }
968c5e6a 2762
051a06d4
KW
2763 return result;
2764 }
2765
2766 /* Here, used locale rules. Convert back to utf8 */
2767 if (UTF8_IS_INVARIANT(result)) {
2768 *ustrp = (U8) result;
2769 *lenp = 1;
2770 }
2771 else {
2772 *ustrp = UTF8_EIGHT_BIT_HI(result);
2773 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2774 *lenp = 2;
2775 }
2776
2777 if (tainted_ptr) {
2778 *tainted_ptr = TRUE;
2779 }
2780 return result;
b4e400f9
JH
2781}
2782
d3e79532 2783/*
87cea99e 2784=for apidoc to_utf8_fold
d3e79532 2785
a1433954
KW
2786Convert the UTF-8 encoded character at C<p> to its foldcase version and
2787store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2788that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
d3e79532
JH
2789foldcase version may be longer than the original character (up to
2790three characters).
2791
2792The first character of the foldcased version is returned
2793(but note, as explained above, that there may be more.)
2794
a1433954
KW
2795The character at C<p> is assumed by this routine to be well-formed.
2796
d3e79532
JH
2797=cut */
2798
051a06d4
KW
2799/* Not currently externally documented, and subject to change,
2800 * in <flags>
2801 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2802 * points < 256. Since foldcase is not defined in
2803 * POSIX, lowercase is used instead
2804 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2805 * otherwise simple folds
a0270393
KW
2806 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2807 * prohibited
051a06d4
KW
2808 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2809 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2810
b4e400f9 2811UV
051a06d4 2812Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2813{
97aff369 2814 dVAR;
7918f24d 2815
051a06d4
KW
2816 UV result;
2817
36bb2ab6 2818 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2819
a0270393
KW
2820 /* These are mutually exclusive */
2821 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2822
50ba90ff
KW
2823 assert(p != ustrp); /* Otherwise overwrites */
2824
a1dde8de 2825 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2826 if (flags & FOLD_FLAGS_LOCALE) {
2827 result = toLOWER_LC(*p);
2828 }
2829 else {
81c6c7ce
KW
2830 return _to_fold_latin1(*p, ustrp, lenp,
2831 cBOOL(flags & FOLD_FLAGS_FULL));
051a06d4 2832 }
a1dde8de
KW
2833 }
2834 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2835 if (flags & FOLD_FLAGS_LOCALE) {
2836 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2837 }
2838 else {
81c6c7ce 2839 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
a0270393
KW
2840 ustrp, lenp,
2841 cBOOL((flags & FOLD_FLAGS_FULL
2842 /* If ASCII safe, don't allow full
2843 * folding, as that could include SHARP
2844 * S => ss; otherwise there is no
2845 * crossing of ascii/non-ascii in the
2846 * latin1 range */
2847 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
051a06d4 2848 }
a1dde8de 2849 }
051a06d4 2850 else { /* utf8, ord above 255 */
a0270393 2851 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2852
051a06d4 2853 if ((flags & FOLD_FLAGS_LOCALE)) {
a0270393 2854 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2855 }
a0270393
KW
2856 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2857 return result;
2858 }
2859 else {
2860 /* This is called when changing the case of a utf8-encoded
2861 * character above the Latin1 range, and the result should not
2862 * contain an ASCII character. */
2863
2864 UV original; /* To store the first code point of <p> */
2865
2866 /* Look at every character in the result; if any cross the
2867 * boundary, the whole thing is disallowed */
2868 U8* s = ustrp;
2869 U8* e = ustrp + *lenp;
2870 while (s < e) {
2871 if (isASCII(*s)) {
2872 /* Crossed, have to return the original */
2873 original = valid_utf8_to_uvchr(p, lenp);
2874 Copy(p, ustrp, *lenp, char);
2875 return original;
2876 }
2877 s += UTF8SKIP(s);
2878 }
051a06d4 2879
a0270393
KW
2880 /* Here, no characters crossed, result is ok as-is */
2881 return result;
2882 }
051a06d4
KW
2883 }
2884
2885 /* Here, used locale rules. Convert back to utf8 */
2886 if (UTF8_IS_INVARIANT(result)) {
2887 *ustrp = (U8) result;
2888 *lenp = 1;
2889 }
2890 else {
2891 *ustrp = UTF8_EIGHT_BIT_HI(result);
2892 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2893 *lenp = 2;
2894 }
2895
2896 if (tainted_ptr) {
2897 *tainted_ptr = TRUE;
2898 }
2899 return result;
a0ed51b3
LW
2900}
2901
711a919c 2902/* Note:
f90a9a02 2903 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2904 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2905 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2906 */
c4a5db0c 2907
a0ed51b3 2908SV*
7fc63493 2909Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2910{
c4a5db0c
KW
2911 PERL_ARGS_ASSERT_SWASH_INIT;
2912
2913 /* Returns a copy of a swash initiated by the called function. This is the
2914 * public interface, and returning a copy prevents others from doing
2915 * mischief on the original */
2916
83199d38 2917 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, FALSE, NULL, NULL));
c4a5db0c
KW
2918}
2919
2920SV*
83199d38 2921Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, bool return_if_undef, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2922{
2923 /* Initialize and return a swash, creating it if necessary. It does this
9a53f6cf 2924 * by calling utf8_heavy.pl in the general case.
