This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
podcheck.t: Allow checking old perldelta
[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
LW
33#include "perl.h"
34
a0c21aa1 35#ifndef EBCDIC
970ea3cb 36/* Separate prototypes needed because in ASCII systems these are
a0c21aa1
JH
37 * usually macros but they still are compiled as code, too. */
38PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
39PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
40#endif
41
27da23d5
JH
42static const char unees[] =
43 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 44
48ef279e 45/*
ccfc67b7 46=head1 Unicode Support
a0ed51b3 47
166f8a29
DM
48This file contains various utility functions for manipulating UTF8-encoded
49strings. For the uninitiated, this is a method of representing arbitrary
61296642 50Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
51characters in the ASCII range are unmodified, and a zero byte never appears
52within non-zero characters.
166f8a29 53
eaf7a4d2
CS
54=cut
55*/
56
57/*
58=for apidoc is_ascii_string
59
a1433954 60Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
61or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
62is, if they are invariant. On ASCII-ish machines, only ASCII characters
63fit this definition, hence the function's name.
eaf7a4d2 64
9f7e3d64
MH
65If C<len> is 0, it will be calculated using C<strlen(s)>.
66
a1433954 67See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
68
69=cut
70*/
71
72bool
668b6d8d 73Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
74{
75 const U8* const send = s + (len ? len : strlen((const char *)s));
76 const U8* x = s;
77
78 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
79
80 for (; x < send; ++x) {
81 if (!UTF8_IS_INVARIANT(*x))
82 break;
83 }
84
85 return x == send;
86}
87
88/*
87cea99e 89=for apidoc uvuni_to_utf8_flags
eebe1485 90
6ee84de2
KW
91Adds the UTF-8 representation of the code point C<uv> to the end
92of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
eebe1485 93bytes available. The return value is the pointer to the byte after the
9041c2e3 94end of the new character. In other words,
eebe1485 95
b851fbc1
JH
96 d = uvuni_to_utf8_flags(d, uv, flags);
97
98or, in most cases,
99
9041c2e3 100 d = uvuni_to_utf8(d, uv);
eebe1485 101
b851fbc1
JH
102(which is equivalent to)
103
104 d = uvuni_to_utf8_flags(d, uv, 0);
105
949cf498 106This is the recommended Unicode-aware way of saying
eebe1485
SC
107
108 *(d++) = uv;
109
949cf498
KW
110This function will convert to UTF-8 (and not warn) even code points that aren't
111legal Unicode or are problematic, unless C<flags> contains one or more of the
a1433954
KW
112following flags:
113
949cf498
KW
114If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
115the function will raise a warning, provided UTF8 warnings are enabled. If instead
116UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
117If both flags are set, the function will both warn and return NULL.
118
119The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
120affect how the function handles a Unicode non-character. And, likewise for the
121UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
122above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
123even less portable) can be warned and/or disallowed even if other above-Unicode
124code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
125flags.
126
127And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
128above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
129DISALLOW flags.
130
131
eebe1485
SC
132=cut
133*/
134
dfe13c55 135U8 *
b851fbc1 136Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 137{
7918f24d
NC
138 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
139
979f77b6
KW
140 /* The first problematic code point is the first surrogate */
141 if (uv >= UNICODE_SURROGATE_FIRST
142 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
143 {
949cf498
KW
144 if (UNICODE_IS_SURROGATE(uv)) {
145 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 146 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
147 "UTF-16 surrogate U+%04"UVXf, uv);
148 }
149 if (flags & UNICODE_DISALLOW_SURROGATE) {
150 return NULL;
151 }
152 }
153 else if (UNICODE_IS_SUPER(uv)) {
154 if (flags & UNICODE_WARN_SUPER
155 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
156 {
8457b38f 157 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
158 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
159 }
160 if (flags & UNICODE_DISALLOW_SUPER
161 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
162 {
163 return NULL;
164 }
165 }
166 else if (UNICODE_IS_NONCHAR(uv)) {
167 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 168 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
169 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
170 uv);
171 }
172 if (flags & UNICODE_DISALLOW_NONCHAR) {
173 return NULL;
174 }
175 }
507b9800 176 }
c4d5f83a 177 if (UNI_IS_INVARIANT(uv)) {
eb160463 178 *d++ = (U8)UTF_TO_NATIVE(uv);
a0ed51b3
LW
179 return d;
180 }
2d331972 181#if defined(EBCDIC)
1d72bdf6
NIS
182 else {
183 STRLEN len = UNISKIP(uv);
184 U8 *p = d+len-1;
185 while (p > d) {
eb160463 186 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
187 uv >>= UTF_ACCUMULATION_SHIFT;
188 }
eb160463 189 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
190 return d+len;
191 }
192#else /* Non loop style */
a0ed51b3 193 if (uv < 0x800) {
eb160463
GS
194 *d++ = (U8)(( uv >> 6) | 0xc0);
195 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
196 return d;
197 }
198 if (uv < 0x10000) {
eb160463
GS
199 *d++ = (U8)(( uv >> 12) | 0xe0);
200 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
201 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
202 return d;
203 }
204 if (uv < 0x200000) {
eb160463
GS
205 *d++ = (U8)(( uv >> 18) | 0xf0);
206 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
207 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
208 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
209 return d;
210 }
211 if (uv < 0x4000000) {
eb160463
GS
212 *d++ = (U8)(( uv >> 24) | 0xf8);
213 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
216 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
217 return d;
218 }
219 if (uv < 0x80000000) {
eb160463
GS
220 *d++ = (U8)(( uv >> 30) | 0xfc);
221 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
222 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
223 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
225 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
226 return d;
227 }
6b8eaf93 228#ifdef HAS_QUAD
d7578b48 229 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
230#endif
231 {
eb160463
GS
232 *d++ = 0xfe; /* Can't match U+FEFF! */
233 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
234 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
235 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
238 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
239 return d;
240 }
6b8eaf93 241#ifdef HAS_QUAD
a0ed51b3 242 {
eb160463
GS
243 *d++ = 0xff; /* Can't match U+FFFE! */
244 *d++ = 0x80; /* 6 Reserved bits */
245 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
246 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
247 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
248 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
255 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
256 return d;
257 }
258#endif
1d72bdf6 259#endif /* Loop style */
a0ed51b3 260}
9041c2e3 261
646ca15d
JH
262/*
263
f7d739d1 264Tests if the first C<len> bytes of string C<s> form a valid UTF-8
646ca15d 265character. Note that an INVARIANT (i.e. ASCII) character is a valid
f7d739d1 266UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
267will be returned if it is valid, otherwise 0.
268
269This is the "slow" version as opposed to the "fast" version which is
270the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
271difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
272or less you should use the IS_UTF8_CHAR(), for lengths of five or more
273you should use the _slow(). In practice this means that the _slow()
274will be used very rarely, since the maximum Unicode code point (as of
275Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
276the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
277five bytes or more.
278
279=cut */
c053b435 280STATIC STRLEN
5f66b61c 281S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 282{
cd7e6c88 283 dTHX; /* The function called below requires thread context */
646ca15d 284
cd7e6c88 285 STRLEN actual_len;
646ca15d 286
cd7e6c88 287 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 288
cd7e6c88 289 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 290
cd7e6c88 291 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 292}
9041c2e3
NIS
293
294/*
492a624f
KW
295=for apidoc is_utf8_char_buf
296
297Returns the number of bytes that comprise the first UTF-8 encoded character in
298buffer C<buf>. C<buf_end> should point to one position beyond the end of the
299buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
300encoded character.
301
302Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
303machines) is a valid UTF-8 character.
304
305=cut */
306
307STRLEN
308Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
309{
310
311 STRLEN len;
312
313 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
314
315 if (buf_end <= buf) {
316 return 0;
317 }
318
319 len = buf_end - buf;
320 if (len > UTF8SKIP(buf)) {
321 len = UTF8SKIP(buf);
322 }
323
324#ifdef IS_UTF8_CHAR
325 if (IS_UTF8_CHAR_FAST(len))
326 return IS_UTF8_CHAR(buf, len) ? len : 0;
327#endif /* #ifdef IS_UTF8_CHAR */
328 return is_utf8_char_slow(buf, len);
329}
330
331/*
87cea99e 332=for apidoc is_utf8_char
eebe1485 333
76848387
KW
334DEPRECATED!
335
5da9da9e 336Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
337character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
338character is a valid UTF-8 character. The actual number of bytes in the UTF-8
339character will be returned if it is valid, otherwise 0.
9041c2e3 340
76848387 341This function is deprecated due to the possibility that malformed input could
a1433954 342cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 343instead.
e0328548 344
82686b01 345=cut */
76848387 346
067a85ef 347STRLEN
668b6d8d 348Perl_is_utf8_char(const U8 *s)
386d01d6 349{
7918f24d 350 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 351
76848387 352 /* Assumes we have enough space, which is why this is deprecated */
492a624f 353 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
354}
355
eaf7a4d2 356
6662521e 357/*
87cea99e 358=for apidoc is_utf8_string
6662521e 359
a1433954 360Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 361UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
362using C<strlen(s)> (which means if you use this option, that C<s> has to have a
363terminating NUL byte). Note that all characters being ASCII constitute 'a
364valid UTF-8 string'.
6662521e 365
a1433954 366See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 367
6662521e
GS
368=cut
369*/
370
8e84507e 371bool
668b6d8d 372Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 373{
35da51f7 374 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 375 const U8* x = s;
067a85ef 376
7918f24d 377 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 378
6662521e 379 while (x < send) {
1acdb0da 380 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
381 if (UTF8_IS_INVARIANT(*x)) {
382 x++;
383 }
1acdb0da 384 else if (!UTF8_IS_START(*x))
e0328548 385 return FALSE;
1acdb0da
JH
386 else {
387 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
388 const STRLEN c = UTF8SKIP(x);
389 const U8* const next_char_ptr = x + c;
390
391 if (next_char_ptr > send) {
392 return FALSE;
393 }
394
768c67ee
JH
395 if (IS_UTF8_CHAR_FAST(c)) {
396 if (!IS_UTF8_CHAR(x, c))
e0328548 397 return FALSE;
3c614e38 398 }
e0328548
KW
399 else if (! is_utf8_char_slow(x, c)) {
400 return FALSE;
401 }
402 x = next_char_ptr;
1acdb0da 403 }
6662521e 404 }
768c67ee 405
067a85ef 406 return TRUE;
6662521e
GS
407}
408
67e989fb 409/*
814fafa7
NC
410Implemented as a macro in utf8.h
411
87cea99e 412=for apidoc is_utf8_string_loc
814fafa7 413
a1433954
KW
414Like L</is_utf8_string> but stores the location of the failure (in the
415case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
416"utf8ness success") in the C<ep>.
417
a1433954 418See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 419
87cea99e 420=for apidoc is_utf8_string_loclen
81cd54e3 421
a1433954
KW
422Like L</is_utf8_string>() but stores the location of the failure (in the
423case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
424"utf8ness success") in the C<ep>, and the number of UTF-8
425encoded characters in the C<el>.
426
a1433954 427See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
428
429=cut
430*/
431
432bool
668b6d8d 433Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 434{
35da51f7 435 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 436 const U8* x = s;
81cd54e3 437 STRLEN c;
3ebfea28 438 STRLEN outlen = 0;
7918f24d
NC
439
440 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 441
81cd54e3 442 while (x < send) {
e0328548
KW
443 const U8* next_char_ptr;
444
81cd54e3
JH
445 /* Inline the easy bits of is_utf8_char() here for speed... */
446 if (UTF8_IS_INVARIANT(*x))
e0328548 447 next_char_ptr = x + 1;
768c67ee
JH
448 else if (!UTF8_IS_START(*x))
449 goto out;
81cd54e3 450 else {
768c67ee 451 /* ... and call is_utf8_char() only if really needed. */
768c67ee 452 c = UTF8SKIP(x);
e0328548
KW
453 next_char_ptr = c + x;
454 if (next_char_ptr > send) {
455 goto out;
456 }
768c67ee
JH
457 if (IS_UTF8_CHAR_FAST(c)) {
458 if (!IS_UTF8_CHAR(x, c))
459 c = 0;
460 } else
461 c = is_utf8_char_slow(x, c);
768c67ee
JH
462 if (!c)
463 goto out;
81cd54e3 464 }
e0328548 465 x = next_char_ptr;
3ebfea28 466 outlen++;
81cd54e3 467 }
768c67ee
JH
468
469 out:
3ebfea28
AL
470 if (el)
471 *el = outlen;
472
768c67ee
JH
473 if (ep)
474 *ep = x;
3ebfea28 475 return (x == send);
81cd54e3
JH
476}
477
478/*
768c67ee 479
87cea99e 480=for apidoc utf8n_to_uvuni
67e989fb 481
9041c2e3 482Bottom level UTF-8 decode routine.
746afd53
KW
483Returns the code point value of the first character in the string C<s>,
484which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
485C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
486the length, in bytes, of that character.
949cf498
KW
487
488The value of C<flags> determines the behavior when C<s> does not point to a
489well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
490zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
491next possible position in C<s> that could begin a non-malformed character.
492Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
493
494Various ALLOW flags can be set in C<flags> to allow (and not warn on)
495individual types of malformations, such as the sequence being overlong (that
496is, when there is a shorter sequence that can express the same code point;
497overlong sequences are expressly forbidden in the UTF-8 standard due to
498potential security issues). Another malformation example is the first byte of
499a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
500flags. For allowed 0 length strings, this function returns 0; for allowed
501overlong sequences, the computed code point is returned; for all other allowed
502malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
503determinable reasonable value.
949cf498
KW
504
505The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
506flags) malformation is found. If this flag is set, the routine assumes that
507the caller will raise a warning, and this function will silently just set
508C<retlen> to C<-1> and return zero.
509
510Certain code points are considered problematic. These are Unicode surrogates,
746afd53 511Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 512By default these are considered regular code points, but certain situations
5eafe189 513warrant special handling for them. If C<flags> contains
949cf498
KW
514UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
515malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
516UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
517maximum) can be set to disallow these categories individually.
518
519The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
520UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
521for their respective categories, but otherwise the code points are considered
522valid (not malformations). To get a category to both be treated as a
523malformation and raise a warning, specify both the WARN and DISALLOW flags.
524(But note that warnings are not raised if lexically disabled nor if
525UTF8_CHECK_ONLY is also specified.)
526
527Very large code points (above 0x7FFF_FFFF) are considered more problematic than
528the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
529reasons: they requre at least 32 bits to represent them on ASCII platforms, are
530not representable at all on EBCDIC platforms, and the original UTF-8
531specification never went above this number (the current 0x10FFFF limit was
532imposed later). (The smaller ones, those that fit into 32 bits, are
533representable by a UV on ASCII platforms, but not by an IV, which means that
534the number of operations that can be performed on them is quite restricted.)
