This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
charnames pod: Nits
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
a0ed51b3 35
a0c21aa1 36#ifndef EBCDIC
970ea3cb 37/* Separate prototypes needed because in ASCII systems these are
a0c21aa1
JH
38 * usually macros but they still are compiled as code, too. */
39PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
6dd9dce9 40PERL_CALLCONV UV Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen);
a0c21aa1
JH
41PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
42#endif
43
27da23d5
JH
44static const char unees[] =
45 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 46
48ef279e 47/*
ccfc67b7 48=head1 Unicode Support
a0ed51b3 49
166f8a29
DM
50This file contains various utility functions for manipulating UTF8-encoded
51strings. For the uninitiated, this is a method of representing arbitrary
61296642 52Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
53characters in the ASCII range are unmodified, and a zero byte never appears
54within non-zero characters.
166f8a29 55
eaf7a4d2
CS
56=cut
57*/
58
59/*
60=for apidoc is_ascii_string
61
a1433954 62Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
63or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
64is, if they are invariant. On ASCII-ish machines, only ASCII characters
65fit this definition, hence the function's name.
eaf7a4d2 66
9f7e3d64
MH
67If C<len> is 0, it will be calculated using C<strlen(s)>.
68
a1433954 69See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
70
71=cut
72*/
73
74bool
668b6d8d 75Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
76{
77 const U8* const send = s + (len ? len : strlen((const char *)s));
78 const U8* x = s;
79
80 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
81
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
84 break;
85 }
86
87 return x == send;
88}
89
90/*
87cea99e 91=for apidoc uvuni_to_utf8_flags
eebe1485 92
6ee84de2
KW
93Adds the UTF-8 representation of the code point C<uv> to the end
94of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
eebe1485 95bytes available. The return value is the pointer to the byte after the
9041c2e3 96end of the new character. In other words,
eebe1485 97
b851fbc1
JH
98 d = uvuni_to_utf8_flags(d, uv, flags);
99
100or, in most cases,
101
9041c2e3 102 d = uvuni_to_utf8(d, uv);
eebe1485 103
b851fbc1
JH
104(which is equivalent to)
105
106 d = uvuni_to_utf8_flags(d, uv, 0);
107
949cf498 108This is the recommended Unicode-aware way of saying
eebe1485
SC
109
110 *(d++) = uv;
111
949cf498
KW
112This function will convert to UTF-8 (and not warn) even code points that aren't
113legal Unicode or are problematic, unless C<flags> contains one or more of the
a1433954
KW
114following flags:
115
949cf498
KW
116If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
117the function will raise a warning, provided UTF8 warnings are enabled. If instead
118UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
119If both flags are set, the function will both warn and return NULL.
120
121The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
122affect how the function handles a Unicode non-character. And, likewise for the
123UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
124above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
125even less portable) can be warned and/or disallowed even if other above-Unicode
126code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
127flags.
128
129And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
130above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
131DISALLOW flags.
132
133
eebe1485
SC
134=cut
135*/
136
dfe13c55 137U8 *
b851fbc1 138Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 139{
7918f24d
NC
140 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
141
979f77b6
KW
142 /* The first problematic code point is the first surrogate */
143 if (uv >= UNICODE_SURROGATE_FIRST
144 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
145 {
949cf498
KW
146 if (UNICODE_IS_SURROGATE(uv)) {
147 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 148 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
149 "UTF-16 surrogate U+%04"UVXf, uv);
150 }
151 if (flags & UNICODE_DISALLOW_SURROGATE) {
152 return NULL;
153 }
154 }
155 else if (UNICODE_IS_SUPER(uv)) {
156 if (flags & UNICODE_WARN_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
158 {
8457b38f 159 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
160 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
161 }
162 if (flags & UNICODE_DISALLOW_SUPER
163 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
164 {
165 return NULL;
166 }
167 }
168 else if (UNICODE_IS_NONCHAR(uv)) {
169 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 170 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
171 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
172 uv);
173 }
174 if (flags & UNICODE_DISALLOW_NONCHAR) {
175 return NULL;
176 }
177 }
507b9800 178 }
c4d5f83a 179 if (UNI_IS_INVARIANT(uv)) {
eb160463 180 *d++ = (U8)UTF_TO_NATIVE(uv);
a0ed51b3
LW
181 return d;
182 }
2d331972 183#if defined(EBCDIC)
1d72bdf6
NIS
184 else {
185 STRLEN len = UNISKIP(uv);
186 U8 *p = d+len-1;
187 while (p > d) {
eb160463 188 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
189 uv >>= UTF_ACCUMULATION_SHIFT;
190 }
eb160463 191 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
192 return d+len;
193 }
194#else /* Non loop style */
a0ed51b3 195 if (uv < 0x800) {
eb160463
GS
196 *d++ = (U8)(( uv >> 6) | 0xc0);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
198 return d;
199 }
200 if (uv < 0x10000) {
eb160463
GS
201 *d++ = (U8)(( uv >> 12) | 0xe0);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
204 return d;
205 }
206 if (uv < 0x200000) {
eb160463
GS
207 *d++ = (U8)(( uv >> 18) | 0xf0);
208 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
209 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
210 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
211 return d;
212 }
213 if (uv < 0x4000000) {
eb160463
GS
214 *d++ = (U8)(( uv >> 24) | 0xf8);
215 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
218 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
219 return d;
220 }
221 if (uv < 0x80000000) {
eb160463
GS
222 *d++ = (U8)(( uv >> 30) | 0xfc);
223 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
227 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
228 return d;
229 }
6b8eaf93 230#ifdef HAS_QUAD
d7578b48 231 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
232#endif
233 {
eb160463
GS
234 *d++ = 0xfe; /* Can't match U+FEFF! */
235 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
239 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
240 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
241 return d;
242 }
6b8eaf93 243#ifdef HAS_QUAD
a0ed51b3 244 {
eb160463
GS
245 *d++ = 0xff; /* Can't match U+FFFE! */
246 *d++ = 0x80; /* 6 Reserved bits */
247 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
248 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
257 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
258 return d;
259 }
260#endif
1d72bdf6 261#endif /* Loop style */
a0ed51b3 262}
9041c2e3 263
646ca15d
JH
264/*
265
f7d739d1 266Tests if the first C<len> bytes of string C<s> form a valid UTF-8
646ca15d 267character. Note that an INVARIANT (i.e. ASCII) character is a valid
f7d739d1 268UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
269will be returned if it is valid, otherwise 0.
270
271This is the "slow" version as opposed to the "fast" version which is
272the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
273difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
274or less you should use the IS_UTF8_CHAR(), for lengths of five or more
275you should use the _slow(). In practice this means that the _slow()
276will be used very rarely, since the maximum Unicode code point (as of
277Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
278the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
279five bytes or more.
280
281=cut */
c053b435 282STATIC STRLEN
5f66b61c 283S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 284{
cd7e6c88 285 dTHX; /* The function called below requires thread context */
646ca15d 286
cd7e6c88 287 STRLEN actual_len;
646ca15d 288
cd7e6c88 289 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 290
cd7e6c88 291 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 292
cd7e6c88 293 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 294}
9041c2e3
NIS
295
296/*
492a624f
KW
297=for apidoc is_utf8_char_buf
298
299Returns the number of bytes that comprise the first UTF-8 encoded character in
300buffer C<buf>. C<buf_end> should point to one position beyond the end of the
301buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
302encoded character.
303
304Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
305machines) is a valid UTF-8 character.
306
307=cut */
308
309STRLEN
310Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
311{
312
313 STRLEN len;
314
315 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
316
317 if (buf_end <= buf) {
318 return 0;
319 }
320
321 len = buf_end - buf;
322 if (len > UTF8SKIP(buf)) {
323 len = UTF8SKIP(buf);
324 }
325
326#ifdef IS_UTF8_CHAR
327 if (IS_UTF8_CHAR_FAST(len))
328 return IS_UTF8_CHAR(buf, len) ? len : 0;
329#endif /* #ifdef IS_UTF8_CHAR */
330 return is_utf8_char_slow(buf, len);
331}
332
333/*
87cea99e 334=for apidoc is_utf8_char
eebe1485 335
76848387
KW
336DEPRECATED!
337
5da9da9e 338Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
339character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
340character is a valid UTF-8 character. The actual number of bytes in the UTF-8
341character will be returned if it is valid, otherwise 0.
9041c2e3 342
76848387 343This function is deprecated due to the possibility that malformed input could
a1433954 344cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 345instead.
e0328548 346
82686b01 347=cut */
76848387 348
067a85ef 349STRLEN
668b6d8d 350Perl_is_utf8_char(const U8 *s)
386d01d6 351{
7918f24d 352 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 353
76848387 354 /* Assumes we have enough space, which is why this is deprecated */
492a624f 355 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
356}
357
eaf7a4d2 358
6662521e 359/*
87cea99e 360=for apidoc is_utf8_string
6662521e 361
a1433954 362Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 363UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
364using C<strlen(s)> (which means if you use this option, that C<s> has to have a
365terminating NUL byte). Note that all characters being ASCII constitute 'a
366valid UTF-8 string'.