c4a5db0c
KW
2925 *
2926 * This interface should only be used by functions that won't destroy or
2927 * adversely change the swash, as doing so affects all other uses of the
2928 * swash in the program; the general public should use 'Perl_swash_init'
2929 * instead.
2930 *
2931 * pkg is the name of the package that <name> should be in.
2932 * name is the name of the swash to find. Typically it is a Unicode
2933 * property name, including user-defined ones
2934 * listsv is a string to initialize the swash with. It must be of the form
2935 * documented as the subroutine return value in
2936 * L<perlunicode/User-Defined Character Properties>
2937 * minbits is the number of bits required to represent each data element.
2938 * It is '1' for binary properties.
2939 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf
KW
2940 * return_if_undef is TRUE if the routine shouldn't croak if it can't find
2941 * the requested property
2942 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2943 * flags_p if non-NULL is the address of various input and output flag bits
2944 * to the routine, as follows: ('I' means is input to the routine;
2945 * 'O' means output from the routine. Only flags marked O are
2946 * meaningful on return.)
2947 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2948 * came from a user-defined property. (I O)
9a53f6cf
KW
2949 *
2950 * Thus there are three possible inputs to find the swash: <name>,
2951 * <listsv>, and <invlist>. At least one must be specified. The result
2952 * will be the union of the specified ones, although <listsv>'s various
2953 * actions can intersect, etc. what <name> gives.
2954 *
2955 * <invlist> is only valid for binary properties */
c4a5db0c 2956
27da23d5 2957 dVAR;
c4a5db0c 2958 SV* retval = &PL_sv_undef;
83199d38 2959 HV* swash_hv = NULL;
9a53f6cf
KW
2960
2961 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2962 assert(! invlist || minbits == 1);
2963
2964 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2965 * so */
2966 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2967 dSP;
2968 const size_t pkg_len = strlen(pkg);
2969 const size_t name_len = strlen(name);
2970 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2971 SV* errsv_save;
2972 GV *method;
2973
2974 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2975
2976 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2977 ENTER;
69794297
KW
2978 SAVEHINTS();
2979 save_re_context();
2980 if (PL_parser && PL_parser->error_count)
2981 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2982 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2983 if (!method) { /* demand load utf8 */
2984 ENTER;
2985 errsv_save = newSVsv(ERRSV);
2986 /* It is assumed that callers of this routine are not passing in
2987 * any user derived data. */
2988 /* Need to do this after save_re_context() as it will set
2989 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2990 * in Perl_magic_get). Even line to create errsv_save can turn on
2991 * PL_tainted. */
2992 SAVEBOOL(PL_tainted);
2993 PL_tainted = 0;
2994 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2995 NULL);
2996 if (!SvTRUE(ERRSV))
2997 sv_setsv(ERRSV, errsv_save);
2998 SvREFCNT_dec(errsv_save);
2999 LEAVE;
3000 }
3001 SPAGAIN;
3002 PUSHMARK(SP);
3003 EXTEND(SP,5);
3004 mPUSHp(pkg, pkg_len);
3005 mPUSHp(name, name_len);
3006 PUSHs(listsv);
3007 mPUSHi(minbits);
3008 mPUSHi(none);
3009 PUTBACK;
f8be5cf0 3010 errsv_save = newSVsv(ERRSV);
69794297
KW
3011 /* If we already have a pointer to the method, no need to use
3012 * call_method() to repeat the lookup. */
3013 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
3014 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3015 {
3016 retval = *PL_stack_sp--;
3017 SvREFCNT_inc(retval);
3018 }
f8be5cf0
JH
3019 if (!SvTRUE(ERRSV))
3020 sv_setsv(ERRSV, errsv_save);
3021 SvREFCNT_dec(errsv_save);
ce3b816e 3022 LEAVE;
69794297
KW
3023 POPSTACK;
3024 if (IN_PERL_COMPILETIME) {
3025 CopHINTS_set(PL_curcop, PL_hints);
3026 }
3027 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3028 if (SvPOK(retval))
3029
3030 /* If caller wants to handle missing properties, let them */
3031 if (return_if_undef) {
3032 return NULL;
3033 }