535The UTF-8 encoding on ASCII platforms for these large code points begins with a
536byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
537be treated as malformations, while allowing smaller above-Unicode code points.
538(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
539including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
540the other WARN flags, but applies just to these code points.
949cf498
KW
541
542All other code points corresponding to Unicode characters, including private
543use and those yet to be assigned, are never considered malformed and never
544warn.
67e989fb 545
ec5f19d0 546Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
9041c2e3 547
37607a96
PK
548=cut
549*/
67e989fb 550
a0ed51b3 551UV
7fc63493 552Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 553{
97aff369 554 dVAR;
d4c19fe8 555 const U8 * const s0 = s;
eb83ed87 556 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 557 U8 * send;
eb83ed87
KW
558 UV uv = *s;
559 STRLEN expectlen;
949cf498 560 SV* sv = NULL;
eb83ed87
KW
561 UV outlier_ret = 0; /* return value when input is in error or problematic
562 */
563 UV pack_warn = 0; /* Save result of packWARN() for later */
564 bool unexpected_non_continuation = FALSE;
565 bool overflowed = FALSE;
2f8f112e 566 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 567
eb83ed87 568 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 569
eb83ed87 570 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
a0dbb045 571
eb83ed87
KW
572 /* The order of malformation tests here is important. We should consume as
573 * few bytes as possible in order to not skip any valid character. This is
574 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
575 * http://unicode.org/reports/tr36 for more discussion as to why. For
576 * example, once we've done a UTF8SKIP, we can tell the expected number of
577 * bytes, and could fail right off the bat if the input parameters indicate
578 * that there are too few available. But it could be that just that first
579 * byte is garbled, and the intended character occupies fewer bytes. If we
580 * blindly assumed that the first byte is correct, and skipped based on
581 * that number, we could skip over a valid input character. So instead, we
582 * always examine the sequence byte-by-byte.
583 *
584 * We also should not consume too few bytes, otherwise someone could inject
585 * things. For example, an input could be deliberately designed to
586 * overflow, and if this code bailed out immediately upon discovering that,
587 * returning to the caller *retlen pointing to the very next byte (one
588 * which is actually part of of the overflowing sequence), that could look
589 * legitimate to the caller, which could discard the initial partial
590 * sequence and process the rest, inappropriately */
591
592 /* Zero length strings, if allowed, of necessity are zero */
593 if (curlen == 0) {
594 if (retlen) {
595 *retlen = 0;
596 }
a0dbb045 597
eb83ed87
KW
598 if (flags & UTF8_ALLOW_EMPTY) {
599 return 0;
600 }
601 if (! (flags & UTF8_CHECK_ONLY)) {
602 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
603 }
0c443dc2
JH
604 goto malformed;
605 }
606
eb83ed87
KW
607 expectlen = UTF8SKIP(s);
608
609 /* A well-formed UTF-8 character, as the vast majority of calls to this
610 * function will be for, has this expected length. For efficiency, set
611 * things up here to return it. It will be overriden only in those rare
612 * cases where a malformation is found */
613 if (retlen) {
614 *retlen = expectlen;
615 }
616
617 /* An invariant is trivially well-formed */
1d72bdf6 618 if (UTF8_IS_INVARIANT(uv)) {
c4d5f83a 619 return (UV) (NATIVE_TO_UTF(*s));
a0ed51b3 620 }
67e989fb 621
eb83ed87
KW
622 /* A continuation character can't start a valid sequence */
623 if (UTF8_IS_CONTINUATION(uv)) {
624 if (flags & UTF8_ALLOW_CONTINUATION) {
625 if (retlen) {
626 *retlen = 1;
627 }
628 return UNICODE_REPLACEMENT;
629 }
ba210ebe 630
eb83ed87
KW
631 if (! (flags & UTF8_CHECK_ONLY)) {
632 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
633 }
634 curlen = 1;
ba210ebe
JH
635 goto malformed;
636 }
9041c2e3 637
1d72bdf6 638#ifdef EBCDIC
75383841 639 uv = NATIVE_TO_UTF(uv);
1d72bdf6
NIS
640#endif
641
eb83ed87
KW
642 /* Here is not a continuation byte, nor an invariant. The only thing left
643 * is a start byte (possibly for an overlong) */
ba210ebe 644
eb83ed87
KW
645 /* Remove the leading bits that indicate the number of bytes in the
646 * character's whole UTF-8 sequence, leaving just the bits that are part of
647 * the value */
648 uv &= UTF_START_MASK(expectlen);
ba210ebe 649
eb83ed87
KW
650 /* Now, loop through the remaining bytes in the character's sequence,
651 * accumulating each into the working value as we go. Be sure to not look
652 * past the end of the input string */
0b8d30e8
KW
653 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
654
eb83ed87
KW
655 for (s = s0 + 1; s < send; s++) {
656 if (UTF8_IS_CONTINUATION(*s)) {
657#ifndef EBCDIC /* Can't overflow in EBCDIC */
658 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
659
660 /* The original implementors viewed this malformation as more
661 * serious than the others (though I, khw, don't understand
662 * why, since other malformations also give very very wrong
663 * results), so there is no way to turn off checking for it.
664 * Set a flag, but keep going in the loop, so that we absorb
665 * the rest of the bytes that comprise the character. */
666 overflowed = TRUE;
667 overflow_byte = *s; /* Save for warning message's use */
668 }
669#endif
8850bf83 670 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
671 }
672 else {
673 /* Here, found a non-continuation before processing all expected
674 * bytes. This byte begins a new character, so quit, even if
675 * allowing this malformation. */
676 unexpected_non_continuation = TRUE;
677 break;
678 }
679 } /* End of loop through the character's bytes */
680
681 /* Save how many bytes were actually in the character */
682 curlen = s - s0;
683
684 /* The loop above finds two types of malformations: non-continuation and/or
685 * overflow. The non-continuation malformation is really a too-short
686 * malformation, as it means that the current character ended before it was
687 * expected to (being terminated prematurely by the beginning of the next
688 * character, whereas in the too-short malformation there just are too few
689 * bytes available to hold the character. In both cases, the check below
690 * that we have found the expected number of bytes would fail if executed.)
691 * Thus the non-continuation malformation is really unnecessary, being a
692 * subset of the too-short malformation. But there may be existing
693 * applications that are expecting the non-continuation type, so we retain
694 * it, and return it in preference to the too-short malformation. (If this
695 * code were being written from scratch, the two types might be collapsed
696 * into one.) I, khw, am also giving priority to returning the
697 * non-continuation and too-short malformations over overflow when multiple
698 * ones are present. I don't know of any real reason to prefer one over
699 * the other, except that it seems to me that multiple-byte errors trumps
700 * errors from a single byte */
701 if (unexpected_non_continuation) {
702 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
703 if (! (flags & UTF8_CHECK_ONLY)) {
704 if (curlen == 1) {
705 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
706 }
707 else {
708 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
709 }
710 }
eb83ed87
KW
711 goto malformed;
712 }
713 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
714
715 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
716 * as what the original expectations were. */
717 do_overlong_test = FALSE;
eb83ed87
KW
718 if (retlen) {
719 *retlen = curlen;
720 }
721 }
722 else if (curlen < expectlen) {
723 if (! (flags & UTF8_ALLOW_SHORT)) {
724 if (! (flags & UTF8_CHECK_ONLY)) {
725 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 726 }
eb83ed87
KW
727 goto malformed;
728 }
729 uv = UNICODE_REPLACEMENT;
2f8f112e 730 do_overlong_test = FALSE;
eb83ed87
KW
731 if (retlen) {
732 *retlen = curlen;
733 }
734 }
735
736#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 737 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
738 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
739 {
740 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
741 * generation of the sv, since no warnings are raised under CHECK */
742 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
743 && ckWARN_d(WARN_UTF8))
744 {
745 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
746 pack_warn = packWARN(WARN_UTF8);
747 }
748 if (flags & UTF8_DISALLOW_FE_FF) {
749 goto malformed;
ba210ebe 750 }
ba210ebe 751 }
2f8f112e 752 if (overflowed) {
ba210ebe 753
eb83ed87
KW
754 /* If the first byte is FF, it will overflow a 32-bit word. If the
755 * first byte is FE, it will overflow a signed 32-bit word. The
756 * above preserves backward compatibility, since its message was used
757 * in earlier versions of this code in preference to overflow */
758 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 759 goto malformed;
eb83ed87
KW
760 }
761#endif
762
2f8f112e
KW
763 if (do_overlong_test
764 && expectlen > (STRLEN)UNISKIP(uv)
765 && ! (flags & UTF8_ALLOW_LONG))
766 {
eb83ed87
KW
767 /* The overlong malformation has lower precedence than the others.
768 * Note that if this malformation is allowed, we return the actual
769 * value, instead of the replacement character. This is because this
770 * value is actually well-defined. */
771 if (! (flags & UTF8_CHECK_ONLY)) {
772 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));
773 }
774 goto malformed;
775 }
776
777 /* Here, the input is considered to be well-formed , but could be a
778 * problematic code point that is not allowed by the input parameters. */
779 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
780 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
781 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
782 {
949cf498 783 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
784 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
785 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
786 {
111d382d 787 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 788 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
789 }
790 if (flags & UTF8_DISALLOW_SURROGATE) {
791 goto disallowed;
792 }
793 }
794 else if (UNICODE_IS_NONCHAR(uv)) {
eb83ed87
KW
795 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
796 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
797 {
111d382d 798 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
eb83ed87 799 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
949cf498
KW
800 }
801 if (flags & UTF8_DISALLOW_NONCHAR) {
802 goto disallowed;
803 }
804 }
805 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
806 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
807 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
808 {
111d382d 809 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 810 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
811 }
812 if (flags & UTF8_DISALLOW_SUPER) {
813 goto disallowed;
814 }
815 }
816
eb83ed87
KW
817 if (sv) {
818 outlier_ret = uv;
819 goto do_warn;
820 }
821
949cf498
KW
822 /* Here, this is not considered a malformed character, so drop through
823 * to return it */
a0ed51b3 824 }
ba210ebe 825
a0ed51b3 826 return uv;
ba210ebe 827
eb83ed87
KW
828 /* There are three cases which get to beyond this point. In all 3 cases:
829 * <sv> if not null points to a string to print as a warning.
830 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
831 * set.
832 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
833 * This is done by initializing it to 0, and changing it only
834 * for case 1).
835 * The 3 cases are:
836 * 1) The input is valid but problematic, and to be warned about. The
837 * return value is the resultant code point; <*retlen> is set to
838 * <curlen>, the number of bytes that comprise the code point.
839 * <pack_warn> contains the result of packWARN() for the warning
840 * types. The entry point for this case is the label <do_warn>;
841 * 2) The input is a valid code point but disallowed by the parameters to
842 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
843 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
844 * comprise the code point. <pack_warn> contains the result of
845 * packWARN() for the warning types. The entry point for this case is
846 * the label <disallowed>.
847 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
848 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
849 * bytes that comprise the malformation. All such malformations are
850 * assumed to be warning type <utf8>. The entry point for this case
851 * is the label <malformed>.
852 */
949cf498 853
ba210ebe
JH
854malformed:
855
eb83ed87
KW
856 if (sv && ckWARN_d(WARN_UTF8)) {
857 pack_warn = packWARN(WARN_UTF8);
858 }
859
860disallowed:
861
fcc8fcf6 862 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 863 if (retlen)
10edeb5d 864 *retlen = ((STRLEN) -1);
ba210ebe
JH
865 return 0;
866 }
867
eb83ed87 868do_warn:
5b311467 869
eb83ed87
KW
870 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
871 if warnings are to be raised. */
f555bc63 872 const char * const string = SvPVX_const(sv);
a0dbb045 873
f555bc63
KW
874 if (PL_op)
875 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
876 else
877 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
878 }
879
eb83ed87
KW
880 if (retlen) {
881 *retlen = curlen;
882 }
ba210ebe 883
eb83ed87 884 return outlier_ret;
a0ed51b3
LW
885}
886
8e84507e 887/*
ec5f19d0
KW
888=for apidoc utf8_to_uvchr_buf
889
890Returns the native code point of the first character in the string C<s> which
891is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 892C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 893
524080c4
KW
894If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
895enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
896NULL) to -1. If those warnings are off, the computed value if well-defined (or
897the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
898is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
899next possible position in C<s> that could begin a non-malformed character.
900See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
901
902=cut
903*/
904
905
906UV
907Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
908{
909 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
910
911 assert(s < send);
912
913 return utf8n_to_uvchr(s, send - s, retlen,
914 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
915}
916
27d6c58a 917/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c
KW
918 * there are no malformations in the input UTF-8 string C<s>. surrogates,
919 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
920
921UV
922Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
923{
3986bb7c
KW
924 const UV uv = valid_utf8_to_uvuni(s, retlen);
925
27d6c58a
KW
926 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
927
3986bb7c 928 return UNI_TO_NATIVE(uv);
27d6c58a
KW
929}
930
ec5f19d0 931/*
87cea99e 932=for apidoc utf8_to_uvchr
9041c2e3 933
977c1d31
KW
934DEPRECATED!
935
6ee84de2 936Returns the native code point of the first character in the string C<s>
1e54db1a 937which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
938length, in bytes, of that character.
939
4b88fb76 940Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
941malformed input could cause reading beyond the end of the input buffer, which
942is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 943
524080c4
KW
944If C<s> points to one of the detected malformations, and UTF8 warnings are
945enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
946NULL) to -1. If those warnings are off, the computed value if well-defined (or
947the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
948is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
949next possible position in C<s> that could begin a non-malformed character.
950See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
951
952=cut
953*/
954
955UV
7fc63493 956Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 957{
7918f24d
NC
958 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
959
2ff6c191 960 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
961}
962
963/*
ec5f19d0
KW
964=for apidoc utf8_to_uvuni_buf
965
966Returns the Unicode code point of the first character in the string C<s> which
967is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
968C<retlen> will be set to the length, in bytes, of that character.
969
970This function should only be used when the returned UV is considered
971an index into the Unicode semantic tables (e.g. swashes).
972
524080c4
KW
973If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
974enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
975NULL) to -1. If those warnings are off, the computed value if well-defined (or
976the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
977is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
978next possible position in C<s> that could begin a non-malformed character.