6662521e 367
a1433954 368See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 369
6662521e
GS
370=cut
371*/
372
8e84507e 373bool
668b6d8d 374Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 375{
35da51f7 376 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 377 const U8* x = s;
067a85ef 378
7918f24d 379 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 380
6662521e 381 while (x < send) {
1acdb0da 382 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
383 if (UTF8_IS_INVARIANT(*x)) {
384 x++;
385 }
1acdb0da 386 else if (!UTF8_IS_START(*x))
e0328548 387 return FALSE;
1acdb0da
JH
388 else {
389 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
390 const STRLEN c = UTF8SKIP(x);
391 const U8* const next_char_ptr = x + c;
392
393 if (next_char_ptr > send) {
394 return FALSE;
395 }
396
768c67ee
JH
397 if (IS_UTF8_CHAR_FAST(c)) {
398 if (!IS_UTF8_CHAR(x, c))
e0328548 399 return FALSE;
3c614e38 400 }
e0328548
KW
401 else if (! is_utf8_char_slow(x, c)) {
402 return FALSE;
403 }
404 x = next_char_ptr;
1acdb0da 405 }
6662521e 406 }
768c67ee 407
067a85ef 408 return TRUE;
6662521e
GS
409}
410
67e989fb 411/*
814fafa7
NC
412Implemented as a macro in utf8.h
413
87cea99e 414=for apidoc is_utf8_string_loc
814fafa7 415
a1433954
KW
416Like L</is_utf8_string> but stores the location of the failure (in the
417case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
418"utf8ness success") in the C<ep>.
419
a1433954 420See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 421
87cea99e 422=for apidoc is_utf8_string_loclen
81cd54e3 423
a1433954
KW
424Like L</is_utf8_string>() but stores the location of the failure (in the
425case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
426"utf8ness success") in the C<ep>, and the number of UTF-8
427encoded characters in the C<el>.
428
a1433954 429See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
430
431=cut
432*/
433
434bool
668b6d8d 435Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 436{
35da51f7 437 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 438 const U8* x = s;
81cd54e3 439 STRLEN c;
3ebfea28 440 STRLEN outlen = 0;
7918f24d
NC
441
442 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 443
81cd54e3 444 while (x < send) {
e0328548
KW
445 const U8* next_char_ptr;
446
81cd54e3
JH
447 /* Inline the easy bits of is_utf8_char() here for speed... */
448 if (UTF8_IS_INVARIANT(*x))
e0328548 449 next_char_ptr = x + 1;
768c67ee
JH
450 else if (!UTF8_IS_START(*x))
451 goto out;
81cd54e3 452 else {
768c67ee 453 /* ... and call is_utf8_char() only if really needed. */
768c67ee 454 c = UTF8SKIP(x);
e0328548
KW
455 next_char_ptr = c + x;
456 if (next_char_ptr > send) {
457 goto out;
458 }
768c67ee
JH
459 if (IS_UTF8_CHAR_FAST(c)) {
460 if (!IS_UTF8_CHAR(x, c))
461 c = 0;
462 } else
463 c = is_utf8_char_slow(x, c);
768c67ee
JH
464 if (!c)
465 goto out;
81cd54e3 466 }
e0328548 467 x = next_char_ptr;
3ebfea28 468 outlen++;
81cd54e3 469 }
768c67ee
JH
470
471 out:
3ebfea28
AL
472 if (el)
473 *el = outlen;
474
768c67ee
JH
475 if (ep)
476 *ep = x;
3ebfea28 477 return (x == send);
81cd54e3
JH
478}
479
480/*
768c67ee 481
87cea99e 482=for apidoc utf8n_to_uvuni
67e989fb 483
9041c2e3 484Bottom level UTF-8 decode routine.
746afd53
KW
485Returns the code point value of the first character in the string C<s>,
486which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
487C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
488the length, in bytes, of that character.
949cf498
KW
489
490The value of C<flags> determines the behavior when C<s> does not point to a
491well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
492zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
493next possible position in C<s> that could begin a non-malformed character.
494Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
495
496Various ALLOW flags can be set in C<flags> to allow (and not warn on)
497individual types of malformations, such as the sequence being overlong (that
498is, when there is a shorter sequence that can express the same code point;
499overlong sequences are expressly forbidden in the UTF-8 standard due to
500potential security issues). Another malformation example is the first byte of
501a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
502flags. For allowed 0 length strings, this function returns 0; for allowed
503overlong sequences, the computed code point is returned; for all other allowed
504malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
505determinable reasonable value.
949cf498
KW
506
507The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
508flags) malformation is found. If this flag is set, the routine assumes that
509the caller will raise a warning, and this function will silently just set
510C<retlen> to C<-1> and return zero.
511
512Certain code points are considered problematic. These are Unicode surrogates,
746afd53 513Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 514By default these are considered regular code points, but certain situations
5eafe189 515warrant special handling for them. If C<flags> contains
949cf498
KW
516UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
517malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
518UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
519maximum) can be set to disallow these categories individually.
520
521The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
522UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
523for their respective categories, but otherwise the code points are considered
524valid (not malformations). To get a category to both be treated as a
525malformation and raise a warning, specify both the WARN and DISALLOW flags.
526(But note that warnings are not raised if lexically disabled nor if
527UTF8_CHECK_ONLY is also specified.)
528
529Very large code points (above 0x7FFF_FFFF) are considered more problematic than
530the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
531reasons: they requre at least 32 bits to represent them on ASCII platforms, are
532not representable at all on EBCDIC platforms, and the original UTF-8
533specification never went above this number (the current 0x10FFFF limit was
534imposed later). (The smaller ones, those that fit into 32 bits, are
535representable by a UV on ASCII platforms, but not by an IV, which means that
536the number of operations that can be performed on them is quite restricted.)
537The UTF-8 encoding on ASCII platforms for these large code points begins with a
538byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
539be treated as malformations, while allowing smaller above-Unicode code points.
540(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
541including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
542the other WARN flags, but applies just to these code points.
949cf498
KW
543
544All other code points corresponding to Unicode characters, including private
545use and those yet to be assigned, are never considered malformed and never
546warn.
67e989fb 547
ec5f19d0 548Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
9041c2e3 549
37607a96
PK
550=cut
551*/
67e989fb 552
a0ed51b3 553UV
7fc63493 554Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 555{
97aff369 556 dVAR;
d4c19fe8 557 const U8 * const s0 = s;
eb83ed87 558 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 559 U8 * send;
eb83ed87
KW
560 UV uv = *s;
561 STRLEN expectlen;
949cf498 562 SV* sv = NULL;
eb83ed87
KW
563 UV outlier_ret = 0; /* return value when input is in error or problematic
564 */
565 UV pack_warn = 0; /* Save result of packWARN() for later */
566 bool unexpected_non_continuation = FALSE;
567 bool overflowed = FALSE;
2f8f112e 568 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 569
eb83ed87 570 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 571
eb83ed87 572 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
a0dbb045 573
eb83ed87
KW
574 /* The order of malformation tests here is important. We should consume as
575 * few bytes as possible in order to not skip any valid character. This is
576 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
577 * http://unicode.org/reports/tr36 for more discussion as to why. For
578 * example, once we've done a UTF8SKIP, we can tell the expected number of
579 * bytes, and could fail right off the bat if the input parameters indicate
580 * that there are too few available. But it could be that just that first
581 * byte is garbled, and the intended character occupies fewer bytes. If we
582 * blindly assumed that the first byte is correct, and skipped based on
583 * that number, we could skip over a valid input character. So instead, we
584 * always examine the sequence byte-by-byte.