3034 Perl_croak(aTHX_
3035 "Can't find Unicode property definition \"%"SVf"\"",
3036 SVfARG(retval));
3037 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3038 }
9a53f6cf 3039 } /* End of calling the module to find the swash */
36eb48b4 3040
83199d38
KW
3041 /* If this operation fetched a swash, and we will need it later, get it */
3042 if (retval != &PL_sv_undef
3043 && (minbits == 1 || (flags_p
3044 && ! (*flags_p
3045 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3046 {
3047 swash_hv = MUTABLE_HV(SvRV(retval));
3048
3049 /* If we don't already know that there is a user-defined component to
3050 * this swash, and the user has indicated they wish to know if there is
3051 * one (by passing <flags_p>), find out */
3052 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3053 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3054 if (user_defined && SvUV(*user_defined)) {
3055 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3056 }
3057 }
3058 }
3059
36eb48b4
KW
3060 /* Make sure there is an inversion list for binary properties */
3061 if (minbits == 1) {
3062 SV** swash_invlistsvp = NULL;
3063 SV* swash_invlist = NULL;
9a53f6cf 3064 bool invlist_in_swash_is_valid = FALSE;
36eb48b4 3065
9a53f6cf 3066 /* If this operation fetched a swash, get its already existing
83199d38 3067 * inversion list, or create one for it */
36eb48b4 3068
83199d38
KW
3069 if (swash_hv) {
3070 swash_invlistsvp = hv_fetchs(swash_hv, "I", FALSE);
9a53f6cf
KW
3071 if (swash_invlistsvp) {
3072 swash_invlist = *swash_invlistsvp;
3073 invlist_in_swash_is_valid = TRUE;
3074 }
3075 else {
36eb48b4 3076 swash_invlist = _swash_to_invlist(retval);
9a53f6cf
KW
3077 }
3078 }
3079
3080 /* If an inversion list was passed in, have to include it */
3081 if (invlist) {
3082
3083 /* Any fetched swash will by now have an inversion list in it;
3084 * otherwise <swash_invlist> will be NULL, indicating that we
3085 * didn't fetch a swash */
3086 if (swash_invlist) {
3087
3088 /* Add the passed-in inversion list, which invalidates the one
3089 * already stored in the swash */
3090 invlist_in_swash_is_valid = FALSE;
3091 _invlist_union(invlist, swash_invlist, &swash_invlist);
3092 }
3093 else {
3094
3095 /* Here, there is no swash already. Set up a minimal one */
3096 swash_hv = newHV();
3097 retval = newRV_inc(MUTABLE_SV(swash_hv));
3098 swash_invlist = invlist;
3099 }
9a53f6cf
KW
3100 }
3101
3102 /* Here, we have computed the union of all the passed-in data. It may
3103 * be that there was an inversion list in the swash which didn't get
3104 * touched; otherwise save the one computed one */
3105 if (! invlist_in_swash_is_valid) {
69794297
KW
3106 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "INVLIST", swash_invlist))
3107 {
3108 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3109 }
9a53f6cf 3110 }
36eb48b4
KW
3111 }
3112
a0ed51b3
LW
3113 return retval;
3114}
3115
035d37be
JH
3116
3117/* This API is wrong for special case conversions since we may need to
3118 * return several Unicode characters for a single Unicode character
3119 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3120 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
3121 * multiple values. --jhi
3122 * For those, you should use to_utf8_case() instead */
b0e3252e 3123/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3124
3125/* Note:
3126 * Returns the value of property/mapping C<swash> for the first character
3127 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3128 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3129 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3130 *
3131 * A "swash" is a hash which contains initially the keys/values set up by
3132 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3133 * property for all possible code points. Things are stored in a compact form
3134 * (see utf8_heavy.pl) so that calculation is required to find the actual
3135 * property value for a given code point. As code points are looked up, new
3136 * key/value pairs are added to the hash, so that the calculation doesn't have
3137 * to ever be re-done. Further, each calculation is done, not just for the
3138 * desired one, but for a whole block of code points adjacent to that one.