979See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
980
981=cut
982*/
983
984UV
985Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
986{
987 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
988
989 assert(send > s);
990
991 /* Call the low level routine asking for checks */
992 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
993 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
994}
995
27d6c58a 996/* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
3986bb7c
KW
997 * there are no malformations in the input UTF-8 string C<s>. surrogates,
998 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
999
1000UV
1001Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1002{
3986bb7c
KW
1003 UV expectlen = UTF8SKIP(s);
1004 const U8* send = s + expectlen;
1005 UV uv = NATIVE_TO_UTF(*s);
1006
27d6c58a
KW
1007 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1008
3986bb7c
KW
1009 if (retlen) {
1010 *retlen = expectlen;
1011 }
1012
1013 /* An invariant is trivially returned */
1014 if (expectlen == 1) {
1015 return uv;
1016 }
1017
1018 /* Remove the leading bits that indicate the number of bytes, leaving just
1019 * the bits that are part of the value */
1020 uv &= UTF_START_MASK(expectlen);
1021
1022 /* Now, loop through the remaining bytes, accumulating each into the
1023 * working total as we go */
1024 for (++s; s < send; s++) {
1025 uv = UTF8_ACCUMULATE(uv, *s);
1026 }
1027
1028 return uv;
27d6c58a
KW
1029}
1030
ec5f19d0 1031/*
87cea99e 1032=for apidoc utf8_to_uvuni
9041c2e3 1033
977c1d31
KW
1034DEPRECATED!
1035
9041c2e3 1036Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1037which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1038length, in bytes, of that character.
1039
39e518fd
KW
1040This function should only be used when the returned UV is considered
1041an index into the Unicode semantic tables (e.g. swashes).
1042
4b88fb76 1043Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1044malformed input could cause reading beyond the end of the input buffer, which
1045is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
9041c2e3 1046
524080c4
KW
1047If C<s> points to one of the detected malformations, and UTF8 warnings are
1048enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1049NULL) to -1. If those warnings are off, the computed value if well-defined (or
1050the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1051is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1052next possible position in C<s> that could begin a non-malformed character.
1053See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1054
1055=cut
1056*/
1057
1058UV
7fc63493 1059Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1060{
7918f24d
NC
1061 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1062
4b88fb76 1063 return valid_utf8_to_uvuni(s, retlen);
8e84507e
NIS
1064}
1065
b76347f2 1066/*
87cea99e 1067=for apidoc utf8_length
b76347f2
JH
1068
1069Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1070Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1071up past C<e>, croaks.
b76347f2
JH
1072
1073=cut
1074*/
1075
1076STRLEN
35a4481c 1077Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1078{
97aff369 1079 dVAR;
b76347f2
JH
1080 STRLEN len = 0;
1081
7918f24d
NC
1082 PERL_ARGS_ASSERT_UTF8_LENGTH;
1083
8850bf83
JH
1084 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1085 * the bitops (especially ~) can create illegal UTF-8.
1086 * In other words: in Perl UTF-8 is not just for Unicode. */
1087
a3b680e6
AL
1088 if (e < s)
1089 goto warn_and_return;
b76347f2 1090 while (s < e) {
8e91ec7f
AV
1091 if (!UTF8_IS_INVARIANT(*s))
1092 s += UTF8SKIP(s);
1093 else
1094 s++;
1095 len++;
1096 }
1097
1098 if (e != s) {
1099 len--;
1100 warn_and_return:
9b387841
NC
1101 if (PL_op)
1102 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1103 "%s in %s", unees, OP_DESC(PL_op));
1104 else
61a12c31 1105 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1106 }
1107
1108 return len;
1109}
1110
b06226ff 1111/*
87cea99e 1112=for apidoc utf8_distance
b06226ff 1113
1e54db1a 1114Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1115and C<b>.
1116
1117WARNING: use only if you *know* that the pointers point inside the
1118same UTF-8 buffer.
1119
37607a96
PK
1120=cut
1121*/
a0ed51b3 1122
02eb7b47 1123IV
35a4481c 1124Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1125{
7918f24d
NC
1126 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1127
bf1665bc 1128 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1129}
1130
b06226ff 1131/*
87cea99e 1132=for apidoc utf8_hop
b06226ff 1133
8850bf83
JH
1134Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1135forward or backward.
b06226ff
JH
1136
1137WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1138the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1139on the first byte of character or just after the last byte of a character.
b06226ff 1140
37607a96
PK
1141=cut
1142*/
a0ed51b3
LW
1143
1144U8 *
4373e329 1145Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1146{
7918f24d
NC
1147 PERL_ARGS_ASSERT_UTF8_HOP;
1148
96a5add6 1149 PERL_UNUSED_CONTEXT;
8850bf83
JH
1150 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1151 * the bitops (especially ~) can create illegal UTF-8.
1152 * In other words: in Perl UTF-8 is not just for Unicode. */
1153
a0ed51b3
LW
1154 if (off >= 0) {
1155 while (off--)
1156 s += UTF8SKIP(s);
1157 }
1158 else {
1159 while (off++) {
1160 s--;
8850bf83
JH
1161 while (UTF8_IS_CONTINUATION(*s))
1162 s--;
a0ed51b3
LW
1163 }
1164 }
4373e329 1165 return (U8 *)s;
a0ed51b3
LW
1166}
1167
6940069f 1168/*
fed3ba5d
NC
1169=for apidoc bytes_cmp_utf8
1170
a1433954
KW
1171Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1172sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1173equal, -1 or -2 if the first string is less than the second string, +1 or +2
1174if the first string is greater than the second string.
1175
1176-1 or +1 is returned if the shorter string was identical to the start of the
1177longer string. -2 or +2 is returned if the was a difference between characters
1178within the strings.
1179
1180=cut
1181*/
1182
1183int
1184Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1185{
1186 const U8 *const bend = b + blen;
1187 const U8 *const uend = u + ulen;
1188
1189 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1190
1191 PERL_UNUSED_CONTEXT;
1192
1193 while (b < bend && u < uend) {
1194 U8 c = *u++;
1195 if (!UTF8_IS_INVARIANT(c)) {
1196 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1197 if (u < uend) {
1198 U8 c1 = *u++;
1199 if (UTF8_IS_CONTINUATION(c1)) {
356979f4 1200 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
fed3ba5d
NC
1201 } else {
1202 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1203 "Malformed UTF-8 character "
1204 "(unexpected non-continuation byte 0x%02x"
1205 ", immediately after start byte 0x%02x)"
1206 /* Dear diag.t, it's in the pod. */
1207 "%s%s", c1, c,
1208 PL_op ? " in " : "",
1209 PL_op ? OP_DESC(PL_op) : "");
1210 return -2;
1211 }
1212 } else {
1213 if (PL_op)
1214 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1215 "%s in %s", unees, OP_DESC(PL_op));
1216 else
61a12c31 1217 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1218 return -2; /* Really want to return undef :-) */
1219 }
1220 } else {
1221 return -2;
1222 }
1223 }
1224 if (*b != c) {
1225 return *b < c ? -2 : +2;
1226 }
1227 ++b;
1228 }
1229
1230 if (b == bend && u == uend)
1231 return 0;
1232
1233 return b < bend ? +1 : -1;
1234}
1235
1236/*
87cea99e 1237=for apidoc utf8_to_bytes
6940069f 1238
2bbc8d55 1239Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1240Unlike L</bytes_to_utf8>, this over-writes the original string, and
1241updates C<len> to contain the new length.
67e989fb 1242Returns zero on failure, setting C<len> to -1.
6940069f 1243
a1433954 1244If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1245
6940069f
GS
1246=cut
1247*/
1248
1249U8 *
37607a96 1250Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1251{
d4c19fe8
AL
1252 U8 * const save = s;
1253 U8 * const send = s + *len;
6940069f 1254 U8 *d;
246fae53 1255
7918f24d
NC
1256 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1257
1e54db1a 1258 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1259 while (s < send) {
dcad2880
JH
1260 U8 c = *s++;
1261
1d72bdf6
NIS
1262 if (!UTF8_IS_INVARIANT(c) &&
1263 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1264 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
10edeb5d 1265 *len = ((STRLEN) -1);
dcad2880
JH
1266 return 0;
1267 }
246fae53 1268 }
dcad2880
JH
1269
1270 d = s = save;
6940069f 1271 while (s < send) {
ed646e6e 1272 STRLEN ulen;
4b88fb76 1273 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
ed646e6e 1274 s += ulen;
6940069f
GS
1275 }
1276 *d = '\0';
246fae53 1277 *len = d - save;
6940069f
GS
1278 return save;
1279}
1280
1281/*
87cea99e 1282=for apidoc bytes_from_utf8
f9a63242 1283
2bbc8d55 1284Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1285Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1286the newly-created string, and updates C<len> to contain the new
1287length. Returns the original string if no conversion occurs, C<len>
1288is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
12890 if C<s> is converted or consisted entirely of characters that are invariant
1290in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1291
37607a96
PK
1292=cut
1293*/
f9a63242
JH
1294
1295U8 *
e1ec3a88 1296Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1297{
f9a63242 1298 U8 *d;
e1ec3a88
AL
1299 const U8 *start = s;
1300 const U8 *send;
f9a63242
JH
1301 I32 count = 0;
1302
7918f24d
NC
1303 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1304
96a5add6 1305 PERL_UNUSED_CONTEXT;
f9a63242 1306 if (!*is_utf8)
73d840c0 1307 return (U8 *)start;
f9a63242 1308
1e54db1a 1309 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1310 for (send = s + *len; s < send;) {
e1ec3a88 1311 U8 c = *s++;
1d72bdf6 1312 if (!UTF8_IS_INVARIANT(c)) {
db42d148
NIS
1313 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1314 (c = *s++) && UTF8_IS_CONTINUATION(c))
1315 count++;
1316 else
73d840c0 1317 return (U8 *)start;
db42d148 1318 }
f9a63242
JH
1319 }
1320
35da51f7 1321 *is_utf8 = FALSE;
f9a63242 1322
212542aa 1323 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1324 s = start; start = d;
f9a63242
JH
1325 while (s < send) {
1326 U8 c = *s++;
c4d5f83a
NIS
1327 if (!UTF8_IS_INVARIANT(c)) {
1328 /* Then it is two-byte encoded */
356979f4 1329 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
c4d5f83a
NIS
1330 }
1331 *d++ = c;
f9a63242
JH
1332 }
1333 *d = '\0';
1334 *len = d - start;
73d840c0 1335 return (U8 *)start;
f9a63242
JH
1336}
1337
1338/*
87cea99e 1339=for apidoc bytes_to_utf8
6940069f 1340
ff97e5cf
KW
1341Converts a string C<s> of length C<len> bytes from the native encoding into
1342UTF-8.
6662521e 1343Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1344reflect the new length in bytes.
6940069f 1345
2bbc8d55
SP
1346A NUL character will be written after the end of the string.
1347
1348If you want to convert to UTF-8 from encodings other than
1349the native (Latin1 or EBCDIC),
a1433954 1350see L</sv_recode_to_utf8>().
c9ada85f 1351
497711e7 1352=cut
6940069f
GS
1353*/
1354
c682ebef
FC
1355/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1356 likewise need duplication. */
1357
6940069f 1358U8*
35a4481c 1359Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1360{
35a4481c 1361 const U8 * const send = s + (*len);
6940069f
GS
1362 U8 *d;
1363 U8 *dst;
7918f24d
NC
1364
1365 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1366 PERL_UNUSED_CONTEXT;
6940069f 1367
212542aa 1368 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1369 dst = d;
1370
1371 while (s < send) {
35a4481c 1372 const UV uv = NATIVE_TO_ASCII(*s++);
c4d5f83a 1373 if (UNI_IS_INVARIANT(uv))
eb160463 1374 *d++ = (U8)UTF_TO_NATIVE(uv);
6940069f 1375 else {
eb160463
GS
1376 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1377 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
6940069f
GS
1378 }
1379 }
1380 *d = '\0';
6662521e 1381 *len = d-dst;
6940069f
GS
1382 return dst;
1383}
1384
a0ed51b3 1385/*
dea0fc0b 1386 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1387 *
1388 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1389 * We optimize for native, for obvious reasons. */
1390
1391U8*
dea0fc0b 1392Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1393{
dea0fc0b
JH
1394 U8* pend;
1395 U8* dstart = d;
1396
7918f24d
NC
1397 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1398
dea0fc0b 1399 if (bytelen & 1)
f5992bc4 1400 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1401
1402 pend = p + bytelen;
1403
a0ed51b3 1404 while (p < pend) {
dea0fc0b
JH
1405 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1406 p += 2;
a0ed51b3 1407 if (uv < 0x80) {
e294cc5d
JH
1408#ifdef EBCDIC
1409 *d++ = UNI_TO_NATIVE(uv);
1410#else
eb160463 1411 *d++ = (U8)uv;
e294cc5d 1412#endif
a0ed51b3
LW
1413 continue;
1414 }
1415 if (uv < 0x800) {
eb160463
GS
1416 *d++ = (U8)(( uv >> 6) | 0xc0);
1417 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1418 continue;
1419 }
52b9aa85 1420 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
01ea242b 1421 if (p >= pend) {
dea0fc0b 1422 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1423 } else {
1424 UV low = (p[0] << 8) + p[1];
1425 p += 2;
52b9aa85 1426 if (low < 0xdc00 || low > 0xdfff)
01ea242b
NC
1427 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1428 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1429 }
dbde1951
NC
1430 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1431 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3
LW
1432 }
1433 if (uv < 0x10000) {
eb160463
GS
1434 *d++ = (U8)(( uv >> 12) | 0xe0);
1435 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1436 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1437 continue;
1438 }
1439 else {
eb160463
GS
1440 *d++ = (U8)(( uv >> 18) | 0xf0);
1441 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1442 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1443 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1444 continue;
1445 }
1446 }
dea0fc0b 1447 *newlen = d - dstart;
a0ed51b3
LW
1448 return d;
1449}
1450
1451/* Note: this one is slightly destructive of the source. */
1452
1453U8*
dea0fc0b 1454Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1455{
1456 U8* s = (U8*)p;
d4c19fe8 1457 U8* const send = s + bytelen;
7918f24d
NC
1458
1459 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1460
e0ea5e2d
NC
1461 if (bytelen & 1)
1462 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1463 (UV)bytelen);
1464
a0ed51b3 1465 while (s < send) {
d4c19fe8 1466 const U8 tmp = s[0];
a0ed51b3
LW
1467 s[0] = s[1];
1468 s[1] = tmp;
1469 s += 2;
1470 }
dea0fc0b 1471 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1472}
1473
c3fd2246
KW
1474/* for now these are all defined (inefficiently) in terms of the utf8 versions.