585 *
586 * We also should not consume too few bytes, otherwise someone could inject
587 * things. For example, an input could be deliberately designed to
588 * overflow, and if this code bailed out immediately upon discovering that,
589 * returning to the caller *retlen pointing to the very next byte (one
590 * which is actually part of of the overflowing sequence), that could look
591 * legitimate to the caller, which could discard the initial partial
592 * sequence and process the rest, inappropriately */
593
594 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 595 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
596 if (retlen) {
597 *retlen = 0;
598 }
a0dbb045 599
eb83ed87
KW
600 if (flags & UTF8_ALLOW_EMPTY) {
601 return 0;
602 }
603 if (! (flags & UTF8_CHECK_ONLY)) {
604 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
605 }
0c443dc2
JH
606 goto malformed;
607 }
608
eb83ed87
KW
609 expectlen = UTF8SKIP(s);
610
611 /* A well-formed UTF-8 character, as the vast majority of calls to this
612 * function will be for, has this expected length. For efficiency, set
613 * things up here to return it. It will be overriden only in those rare
614 * cases where a malformation is found */
615 if (retlen) {
616 *retlen = expectlen;
617 }
618
619 /* An invariant is trivially well-formed */
1d72bdf6 620 if (UTF8_IS_INVARIANT(uv)) {
c4d5f83a 621 return (UV) (NATIVE_TO_UTF(*s));
a0ed51b3 622 }
67e989fb 623
eb83ed87 624 /* A continuation character can't start a valid sequence */
b5b9af04 625 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
626 if (flags & UTF8_ALLOW_CONTINUATION) {
627 if (retlen) {
628 *retlen = 1;
629 }
630 return UNICODE_REPLACEMENT;
631 }
ba210ebe 632
eb83ed87
KW
633 if (! (flags & UTF8_CHECK_ONLY)) {
634 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
635 }
636 curlen = 1;
ba210ebe
JH
637 goto malformed;
638 }
9041c2e3 639
1d72bdf6 640#ifdef EBCDIC
75383841 641 uv = NATIVE_TO_UTF(uv);
1d72bdf6
NIS
642#endif
643
eb83ed87
KW
644 /* Here is not a continuation byte, nor an invariant. The only thing left
645 * is a start byte (possibly for an overlong) */
ba210ebe 646
eb83ed87
KW
647 /* Remove the leading bits that indicate the number of bytes in the
648 * character's whole UTF-8 sequence, leaving just the bits that are part of
649 * the value */
650 uv &= UTF_START_MASK(expectlen);
ba210ebe 651
eb83ed87
KW
652 /* Now, loop through the remaining bytes in the character's sequence,
653 * accumulating each into the working value as we go. Be sure to not look
654 * past the end of the input string */
0b8d30e8
KW
655 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
656
eb83ed87 657 for (s = s0 + 1; s < send; s++) {
b5b9af04 658 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
659#ifndef EBCDIC /* Can't overflow in EBCDIC */
660 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
661
662 /* The original implementors viewed this malformation as more
663 * serious than the others (though I, khw, don't understand
664 * why, since other malformations also give very very wrong
665 * results), so there is no way to turn off checking for it.
666 * Set a flag, but keep going in the loop, so that we absorb
667 * the rest of the bytes that comprise the character. */
668 overflowed = TRUE;
669 overflow_byte = *s; /* Save for warning message's use */
670 }
671#endif
8850bf83 672 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
673 }
674 else {
675 /* Here, found a non-continuation before processing all expected
676 * bytes. This byte begins a new character, so quit, even if
677 * allowing this malformation. */
678 unexpected_non_continuation = TRUE;
679 break;
680 }
681 } /* End of loop through the character's bytes */
682
683 /* Save how many bytes were actually in the character */
684 curlen = s - s0;
685
686 /* The loop above finds two types of malformations: non-continuation and/or
687 * overflow. The non-continuation malformation is really a too-short
688 * malformation, as it means that the current character ended before it was
689 * expected to (being terminated prematurely by the beginning of the next
690 * character, whereas in the too-short malformation there just are too few
691 * bytes available to hold the character. In both cases, the check below
692 * that we have found the expected number of bytes would fail if executed.)
693 * Thus the non-continuation malformation is really unnecessary, being a
694 * subset of the too-short malformation. But there may be existing
695 * applications that are expecting the non-continuation type, so we retain
696 * it, and return it in preference to the too-short malformation. (If this
697 * code were being written from scratch, the two types might be collapsed
698 * into one.) I, khw, am also giving priority to returning the
699 * non-continuation and too-short malformations over overflow when multiple
700 * ones are present. I don't know of any real reason to prefer one over
701 * the other, except that it seems to me that multiple-byte errors trumps
702 * errors from a single byte */
b5b9af04 703 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
704 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
705 if (! (flags & UTF8_CHECK_ONLY)) {
706 if (curlen == 1) {
707 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
708 }
709 else {
710 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, %d bytes after start byte 0x%02x, expected %d bytes)", malformed_text, *s, (int) curlen, *s0, (int)expectlen));
a0dbb045
JH
711 }
712 }
eb83ed87
KW
713 goto malformed;
714 }
715 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
716
717 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
718 * as what the original expectations were. */
719 do_overlong_test = FALSE;
eb83ed87
KW
720 if (retlen) {
721 *retlen = curlen;
722 }
723 }
b5b9af04 724 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
725 if (! (flags & UTF8_ALLOW_SHORT)) {
726 if (! (flags & UTF8_CHECK_ONLY)) {
727 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, *s0));
a0dbb045 728 }
eb83ed87
KW
729 goto malformed;
730 }
731 uv = UNICODE_REPLACEMENT;
2f8f112e 732 do_overlong_test = FALSE;
eb83ed87
KW
733 if (retlen) {
734 *retlen = curlen;
735 }
736 }
737
738#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
741 {
742 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
743 * generation of the sv, since no warnings are raised under CHECK */
744 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
745 && ckWARN_d(WARN_UTF8))
746 {
42303544
KW
747 /* This message is deliberately not of the same syntax as the other
748 * messages for malformations, for backwards compatibility in the
749 * unlikely event that code is relying on its precise earlier text
750 */
eb83ed87
KW
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
752 pack_warn = packWARN(WARN_UTF8);
753 }
754 if (flags & UTF8_DISALLOW_FE_FF) {
755 goto malformed;
ba210ebe 756 }
ba210ebe 757 }
b5b9af04 758 if (UNLIKELY(overflowed)) {
ba210ebe 759
eb83ed87
KW
760 /* If the first byte is FF, it will overflow a 32-bit word. If the
761 * first byte is FE, it will overflow a signed 32-bit word. The
762 * above preserves backward compatibility, since its message was used
763 * in earlier versions of this code in preference to overflow */
764 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 765 goto malformed;
eb83ed87
KW
766 }
767#endif
768
2f8f112e
KW
769 if (do_overlong_test
770 && expectlen > (STRLEN)UNISKIP(uv)
771 && ! (flags & UTF8_ALLOW_LONG))
772 {
eb83ed87
KW
773 /* The overlong malformation has lower precedence than the others.
774 * Note that if this malformation is allowed, we return the actual
775 * value, instead of the replacement character. This is because this
776 * value is actually well-defined. */
777 if (! (flags & UTF8_CHECK_ONLY)) {
778 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), *s0));
779 }
780 goto malformed;
781 }
782
783 /* Here, the input is considered to be well-formed , but could be a
784 * problematic code point that is not allowed by the input parameters. */
785 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
786 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
787 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
788 {
949cf498 789 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
790 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
791 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
792 {
111d382d 793 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 794 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
795 }
796 if (flags & UTF8_DISALLOW_SURROGATE) {
797 goto disallowed;
798 }
799 }
949cf498 800 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
801 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
802 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
803 {
111d382d 804 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 805 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
806 }
807 if (flags & UTF8_DISALLOW_SUPER) {
808 goto disallowed;
809 }
810 }
4190d317
KW
811 else if (UNICODE_IS_NONCHAR(uv)) {
812 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
813 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
814 {
815 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
816 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
817 }
818 if (flags & UTF8_DISALLOW_NONCHAR) {
819 goto disallowed;
820 }
821 }
949cf498 822
eb83ed87
KW
823 if (sv) {
824 outlier_ret = uv;
825 goto do_warn;
826 }
827
949cf498
KW
828 /* Here, this is not considered a malformed character, so drop through
829 * to return it */
a0ed51b3 830 }
ba210ebe 831
a0ed51b3 832 return uv;
ba210ebe 833
eb83ed87
KW
834 /* There are three cases which get to beyond this point. In all 3 cases:
835 * <sv> if not null points to a string to print as a warning.
836 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
837 * set.
838 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
839 * This is done by initializing it to 0, and changing it only
840 * for case 1).
841 * The 3 cases are:
842 * 1) The input is valid but problematic, and to be warned about. The
843 * return value is the resultant code point; <*retlen> is set to
844 * <curlen>, the number of bytes that comprise the code point.
845 * <pack_warn> contains the result of packWARN() for the warning
846 * types. The entry point for this case is the label <do_warn>;
847 * 2) The input is a valid code point but disallowed by the parameters to
848 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
849 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
850 * comprise the code point. <pack_warn> contains the result of
851 * packWARN() for the warning types. The entry point for this case is
852 * the label <disallowed>.
853 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
854 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
855 * bytes that comprise the malformation. All such malformations are
856 * assumed to be warning type <utf8>. The entry point for this case
857 * is the label <malformed>.