3139 * For binary properties on ASCII machines, the block is usually for 64 code
3140 * points, starting with a code point evenly divisible by 64. Thus if the
3141 * property value for code point 257 is requested, the code goes out and
3142 * calculates the property values for all 64 code points between 256 and 319,
3143 * and stores these as a single 64-bit long bit vector, called a "swatch",
3144 * under the key for code point 256. The key is the UTF-8 encoding for code
3145 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3146 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3147 * for code point 258 is then requested, this code realizes that it would be
3148 * stored under the key for 256, and would find that value and extract the
3149 * relevant bit, offset from 256.
3150 *
3151 * Non-binary properties are stored in as many bits as necessary to represent
3152 * their values (32 currently, though the code is more general than that), not
3153 * as single bits, but the principal is the same: the value for each key is a
3154 * vector that encompasses the property values for all code points whose UTF-8
3155 * representations are represented by the key. That is, for all code points
3156 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3157 * bytes of that.
680c470c 3158 */
a0ed51b3 3159UV
680c470c 3160Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3161{
27da23d5 3162 dVAR;
ef8f7699 3163 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3164 U32 klen;
3165 U32 off;
a0ed51b3 3166 STRLEN slen;
7d85a32c 3167 STRLEN needents;
cfd0369c 3168 const U8 *tmps = NULL;
a0ed51b3 3169 U32 bit;
979f2922 3170 SV *swatch;
3568d838 3171 U8 tmputf8[2];
35da51f7 3172 const UV c = NATIVE_TO_ASCII(*ptr);
3568d838 3173
7918f24d
NC
3174 PERL_ARGS_ASSERT_SWASH_FETCH;
3175
dbe7a391 3176 /* Convert to utf8 if not already */
3568d838 3177 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
979f2922
ST
3178 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3179 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3180 ptr = tmputf8;
3568d838
JH
3181 }
3182 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
37e2e78e 3183 * then the "swatch" is a vec() for all the chars which start
3568d838
JH
3184 * with 0xAA..0xYY
3185 * So the key in the hash (klen) is length of encoded char -1
3186 */
3187 klen = UTF8SKIP(ptr) - 1;
3188 off = ptr[klen];
a0ed51b3 3189
979f2922 3190 if (klen == 0) {
37e2e78e 3191 /* If char is invariant then swatch is for all the invariant chars
1e54db1a 3192 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
7d85a32c 3193 */
979f2922
ST
3194 needents = UTF_CONTINUATION_MARK;
3195 off = NATIVE_TO_UTF(ptr[klen]);
3196 }
3197 else {
7d85a32c 3198 /* If char is encoded then swatch is for the prefix */
979f2922
ST
3199 needents = (1 << UTF_ACCUMULATION_SHIFT);
3200 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3201 }
7d85a32c 3202
a0ed51b3
LW
3203 /*
3204 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3205 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3206 * it's nothing to sniff at.) Pity we usually come through at least
3207 * two function calls to get here...
3208 *
3209 * NB: this code assumes that swatches are never modified, once generated!
3210 */
3211
3568d838 3212 if (hv == PL_last_swash_hv &&
a0ed51b3 3213 klen == PL_last_swash_klen &&
27da23d5 3214 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3215 {
3216 tmps = PL_last_swash_tmps;
3217 slen = PL_last_swash_slen;
3218 }
3219 else {
3220 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3221 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3222
b0e3252e 3223 /* If not cached, generate it via swatch_get */
979f2922
ST
3224 if (!svp || !SvPOK(*svp)
3225 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2b9d42f0
NIS
3226 /* We use utf8n_to_uvuni() as we want an index into
3227 Unicode tables, not a native character number.
3228 */
aec46f14 3229 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
872c91ae
JH
3230 ckWARN(WARN_UTF8) ?