1475 * Note that the macros in handy.h that call these short-circuit calling them
1476 * for Latin-1 range inputs */
a0ed51b3
LW
1477
1478bool
84afefe6 1479Perl_is_uni_alnum(pTHX_ UV c)
a0ed51b3 1480{
89ebb4a3 1481 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1482 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1483 return is_utf8_alnum(tmpbuf);
1484}
1485
1486bool
84afefe6 1487Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1488{
89ebb4a3 1489 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1490 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1491 return is_utf8_idfirst(tmpbuf);
1492}
1493
1494bool
84afefe6 1495Perl_is_uni_alpha(pTHX_ UV c)
a0ed51b3 1496{
89ebb4a3 1497 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1498 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1499 return is_utf8_alpha(tmpbuf);
1500}
1501
1502bool
84afefe6 1503Perl_is_uni_ascii(pTHX_ UV c)
4d61ec05 1504{
bc39fe24 1505 return isASCII(c);
4d61ec05
GS
1506}
1507
1508bool
84afefe6 1509Perl_is_uni_space(pTHX_ UV c)
a0ed51b3 1510{
89ebb4a3 1511 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1512 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1513 return is_utf8_space(tmpbuf);
1514}
1515
1516bool
84afefe6 1517Perl_is_uni_digit(pTHX_ UV c)
a0ed51b3 1518{
89ebb4a3 1519 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1520 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1521 return is_utf8_digit(tmpbuf);
1522}
1523
1524bool
84afefe6 1525Perl_is_uni_upper(pTHX_ UV c)
a0ed51b3 1526{
89ebb4a3 1527 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1528 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1529 return is_utf8_upper(tmpbuf);
1530}
1531
1532bool
84afefe6 1533Perl_is_uni_lower(pTHX_ UV c)
a0ed51b3 1534{
89ebb4a3 1535 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1536 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1537 return is_utf8_lower(tmpbuf);
1538}
1539
1540bool
84afefe6 1541Perl_is_uni_cntrl(pTHX_ UV c)
b8c5462f 1542{
7b952154 1543 return isCNTRL_L1(c);
b8c5462f
JH
1544}
1545
1546bool
84afefe6 1547Perl_is_uni_graph(pTHX_ UV c)
b8c5462f 1548{
89ebb4a3 1549 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1550 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1551 return is_utf8_graph(tmpbuf);
1552}
1553
1554bool
84afefe6 1555Perl_is_uni_print(pTHX_ UV c)
a0ed51b3 1556{
89ebb4a3 1557 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1558 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1559 return is_utf8_print(tmpbuf);
1560}
1561
b8c5462f 1562bool
84afefe6 1563Perl_is_uni_punct(pTHX_ UV c)
b8c5462f 1564{
89ebb4a3 1565 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1566 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1567 return is_utf8_punct(tmpbuf);
1568}
1569
4d61ec05 1570bool
84afefe6 1571Perl_is_uni_xdigit(pTHX_ UV c)
4d61ec05 1572{
89ebb4a3 1573 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
230880c1 1574 uvchr_to_utf8(tmpbuf, c);
4d61ec05
GS
1575 return is_utf8_xdigit(tmpbuf);
1576}
1577
3a4c58c9
KW
1578UV
1579Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1580{
1581 /* We have the latin1-range values compiled into the core, so just use
1582 * those, converting the result to utf8. The only difference between upper
1583 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1584 * either "SS" or "Ss". Which one to use is passed into the routine in
1585 * 'S_or_s' to avoid a test */
1586
1587 UV converted = toUPPER_LATIN1_MOD(c);
1588
1589 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1590
1591 assert(S_or_s == 'S' || S_or_s == 's');
1592
1593 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1594 characters in this range */
1595 *p = (U8) converted;
1596 *lenp = 1;
1597 return converted;
1598 }
1599
1600 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1601 * which it maps to one of them, so as to only have to have one check for
1602 * it in the main case */
1603 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1604 switch (c) {
1605 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1606 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1607 break;
1608 case MICRO_SIGN:
1609 converted = GREEK_CAPITAL_LETTER_MU;
1610 break;
1611 case LATIN_SMALL_LETTER_SHARP_S:
1612 *(p)++ = 'S';
1613 *p = S_or_s;
1614 *lenp = 2;
1615 return 'S';
1616 default:
1617 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
1618 /* NOTREACHED */
1619 }
1620 }
1621
1622 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1623 *p = UTF8_TWO_BYTE_LO(converted);
1624 *lenp = 2;
1625
1626 return converted;
1627}
1628
50bda2c3
KW
1629/* Call the function to convert a UTF-8 encoded character to the specified case.
1630 * Note that there may be more than one character in the result.
1631 * INP is a pointer to the first byte of the input character
1632 * OUTP will be set to the first byte of the string of changed characters. It
1633 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1634 * LENP will be set to the length in bytes of the string of changed characters
1635 *
1636 * The functions return the ordinal of the first character in the string of OUTP */
f90a9a02
KW
1637#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1638#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1639#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
50bda2c3
KW
1640
1641/* This additionally has the input parameter SPECIALS, which if non-zero will
1642 * cause this to use the SPECIALS hash for folding (meaning get full case
1643 * folding); otherwise, when zero, this implies a simple case fold */
f90a9a02 1644#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 1645
84afefe6
JH
1646UV
1647Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1648{
3a4c58c9
KW
1649 dVAR;
1650
a1433954
KW
1651 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1652 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1653 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1654 * the changed version may be longer than the original character.
1655 *
1656 * The ordinal of the first character of the changed version is returned
1657 * (but note, as explained above, that there may be more.) */
1658
7918f24d
NC
1659 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1660
3a4c58c9
KW
1661 if (c < 256) {
1662 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1663 }
1664
0ebc6274 1665 uvchr_to_utf8(p, c);
3a4c58c9 1666 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1667}
1668
84afefe6
JH
1669UV
1670Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1671{
3a4c58c9
KW
1672 dVAR;
1673
7918f24d
NC
1674 PERL_ARGS_ASSERT_TO_UNI_TITLE;
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_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1682}
1683
afc16117
KW
1684STATIC U8
1685S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1686{
1687 /* We have the latin1-range values compiled into the core, so just use
1688 * those, converting the result to utf8. Since the result is always just
a1433954 1689 * one character, we allow <p> to be NULL */
afc16117
KW
1690
1691 U8 converted = toLOWER_LATIN1(c);
1692
1693 if (p != NULL) {
1694 if (UNI_IS_INVARIANT(converted)) {
1695 *p = converted;
1696 *lenp = 1;
1697 }
1698 else {
1699 *p = UTF8_TWO_BYTE_HI(converted);
1700 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1701 *lenp = 2;
1702 }
1703 }
1704 return converted;
1705}
1706
84afefe6
JH
1707UV
1708Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1709{
968c5e6a
KW
1710 dVAR;
1711
7918f24d
NC
1712 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1713
afc16117
KW
1714 if (c < 256) {
1715 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1716 }
1717
afc16117 1718 uvchr_to_utf8(p, c);
968c5e6a 1719 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1720}
1721
84afefe6 1722UV
f673fad4 1723Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
a1dde8de 1724{
a1433954 1725 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
f673fad4
KW
1726 * folding */
1727
a1dde8de
KW
1728 UV converted;
1729
1730 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1731
1732 if (c == MICRO_SIGN) {
1733 converted = GREEK_SMALL_LETTER_MU;
1734 }
1735 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1736 *(p)++ = 's';
1737 *p = 's';
1738 *lenp = 2;
1739 return 's';
1740 }
1741 else { /* In this range the fold of all other characters is their lower
1742 case */
1743 converted = toLOWER_LATIN1(c);
1744 }
1745
1746 if (UNI_IS_INVARIANT(converted)) {
1747 *p = (U8) converted;
1748 *lenp = 1;
1749 }
1750 else {
1751 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1752 *p = UTF8_TWO_BYTE_LO(converted);
1753 *lenp = 2;
1754 }
1755
1756 return converted;
1757}
1758
1759UV
f673fad4 1760Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const bool flags)
84afefe6 1761{
4b593389
KW
1762
1763 /* Not currently externally documented, and subject to change, <flags> is
f673fad4 1764 * TRUE iff full folding is to be used */
4b593389 1765
36bb2ab6 1766 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1767
a1dde8de
KW
1768 if (c < 256) {
1769 return _to_fold_latin1((U8) c, p, lenp, flags);
1770 }
1771
0ebc6274 1772 uvchr_to_utf8(p, c);
a1dde8de 1773 return CALL_FOLD_CASE(p, p, lenp, flags);
84afefe6
JH
1774}
1775
ea317ccb
KW
1776/* for now these all assume no locale info available for Unicode > 255; and
1777 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1778 * called instead, so that these don't get called for < 255 */
a0ed51b3
LW
1779
1780bool
84afefe6 1781Perl_is_uni_alnum_lc(pTHX_ UV c)
a0ed51b3
LW
1782{
1783 return is_uni_alnum(c); /* XXX no locale support yet */
1784}
1785
1786bool
84afefe6 1787Perl_is_uni_idfirst_lc(pTHX_ UV c)
a0ed51b3
LW
1788{
1789 return is_uni_idfirst(c); /* XXX no locale support yet */
1790}
1791
1792bool
84afefe6 1793Perl_is_uni_alpha_lc(pTHX_ UV c)
a0ed51b3
LW
1794{
1795 return is_uni_alpha(c); /* XXX no locale support yet */
1796}
1797
1798bool
84afefe6 1799Perl_is_uni_ascii_lc(pTHX_ UV c)
4d61ec05
GS
1800{
1801 return is_uni_ascii(c); /* XXX no locale support yet */
1802}
1803
1804bool
84afefe6 1805Perl_is_uni_space_lc(pTHX_ UV c)
a0ed51b3
LW
1806{
1807 return is_uni_space(c); /* XXX no locale support yet */
1808}
1809
1810bool
84afefe6 1811Perl_is_uni_digit_lc(pTHX_ UV c)
a0ed51b3
LW
1812{
1813 return is_uni_digit(c); /* XXX no locale support yet */
1814}
1815
1816bool
84afefe6 1817Perl_is_uni_upper_lc(pTHX_ UV c)
a0ed51b3
LW
1818{
1819 return is_uni_upper(c); /* XXX no locale support yet */
1820}
1821
1822bool
84afefe6 1823Perl_is_uni_lower_lc(pTHX_ UV c)
a0ed51b3
LW
1824{
1825 return is_uni_lower(c); /* XXX no locale support yet */
1826}
1827
1828bool
84afefe6 1829Perl_is_uni_cntrl_lc(pTHX_ UV c)
b8c5462f
JH
1830{
1831 return is_uni_cntrl(c); /* XXX no locale support yet */
1832}
1833
1834bool
84afefe6 1835Perl_is_uni_graph_lc(pTHX_ UV c)
b8c5462f
JH
1836{
1837 return is_uni_graph(c); /* XXX no locale support yet */
1838}
1839
1840bool
84afefe6 1841Perl_is_uni_print_lc(pTHX_ UV c)
a0ed51b3
LW
1842{
1843 return is_uni_print(c); /* XXX no locale support yet */
1844}
1845
b8c5462f 1846bool
84afefe6 1847Perl_is_uni_punct_lc(pTHX_ UV c)
b8c5462f
JH
1848{
1849 return is_uni_punct(c); /* XXX no locale support yet */
1850}
1851
4d61ec05 1852bool
84afefe6 1853Perl_is_uni_xdigit_lc(pTHX_ UV c)
4d61ec05
GS
1854{
1855 return is_uni_xdigit(c); /* XXX no locale support yet */
1856}
1857
b7ac61fa
JH
1858U32
1859Perl_to_uni_upper_lc(pTHX_ U32 c)
1860{
ee099d14
JH
1861 /* XXX returns only the first character -- do not use XXX */
1862 /* XXX no locale support yet */
1863 STRLEN len;
89ebb4a3 1864 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1865 return (U32)to_uni_upper(c, tmpbuf, &len);
b7ac61fa
JH
1866}
1867
1868U32
1869Perl_to_uni_title_lc(pTHX_ U32 c)
1870{
ee099d14
JH
1871 /* XXX returns only the first character XXX -- do not use XXX */
1872 /* XXX no locale support yet */
1873 STRLEN len;
89ebb4a3 1874 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1875 return (U32)to_uni_title(c, tmpbuf, &len);
b7ac61fa
JH
1876}
1877
1878U32
1879Perl_to_uni_lower_lc(pTHX_ U32 c)
1880{
ee099d14
JH
1881 /* XXX returns only the first character -- do not use XXX */
1882 /* XXX no locale support yet */
1883 STRLEN len;
89ebb4a3 1884 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1885 return (U32)to_uni_lower(c, tmpbuf, &len);
b7ac61fa
JH
1886}
1887
7452cf6a 1888static bool
5141f98e 1889S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
1890 const char *const swashname)
1891{
ea317ccb
KW
1892 /* returns a boolean giving whether or not the UTF8-encoded character that
1893 * starts at <p> is in the swash indicated by <swashname>. <swash>
1894 * contains a pointer to where the swash indicated by <swashname>
1895 * is to be stored; which this routine will do, so that future calls will
1896 * look at <*swash> and only generate a swash if it is not null
1897 *
1898 * Note that it is assumed that the buffer length of <p> is enough to
1899 * contain all the bytes that comprise the character. Thus, <*p> should
1900 * have been checked before this call for mal-formedness enough to assure
1901 * that. */
1902
97aff369 1903 dVAR;
7918f24d
NC
1904
1905 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1906
492a624f
KW
1907 /* The API should have included a length for the UTF-8 character in <p>,
1908 * but it doesn't. We therefor assume that p has been validated at least
1909 * as far as there being enough bytes available in it to accommodate the
1910 * character without reading beyond the end, and pass that number on to the
1911 * validating routine */
1912 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
bde6a22d
NC
1913 return FALSE;
1914 if (!*swash)
711a919c 1915 *swash = swash_init("utf8", swashname, &PL_sv_undef, 1, 0);
bde6a22d
NC
1916 return swash_fetch(*swash, p, TRUE) != 0;
1917}
1918
1919bool
7fc63493 1920Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 1921{
97aff369 1922 dVAR;
7918f24d
NC
1923
1924 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1925
671c33bf
NC
1926 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1927 * descendant of isalnum(3), in other words, it doesn't
1928 * contain the '_'. --jhi */
d4c19fe8 1929 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
a0ed51b3
LW
1930}
1931
1932bool
7fc63493 1933Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1934{
97aff369 1935 dVAR;
7918f24d
NC
1936
1937 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1938
82686b01
JH
1939 if (*p == '_')
1940 return TRUE;
bde6a22d 1941 /* is_utf8_idstart would be more logical. */
d4c19fe8 1942 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
82686b01
JH
1943}
1944
1945bool
c11ff943
KW
1946Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1947{
1948 dVAR;
1949
1950 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1951
1952 if (*p == '_')