858 */
949cf498 859
ba210ebe
JH
860malformed:
861
eb83ed87
KW
862 if (sv && ckWARN_d(WARN_UTF8)) {
863 pack_warn = packWARN(WARN_UTF8);
864 }
865
866disallowed:
867
fcc8fcf6 868 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 869 if (retlen)
10edeb5d 870 *retlen = ((STRLEN) -1);
ba210ebe
JH
871 return 0;
872 }
873
eb83ed87 874do_warn:
5b311467 875
eb83ed87
KW
876 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
877 if warnings are to be raised. */
f555bc63 878 const char * const string = SvPVX_const(sv);
a0dbb045 879
f555bc63
KW
880 if (PL_op)
881 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
882 else
883 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
884 }
885
eb83ed87
KW
886 if (retlen) {
887 *retlen = curlen;
888 }
ba210ebe 889
eb83ed87 890 return outlier_ret;
a0ed51b3
LW
891}
892
8e84507e 893/*
ec5f19d0
KW
894=for apidoc utf8_to_uvchr_buf
895
896Returns the native code point of the first character in the string C<s> which
897is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 898C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 899
524080c4
KW
900If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
901enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
902NULL) to -1. If those warnings are off, the computed value if well-defined (or
903the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
904is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
905next possible position in C<s> that could begin a non-malformed character.
906See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
907
908=cut
909*/
910
911
912UV
913Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
914{
915 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
916
917 assert(s < send);
918
919 return utf8n_to_uvchr(s, send - s, retlen,
920 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
921}
922
27d6c58a 923/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 924 * there are no malformations in the input UTF-8 string C<s>. surrogates,
6dd9dce9
KW
925 * non-character code points, and non-Unicode code points are allowed. A macro
926 * in utf8.h is used to normally avoid this function wrapper */
27d6c58a
KW
927
928UV
929Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
930{
3986bb7c
KW
931 const UV uv = valid_utf8_to_uvuni(s, retlen);
932
27d6c58a
KW
933 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
934
3986bb7c 935 return UNI_TO_NATIVE(uv);
27d6c58a
KW
936}
937
ec5f19d0 938/*
87cea99e 939=for apidoc utf8_to_uvchr
9041c2e3 940
977c1d31
KW
941DEPRECATED!
942
6ee84de2 943Returns the native code point of the first character in the string C<s>
1e54db1a 944which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
945length, in bytes, of that character.
946
4b88fb76 947Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
948malformed input could cause reading beyond the end of the input buffer, which
949is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 950
524080c4
KW
951If C<s> points to one of the detected malformations, and UTF8 warnings are
952enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
953NULL) to -1. If those warnings are off, the computed value if well-defined (or
954the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
955is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
956next possible position in C<s> that could begin a non-malformed character.
957See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
958
959=cut
960*/
961
962UV
7fc63493 963Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 964{
7918f24d
NC
965 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
966
2ff6c191 967 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
968}
969
970/*
ec5f19d0
KW
971=for apidoc utf8_to_uvuni_buf
972
973Returns the Unicode code point of the first character in the string C<s> which
974is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
975C<retlen> will be set to the length, in bytes, of that character.
976
977This function should only be used when the returned UV is considered
978an index into the Unicode semantic tables (e.g. swashes).
979
524080c4
KW
980If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
981enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
982NULL) to -1. If those warnings are off, the computed value if well-defined (or
983the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
984is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
985next possible position in C<s> that could begin a non-malformed character.
986See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
987
988=cut
989*/
990
991UV
992Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
993{
994 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
995
996 assert(send > s);
997
998 /* Call the low level routine asking for checks */
999 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1000 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1001}
1002
27d6c58a 1003/* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1004 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1005 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1006
1007UV
1008Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1009{
3986bb7c
KW
1010 UV expectlen = UTF8SKIP(s);
1011 const U8* send = s + expectlen;
1012 UV uv = NATIVE_TO_UTF(*s);
1013
27d6c58a
KW
1014 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1015
3986bb7c
KW
1016 if (retlen) {
1017 *retlen = expectlen;
1018 }
1019
1020 /* An invariant is trivially returned */
1021 if (expectlen == 1) {
1022 return uv;
1023 }
1024
1025 /* Remove the leading bits that indicate the number of bytes, leaving just
1026 * the bits that are part of the value */
1027 uv &= UTF_START_MASK(expectlen);
1028
1029 /* Now, loop through the remaining bytes, accumulating each into the
08bc774e
KW
1030 * working total as we go. (I khw tried unrolling the loop for up to 4
1031 * bytes, but there was no performance improvement) */
3986bb7c
KW
1032 for (++s; s < send; s++) {
1033 uv = UTF8_ACCUMULATE(uv, *s);
1034 }
1035
1036 return uv;
27d6c58a
KW
1037}
1038
ec5f19d0 1039/*
87cea99e 1040=for apidoc utf8_to_uvuni
9041c2e3 1041
977c1d31
KW
1042DEPRECATED!
1043
9041c2e3 1044Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1045which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1046length, in bytes, of that character.
1047
39e518fd
KW
1048This function should only be used when the returned UV is considered
1049an index into the Unicode semantic tables (e.g. swashes).
1050
4b88fb76 1051Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1052malformed input could cause reading beyond the end of the input buffer, which
1053is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
9041c2e3 1054
524080c4
KW
1055If C<s> points to one of the detected malformations, and UTF8 warnings are
1056enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1057NULL) to -1. If those warnings are off, the computed value if well-defined (or
1058the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1059is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1060next possible position in C<s> that could begin a non-malformed character.
1061See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1062
1063=cut
1064*/
1065
1066UV
7fc63493 1067Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1068{
7918f24d
NC
1069 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1070
4b88fb76 1071 return valid_utf8_to_uvuni(s, retlen);
8e84507e
NIS
1072}
1073
b76347f2 1074/*
87cea99e 1075=for apidoc utf8_length
b76347f2
JH
1076
1077Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1078Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1079up past C<e>, croaks.
b76347f2
JH
1080
1081=cut
1082*/
1083
1084STRLEN
35a4481c 1085Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1086{
97aff369 1087 dVAR;
b76347f2
JH
1088 STRLEN len = 0;
1089
7918f24d
NC
1090 PERL_ARGS_ASSERT_UTF8_LENGTH;
1091
8850bf83
JH
1092 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1093 * the bitops (especially ~) can create illegal UTF-8.
1094 * In other words: in Perl UTF-8 is not just for Unicode. */
1095
a3b680e6
AL
1096 if (e < s)
1097 goto warn_and_return;
b76347f2 1098 while (s < e) {
4cbf4130 1099 s += UTF8SKIP(s);
8e91ec7f
AV
1100 len++;
1101 }
1102
1103 if (e != s) {
1104 len--;
1105 warn_and_return:
9b387841
NC
1106 if (PL_op)
1107 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1108 "%s in %s", unees, OP_DESC(PL_op));
1109 else
61a12c31 1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1111 }
1112
1113 return len;
1114}
1115
b06226ff 1116/*
87cea99e 1117=for apidoc utf8_distance
b06226ff 1118
1e54db1a 1119Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1120and C<b>.
1121
1122WARNING: use only if you *know* that the pointers point inside the
1123same UTF-8 buffer.
1124
37607a96
PK
1125=cut
1126*/
a0ed51b3 1127
02eb7b47 1128IV
35a4481c 1129Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1130{
7918f24d
NC
1131 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1132
bf1665bc 1133 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1134}
1135
b06226ff 1136/*
87cea99e 1137=for apidoc utf8_hop
b06226ff 1138
8850bf83
JH
1139Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1140forward or backward.
b06226ff
JH
1141
1142WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1143the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1144on the first byte of character or just after the last byte of a character.
b06226ff 1145
37607a96
PK
1146=cut
1147*/
a0ed51b3
LW
1148
1149U8 *
4373e329 1150Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1151{
7918f24d
NC
1152 PERL_ARGS_ASSERT_UTF8_HOP;
1153
96a5add6 1154 PERL_UNUSED_CONTEXT;
8850bf83
JH
1155 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1156 * the bitops (especially ~) can create illegal UTF-8.
1157 * In other words: in Perl UTF-8 is not just for Unicode. */
1158
a0ed51b3
LW
1159 if (off >= 0) {
1160 while (off--)
1161 s += UTF8SKIP(s);
1162 }
1163 else {
1164 while (off++) {
1165 s--;
8850bf83
JH
1166 while (UTF8_IS_CONTINUATION(*s))
1167 s--;
a0ed51b3
LW
1168 }
1169 }
4373e329 1170 return (U8 *)s;
a0ed51b3
LW
1171}
1172
6940069f 1173/*
fed3ba5d
NC
1174=for apidoc bytes_cmp_utf8
1175
a1433954
KW
1176Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1177sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1178equal, -1 or -2 if the first string is less than the second string, +1 or +2
1179if the first string is greater than the second string.
1180
1181-1 or +1 is returned if the shorter string was identical to the start of the
1182longer string. -2 or +2 is returned if the was a difference between characters
1183within the strings.