3231 0 : UTF8_ALLOW_ANY);
b0e3252e 3232 swatch = swatch_get(swash,
979f2922 3233 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
361ee0fe 3234 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
979f2922
ST
3235 needents);
3236
923e4eb5 3237 if (IN_PERL_COMPILETIME)
623e6609 3238 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3239
979f2922 3240 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3241
979f2922
ST
3242 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3243 || (slen << 3) < needents)
5637ef5b
NC
3244 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3245 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3246 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3247 }
3248
3249 PL_last_swash_hv = hv;
16d8f38a 3250 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3251 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3252 /* FIXME change interpvar.h? */
3253 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3254 PL_last_swash_slen = slen;
3255 if (klen)
3256 Copy(ptr, PL_last_swash_key, klen, U8);
3257 }
3258
9faf8d75 3259 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3260 case 1:
3261 bit = 1 << (off & 7);
3262 off >>= 3;
3263 return (tmps[off] & bit) != 0;
3264 case 8:
3265 return tmps[off];
3266 case 16:
3267 off <<= 1;
3268 return (tmps[off] << 8) + tmps[off + 1] ;
3269 case 32:
3270 off <<= 2;
3271 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3272 }
5637ef5b
NC
3273 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3274 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3275 NORETURN_FUNCTION_END;
a0ed51b3 3276}
2b9d42f0 3277
319009ee
KW
3278/* Read a single line of the main body of the swash input text. These are of
3279 * the form:
3280 * 0053 0056 0073
3281 * where each number is hex. The first two numbers form the minimum and
3282 * maximum of a range, and the third is the value associated with the range.
3283 * Not all swashes should have a third number
3284 *
3285 * On input: l points to the beginning of the line to be examined; it points
3286 * to somewhere in the string of the whole input text, and is
3287 * terminated by a \n or the null string terminator.
3288 * lend points to the null terminator of that string
3289 * wants_value is non-zero if the swash expects a third number
3290 * typestr is the name of the swash's mapping, like 'ToLower'
3291 * On output: *min, *max, and *val are set to the values read from the line.
3292 * returns a pointer just beyond the line examined. If there was no
3293 * valid min number on the line, returns lend+1
3294 */
3295
3296STATIC U8*
3297S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3298 const bool wants_value, const U8* const typestr)
3299{
3300 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3301 STRLEN numlen; /* Length of the number */
02470786
KW
3302 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3303 | PERL_SCAN_DISALLOW_PREFIX
3304 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3305
3306 /* nl points to the next \n in the scan */
3307 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3308
3309 /* Get the first number on the line: the range minimum */
3310 numlen = lend - l;
3311 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3312 if (numlen) /* If found a hex number, position past it */
3313 l += numlen;
3314 else if (nl) { /* Else, go handle next line, if any */
3315 return nl + 1; /* 1 is length of "\n" */
3316 }
3317 else { /* Else, no next line */
3318 return lend + 1; /* to LIST's end at which \n is not found */
3319 }
3320
3321 /* The max range value follows, separated by a BLANK */
3322 if (isBLANK(*l)) {
3323 ++l;
02470786
KW
3324 flags = PERL_SCAN_SILENT_ILLDIGIT
3325 | PERL_SCAN_DISALLOW_PREFIX
3326 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3327 numlen = lend - l;
3328 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3329 if (numlen)
3330 l += numlen;
3331 else /* If no value here, it is a single element range */
3332 *max = *min;
3333
3334 /* Non-binary tables have a third entry: what the first element of the
3335 * range maps to */
3336 if (wants_value) {
3337 if (isBLANK(*l)) {
3338 ++l;
f90a9a02
KW
3339
3340 /* The ToLc, etc table mappings are not in hex, and must be
3341 * corrected by adding the code point to them */
3342 if (typeto) {
3343 char *after_strtol = (char *) lend;
3344 *val = Strtol((char *)l, &after_strtol, 10);
3345 l = (U8 *) after_strtol;
f90a9a02
KW
3346 }
3347 else { /* Other tables are in hex, and are the correct result
3348 without tweaking */
a9d188b3
KW
3349 flags = PERL_SCAN_SILENT_ILLDIGIT
3350 | PERL_SCAN_DISALLOW_PREFIX
3351 | PERL_SCAN_SILENT_NON_PORTABLE;
3352 numlen = lend - l;
3353 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3354 if (numlen)
3355 l += numlen;
3356 else
3357 *val = 0;
f90a9a02 3358 }
319009ee
KW
3359 }
3360 else {
3361 *val = 0;
3362 if (typeto) {
dcbac5bb 3363 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3364 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3365 typestr, l);
3366 }
3367 }
3368 }
3369 else
3370 *val = 0; /* bits == 1, then any val should be ignored */
3371 }
3372 else { /* Nothing following range min, should be single element with no
3373 mapping expected */
3374 *max = *min;
3375 if (wants_value) {
3376 *val = 0;
3377 if (typeto) {
dcbac5bb 3378 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3379 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3380 }
3381 }
3382 else
3383 *val = 0; /* bits == 1, then val should be ignored */
3384 }
3385
3386 /* Position to next line if any, or EOF */
3387 if (nl)
3388 l = nl + 1;
3389 else
3390 l = lend;
3391
3392 return l;
3393}
3394
979f2922
ST
3395/* Note:
3396 * Returns a swatch (a bit vector string) for a code point sequence
3397 * that starts from the value C<start> and comprises the number C<span>.