1953 return TRUE;
1954 /* is_utf8_idstart would be more logical. */
1955 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
1956}
1957
1958bool
b6912c02
KW
1959Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
1960{
1961 dVAR;
1962
1963 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
1964
1965 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
1966}
1967
1968bool
7fc63493 1969Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1970{
97aff369 1971 dVAR;
7918f24d
NC
1972
1973 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
1974
d4c19fe8 1975 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
1976}
1977
1978bool
c11ff943
KW
1979Perl_is_utf8_xidcont(pTHX_ const U8 *p)
1980{
1981 dVAR;
1982
1983 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
1984
c11ff943
KW
1985 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
1986}
1987
1988bool
7fc63493 1989Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 1990{
97aff369 1991 dVAR;
7918f24d
NC
1992
1993 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
1994
d4c19fe8 1995 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
a0ed51b3
LW
1996}
1997
1998bool
7fc63493 1999Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2000{
97aff369 2001 dVAR;
7918f24d
NC
2002
2003 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2004
bc39fe24
KW
2005 /* ASCII characters are the same whether in utf8 or not. So the macro
2006 * works on both utf8 and non-utf8 representations. */
2007 return isASCII(*p);
b8c5462f
JH
2008}
2009
2010bool
7fc63493 2011Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2012{
97aff369 2013 dVAR;
7918f24d
NC
2014
2015 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2016
a34094a9 2017 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
a0ed51b3
LW
2018}
2019
2020bool
d1eb3177
YO
2021Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2022{
2023 dVAR;
2024
2025 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2026
c4428693
KW
2027 /* Only true if is an ASCII space-like character, and ASCII is invariant
2028 * under utf8, so can just use the macro */
2029 return isSPACE_A(*p);
d1eb3177
YO
2030}
2031
2032bool
2033Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2034{
2035 dVAR;
2036
2037 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2038
c4428693
KW
2039 /* Only true if is an ASCII word character, and ASCII is invariant
2040 * under utf8, so can just use the macro */
2041 return isWORDCHAR_A(*p);
d1eb3177
YO
2042}
2043
2044bool
7fc63493 2045Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2046{
97aff369 2047 dVAR;
7918f24d
NC
2048
2049 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2050
d4c19fe8 2051 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
a0ed51b3
LW
2052}
2053
2054bool
d1eb3177
YO
2055Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2056{
2057 dVAR;
2058
2059 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2060
c4428693
KW
2061 /* Only true if is an ASCII digit character, and ASCII is invariant
2062 * under utf8, so can just use the macro */
2063 return isDIGIT_A(*p);
d1eb3177
YO
2064}
2065
2066bool
7fc63493 2067Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2068{
97aff369 2069 dVAR;
7918f24d
NC
2070
2071 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2072
d4c19fe8 2073 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
a0ed51b3
LW
2074}
2075
2076bool
7fc63493 2077Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2078{
97aff369 2079 dVAR;
7918f24d
NC
2080
2081 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2082
d4c19fe8 2083 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
a0ed51b3
LW
2084}
2085
2086bool
7fc63493 2087Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2088{
97aff369 2089 dVAR;
7918f24d
NC
2090
2091 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2092
7b952154
KW
2093 if (isASCII(*p)) {
2094 return isCNTRL_A(*p);
2095 }
2096
2097 /* All controls are in Latin1 */
2098 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2099 return 0;
2100 }
2101 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
b8c5462f
JH
2102}
2103
2104bool
7fc63493 2105Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2106{
97aff369 2107 dVAR;
7918f24d
NC
2108
2109 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2110
d4c19fe8 2111 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
b8c5462f
JH
2112}
2113
2114bool
7fc63493 2115Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2116{
97aff369 2117 dVAR;
7918f24d
NC
2118
2119 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2120
d4c19fe8 2121 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
a0ed51b3
LW
2122}
2123
2124bool
7fc63493 2125Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2126{
97aff369 2127 dVAR;
7918f24d
NC
2128
2129 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2130
d4c19fe8 2131 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
b8c5462f
JH
2132}
2133
2134bool
7fc63493 2135Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2136{
97aff369 2137 dVAR;
7918f24d
NC
2138
2139 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2140
d1eb3177 2141 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
b8c5462f
JH
2142}
2143
2144bool
7fc63493 2145Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2146{
97aff369 2147 dVAR;
7918f24d
NC
2148
2149 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2150
d4c19fe8 2151 return is_utf8_common(p, &PL_utf8_mark, "IsM");
a0ed51b3
LW
2152}
2153
37e2e78e
KW
2154bool
2155Perl_is_utf8_X_begin(pTHX_ const U8 *p)
2156{
2157 dVAR;
2158
2159 PERL_ARGS_ASSERT_IS_UTF8_X_BEGIN;
2160
2161 return is_utf8_common(p, &PL_utf8_X_begin, "_X_Begin");
2162}
2163
2164bool
2165Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2166{
2167 dVAR;
2168
2169 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2170
2171 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2172}
2173
2174bool
2175Perl_is_utf8_X_prepend(pTHX_ const U8 *p)
2176{
2177 dVAR;
2178
2179 PERL_ARGS_ASSERT_IS_UTF8_X_PREPEND;
2180
2181 return is_utf8_common(p, &PL_utf8_X_prepend, "GCB=Prepend");
2182}
2183
2184bool
2185Perl_is_utf8_X_non_hangul(pTHX_ const U8 *p)
2186{
2187 dVAR;
2188
2189 PERL_ARGS_ASSERT_IS_UTF8_X_NON_HANGUL;
2190
2191 return is_utf8_common(p, &PL_utf8_X_non_hangul, "HST=Not_Applicable");
2192}
2193
2194bool
2195Perl_is_utf8_X_L(pTHX_ const U8 *p)
2196{
2197 dVAR;
2198
2199 PERL_ARGS_ASSERT_IS_UTF8_X_L;
2200
2201 return is_utf8_common(p, &PL_utf8_X_L, "GCB=L");
2202}
2203
2204bool
2205Perl_is_utf8_X_LV(pTHX_ const U8 *p)
2206{
2207 dVAR;
2208
2209 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
2210
2211 return is_utf8_common(p, &PL_utf8_X_LV, "GCB=LV");
2212}
2213
2214bool
2215Perl_is_utf8_X_LVT(pTHX_ const U8 *p)
2216{
2217 dVAR;
2218
2219 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
2220
2221 return is_utf8_common(p, &PL_utf8_X_LVT, "GCB=LVT");
2222}
2223
2224bool
2225Perl_is_utf8_X_T(pTHX_ const U8 *p)
2226{
2227 dVAR;
2228
2229 PERL_ARGS_ASSERT_IS_UTF8_X_T;
2230
2231 return is_utf8_common(p, &PL_utf8_X_T, "GCB=T");
2232}
2233
2234bool
2235Perl_is_utf8_X_V(pTHX_ const U8 *p)
2236{
2237 dVAR;
2238
2239 PERL_ARGS_ASSERT_IS_UTF8_X_V;
2240
2241 return is_utf8_common(p, &PL_utf8_X_V, "GCB=V");
2242}
2243
2244bool
2245Perl_is_utf8_X_LV_LVT_V(pTHX_ const U8 *p)
2246{
2247 dVAR;
2248
2249 PERL_ARGS_ASSERT_IS_UTF8_X_LV_LVT_V;
2250
2251 return is_utf8_common(p, &PL_utf8_X_LV_LVT_V, "_X_LV_LVT_V");
2252}
2253
2e2b2571
KW
2254bool
2255Perl__is_utf8_quotemeta(pTHX_ const U8 *p)
2256{
2257 /* For exclusive use of pp_quotemeta() */
2258
2259 dVAR;
2260
2261 PERL_ARGS_ASSERT__IS_UTF8_QUOTEMETA;
2262
2263 return is_utf8_common(p, &PL_utf8_quotemeta, "_Perl_Quotemeta");
2264}
2265
6b5c0936 2266/*
87cea99e 2267=for apidoc to_utf8_case
6b5c0936 2268
a1433954
KW
2269The C<p> contains the pointer to the UTF-8 string encoding
2270the character that is being converted. This routine assumes that the character
2271at C<p> is well-formed.
6b5c0936 2272
a1433954
KW
2273The C<ustrp> is a pointer to the character buffer to put the
2274conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2275of the result.
2276
a1433954 2277The C<swashp> is a pointer to the swash to use.
6b5c0936 2278
a1433954
KW
2279Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2280and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2281but not always, a multicharacter mapping), is tried first.
6b5c0936 2282
a1433954 2283The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2284hash %utf8::ToSpecLower. The access to the hash is through
2285Perl_to_utf8_case().
6b5c0936 2286
a1433954 2287The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2288%utf8::ToLower.
2289
2290=cut */
6b5c0936 2291
2104c8d9 2292UV
9a957fbc
AL
2293Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2294 SV **swashp, const char *normal, const char *special)
a0ed51b3 2295{
97aff369 2296 dVAR;
89ebb4a3 2297 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
0134edef 2298 STRLEN len = 0;
4b88fb76 2299 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
1feea2c7
JH
2300 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2301 * are necessary in EBCDIC, they are redundant no-ops
2302 * in ASCII-ish platforms, and hopefully optimized away. */
f54cb97a 2303 const UV uv1 = NATIVE_TO_UNI(uv0);
7918f24d
NC
2304
2305 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2306
9ae3ac1a
KW
2307 /* Note that swash_fetch() doesn't output warnings for these because it
2308 * assumes we will */
8457b38f 2309 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2310 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2311 if (ckWARN_d(WARN_SURROGATE)) {
2312 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2313 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2314 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2315 }
9ae3ac1a
KW
2316 }
2317 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2318 if (ckWARN_d(WARN_NON_UNICODE)) {
2319 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2320 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2321 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2322 }
9ae3ac1a
KW
2323 }
2324
2325 /* Note that non-characters are perfectly legal, so no warning should
2326 * be given */
2327 }
2328
1feea2c7 2329 uvuni_to_utf8(tmpbuf, uv1);
0134edef
JH
2330
2331 if (!*swashp) /* load on-demand */
2332 *swashp = swash_init("utf8", normal, &PL_sv_undef, 4, 0);
2333
a6f87d8c 2334 if (special) {
0134edef 2335 /* It might be "special" (sometimes, but not always,
2a37f04d 2336 * a multicharacter mapping) */
6673a63c 2337 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2338 SV **svp;
2339
35da51f7 2340 if (hv &&
b08cf34e
JH
2341 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2342 (*svp)) {
cfd0369c 2343 const char *s;
47654450 2344
cfd0369c 2345 s = SvPV_const(*svp, len);
47654450
JH
2346 if (len == 1)
2347 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2a37f04d 2348 else {
2f9475ad
JH
2349#ifdef EBCDIC
2350 /* If we have EBCDIC we need to remap the characters
2351 * since any characters in the low 256 are Unicode
2352 * code points, not EBCDIC. */
7cda7a3d 2353 U8 *t = (U8*)s, *tend = t + len, *d;
2f9475ad
JH
2354
2355 d = tmpbuf;
b08cf34e 2356 if (SvUTF8(*svp)) {
2f9475ad
JH
2357 STRLEN tlen = 0;
2358
2359 while (t < tend) {
6bd1c396 2360 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2f9475ad
JH
2361 if (tlen > 0) {
2362 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2363 t += tlen;
2364 }
2365 else
2366 break;
2367 }
2368 }
2369 else {
36fec512
JH
2370 while (t < tend) {
2371 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2372 t++;
2373 }
2f9475ad
JH
2374 }
2375 len = d - tmpbuf;
2376 Copy(tmpbuf, ustrp, len, U8);
2377#else
d2dcd0fb 2378 Copy(s, ustrp, len, U8);
2f9475ad 2379#endif
29e98929 2380 }
983ffd37 2381 }
0134edef
JH
2382 }
2383
2384 if (!len && *swashp) {
d4c19fe8
AL
2385 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE);
2386
0134edef
JH
2387 if (uv2) {
2388 /* It was "normal" (a single character mapping). */
d4c19fe8 2389 const UV uv3 = UNI_TO_NATIVE(uv2);
e9101d72 2390 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2a37f04d
JH
2391 }
2392 }
1feea2c7 2393
37e2e78e
KW
2394 if (!len) /* Neither: just copy. In other words, there was no mapping
2395 defined, which means that the code point maps to itself */
0134edef
JH
2396 len = uvchr_to_utf8(ustrp, uv0) - ustrp;
2397
2a37f04d
JH
2398 if (lenp)
2399 *lenp = len;
2400
6bd1c396 2401 return len ? valid_utf8_to_uvchr(ustrp, 0) : 0;
a0ed51b3
LW
2402}
2403
051a06d4
KW
2404STATIC UV
2405S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2406{
2407 /* This is called when changing the case of a utf8-encoded character above
2408 * the Latin1 range, and the operation is in locale. If the result
2409 * contains a character that crosses the 255/256 boundary, disallow the
2410 * change, and return the original code point. See L<perlfunc/lc> for why;
2411 *
a1433954
KW
2412 * p points to the original string whose case was changed; assumed
2413 * by this routine to be well-formed
051a06d4
KW
2414 * result the code point of the first character in the changed-case string
2415 * ustrp points to the changed-case string (<result> represents its first char)
2416 * lenp points to the length of <ustrp> */
2417
2418 UV original; /* To store the first code point of <p> */
2419
2420 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2421
2422 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2423
2424 /* We know immediately if the first character in the string crosses the
2425 * boundary, so can skip */
2426 if (result > 255) {
2427
2428 /* Look at every character in the result; if any cross the
2429 * boundary, the whole thing is disallowed */
2430 U8* s = ustrp + UTF8SKIP(ustrp);
2431 U8* e = ustrp + *lenp;
2432 while (s < e) {
2433 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2434 {
2435 goto bad_crossing;
2436 }
2437 s += UTF8SKIP(s);
2438 }
2439
2440 /* Here, no characters crossed, result is ok as-is */
2441 return result;
2442 }
2443
2444bad_crossing:
2445
2446 /* Failed, have to return the original */
4b88fb76 2447 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2448 Copy(p, ustrp, *lenp, char);
2449 return original;
2450}
2451
d3e79532 2452/*
87cea99e 2453=for apidoc to_utf8_upper
d3e79532 2454
a1433954
KW
2455Convert the UTF-8 encoded character at C<p> to its uppercase version and
2456store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
89ebb4a3
JH
2457that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2458the uppercase version may be longer than the original character.
d3e79532
JH
2459
2460The first character of the uppercased version is returned
2461(but note, as explained above, that there may be more.)
2462
a1433954
KW
2463The character at C<p> is assumed by this routine to be well-formed.