1184
1185=cut
1186*/
1187
1188int
1189Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1190{
1191 const U8 *const bend = b + blen;
1192 const U8 *const uend = u + ulen;
1193
1194 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1195
1196 PERL_UNUSED_CONTEXT;
1197
1198 while (b < bend && u < uend) {
1199 U8 c = *u++;
1200 if (!UTF8_IS_INVARIANT(c)) {
1201 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1202 if (u < uend) {
1203 U8 c1 = *u++;
1204 if (UTF8_IS_CONTINUATION(c1)) {
356979f4 1205 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
fed3ba5d
NC
1206 } else {
1207 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1208 "Malformed UTF-8 character "
1209 "(unexpected non-continuation byte 0x%02x"
1210 ", immediately after start byte 0x%02x)"
1211 /* Dear diag.t, it's in the pod. */
1212 "%s%s", c1, c,
1213 PL_op ? " in " : "",
1214 PL_op ? OP_DESC(PL_op) : "");
1215 return -2;
1216 }
1217 } else {
1218 if (PL_op)
1219 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1220 "%s in %s", unees, OP_DESC(PL_op));
1221 else
61a12c31 1222 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1223 return -2; /* Really want to return undef :-) */
1224 }
1225 } else {
1226 return -2;
1227 }
1228 }
1229 if (*b != c) {
1230 return *b < c ? -2 : +2;
1231 }
1232 ++b;
1233 }
1234
1235 if (b == bend && u == uend)
1236 return 0;
1237
1238 return b < bend ? +1 : -1;
1239}
1240
1241/*
87cea99e 1242=for apidoc utf8_to_bytes
6940069f 1243
2bbc8d55 1244Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1245Unlike L</bytes_to_utf8>, this over-writes the original string, and
1246updates C<len> to contain the new length.
67e989fb 1247Returns zero on failure, setting C<len> to -1.
6940069f 1248
a1433954 1249If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1250
6940069f
GS
1251=cut
1252*/
1253
1254U8 *
37607a96 1255Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1256{
d4c19fe8
AL
1257 U8 * const save = s;
1258 U8 * const send = s + *len;
6940069f 1259 U8 *d;
246fae53 1260
7918f24d
NC
1261 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1262
1e54db1a 1263 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1264 while (s < send) {
dcad2880
JH
1265 U8 c = *s++;
1266
1d72bdf6
NIS
1267 if (!UTF8_IS_INVARIANT(c) &&
1268 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1269 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
10edeb5d 1270 *len = ((STRLEN) -1);
dcad2880
JH
1271 return 0;
1272 }
246fae53 1273 }
dcad2880
JH
1274
1275 d = s = save;
6940069f 1276 while (s < send) {
ed646e6e 1277 STRLEN ulen;
4b88fb76 1278 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
ed646e6e 1279 s += ulen;
6940069f
GS
1280 }
1281 *d = '\0';
246fae53 1282 *len = d - save;
6940069f
GS
1283 return save;
1284}
1285
1286/*
87cea99e 1287=for apidoc bytes_from_utf8
f9a63242 1288
2bbc8d55 1289Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1290Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1291the newly-created string, and updates C<len> to contain the new
1292length. Returns the original string if no conversion occurs, C<len>
1293is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
12940 if C<s> is converted or consisted entirely of characters that are invariant
1295in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1296
37607a96
PK
1297=cut
1298*/
f9a63242
JH
1299
1300U8 *
e1ec3a88 1301Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1302{
f9a63242 1303 U8 *d;
e1ec3a88
AL
1304 const U8 *start = s;
1305 const U8 *send;
f9a63242
JH
1306 I32 count = 0;
1307
7918f24d
NC
1308 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1309
96a5add6 1310 PERL_UNUSED_CONTEXT;
f9a63242 1311 if (!*is_utf8)
73d840c0 1312 return (U8 *)start;
f9a63242 1313
1e54db1a 1314 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1315 for (send = s + *len; s < send;) {
e1ec3a88 1316 U8 c = *s++;
1d72bdf6 1317 if (!UTF8_IS_INVARIANT(c)) {
db42d148
NIS
1318 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1319 (c = *s++) && UTF8_IS_CONTINUATION(c))
1320 count++;
1321 else
73d840c0 1322 return (U8 *)start;
db42d148 1323 }
f9a63242
JH
1324 }
1325
35da51f7 1326 *is_utf8 = FALSE;
f9a63242 1327
212542aa 1328 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1329 s = start; start = d;
f9a63242
JH
1330 while (s < send) {
1331 U8 c = *s++;
c4d5f83a
NIS
1332 if (!UTF8_IS_INVARIANT(c)) {
1333 /* Then it is two-byte encoded */
356979f4 1334 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
c4d5f83a
NIS
1335 }
1336 *d++ = c;
f9a63242
JH
1337 }
1338 *d = '\0';
1339 *len = d - start;
73d840c0 1340 return (U8 *)start;
f9a63242
JH
1341}
1342
1343/*
87cea99e 1344=for apidoc bytes_to_utf8
6940069f 1345
ff97e5cf
KW
1346Converts a string C<s> of length C<len> bytes from the native encoding into
1347UTF-8.
6662521e 1348Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1349reflect the new length in bytes.
6940069f 1350
2bbc8d55
SP
1351A NUL character will be written after the end of the string.
1352
1353If you want to convert to UTF-8 from encodings other than
1354the native (Latin1 or EBCDIC),
a1433954 1355see L</sv_recode_to_utf8>().
c9ada85f 1356
497711e7 1357=cut
6940069f
GS
1358*/
1359
c682ebef
FC
1360/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1361 likewise need duplication. */
1362
6940069f 1363U8*
35a4481c 1364Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1365{
35a4481c 1366 const U8 * const send = s + (*len);
6940069f
GS
1367 U8 *d;
1368 U8 *dst;
7918f24d
NC
1369
1370 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1371 PERL_UNUSED_CONTEXT;
6940069f 1372
212542aa 1373 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1374 dst = d;
1375
1376 while (s < send) {
35a4481c 1377 const UV uv = NATIVE_TO_ASCII(*s++);
c4d5f83a 1378 if (UNI_IS_INVARIANT(uv))
eb160463 1379 *d++ = (U8)UTF_TO_NATIVE(uv);
6940069f 1380 else {
eb160463
GS
1381 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1382 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
6940069f
GS
1383 }
1384 }
1385 *d = '\0';
6662521e 1386 *len = d-dst;
6940069f
GS
1387 return dst;
1388}
1389
a0ed51b3 1390/*
dea0fc0b 1391 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1392 *
1393 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1394 * We optimize for native, for obvious reasons. */
1395
1396U8*
dea0fc0b 1397Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1398{
dea0fc0b
JH
1399 U8* pend;
1400 U8* dstart = d;
1401
7918f24d
NC
1402 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1403
dea0fc0b 1404 if (bytelen & 1)
f5992bc4 1405 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1406
1407 pend = p + bytelen;
1408
a0ed51b3 1409 while (p < pend) {
dea0fc0b
JH
1410 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1411 p += 2;
a0ed51b3 1412 if (uv < 0x80) {
e294cc5d
JH
1413#ifdef EBCDIC
1414 *d++ = UNI_TO_NATIVE(uv);
1415#else
eb160463 1416 *d++ = (U8)uv;
e294cc5d 1417#endif
a0ed51b3
LW
1418 continue;
1419 }
1420 if (uv < 0x800) {
eb160463
GS
1421 *d++ = (U8)(( uv >> 6) | 0xc0);
1422 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1423 continue;
1424 }
52b9aa85 1425 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
01ea242b 1426 if (p >= pend) {
dea0fc0b 1427 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1428 } else {
1429 UV low = (p[0] << 8) + p[1];
1430 p += 2;
52b9aa85 1431 if (low < 0xdc00 || low > 0xdfff)
01ea242b
NC
1432 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1433 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1434 }
dbde1951
NC
1435 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1436 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3
LW
1437 }
1438 if (uv < 0x10000) {
eb160463
GS
1439 *d++ = (U8)(( uv >> 12) | 0xe0);
1440 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1441 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1442 continue;
1443 }
1444 else {
eb160463
GS
1445 *d++ = (U8)(( uv >> 18) | 0xf0);
1446 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1447 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1448 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1449 continue;
1450 }
1451 }
dea0fc0b 1452 *newlen = d - dstart;
a0ed51b3
LW
1453 return d;
1454}
1455
1456/* Note: this one is slightly destructive of the source. */
1457
1458U8*
dea0fc0b 1459Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1460{
1461 U8* s = (U8*)p;
d4c19fe8 1462 U8* const send = s + bytelen;
7918f24d
NC
1463
1464 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1465
e0ea5e2d
NC
1466 if (bytelen & 1)
1467 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1468 (UV)bytelen);
1469
a0ed51b3 1470 while (s < send) {
d4c19fe8 1471 const U8 tmp = s[0];
a0ed51b3
LW
1472 s[0] = s[1];
1473 s[1] = tmp;
1474 s += 2;
1475 }
dea0fc0b 1476 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1477}
1478
c3fd2246
KW
1479/* for now these are all defined (inefficiently) in terms of the utf8 versions.