3398 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3399 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3400 */
3401STATIC SV*
b0e3252e 3402S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3403{
3404 SV *swatch;
77f9f126 3405 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3406 STRLEN lcur, xcur, scur;
ef8f7699 3407 HV *const hv = MUTABLE_HV(SvRV(swash));
36eb48b4
KW
3408 SV** const invlistsvp = hv_fetchs(hv, "INVLIST", FALSE);
3409
88d45d28
KW
3410 SV** listsvp = NULL; /* The string containing the main body of the table */
3411 SV** extssvp = NULL;
3412 SV** invert_it_svp = NULL;
3413 U8* typestr = NULL;
786861f5
KW
3414 STRLEN bits;
3415 STRLEN octets; /* if bits == 1, then octets == 0 */
3416 UV none;
3417 UV end = start + span;
972dd592 3418
36eb48b4 3419 if (invlistsvp == NULL) {
786861f5
KW
3420 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3421 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3422 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3423 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3424 listsvp = hv_fetchs(hv, "LIST", FALSE);
3425 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3426
3427 bits = SvUV(*bitssvp);
3428 none = SvUV(*nonesvp);
3429 typestr = (U8*)SvPV_nolen(*typesvp);
3430 }
36eb48b4
KW
3431 else {
3432 bits = 1;
3433 none = 0;
3434 }
786861f5 3435 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3436
b0e3252e 3437 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3438
979f2922 3439 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3440 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3441 (UV)bits);
979f2922
ST
3442 }
3443
84ea5ef6
KW
3444 /* If overflowed, use the max possible */
3445 if (end < start) {
3446 end = UV_MAX;
3447 span = end - start;
3448 }
3449
979f2922 3450 /* create and initialize $swatch */
979f2922 3451 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3452 swatch = newSV(scur);
3453 SvPOK_on(swatch);
979f2922
ST
3454 s = (U8*)SvPVX(swatch);
3455 if (octets && none) {
0bd48802 3456 const U8* const e = s + scur;
979f2922
ST
3457 while (s < e) {
3458 if (bits == 8)
3459 *s++ = (U8)(none & 0xff);
3460 else if (bits == 16) {
3461 *s++ = (U8)((none >> 8) & 0xff);
3462 *s++ = (U8)( none & 0xff);
3463 }
3464 else if (bits == 32) {
3465 *s++ = (U8)((none >> 24) & 0xff);
3466 *s++ = (U8)((none >> 16) & 0xff);
3467 *s++ = (U8)((none >> 8) & 0xff);
3468 *s++ = (U8)( none & 0xff);
3469 }
3470 }
3471 *s = '\0';
3472 }
3473 else {
3474 (void)memzero((U8*)s, scur + 1);
3475 }
3476 SvCUR_set(swatch, scur);
3477 s = (U8*)SvPVX(swatch);
3478
36eb48b4
KW
3479 if (invlistsvp) { /* If has an inversion list set up use that */
3480 _invlist_populate_swatch(*invlistsvp, start, end, s);
3481 return swatch;
3482 }
3483
3484 /* read $swash->{LIST} */
979f2922
ST
3485 l = (U8*)SvPV(*listsvp, lcur);
3486 lend = l + lcur;
3487 while (l < lend) {
8ed25d53 3488 UV min, max, val, upper;
319009ee
KW
3489 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3490 cBOOL(octets), typestr);
3491 if (l > lend) {
979f2922
ST
3492 break;
3493 }
3494
972dd592 3495 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3496 if (max < start)
3497 continue;
3498
8ed25d53
KW
3499 /* <end> is generally 1 beyond where we want to set things, but at the
3500 * platform's infinity, where we can't go any higher, we want to
3501 * include the code point at <end> */
3502 upper = (max < end)
3503 ? max
3504 : (max != UV_MAX || end != UV_MAX)
3505 ? end - 1
3506 : end;
3507
979f2922 3508 if (octets) {
35da51f7 3509 UV key;
979f2922
ST
3510 if (min < start) {
3511 if (!none || val < none) {
3512 val += start - min;
3513 }
3514 min = start;
3515 }
8ed25d53 3516 for (key = min; key <= upper; key++) {