2464
d3e79532
JH
2465=cut */
2466
051a06d4
KW
2467/* Not currently externally documented, and subject to change:
2468 * <flags> is set iff locale semantics are to be used for code points < 256
2469 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2470 * were used in the calculation; otherwise unchanged. */
2471
2104c8d9 2472UV
051a06d4 2473Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2474{
97aff369 2475 dVAR;
7918f24d 2476
051a06d4
KW
2477 UV result;
2478
2479 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2480
3a4c58c9 2481 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2482 if (flags) {
2483 result = toUPPER_LC(*p);
2484 }
2485 else {
81c6c7ce 2486 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2487 }
3a4c58c9
KW
2488 }
2489 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2490 if (flags) {
2491 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2492 }
2493 else {
81c6c7ce
KW
2494 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2495 ustrp, lenp, 'S');
051a06d4
KW
2496 }
2497 }
2498 else { /* utf8, ord above 255 */
2499 result = CALL_UPPER_CASE(p, ustrp, lenp);
2500
2501 if (flags) {
2502 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2503 }
2504 return result;
2505 }
2506
2507 /* Here, used locale rules. Convert back to utf8 */
2508 if (UTF8_IS_INVARIANT(result)) {
2509 *ustrp = (U8) result;
2510 *lenp = 1;
2511 }
2512 else {
2513 *ustrp = UTF8_EIGHT_BIT_HI(result);
2514 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2515 *lenp = 2;
3a4c58c9
KW
2516 }
2517
051a06d4
KW
2518 if (tainted_ptr) {
2519 *tainted_ptr = TRUE;
2520 }
2521 return result;
983ffd37 2522}
a0ed51b3 2523
d3e79532 2524/*
87cea99e 2525=for apidoc to_utf8_title
d3e79532 2526
a1433954
KW
2527Convert the UTF-8 encoded character at C<p> to its titlecase version and
2528store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2529that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2530titlecase version may be longer than the original character.
d3e79532
JH
2531
2532The first character of the titlecased version is returned
2533(but note, as explained above, that there may be more.)
2534
a1433954
KW
2535The character at C<p> is assumed by this routine to be well-formed.
2536
d3e79532
JH
2537=cut */
2538
051a06d4
KW
2539/* Not currently externally documented, and subject to change:
2540 * <flags> is set iff locale semantics are to be used for code points < 256
2541 * Since titlecase is not defined in POSIX, uppercase is used instead
2542 * for these/
2543 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2544 * were used in the calculation; otherwise unchanged. */
2545
983ffd37 2546UV
051a06d4 2547Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2548{
97aff369 2549 dVAR;
7918f24d 2550
051a06d4
KW
2551 UV result;
2552
2553 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2554
3a4c58c9 2555 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2556 if (flags) {
2557 result = toUPPER_LC(*p);
2558 }
2559 else {
81c6c7ce 2560 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2561 }
3a4c58c9
KW
2562 }
2563 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2564 if (flags) {
2565 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2566 }
2567 else {
81c6c7ce
KW
2568 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2569 ustrp, lenp, 's');
051a06d4
KW
2570 }
2571 }
2572 else { /* utf8, ord above 255 */
2573 result = CALL_TITLE_CASE(p, ustrp, lenp);
2574
2575 if (flags) {
2576 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2577 }
2578 return result;
2579 }
2580
2581 /* Here, used locale rules. Convert back to utf8 */
2582 if (UTF8_IS_INVARIANT(result)) {
2583 *ustrp = (U8) result;
2584 *lenp = 1;
2585 }
2586 else {
2587 *ustrp = UTF8_EIGHT_BIT_HI(result);
2588 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2589 *lenp = 2;
3a4c58c9
KW
2590 }
2591
051a06d4
KW
2592 if (tainted_ptr) {
2593 *tainted_ptr = TRUE;
2594 }
2595 return result;
a0ed51b3
LW
2596}
2597
d3e79532 2598/*
87cea99e 2599=for apidoc to_utf8_lower
d3e79532 2600
a1433954
KW
2601Convert the UTF-8 encoded character at C<p> to its lowercase version and
2602store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2603that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2604lowercase version may be longer than the original character.
d3e79532
JH
2605
2606The first character of the lowercased version is returned
2607(but note, as explained above, that there may be more.)
2608
a1433954
KW
2609The character at C<p> is assumed by this routine to be well-formed.
2610
d3e79532
JH
2611=cut */
2612
051a06d4
KW
2613/* Not currently externally documented, and subject to change:
2614 * <flags> is set iff locale semantics are to be used for code points < 256
2615 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2616 * were used in the calculation; otherwise unchanged. */
2617
2104c8d9 2618UV
051a06d4 2619Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2620{
051a06d4
KW
2621 UV result;
2622
97aff369 2623 dVAR;
7918f24d 2624
051a06d4 2625 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2626
968c5e6a 2627 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2628 if (flags) {
2629 result = toLOWER_LC(*p);
2630 }
2631 else {
81c6c7ce 2632 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2633 }
968c5e6a
KW
2634 }
2635 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2636 if (flags) {
2637 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2638 }
2639 else {
81c6c7ce
KW
2640 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2641 ustrp, lenp);
051a06d4 2642 }
968c5e6a 2643 }
051a06d4
KW
2644 else { /* utf8, ord above 255 */
2645 result = CALL_LOWER_CASE(p, ustrp, lenp);
2646
2647 if (flags) {
2648 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2649 }
968c5e6a 2650
051a06d4
KW
2651 return result;
2652 }
2653
2654 /* Here, used locale rules. Convert back to utf8 */
2655 if (UTF8_IS_INVARIANT(result)) {
2656 *ustrp = (U8) result;
2657 *lenp = 1;
2658 }
2659 else {
2660 *ustrp = UTF8_EIGHT_BIT_HI(result);
2661 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2662 *lenp = 2;
2663 }
2664
2665 if (tainted_ptr) {
2666 *tainted_ptr = TRUE;
2667 }
2668 return result;
b4e400f9
JH
2669}
2670
d3e79532 2671/*
87cea99e 2672=for apidoc to_utf8_fold
d3e79532 2673
a1433954
KW
2674Convert the UTF-8 encoded character at C<p> to its foldcase version and
2675store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2676that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
d3e79532
JH
2677foldcase version may be longer than the original character (up to
2678three characters).
2679
2680The first character of the foldcased version is returned
2681(but note, as explained above, that there may be more.)
2682
a1433954
KW
2683The character at C<p> is assumed by this routine to be well-formed.
2684
d3e79532
JH
2685=cut */
2686
051a06d4
KW
2687/* Not currently externally documented, and subject to change,
2688 * in <flags>
2689 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2690 * points < 256. Since foldcase is not defined in
2691 * POSIX, lowercase is used instead
2692 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2693 * otherwise simple folds
2694 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2695 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2696
b4e400f9 2697UV
051a06d4 2698Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2699{
97aff369 2700 dVAR;
7918f24d 2701
051a06d4
KW
2702 UV result;
2703
36bb2ab6 2704 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2705
a1dde8de 2706 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2707 if (flags & FOLD_FLAGS_LOCALE) {
2708 result = toLOWER_LC(*p);
2709 }
2710 else {
81c6c7ce
KW
2711 return _to_fold_latin1(*p, ustrp, lenp,
2712 cBOOL(flags & FOLD_FLAGS_FULL));
051a06d4 2713 }
a1dde8de
KW
2714 }
2715 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2716 if (flags & FOLD_FLAGS_LOCALE) {
2717 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2718 }
2719 else {
81c6c7ce 2720 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
051a06d4
KW
2721 ustrp, lenp, cBOOL(flags & FOLD_FLAGS_FULL));
2722 }
a1dde8de 2723 }
051a06d4
KW
2724 else { /* utf8, ord above 255 */
2725 result = CALL_FOLD_CASE(p, ustrp, lenp, flags);
a1dde8de 2726
051a06d4
KW
2727 if ((flags & FOLD_FLAGS_LOCALE)) {
2728 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2729 }
2730
2731 return result;
2732 }
2733
2734 /* Here, used locale rules. Convert back to utf8 */
2735 if (UTF8_IS_INVARIANT(result)) {
2736 *ustrp = (U8) result;
2737 *lenp = 1;
2738 }
2739 else {
2740 *ustrp = UTF8_EIGHT_BIT_HI(result);
2741 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2742 *lenp = 2;
2743 }
2744
2745 if (tainted_ptr) {
2746 *tainted_ptr = TRUE;
2747 }
2748 return result;
a0ed51b3
LW
2749}
2750
711a919c 2751/* Note:
f90a9a02 2752 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2753 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2754 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2755 */
c4a5db0c 2756
a0ed51b3 2757SV*
7fc63493 2758Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2759{
c4a5db0c
KW
2760 PERL_ARGS_ASSERT_SWASH_INIT;
2761
2762 /* Returns a copy of a swash initiated by the called function. This is the
2763 * public interface, and returning a copy prevents others from doing
2764 * mischief on the original */
2765
9a53f6cf 2766 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, FALSE, NULL, FALSE));
c4a5db0c
KW
2767}
2768
2769SV*
9a53f6cf 2770Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, bool return_if_undef, SV* invlist, bool passed_in_invlist_has_user_defined_property)
c4a5db0c
KW
2771{
2772 /* Initialize and return a swash, creating it if necessary. It does this
9a53f6cf 2773 * by calling utf8_heavy.pl in the general case.
c4a5db0c
KW
2774 *
2775 * This interface should only be used by functions that won't destroy or
2776 * adversely change the swash, as doing so affects all other uses of the
2777 * swash in the program; the general public should use 'Perl_swash_init'
2778 * instead.
2779 *
2780 * pkg is the name of the package that <name> should be in.
2781 * name is the name of the swash to find. Typically it is a Unicode
2782 * property name, including user-defined ones
2783 * listsv is a string to initialize the swash with. It must be of the form
2784 * documented as the subroutine return value in
2785 * L<perlunicode/User-Defined Character Properties>
2786 * minbits is the number of bits required to represent each data element.
2787 * It is '1' for binary properties.
2788 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf
KW
2789 * return_if_undef is TRUE if the routine shouldn't croak if it can't find
2790 * the requested property
2791 * invlist is an inversion list to initialize the swash with (or NULL)
2792 * has_user_defined_property is TRUE if <invlist> has some component that
2793 * came from a user-defined property
2794 *
2795 * Thus there are three possible inputs to find the swash: <name>,
2796 * <listsv>, and <invlist>. At least one must be specified. The result
2797 * will be the union of the specified ones, although <listsv>'s various
2798 * actions can intersect, etc. what <name> gives.
2799 *
2800 * <invlist> is only valid for binary properties */
c4a5db0c 2801
27da23d5 2802 dVAR;
c4a5db0c 2803 SV* retval = &PL_sv_undef;
9a53f6cf
KW
2804
2805 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2806 assert(! invlist || minbits == 1);
2807
2808 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2809 * so */
2810 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2811 dSP;
2812 const size_t pkg_len = strlen(pkg);
2813 const size_t name_len = strlen(name);
2814 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2815 SV* errsv_save;
2816 GV *method;
2817
2818 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2819
2820 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2821 ENTER;
69794297
KW
2822 SAVEHINTS();
2823 save_re_context();
2824 if (PL_parser && PL_parser->error_count)
2825 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2826 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2827 if (!method) { /* demand load utf8 */
2828 ENTER;
2829 errsv_save = newSVsv(ERRSV);
2830 /* It is assumed that callers of this routine are not passing in
2831 * any user derived data. */
2832 /* Need to do this after save_re_context() as it will set
2833 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2834 * in Perl_magic_get). Even line to create errsv_save can turn on
2835 * PL_tainted. */
2836 SAVEBOOL(PL_tainted);
2837 PL_tainted = 0;
2838 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2839 NULL);
2840 if (!SvTRUE(ERRSV))
2841 sv_setsv(ERRSV, errsv_save);
2842 SvREFCNT_dec(errsv_save);
2843 LEAVE;
2844 }
2845 SPAGAIN;
2846 PUSHMARK(SP);
2847 EXTEND(SP,5);
2848 mPUSHp(pkg, pkg_len);
2849 mPUSHp(name, name_len);
2850 PUSHs(listsv);
2851 mPUSHi(minbits);
2852 mPUSHi(none);
2853 PUTBACK;
f8be5cf0 2854 errsv_save = newSVsv(ERRSV);
69794297
KW
2855 /* If we already have a pointer to the method, no need to use
2856 * call_method() to repeat the lookup. */
2857 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
2858 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2859 {
2860 retval = *PL_stack_sp--;
2861 SvREFCNT_inc(retval);
2862 }
f8be5cf0
JH
2863 if (!SvTRUE(ERRSV))
2864 sv_setsv(ERRSV, errsv_save);
2865 SvREFCNT_dec(errsv_save);
ce3b816e 2866 LEAVE;
69794297
KW
2867 POPSTACK;
2868 if (IN_PERL_COMPILETIME) {
2869 CopHINTS_set(PL_curcop, PL_hints);
2870 }
2871 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2872 if (SvPOK(retval))
2873
2874 /* If caller wants to handle missing properties, let them */
2875 if (return_if_undef) {
2876 return NULL;
2877 }
2878 Perl_croak(aTHX_
2879 "Can't find Unicode property definition \"%"SVf"\"",
2880 SVfARG(retval));
2881 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2882 }
9a53f6cf 2883 } /* End of calling the module to find the swash */
36eb48b4
KW
2884
2885 /* Make sure there is an inversion list for binary properties */
2886 if (minbits == 1) {
2887 SV** swash_invlistsvp = NULL;
2888 SV* swash_invlist = NULL;
9a53f6cf 2889 bool invlist_in_swash_is_valid = FALSE;
88d45d28 2890 HV* swash_hv = NULL;
36eb48b4 2891
9a53f6cf
KW
2892 /* If this operation fetched a swash, get its already existing
2893 * inversion list or create one for it */
2894 if (retval != &PL_sv_undef) {
36eb48b4
KW
2895 swash_hv = MUTABLE_HV(SvRV(retval));
2896
2897 swash_invlistsvp = hv_fetchs(swash_hv, "INVLIST", FALSE);
9a53f6cf
KW
2898 if (swash_invlistsvp) {
2899 swash_invlist = *swash_invlistsvp;
2900 invlist_in_swash_is_valid = TRUE;
2901 }
2902 else {
36eb48b4 2903 swash_invlist = _swash_to_invlist(retval);
9a53f6cf
KW
2904 }
2905 }
2906
2907 /* If an inversion list was passed in, have to include it */
2908 if (invlist) {
2909
2910 /* Any fetched swash will by now have an inversion list in it;
2911 * otherwise <swash_invlist> will be NULL, indicating that we
2912 * didn't fetch a swash */
2913 if (swash_invlist) {
2914
2915 /* Add the passed-in inversion list, which invalidates the one
2916 * already stored in the swash */
2917 invlist_in_swash_is_valid = FALSE;
2918 _invlist_union(invlist, swash_invlist, &swash_invlist);
2919 }
2920 else {
2921
2922 /* Here, there is no swash already. Set up a minimal one */
2923 swash_hv = newHV();
2924 retval = newRV_inc(MUTABLE_SV(swash_hv));
2925 swash_invlist = invlist;
2926 }
2927
2928 if (passed_in_invlist_has_user_defined_property) {
2929 if (! hv_stores(swash_hv, "USER_DEFINED", newSVuv(1))) {
2930 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2931 }
2932 }
2933 }
2934
2935 /* Here, we have computed the union of all the passed-in data. It may
2936 * be that there was an inversion list in the swash which didn't get
2937 * touched; otherwise save the one computed one */
2938 if (! invlist_in_swash_is_valid) {
69794297
KW
2939 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "INVLIST", swash_invlist))
2940 {
2941 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2942 }
9a53f6cf 2943 }
36eb48b4
KW
2944 }
2945
a0ed51b3
LW
2946 return retval;
2947}
2948
035d37be
JH
2949
2950/* This API is wrong for special case conversions since we may need to
2951 * return several Unicode characters for a single Unicode character
2952 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2953 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2954 * multiple values. --jhi
2955 * For those, you should use to_utf8_case() instead */
b0e3252e 2956/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2957
2958/* Note:
2959 * Returns the value of property/mapping C<swash> for the first character
2960 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
2961 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
2962 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2963 *
2964 * A "swash" is a hash which contains initially the keys/values set up by
2965 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2966 * property for all possible code points. Things are stored in a compact form
2967 * (see utf8_heavy.pl) so that calculation is required to find the actual
2968 * property value for a given code point. As code points are looked up, new
2969 * key/value pairs are added to the hash, so that the calculation doesn't have
2970 * to ever be re-done. Further, each calculation is done, not just for the
2971 * desired one, but for a whole block of code points adjacent to that one.