1480 * Note that the macros in handy.h that call these short-circuit calling them
1481 * for Latin-1 range inputs */
a0ed51b3
LW
1482
1483bool
84afefe6 1484Perl_is_uni_alnum(pTHX_ UV c)
a0ed51b3 1485{
89ebb4a3 1486 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1487 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1488 return is_utf8_alnum(tmpbuf);
1489}
1490
1491bool
84afefe6 1492Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1493{
89ebb4a3 1494 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1495 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1496 return is_utf8_idfirst(tmpbuf);
1497}
1498
1499bool
84afefe6 1500Perl_is_uni_alpha(pTHX_ UV c)
a0ed51b3 1501{
89ebb4a3 1502 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1503 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1504 return is_utf8_alpha(tmpbuf);
1505}
1506
1507bool
84afefe6 1508Perl_is_uni_ascii(pTHX_ UV c)
4d61ec05 1509{
bc39fe24 1510 return isASCII(c);
4d61ec05
GS
1511}
1512
1513bool
bdd8600f
KW
1514Perl_is_uni_blank(pTHX_ UV c)
1515{
1516 U8 tmpbuf[UTF8_MAXBYTES+1];
1517 uvchr_to_utf8(tmpbuf, c);
1518 return is_utf8_blank(tmpbuf);
1519}
1520
1521bool
84afefe6 1522Perl_is_uni_space(pTHX_ UV c)
a0ed51b3 1523{
89ebb4a3 1524 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1525 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1526 return is_utf8_space(tmpbuf);
1527}
1528
1529bool
84afefe6 1530Perl_is_uni_digit(pTHX_ UV c)
a0ed51b3 1531{
89ebb4a3 1532 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1533 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1534 return is_utf8_digit(tmpbuf);
1535}
1536
1537bool
84afefe6 1538Perl_is_uni_upper(pTHX_ UV c)
a0ed51b3 1539{
89ebb4a3 1540 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1541 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1542 return is_utf8_upper(tmpbuf);
1543}
1544
1545bool
84afefe6 1546Perl_is_uni_lower(pTHX_ UV c)
a0ed51b3 1547{
89ebb4a3 1548 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1549 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1550 return is_utf8_lower(tmpbuf);
1551}
1552
1553bool
84afefe6 1554Perl_is_uni_cntrl(pTHX_ UV c)
b8c5462f 1555{
7b952154 1556 return isCNTRL_L1(c);
b8c5462f
JH
1557}
1558
1559bool
84afefe6 1560Perl_is_uni_graph(pTHX_ UV c)
b8c5462f 1561{
89ebb4a3 1562 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1563 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1564 return is_utf8_graph(tmpbuf);
1565}
1566
1567bool
84afefe6 1568Perl_is_uni_print(pTHX_ UV c)
a0ed51b3 1569{
89ebb4a3 1570 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1571 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1572 return is_utf8_print(tmpbuf);
1573}
1574
b8c5462f 1575bool
84afefe6 1576Perl_is_uni_punct(pTHX_ UV c)
b8c5462f 1577{
89ebb4a3 1578 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1579 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1580 return is_utf8_punct(tmpbuf);
1581}
1582
4d61ec05 1583bool
84afefe6 1584Perl_is_uni_xdigit(pTHX_ UV c)
4d61ec05 1585{
89ebb4a3 1586 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
230880c1 1587 uvchr_to_utf8(tmpbuf, c);
4d61ec05
GS
1588 return is_utf8_xdigit(tmpbuf);
1589}
1590
3a4c58c9
KW
1591UV
1592Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1593{
1594 /* We have the latin1-range values compiled into the core, so just use
1595 * those, converting the result to utf8. The only difference between upper
1596 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1597 * either "SS" or "Ss". Which one to use is passed into the routine in
1598 * 'S_or_s' to avoid a test */
1599
1600 UV converted = toUPPER_LATIN1_MOD(c);
1601
1602 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1603
1604 assert(S_or_s == 'S' || S_or_s == 's');
1605
1606 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1607 characters in this range */
1608 *p = (U8) converted;
1609 *lenp = 1;
1610 return converted;
1611 }
1612
1613 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1614 * which it maps to one of them, so as to only have to have one check for
1615 * it in the main case */
1616 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1617 switch (c) {
1618 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1619 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1620 break;
1621 case MICRO_SIGN:
1622 converted = GREEK_CAPITAL_LETTER_MU;
1623 break;
1624 case LATIN_SMALL_LETTER_SHARP_S:
1625 *(p)++ = 'S';
1626 *p = S_or_s;
1627 *lenp = 2;
1628 return 'S';
1629 default:
1630 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1631 assert(0); /* NOTREACHED */
3a4c58c9
KW
1632 }
1633 }
1634
1635 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1636 *p = UTF8_TWO_BYTE_LO(converted);
1637 *lenp = 2;
1638
1639 return converted;
1640}
1641
50bda2c3
KW
1642/* Call the function to convert a UTF-8 encoded character to the specified case.
1643 * Note that there may be more than one character in the result.
1644 * INP is a pointer to the first byte of the input character
1645 * OUTP will be set to the first byte of the string of changed characters. It
1646 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1647 * LENP will be set to the length in bytes of the string of changed characters
1648 *
1649 * The functions return the ordinal of the first character in the string of OUTP */
f90a9a02
KW
1650#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1651#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1652#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
50bda2c3
KW
1653
1654/* This additionally has the input parameter SPECIALS, which if non-zero will
1655 * cause this to use the SPECIALS hash for folding (meaning get full case
1656 * folding); otherwise, when zero, this implies a simple case fold */
f90a9a02 1657#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 1658
84afefe6
JH
1659UV
1660Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1661{
3a4c58c9
KW
1662 dVAR;
1663
a1433954
KW
1664 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1665 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1666 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1667 * the changed version may be longer than the original character.
1668 *
1669 * The ordinal of the first character of the changed version is returned
1670 * (but note, as explained above, that there may be more.) */
1671
7918f24d
NC
1672 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1673
3a4c58c9
KW
1674 if (c < 256) {
1675 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1676 }
1677
0ebc6274 1678 uvchr_to_utf8(p, c);
3a4c58c9 1679 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1680}
1681
84afefe6
JH
1682UV
1683Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1684{
3a4c58c9
KW
1685 dVAR;
1686
7918f24d
NC
1687 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1688
3a4c58c9
KW
1689 if (c < 256) {
1690 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1691 }
1692
0ebc6274 1693 uvchr_to_utf8(p, c);
3a4c58c9 1694 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1695}
1696
afc16117
KW
1697STATIC U8
1698S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1699{
1700 /* We have the latin1-range values compiled into the core, so just use
1701 * those, converting the result to utf8. Since the result is always just
a1433954 1702 * one character, we allow <p> to be NULL */
afc16117
KW
1703
1704 U8 converted = toLOWER_LATIN1(c);
1705
1706 if (p != NULL) {
1707 if (UNI_IS_INVARIANT(converted)) {
1708 *p = converted;
1709 *lenp = 1;
1710 }
1711 else {
1712 *p = UTF8_TWO_BYTE_HI(converted);
1713 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1714 *lenp = 2;
1715 }
1716 }
1717 return converted;
1718}
1719
84afefe6
JH
1720UV
1721Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1722{
968c5e6a
KW
1723 dVAR;
1724
7918f24d
NC
1725 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1726
afc16117
KW
1727 if (c < 256) {
1728 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1729 }
1730
afc16117 1731 uvchr_to_utf8(p, c);
968c5e6a 1732 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1733}
1734
84afefe6 1735UV
f673fad4 1736Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
a1dde8de 1737{
a1433954 1738 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
f673fad4
KW
1739 * folding */
1740
a1dde8de
KW
1741 UV converted;
1742
1743 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1744
1745 if (c == MICRO_SIGN) {
1746 converted = GREEK_SMALL_LETTER_MU;
1747 }
1748 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1749 *(p)++ = 's';
1750 *p = 's';
1751 *lenp = 2;
1752 return 's';
1753 }
1754 else { /* In this range the fold of all other characters is their lower
1755 case */
1756 converted = toLOWER_LATIN1(c);
1757 }
1758
1759 if (UNI_IS_INVARIANT(converted)) {
1760 *p = (U8) converted;
1761 *lenp = 1;
1762 }
1763 else {
1764 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1765 *p = UTF8_TWO_BYTE_LO(converted);
1766 *lenp = 2;
1767 }
1768
1769 return converted;
1770}
1771
1772UV
a0270393 1773Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
84afefe6 1774{
4b593389 1775
a0270393
KW
1776 /* Not currently externally documented, and subject to change
1777 * <flags> bits meanings:
1778 * FOLD_FLAGS_FULL iff full folding is to be used;
1779 * FOLD_FLAGS_LOCALE iff in locale
1780 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1781 */
4b593389 1782
36bb2ab6 1783 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1784
a1dde8de 1785 if (c < 256) {
a0270393
KW
1786 UV result = _to_fold_latin1((U8) c, p, lenp,
1787 cBOOL(((flags & FOLD_FLAGS_FULL)
1788 /* If ASCII-safe, don't allow full folding,
1789 * as that could include SHARP S => ss;
1790 * otherwise there is no crossing of
1791 * ascii/non-ascii in the latin1 range */
1792 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1793 /* It is illegal for the fold to cross the 255/256 boundary under
1794 * locale; in this case return the original */
1795 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1796 ? c
1797 : result;
a1dde8de
KW
1798 }
1799
a0270393
KW
1800 /* If no special needs, just use the macro */
1801 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1802 uvchr_to_utf8(p, c);
1803 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1804 }
1805 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1806 the special flags. */
1807 U8 utf8_c[UTF8_MAXBYTES + 1];
1808 uvchr_to_utf8(utf8_c, c);
1809 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1810 }
84afefe6
JH
1811}
1812
ea317ccb
KW
1813/* for now these all assume no locale info available for Unicode > 255; and
1814 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1815 * called instead, so that these don't get called for < 255 */
a0ed51b3
LW
1816
1817bool
84afefe6 1818Perl_is_uni_alnum_lc(pTHX_ UV c)
a0ed51b3
LW
1819{
1820 return is_uni_alnum(c); /* XXX no locale support yet */
1821}
1822
1823bool
84afefe6 1824Perl_is_uni_idfirst_lc(pTHX_ UV c)
a0ed51b3
LW
1825{
1826 return is_uni_idfirst(c); /* XXX no locale support yet */
1827}
1828
1829bool
84afefe6 1830Perl_is_uni_alpha_lc(pTHX_ UV c)
a0ed51b3
LW
1831{
1832 return is_uni_alpha(c); /* XXX no locale support yet */
1833}
1834
1835bool
84afefe6 1836Perl_is_uni_ascii_lc(pTHX_ UV c)
4d61ec05
GS
1837{
1838 return is_uni_ascii(c); /* XXX no locale support yet */
1839}
1840
1841bool
bdd8600f
KW
1842Perl_is_uni_blank_lc(pTHX_ UV c)
1843{
1844 return is_uni_blank(c); /* XXX no locale support yet */
1845}
1846
1847bool
84afefe6 1848Perl_is_uni_space_lc(pTHX_ UV c)
a0ed51b3
LW
1849{
1850 return is_uni_space(c); /* XXX no locale support yet */
1851}
1852
1853bool
84afefe6 1854Perl_is_uni_digit_lc(pTHX_ UV c)
a0ed51b3
LW
1855{
1856 return is_uni_digit(c); /* XXX no locale support yet */
1857}
1858
1859bool
84afefe6 1860Perl_is_uni_upper_lc(pTHX_ UV c)
a0ed51b3
LW
1861{
1862 return is_uni_upper(c); /* XXX no locale support yet */
1863}
1864
1865bool
84afefe6 1866Perl_is_uni_lower_lc(pTHX_ UV c)
a0ed51b3
LW
1867{
1868 return is_uni_lower(c); /* XXX no locale support yet */
1869}
1870
1871bool
84afefe6 1872Perl_is_uni_cntrl_lc(pTHX_ UV c)
b8c5462f
JH
1873{
1874 return is_uni_cntrl(c); /* XXX no locale support yet */
1875}
1876
1877bool
84afefe6 1878Perl_is_uni_graph_lc(pTHX_ UV c)
b8c5462f
JH
1879{
1880 return is_uni_graph(c); /* XXX no locale support yet */
1881}
1882
1883bool
84afefe6 1884Perl_is_uni_print_lc(pTHX_ UV c)
a0ed51b3
LW
1885{
1886 return is_uni_print(c); /* XXX no locale support yet */
1887}
1888
b8c5462f 1889bool
84afefe6 1890Perl_is_uni_punct_lc(pTHX_ UV c)
b8c5462f
JH
1891{
1892 return is_uni_punct(c); /* XXX no locale support yet */
1893}
1894
4d61ec05 1895bool
84afefe6 1896Perl_is_uni_xdigit_lc(pTHX_ UV c)
4d61ec05
GS
1897{
1898 return is_uni_xdigit(c); /* XXX no locale support yet */
1899}
1900
b7ac61fa
JH
1901U32
1902Perl_to_uni_upper_lc(pTHX_ U32 c)
1903{
ee099d14
JH
1904 /* XXX returns only the first character -- do not use XXX */
1905 /* XXX no locale support yet */
1906 STRLEN len;
89ebb4a3 1907 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1908 return (U32)to_uni_upper(c, tmpbuf, &len);
b7ac61fa
JH
1909}
1910
1911U32
1912Perl_to_uni_title_lc(pTHX_ U32 c)
1913{
ee099d14
JH
1914 /* XXX returns only the first character XXX -- do not use XXX */
1915 /* XXX no locale support yet */
1916 STRLEN len;
89ebb4a3 1917 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1918 return (U32)to_uni_title(c, tmpbuf, &len);
b7ac61fa
JH
1919}
1920
1921U32
1922Perl_to_uni_lower_lc(pTHX_ U32 c)
1923{
ee099d14
JH
1924 /* XXX returns only the first character -- do not use XXX */
1925 /* XXX no locale support yet */
1926 STRLEN len;
89ebb4a3 1927 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1928 return (U32)to_uni_lower(c, tmpbuf, &len);
b7ac61fa
JH
1929}
1930
7452cf6a 1931static bool
5141f98e 1932S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
1933 const char *const swashname)
1934{
ea317ccb
KW
1935 /* returns a boolean giving whether or not the UTF8-encoded character that
1936 * starts at <p> is in the swash indicated by <swashname>. <swash>
1937 * contains a pointer to where the swash indicated by <swashname>
1938 * is to be stored; which this routine will do, so that future calls will
1939 * look at <*swash> and only generate a swash if it is not null
1940 *
1941 * Note that it is assumed that the buffer length of <p> is enough to
1942 * contain all the bytes that comprise the character. Thus, <*p> should
1943 * have been checked before this call for mal-formedness enough to assure
1944 * that. */
1945
97aff369 1946 dVAR;
7918f24d
NC
1947
1948 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1949
492a624f
KW
1950 /* The API should have included a length for the UTF-8 character in <p>,
1951 * but it doesn't. We therefor assume that p has been validated at least
1952 * as far as there being enough bytes available in it to accommodate the
1953 * character without reading beyond the end, and pass that number on to the
1954 * validating routine */
1955 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
bde6a22d 1956 return FALSE;
87367d5f
KW
1957 if (!*swash) {
1958 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1959 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
1960 }
bde6a22d
NC
1961 return swash_fetch(*swash, p, TRUE) != 0;
1962}
1963
1964bool
7fc63493 1965Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 1966{
97aff369 1967 dVAR;
7918f24d
NC
1968
1969 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1970
671c33bf
NC
1971 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1972 * descendant of isalnum(3), in other words, it doesn't
1973 * contain the '_'. --jhi */
d4c19fe8 1974 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
a0ed51b3
LW
1975}
1976
1977bool
7fc63493 1978Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1979{
97aff369 1980 dVAR;
7918f24d
NC
1981
1982 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1983
82686b01
JH
1984 if (*p == '_')
1985 return TRUE;
bde6a22d 1986 /* is_utf8_idstart would be more logical. */
d4c19fe8 1987 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
82686b01
JH
1988}
1989
1990bool
c11ff943
KW
1991Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1992{
1993 dVAR;
1994
1995 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1996
1997 if (*p == '_')
1998 return TRUE;
1999 /* is_utf8_idstart would be more logical. */
2000 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2001}
2002
2003bool
b6912c02
KW
2004Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2005{
2006 dVAR;
2007
2008 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2009
2010 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2011}
2012
2013bool
7fc63493 2014Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2015{
97aff369 2016 dVAR;
7918f24d
NC
2017
2018 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2019
d4c19fe8 2020 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
2021}
2022
2023bool
c11ff943
KW
2024Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2025{
2026 dVAR;
2027
2028 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2029
c11ff943
KW
2030 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2031}
2032
2033bool
7fc63493 2034Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 2035{
97aff369 2036 dVAR;
7918f24d
NC
2037
2038 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2039
d4c19fe8 2040 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
a0ed51b3
LW
2041}
2042
2043bool
7fc63493 2044Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2045{
97aff369 2046 dVAR;
7918f24d
NC
2047
2048 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2049
bc39fe24
KW
2050 /* ASCII characters are the same whether in utf8 or not. So the macro
2051 * works on both utf8 and non-utf8 representations. */
2052 return isASCII(*p);
b8c5462f
JH
2053}
2054
2055bool
bdd8600f
KW
2056Perl_is_utf8_blank(pTHX_ const U8 *p)
2057{
2058 dVAR;
2059
2060 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2061
2062 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2063}
2064
2065bool
7fc63493 2066Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2067{
97aff369 2068 dVAR;
7918f24d
NC
2069
2070 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2071
a34094a9 2072 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
a0ed51b3
LW
2073}
2074
2075bool
d1eb3177
YO
2076Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2077{
2078 dVAR;
2079
2080 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2081