979f2922 3517 STRLEN offset;
979f2922
ST
3518 /* offset must be non-negative (start <= min <= key < end) */
3519 offset = octets * (key - start);
3520 if (bits == 8)
3521 s[offset] = (U8)(val & 0xff);
3522 else if (bits == 16) {
3523 s[offset ] = (U8)((val >> 8) & 0xff);
3524 s[offset + 1] = (U8)( val & 0xff);
3525 }
3526 else if (bits == 32) {
3527 s[offset ] = (U8)((val >> 24) & 0xff);
3528 s[offset + 1] = (U8)((val >> 16) & 0xff);
3529 s[offset + 2] = (U8)((val >> 8) & 0xff);
3530 s[offset + 3] = (U8)( val & 0xff);
3531 }
3532
3533 if (!none || val < none)
3534 ++val;
3535 }
3536 }
711a919c 3537 else { /* bits == 1, then val should be ignored */
35da51f7 3538 UV key;
979f2922
ST
3539 if (min < start)
3540 min = start;
6cb05c12 3541
8ed25d53 3542 for (key = min; key <= upper; key++) {
0bd48802 3543 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3544 s[offset >> 3] |= 1 << (offset & 7);
3545 }
3546 }
3547 } /* while */
979f2922 3548
9479a769 3549 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3550 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3551
3552 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3553 * be 0, and their inversion should also be 0, as we don't succeed any
3554 * Unicode property matches for non-Unicode code points */
3555 if (start <= PERL_UNICODE_MAX) {
3556
3557 /* The code below assumes that we never cross the
3558 * Unicode/above-Unicode boundary in a range, as otherwise we would
3559 * have to figure out where to stop flipping the bits. Since this
3560 * boundary is divisible by a large power of 2, and swatches comes
3561 * in small powers of 2, this should be a valid assumption */
3562 assert(start + span - 1 <= PERL_UNICODE_MAX);
3563
507a8485
KW
3564 send = s + scur;
3565 while (s < send) {
3566 *s = ~(*s);
3567 s++;
3568 }
0bda3001 3569 }
77f9f126
KW
3570 }
3571
d73c39c5
KW
3572 /* read $swash->{EXTRAS}
3573 * This code also copied to swash_to_invlist() below */
979f2922
ST
3574 x = (U8*)SvPV(*extssvp, xcur);
3575 xend = x + xcur;
3576 while (x < xend) {
3577 STRLEN namelen;
3578 U8 *namestr;
3579 SV** othersvp;
3580 HV* otherhv;
3581 STRLEN otherbits;
3582 SV **otherbitssvp, *other;
711a919c 3583 U8 *s, *o, *nl;
979f2922
ST
3584 STRLEN slen, olen;
3585
35da51f7 3586 const U8 opc = *x++;
979f2922
ST
3587 if (opc == '\n')
3588 continue;
3589
3590 nl = (U8*)memchr(x, '\n', xend - x);
3591
3592 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3593 if (nl) {
3594 x = nl + 1; /* 1 is length of "\n" */
3595 continue;
3596 }
3597 else {
3598 x = xend; /* to EXTRAS' end at which \n is not found */
3599 break;
3600 }
3601 }
3602
3603 namestr = x;
3604 if (nl) {
3605 namelen = nl - namestr;
3606 x = nl + 1;
3607 }
3608 else {
3609 namelen = xend - namestr;
3610 x = xend;
3611 }
3612
3613 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3614 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3615 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3616 otherbits = (STRLEN)SvUV(*otherbitssvp);
3617 if (bits < otherbits)
5637ef5b
NC
3618 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3619 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3620
3621 /* The "other" swatch must be destroyed after. */
b0e3252e 3622 other = swatch_get(*othersvp, start, span);
979f2922
ST
3623 o = (U8*)SvPV(other, olen);
3624
3625 if (!olen)
b0e3252e 3626 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3627
3628 s = (U8*)SvPV(swatch, slen);
3629 if (bits == 1 && otherbits == 1) {
3630 if (slen != olen)
5637ef5b
NC
3631 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3632 "mismatch, slen=%"UVuf", olen=%"UVuf,
3633 (UV)slen, (UV)olen);
979f2922
ST
3634
3635 switch (opc) {
3636 case '+':
3637 while (slen--)
3638 *s++ |= *o++;