2972 * For binary properties on ASCII machines, the block is usually for 64 code
2973 * points, starting with a code point evenly divisible by 64. Thus if the
2974 * property value for code point 257 is requested, the code goes out and
2975 * calculates the property values for all 64 code points between 256 and 319,
2976 * and stores these as a single 64-bit long bit vector, called a "swatch",
2977 * under the key for code point 256. The key is the UTF-8 encoding for code
2978 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2979 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2980 * for code point 258 is then requested, this code realizes that it would be
2981 * stored under the key for 256, and would find that value and extract the
2982 * relevant bit, offset from 256.
2983 *
2984 * Non-binary properties are stored in as many bits as necessary to represent
2985 * their values (32 currently, though the code is more general than that), not
2986 * as single bits, but the principal is the same: the value for each key is a
2987 * vector that encompasses the property values for all code points whose UTF-8
2988 * representations are represented by the key. That is, for all code points
2989 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2990 * bytes of that.
680c470c 2991 */
a0ed51b3 2992UV
680c470c 2993Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2994{
27da23d5 2995 dVAR;
ef8f7699 2996 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2997 U32 klen;
2998 U32 off;
a0ed51b3 2999 STRLEN slen;
7d85a32c 3000 STRLEN needents;
cfd0369c 3001 const U8 *tmps = NULL;
a0ed51b3 3002 U32 bit;
979f2922 3003 SV *swatch;
3568d838 3004 U8 tmputf8[2];
35da51f7 3005 const UV c = NATIVE_TO_ASCII(*ptr);
3568d838 3006
7918f24d
NC
3007 PERL_ARGS_ASSERT_SWASH_FETCH;
3008
dbe7a391 3009 /* Convert to utf8 if not already */
3568d838 3010 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
979f2922
ST
3011 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3012 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3013 ptr = tmputf8;
3568d838
JH
3014 }
3015 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
37e2e78e 3016 * then the "swatch" is a vec() for all the chars which start
3568d838
JH
3017 * with 0xAA..0xYY
3018 * So the key in the hash (klen) is length of encoded char -1
3019 */
3020 klen = UTF8SKIP(ptr) - 1;
3021 off = ptr[klen];
a0ed51b3 3022
979f2922 3023 if (klen == 0) {
37e2e78e 3024 /* If char is invariant then swatch is for all the invariant chars
1e54db1a 3025 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
7d85a32c 3026 */
979f2922
ST
3027 needents = UTF_CONTINUATION_MARK;
3028 off = NATIVE_TO_UTF(ptr[klen]);
3029 }
3030 else {
7d85a32c 3031 /* If char is encoded then swatch is for the prefix */
979f2922
ST
3032 needents = (1 << UTF_ACCUMULATION_SHIFT);
3033 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3034 }
7d85a32c 3035
a0ed51b3
LW
3036 /*
3037 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3038 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3039 * it's nothing to sniff at.) Pity we usually come through at least
3040 * two function calls to get here...
3041 *
3042 * NB: this code assumes that swatches are never modified, once generated!
3043 */
3044
3568d838 3045 if (hv == PL_last_swash_hv &&
a0ed51b3 3046 klen == PL_last_swash_klen &&
27da23d5 3047 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3048 {
3049 tmps = PL_last_swash_tmps;
3050 slen = PL_last_swash_slen;
3051 }
3052 else {
3053 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3054 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3055
b0e3252e 3056 /* If not cached, generate it via swatch_get */
979f2922
ST
3057 if (!svp || !SvPOK(*svp)
3058 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2b9d42f0
NIS
3059 /* We use utf8n_to_uvuni() as we want an index into
3060 Unicode tables, not a native character number.
3061 */
aec46f14 3062 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
872c91ae
JH
3063 ckWARN(WARN_UTF8) ?
3064 0 : UTF8_ALLOW_ANY);
b0e3252e 3065 swatch = swatch_get(swash,
979f2922 3066 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
361ee0fe 3067 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
979f2922
ST
3068 needents);
3069
923e4eb5 3070 if (IN_PERL_COMPILETIME)
623e6609 3071 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3072
979f2922 3073 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3074
979f2922
ST
3075 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3076 || (slen << 3) < needents)
5637ef5b
NC
3077 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3078 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3079 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3080 }
3081
3082 PL_last_swash_hv = hv;
16d8f38a 3083 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3084 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3085 /* FIXME change interpvar.h? */
3086 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3087 PL_last_swash_slen = slen;
3088 if (klen)
3089 Copy(ptr, PL_last_swash_key, klen, U8);
3090 }
3091
a410ec23
KW
3092 if (UTF8_IS_SUPER(ptr) && ckWARN_d(WARN_NON_UNICODE)) {
3093 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3094
3095 /* This outputs warnings for binary properties only, assuming that
3096 * to_utf8_case() will output any for non-binary. Also, surrogates
3097 * aren't checked for, as that would warn on things like /\p{Gc=Cs}/ */
3098
9a53f6cf 3099 if (! bitssvp || SvUV(*bitssvp) == 1) {
a410ec23
KW
3100 /* User-defined properties can silently match above-Unicode */
3101 SV** const user_defined_svp = hv_fetchs(hv, "USER_DEFINED", FALSE);
3102 if (! user_defined_svp || ! SvUV(*user_defined_svp)) {
3103 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0, 0);
3104 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3105 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", code_point);
3106 }
3107 }
3108 }
3109
9faf8d75 3110 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3111 case 1:
3112 bit = 1 << (off & 7);
3113 off >>= 3;
3114 return (tmps[off] & bit) != 0;
3115 case 8:
3116 return tmps[off];
3117 case 16:
3118 off <<= 1;
3119 return (tmps[off] << 8) + tmps[off + 1] ;
3120 case 32:
3121 off <<= 2;
3122 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3123 }
5637ef5b
NC
3124 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3125 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3126 NORETURN_FUNCTION_END;
a0ed51b3 3127}
2b9d42f0 3128
319009ee
KW
3129/* Read a single line of the main body of the swash input text. These are of
3130 * the form:
3131 * 0053 0056 0073
3132 * where each number is hex. The first two numbers form the minimum and
3133 * maximum of a range, and the third is the value associated with the range.
3134 * Not all swashes should have a third number
3135 *
3136 * On input: l points to the beginning of the line to be examined; it points
3137 * to somewhere in the string of the whole input text, and is
3138 * terminated by a \n or the null string terminator.
3139 * lend points to the null terminator of that string
3140 * wants_value is non-zero if the swash expects a third number
3141 * typestr is the name of the swash's mapping, like 'ToLower'
3142 * On output: *min, *max, and *val are set to the values read from the line.
3143 * returns a pointer just beyond the line examined. If there was no
3144 * valid min number on the line, returns lend+1
3145 */
3146
3147STATIC U8*
3148S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3149 const bool wants_value, const U8* const typestr)
3150{
3151 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3152 STRLEN numlen; /* Length of the number */
02470786
KW
3153 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3154 | PERL_SCAN_DISALLOW_PREFIX
3155 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3156
3157 /* nl points to the next \n in the scan */
3158 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3159
3160 /* Get the first number on the line: the range minimum */
3161 numlen = lend - l;
3162 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3163 if (numlen) /* If found a hex number, position past it */
3164 l += numlen;
3165 else if (nl) { /* Else, go handle next line, if any */
3166 return nl + 1; /* 1 is length of "\n" */
3167 }
3168 else { /* Else, no next line */
3169 return lend + 1; /* to LIST's end at which \n is not found */
3170 }
3171
3172 /* The max range value follows, separated by a BLANK */
3173 if (isBLANK(*l)) {
3174 ++l;
02470786
KW
3175 flags = PERL_SCAN_SILENT_ILLDIGIT
3176 | PERL_SCAN_DISALLOW_PREFIX
3177 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3178 numlen = lend - l;
3179 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3180 if (numlen)
3181 l += numlen;
3182 else /* If no value here, it is a single element range */
3183 *max = *min;
3184
3185 /* Non-binary tables have a third entry: what the first element of the
3186 * range maps to */
3187 if (wants_value) {
3188 if (isBLANK(*l)) {
3189 ++l;
f90a9a02
KW
3190
3191 /* The ToLc, etc table mappings are not in hex, and must be
3192 * corrected by adding the code point to them */
3193 if (typeto) {
3194 char *after_strtol = (char *) lend;
3195 *val = Strtol((char *)l, &after_strtol, 10);
3196 l = (U8 *) after_strtol;
f90a9a02
KW
3197 }
3198 else { /* Other tables are in hex, and are the correct result
3199 without tweaking */
a9d188b3
KW
3200 flags = PERL_SCAN_SILENT_ILLDIGIT
3201 | PERL_SCAN_DISALLOW_PREFIX
3202 | PERL_SCAN_SILENT_NON_PORTABLE;
3203 numlen = lend - l;
3204 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3205 if (numlen)
3206 l += numlen;
3207 else
3208 *val = 0;
f90a9a02 3209 }
319009ee
KW
3210 }
3211 else {
3212 *val = 0;
3213 if (typeto) {
dcbac5bb 3214 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3215 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3216 typestr, l);
3217 }
3218 }
3219 }
3220 else
3221 *val = 0; /* bits == 1, then any val should be ignored */
3222 }
3223 else { /* Nothing following range min, should be single element with no
3224 mapping expected */
3225 *max = *min;
3226 if (wants_value) {
3227 *val = 0;
3228 if (typeto) {
dcbac5bb 3229 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3230 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3231 }
3232 }
3233 else
3234 *val = 0; /* bits == 1, then val should be ignored */
3235 }
3236
3237 /* Position to next line if any, or EOF */
3238 if (nl)
3239 l = nl + 1;
3240 else
3241 l = lend;
3242
3243 return l;
3244}
3245
979f2922
ST
3246/* Note:
3247 * Returns a swatch (a bit vector string) for a code point sequence
3248 * that starts from the value C<start> and comprises the number C<span>.