3639 break;
3640 case '!':
3641 while (slen--)
3642 *s++ |= ~*o++;
3643 break;
3644 case '-':
3645 while (slen--)
3646 *s++ &= ~*o++;
3647 break;
3648 case '&':
3649 while (slen--)
3650 *s++ &= *o++;
3651 break;
3652 default:
3653 break;
3654 }
3655 }
711a919c 3656 else {
979f2922
ST
3657 STRLEN otheroctets = otherbits >> 3;
3658 STRLEN offset = 0;
35da51f7 3659 U8* const send = s + slen;
979f2922
ST
3660
3661 while (s < send) {
3662 UV otherval = 0;
3663
3664 if (otherbits == 1) {
3665 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3666 ++offset;
3667 }
3668 else {
3669 STRLEN vlen = otheroctets;
3670 otherval = *o++;
3671 while (--vlen) {
3672 otherval <<= 8;
3673 otherval |= *o++;
3674 }
3675 }
3676
711a919c 3677 if (opc == '+' && otherval)
6f207bd3 3678 NOOP; /* replace with otherval */
979f2922
ST
3679 else if (opc == '!' && !otherval)
3680 otherval = 1;
3681 else if (opc == '-' && otherval)
3682 otherval = 0;
3683 else if (opc == '&' && !otherval)
3684 otherval = 0;
3685 else {
711a919c 3686 s += octets; /* no replacement */
979f2922
ST
3687 continue;
3688 }
3689
3690 if (bits == 8)
3691 *s++ = (U8)( otherval & 0xff);
3692 else if (bits == 16) {
3693 *s++ = (U8)((otherval >> 8) & 0xff);
3694 *s++ = (U8)( otherval & 0xff);
3695 }
3696 else if (bits == 32) {
3697 *s++ = (U8)((otherval >> 24) & 0xff);
3698 *s++ = (U8)((otherval >> 16) & 0xff);
3699 *s++ = (U8)((otherval >> 8) & 0xff);
3700 *s++ = (U8)( otherval & 0xff);
3701 }
3702 }
3703 }
3704 sv_free(other); /* through with it! */
3705 } /* while */
3706 return swatch;
3707}
3708
064c021d 3709HV*
4c2e1131 3710Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3711{
3712
5662e334
KW
3713 /* Subject to change or removal. For use only in one place in regcomp.c.
3714 * Can't be used on a property that is subject to user override, as it
3715 * relies on the value of SPECIALS in the swash which would be set by
3716 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3717 * for overridden properties
064c021d
KW
3718 *
3719 * Returns a hash which is the inversion and closure of a swash mapping.
3720 * For example, consider the input lines:
3721 * 004B 006B
3722 * 004C 006C
3723 * 212A 006B
3724 *
3725 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3726 * 006C. The value for each key is an array. For 006C, the array would
3727 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3728 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3729 *
3730 * Essentially, for any code point, it gives all the code points that map to
3731 * it, or the list of 'froms' for that point.
3732 *
5662e334
KW
3733 * Currently it ignores any additions or deletions from other swashes,
3734 * looking at just the main body of the swash, and if there are SPECIALS
3735 * in the swash, at that hash
3736 *
3737 * The specials hash can be extra code points, and most likely consists of
3738 * maps from single code points to multiple ones (each expressed as a string
3739 * of utf8 characters). This function currently returns only 1-1 mappings.
3740 * However consider this possible input in the specials hash:
3741 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3742 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3743 *
3744 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3745 * currently handle. But it also means that FB05 and FB06 are equivalent in
3746 * a 1-1 mapping which we should handle, and this relationship may not be in
3747 * the main table. Therefore this function examines all the multi-char
3748 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3749
3750 U8 *l, *lend;
3751 STRLEN lcur;
3752 HV *const hv = MUTABLE_HV(SvRV(swash));
3753
3754 /* The string containing the main body of the table */