3249 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3250 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3251 */
3252STATIC SV*
b0e3252e 3253S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3254{
3255 SV *swatch;
77f9f126 3256 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3257 STRLEN lcur, xcur, scur;
ef8f7699 3258 HV *const hv = MUTABLE_HV(SvRV(swash));
36eb48b4
KW
3259 SV** const invlistsvp = hv_fetchs(hv, "INVLIST", FALSE);
3260
88d45d28
KW
3261 SV** listsvp = NULL; /* The string containing the main body of the table */
3262 SV** extssvp = NULL;
3263 SV** invert_it_svp = NULL;
3264 U8* typestr = NULL;
786861f5
KW
3265 STRLEN bits;
3266 STRLEN octets; /* if bits == 1, then octets == 0 */
3267 UV none;
3268 UV end = start + span;
972dd592 3269
36eb48b4 3270 if (invlistsvp == NULL) {
786861f5
KW
3271 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3272 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3273 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3274 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3275 listsvp = hv_fetchs(hv, "LIST", FALSE);
3276 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3277
3278 bits = SvUV(*bitssvp);
3279 none = SvUV(*nonesvp);
3280 typestr = (U8*)SvPV_nolen(*typesvp);
3281 }
36eb48b4
KW
3282 else {
3283 bits = 1;
3284 none = 0;
3285 }
786861f5 3286 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3287
b0e3252e 3288 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3289
979f2922 3290 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3291 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3292 (UV)bits);
979f2922
ST
3293 }
3294
84ea5ef6
KW
3295 /* If overflowed, use the max possible */
3296 if (end < start) {
3297 end = UV_MAX;
3298 span = end - start;
3299 }
3300
979f2922 3301 /* create and initialize $swatch */
979f2922 3302 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3303 swatch = newSV(scur);
3304 SvPOK_on(swatch);
979f2922
ST
3305 s = (U8*)SvPVX(swatch);
3306 if (octets && none) {
0bd48802 3307 const U8* const e = s + scur;
979f2922
ST
3308 while (s < e) {
3309 if (bits == 8)
3310 *s++ = (U8)(none & 0xff);
3311 else if (bits == 16) {
3312 *s++ = (U8)((none >> 8) & 0xff);
3313 *s++ = (U8)( none & 0xff);
3314 }
3315 else if (bits == 32) {
3316 *s++ = (U8)((none >> 24) & 0xff);
3317 *s++ = (U8)((none >> 16) & 0xff);
3318 *s++ = (U8)((none >> 8) & 0xff);
3319 *s++ = (U8)( none & 0xff);
3320 }
3321 }
3322 *s = '\0';
3323 }
3324 else {
3325 (void)memzero((U8*)s, scur + 1);
3326 }
3327 SvCUR_set(swatch, scur);
3328 s = (U8*)SvPVX(swatch);
3329
36eb48b4
KW
3330 if (invlistsvp) { /* If has an inversion list set up use that */
3331 _invlist_populate_swatch(*invlistsvp, start, end, s);
3332 return swatch;
3333 }
3334
3335 /* read $swash->{LIST} */
979f2922
ST
3336 l = (U8*)SvPV(*listsvp, lcur);
3337 lend = l + lcur;
3338 while (l < lend) {
8ed25d53 3339 UV min, max, val, upper;
319009ee
KW
3340 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3341 cBOOL(octets), typestr);
3342 if (l > lend) {
979f2922
ST
3343 break;
3344 }
3345
972dd592 3346 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3347 if (max < start)
3348 continue;
3349
8ed25d53
KW
3350 /* <end> is generally 1 beyond where we want to set things, but at the
3351 * platform's infinity, where we can't go any higher, we want to
3352 * include the code point at <end> */
3353 upper = (max < end)
3354 ? max
3355 : (max != UV_MAX || end != UV_MAX)
3356 ? end - 1
3357 : end;
3358
979f2922 3359 if (octets) {
35da51f7 3360 UV key;
979f2922
ST
3361 if (min < start) {
3362 if (!none || val < none) {
3363 val += start - min;
3364 }
3365 min = start;
3366 }
8ed25d53 3367 for (key = min; key <= upper; key++) {
979f2922 3368 STRLEN offset;
979f2922
ST
3369 /* offset must be non-negative (start <= min <= key < end) */
3370 offset = octets * (key - start);
3371 if (bits == 8)
3372 s[offset] = (U8)(val & 0xff);
3373 else if (bits == 16) {
3374 s[offset ] = (U8)((val >> 8) & 0xff);
3375 s[offset + 1] = (U8)( val & 0xff);
3376 }
3377 else if (bits == 32) {
3378 s[offset ] = (U8)((val >> 24) & 0xff);
3379 s[offset + 1] = (U8)((val >> 16) & 0xff);
3380 s[offset + 2] = (U8)((val >> 8) & 0xff);
3381 s[offset + 3] = (U8)( val & 0xff);
3382 }
3383
3384 if (!none || val < none)
3385 ++val;
3386 }
3387 }
711a919c 3388 else { /* bits == 1, then val should be ignored */
35da51f7 3389 UV key;
979f2922
ST
3390 if (min < start)
3391 min = start;
6cb05c12 3392
8ed25d53 3393 for (key = min; key <= upper; key++) {
0bd48802 3394 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3395 s[offset >> 3] |= 1 << (offset & 7);
3396 }
3397 }
3398 } /* while */
979f2922 3399
9479a769 3400 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3401 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3402
3403 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3404 * be 0, and their inversion should also be 0, as we don't succeed any
3405 * Unicode property matches for non-Unicode code points */
3406 if (start <= PERL_UNICODE_MAX) {
3407
3408 /* The code below assumes that we never cross the
3409 * Unicode/above-Unicode boundary in a range, as otherwise we would
3410 * have to figure out where to stop flipping the bits. Since this
3411 * boundary is divisible by a large power of 2, and swatches comes
3412 * in small powers of 2, this should be a valid assumption */
3413 assert(start + span - 1 <= PERL_UNICODE_MAX);
3414
507a8485
KW
3415 send = s + scur;
3416 while (s < send) {
3417 *s = ~(*s);
3418 s++;
3419 }
0bda3001 3420 }
77f9f126
KW
3421 }
3422
d73c39c5
KW
3423 /* read $swash->{EXTRAS}
3424 * This code also copied to swash_to_invlist() below */
979f2922
ST
3425 x = (U8*)SvPV(*extssvp, xcur);
3426 xend = x + xcur;
3427 while (x < xend) {
3428 STRLEN namelen;
3429 U8 *namestr;
3430 SV** othersvp;
3431 HV* otherhv;
3432 STRLEN otherbits;
3433 SV **otherbitssvp, *other;
711a919c 3434 U8 *s, *o, *nl;
979f2922
ST
3435 STRLEN slen, olen;
3436
35da51f7 3437 const U8 opc = *x++;
979f2922
ST
3438 if (opc == '\n')
3439 continue;
3440
3441 nl = (U8*)memchr(x, '\n', xend - x);
3442
3443 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3444 if (nl) {
3445 x = nl + 1; /* 1 is length of "\n" */
3446 continue;
3447 }
3448 else {
3449 x = xend; /* to EXTRAS' end at which \n is not found */
3450 break;
3451 }
3452 }
3453
3454 namestr = x;
3455 if (nl) {
3456 namelen = nl - namestr;
3457 x = nl + 1;
3458 }
3459 else {
3460 namelen = xend - namestr;
3461 x = xend;
3462 }
3463
3464 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3465 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3466 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3467 otherbits = (STRLEN)SvUV(*otherbitssvp);
3468 if (bits < otherbits)
5637ef5b
NC
3469 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3470 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3471
3472 /* The "other" swatch must be destroyed after. */
b0e3252e 3473 other = swatch_get(*othersvp, start, span);
979f2922
ST
3474 o = (U8*)SvPV(other, olen);
3475
3476 if (!olen)
b0e3252e 3477 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3478
3479 s = (U8*)SvPV(swatch, slen);
3480 if (bits == 1 && otherbits == 1) {
3481 if (slen != olen)
5637ef5b
NC
3482 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3483 "mismatch, slen=%"UVuf", olen=%"UVuf,
3484 (UV)slen, (UV)olen);
979f2922
ST
3485
3486 switch (opc) {
3487 case '+':
3488 while (slen--)
3489 *s++ |= *o++;
3490 break;
3491 case '!':
3492 while (slen--)
3493 *s++ |= ~*o++;
3494 break;
3495 case '-':
3496 while (slen--)
3497 *s++ &= ~*o++;
3498 break;
3499 case '&':
3500 while (slen--)
3501 *s++ &= *o++;
3502 break;
3503 default:
3504 break;
3505 }
3506 }
711a919c 3507 else {
979f2922
ST
3508 STRLEN otheroctets = otherbits >> 3;
3509 STRLEN offset = 0;
35da51f7 3510 U8* const send = s + slen;
979f2922
ST
3511
3512 while (s < send) {
3513 UV otherval = 0;
3514
3515 if (otherbits == 1) {
3516 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3517 ++offset;
3518 }
3519 else {
3520 STRLEN vlen = otheroctets;
3521 otherval = *o++;
3522 while (--vlen) {
3523 otherval <<= 8;
3524 otherval |= *o++;
3525 }
3526 }
3527
711a919c 3528 if (opc == '+' && otherval)
6f207bd3 3529 NOOP; /* replace with otherval */
979f2922
ST
3530 else if (opc == '!' && !otherval)
3531 otherval = 1;
3532 else if (opc == '-' && otherval)
3533 otherval = 0;
3534 else if (opc == '&' && !otherval)
3535 otherval = 0;
3536 else {
711a919c 3537 s += octets; /* no replacement */
979f2922
ST
3538 continue;
3539 }
3540
3541 if (bits == 8)
3542 *s++ = (U8)( otherval & 0xff);
3543 else if (bits == 16) {
3544 *s++ = (U8)((otherval >> 8) & 0xff);
3545 *s++ = (U8)( otherval & 0xff);
3546 }
3547 else if (bits == 32) {
3548 *s++ = (U8)((otherval >> 24) & 0xff);
3549 *s++ = (U8)((otherval >> 16) & 0xff);
3550 *s++ = (U8)((otherval >> 8) & 0xff);
3551 *s++ = (U8)( otherval & 0xff);
3552 }
3553 }
3554 }
3555 sv_free(other); /* through with it! */
3556 } /* while */
3557 return swatch;
3558}
3559
064c021d 3560HV*
4c2e1131 3561Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3562{
3563
5662e334
KW
3564 /* Subject to change or removal. For use only in one place in regcomp.c.
3565 * Can't be used on a property that is subject to user override, as it
3566 * relies on the value of SPECIALS in the swash which would be set by
3567 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3568 * for overridden properties
064c021d
KW
3569 *
3570 * Returns a hash which is the inversion and closure of a swash mapping.
3571 * For example, consider the input lines:
3572 * 004B 006B
3573 * 004C 006C
3574 * 212A 006B
3575 *
3576 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3577 * 006C. The value for each key is an array. For 006C, the array would
3578 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3579 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3580 *
3581 * Essentially, for any code point, it gives all the code points that map to
3582 * it, or the list of 'froms' for that point.
3583 *
5662e334
KW
3584 * Currently it ignores any additions or deletions from other swashes,
3585 * looking at just the main body of the swash, and if there are SPECIALS
3586 * in the swash, at that hash
3587 *
3588 * The specials hash can be extra code points, and most likely consists of
3589 * maps from single code points to multiple ones (each expressed as a string
3590 * of utf8 characters). This function currently returns only 1-1 mappings.
3591 * However consider this possible input in the specials hash:
3592 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3593 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3594 *
3595 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3596 * currently handle. But it also means that FB05 and FB06 are equivalent in
3597 * a 1-1 mapping which we should handle, and this relationship may not be in
3598 * the main table. Therefore this function examines all the multi-char
3599 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3600
3601 U8 *l, *lend;
3602 STRLEN lcur;
3603 HV *const hv = MUTABLE_HV(SvRV(swash));
3604
3605 /* The string containing the main body of the table */
3606 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3607
3608 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3609 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3610 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3611 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3612 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3613 const STRLEN bits = SvUV(*bitssvp);
3614 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3615 const UV none = SvUV(*nonesvp);
5662e334 3616 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3617
3618 HV* ret = newHV();
3619
3620 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3621
3622 /* Must have at least 8 bits to get the mappings */
3623 if (bits != 8 && bits != 16 && bits != 32) {
3624 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3625 (UV)bits);
3626 }
3627
5662e334
KW
3628 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3629 mapping to more than one character */
3630
3631 /* Construct an inverse mapping hash for the specials */
3632 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3633 HV * specials_inverse = newHV();
3634 char *char_from; /* the lhs of the map */
3635 I32 from_len; /* its byte length */
3636 char *char_to; /* the rhs of the map */
3637 I32 to_len; /* its byte length */
3638 SV *sv_to; /* and in a sv */
3639 AV* from_list; /* list of things that map to each 'to' */
3640
3641 hv_iterinit(specials_hv);
3642
3643 /* The keys are the characters (in utf8) that map to the corresponding
3644 * utf8 string value. Iterate through the list creating the inverse
3645 * list. */
3646 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3647 SV** listp;
3648 if (! SvPOK(sv_to)) {
5637ef5b
NC
3649 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3650 "unexpectedly is not a string, flags=%lu",
3651 (unsigned long)SvFLAGS(sv_to));
5662e334 3652 }
4b88fb76 3653 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %"UVXf", First char of to is %"UVXf"\n", valid_utf8_to_uvchr((U8*) char_from, 0), valid_utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
5662e334
KW
3654
3655 /* Each key in the inverse list is a mapped-to value, and the key's
3656 * hash value is a list of the strings (each in utf8) that map to
3657 * it. Those strings are all one character long */
3658 if ((listp = hv_fetch(specials_inverse,
3659 SvPVX(sv_to),
3660 SvCUR(sv_to), 0)))
3661 {
3662 from_list = (AV*) *listp;
3663 }
3664 else { /* No entry yet for it: create one */
3665 from_list = newAV();
3666 if (! hv_store(specials_inverse,
3667 SvPVX(sv_to),
3668 SvCUR(sv_to),
3669 (SV*) from_list, 0))
3670 {
3671 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3672 }
3673 }
3674
3675 /* Here have the list associated with this 'to' (perhaps newly
3676 * created and empty). Just add to it. Note that we ASSUME that
3677 * the input is guaranteed to not have duplications, so we don't
3678 * check for that. Duplications just slow down execution time. */
3679 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3680 }
3681
3682 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3683 * it looking for cases like the FB05/FB06 examples above. There would
3684 * be an entry in the hash like
3685 * 'st' => [ FB05, FB06 ]
3686 * In this example we will create two lists that get stored in the
3687 * returned hash, 'ret':
3688 * FB05 => [ FB05, FB06 ]
3689 * FB06 => [ FB05, FB06 ]
3690 *
3691 * Note that there is nothing to do if the array only has one element.
3692 * (In the normal 1-1 case handled below, we don't have to worry about
3693 * two lists, as everything gets tied to the single list that is
3694 * generated for the single character 'to'. But here, we are omitting
3695 * that list, ('st' in the example), so must have multiple lists.) */
3696 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3697 &char_to, &to_len)))
3698 {
3699 if (av_len(from_list) > 0) {
3700 int i;
3701
3702 /* We iterate over all combinations of i,j to place each code
3703 * point on each list */
3704 for (i = 0; i <= av_len(from_list); i++) {
3705 int j;
3706 AV* i_list = newAV();
3707 SV** entryp = av_fetch(from_list, i, FALSE);
3708 if (entryp == NULL) {
3709 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3710 }
3711 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3712 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3713 }
3714 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3715 (SV*) i_list, FALSE))
3716 {
3717 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3718 }
3719
4b88fb76 3720 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
5662e334
KW
3721 for (j = 0; j <= av_len(from_list); j++) {
3722 entryp = av_fetch(from_list, j, FALSE);
3723 if (entryp == NULL) {
3724 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3725 }
3726
3727 /* When i==j this adds itself to the list */
4b88fb76
KW
3728 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3729 (U8*) SvPVX(*entryp),
3730 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3731 0)));
3732 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Adding %"UVXf" to list for %"UVXf"\n", valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
5662e334
KW
3733 }
3734 }
3735 }
3736 }
3737 SvREFCNT_dec(specials_inverse); /* done with it */
3738 } /* End of specials */
3739
064c021d
KW
3740 /* read $swash->{LIST} */
3741 l = (U8*)SvPV(*listsvp, lcur);
3742 lend = l + lcur;
3743
3744 /* Go through each input line */
3745 while (l < lend) {
3746 UV min, max, val;
3747 UV inverse;
3748 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3749 cBOOL(octets), typestr);
3750 if (l > lend) {
3751 break;
3752 }
3753
3754 /* Each element in the range is to be inverted */
3755 for (inverse = min; inverse <= max; inverse++) {
3756 AV* list;
064c021d
KW
3757 SV** listp;
3758 IV i;
3759 bool found_key = FALSE;
5662e334 3760 bool found_inverse = FALSE;
064c021d
KW
3761
3762 /* The key is the inverse mapping */
3763 char key[UTF8_MAXBYTES+1];
3764 char* key_end = (char *) uvuni_to_utf8((U8*) key, val);
3765 STRLEN key_len = key_end - key;
3766
064c021d
KW
3767 /* Get the list for the map */
3768 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3769 list = (AV*) *listp;
3770 }
3771 else { /* No entry yet for it: create one */
3772 list = newAV();
3773 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3774 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3775 }
3776 }
3777
5662e334
KW
3778 /* Look through list to see if this inverse mapping already is
3779 * listed, or if there is a mapping to itself already */
508f7cfa 3780 for (i = 0; i <= av_len(list); i++) {
064c021d
KW
3781 SV** entryp = av_fetch(list, i, FALSE);
3782 SV* entry;
3783 if (entryp == NULL) {
3784 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3785 }
3786 entry = *entryp;
5662e334 3787 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
56ca34ca 3788 if (SvUV(entry) == val) {
064c021d 3789 found_key = TRUE;