c4428693
KW
2082 /* Only true if is an ASCII space-like character, and ASCII is invariant
2083 * under utf8, so can just use the macro */
2084 return isSPACE_A(*p);
d1eb3177
YO
2085}
2086
2087bool
2088Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2089{
2090 dVAR;
2091
2092 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2093
c4428693
KW
2094 /* Only true if is an ASCII word character, and ASCII is invariant
2095 * under utf8, so can just use the macro */
2096 return isWORDCHAR_A(*p);
d1eb3177
YO
2097}
2098
2099bool
7fc63493 2100Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2101{
97aff369 2102 dVAR;
7918f24d
NC
2103
2104 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2105
d4c19fe8 2106 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
a0ed51b3
LW
2107}
2108
2109bool
d1eb3177
YO
2110Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2111{
2112 dVAR;
2113
2114 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2115
c4428693
KW
2116 /* Only true if is an ASCII digit character, and ASCII is invariant
2117 * under utf8, so can just use the macro */
2118 return isDIGIT_A(*p);
d1eb3177
YO
2119}
2120
2121bool
7fc63493 2122Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2123{
97aff369 2124 dVAR;
7918f24d
NC
2125
2126 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2127
d4c19fe8 2128 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
a0ed51b3
LW
2129}
2130
2131bool
7fc63493 2132Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2133{
97aff369 2134 dVAR;
7918f24d
NC
2135
2136 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2137
d4c19fe8 2138 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
a0ed51b3
LW
2139}
2140
2141bool
7fc63493 2142Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2143{
97aff369 2144 dVAR;
7918f24d
NC
2145
2146 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2147
7b952154
KW
2148 if (isASCII(*p)) {
2149 return isCNTRL_A(*p);
2150 }
2151
2152 /* All controls are in Latin1 */
2153 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2154 return 0;
2155 }
2156 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
b8c5462f
JH
2157}
2158
2159bool
7fc63493 2160Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2161{
97aff369 2162 dVAR;
7918f24d
NC
2163
2164 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2165
d4c19fe8 2166 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
b8c5462f
JH
2167}
2168
2169bool
7fc63493 2170Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2171{
97aff369 2172 dVAR;
7918f24d
NC
2173
2174 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2175
d4c19fe8 2176 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
a0ed51b3
LW
2177}
2178
2179bool
7fc63493 2180Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2181{
97aff369 2182 dVAR;
7918f24d
NC
2183
2184 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2185
d4c19fe8 2186 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
b8c5462f
JH
2187}
2188
2189bool
7fc63493 2190Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2191{
97aff369 2192 dVAR;
7918f24d
NC
2193
2194 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2195
d1eb3177 2196 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
b8c5462f
JH
2197}
2198
2199bool
7fc63493 2200Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2201{
97aff369 2202 dVAR;
7918f24d
NC
2203
2204 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2205
d4c19fe8 2206 return is_utf8_common(p, &PL_utf8_mark, "IsM");
a0ed51b3
LW
2207}
2208
37e2e78e 2209bool
27d4fc33 2210Perl_is_utf8_X_regular_begin(pTHX_ const U8 *p)
37e2e78e
KW
2211{
2212 dVAR;
2213
27d4fc33 2214 PERL_ARGS_ASSERT_IS_UTF8_X_REGULAR_BEGIN;
37e2e78e 2215
27d4fc33 2216 return is_utf8_common(p, &PL_utf8_X_regular_begin, "_X_Regular_Begin");
37e2e78e
KW
2217}
2218
2219bool
2220Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2221{
2222 dVAR;
2223
2224 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2225
2226 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2227}
2228
6b5c0936 2229/*
87cea99e 2230=for apidoc to_utf8_case
6b5c0936 2231
a1433954
KW
2232The C<p> contains the pointer to the UTF-8 string encoding
2233the character that is being converted. This routine assumes that the character
2234at C<p> is well-formed.
6b5c0936 2235
a1433954
KW
2236The C<ustrp> is a pointer to the character buffer to put the
2237conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2238of the result.
2239
a1433954 2240The C<swashp> is a pointer to the swash to use.
6b5c0936 2241
a1433954
KW
2242Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2243and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2244but not always, a multicharacter mapping), is tried first.
6b5c0936 2245
a1433954 2246The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2247hash %utf8::ToSpecLower. The access to the hash is through
2248Perl_to_utf8_case().
6b5c0936 2249
a1433954 2250The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2251%utf8::ToLower.
2252
2253=cut */
6b5c0936 2254
2104c8d9 2255UV
9a957fbc
AL
2256Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2257 SV **swashp, const char *normal, const char *special)
a0ed51b3 2258{
97aff369 2259 dVAR;
89ebb4a3 2260 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
0134edef 2261 STRLEN len = 0;
4b88fb76 2262 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
1feea2c7
JH
2263 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2264 * are necessary in EBCDIC, they are redundant no-ops
2265 * in ASCII-ish platforms, and hopefully optimized away. */
f54cb97a 2266 const UV uv1 = NATIVE_TO_UNI(uv0);
7918f24d
NC
2267
2268 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2269
9ae3ac1a
KW
2270 /* Note that swash_fetch() doesn't output warnings for these because it
2271 * assumes we will */
8457b38f 2272 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2273 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2274 if (ckWARN_d(WARN_SURROGATE)) {
2275 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2276 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2277 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2278 }
9ae3ac1a
KW
2279 }
2280 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2281 if (ckWARN_d(WARN_NON_UNICODE)) {
2282 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2283 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2284 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2285 }
9ae3ac1a
KW
2286 }
2287
2288 /* Note that non-characters are perfectly legal, so no warning should
2289 * be given */
2290 }
2291
1feea2c7 2292 uvuni_to_utf8(tmpbuf, uv1);
0134edef
JH
2293
2294 if (!*swashp) /* load on-demand */
5ab9d2ef 2295 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2296
a6f87d8c 2297 if (special) {
0134edef 2298 /* It might be "special" (sometimes, but not always,
2a37f04d 2299 * a multicharacter mapping) */
6673a63c 2300 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2301 SV **svp;
2302
35da51f7 2303 if (hv &&
b08cf34e
JH
2304 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2305 (*svp)) {
cfd0369c 2306 const char *s;
47654450 2307
cfd0369c 2308 s = SvPV_const(*svp, len);
47654450
JH
2309 if (len == 1)
2310 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2a37f04d 2311 else {
2f9475ad
JH
2312#ifdef EBCDIC
2313 /* If we have EBCDIC we need to remap the characters
2314 * since any characters in the low 256 are Unicode
2315 * code points, not EBCDIC. */
7cda7a3d 2316 U8 *t = (U8*)s, *tend = t + len, *d;
2f9475ad
JH
2317
2318 d = tmpbuf;
b08cf34e 2319 if (SvUTF8(*svp)) {
2f9475ad
JH
2320 STRLEN tlen = 0;
2321
2322 while (t < tend) {
6bd1c396 2323 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2f9475ad
JH
2324 if (tlen > 0) {
2325 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2326 t += tlen;
2327 }
2328 else
2329 break;
2330 }
2331 }
2332 else {
36fec512
JH
2333 while (t < tend) {
2334 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2335 t++;
2336 }
2f9475ad
JH
2337 }
2338 len = d - tmpbuf;
2339 Copy(tmpbuf, ustrp, len, U8);
2340#else
d2dcd0fb 2341 Copy(s, ustrp, len, U8);
2f9475ad 2342#endif
29e98929 2343 }
983ffd37 2344 }
0134edef
JH
2345 }
2346
2347 if (!len && *swashp) {
2114036c 2348 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
d4c19fe8 2349
0134edef
JH
2350 if (uv2) {
2351 /* It was "normal" (a single character mapping). */
d4c19fe8 2352 const UV uv3 = UNI_TO_NATIVE(uv2);
e9101d72 2353 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2a37f04d
JH
2354 }
2355 }
1feea2c7 2356
cbe07460
KW
2357 if (len) {
2358 if (lenp) {
2359 *lenp = len;
2360 }
2361 return valid_utf8_to_uvchr(ustrp, 0);
2362 }
2363
2364 /* Here, there was no mapping defined, which means that the code point maps
2365 * to itself. Return the inputs */
bfdf22ec
KW
2366 len = UTF8SKIP(p);
2367 Copy(p, ustrp, len, U8);
0134edef 2368
2a37f04d
JH
2369 if (lenp)
2370 *lenp = len;
2371
cbe07460
KW
2372 return uv0;
2373
a0ed51b3
LW
2374}
2375
051a06d4
KW
2376STATIC UV
2377S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2378{
2379 /* This is called when changing the case of a utf8-encoded character above
2380 * the Latin1 range, and the operation is in locale. If the result
2381 * contains a character that crosses the 255/256 boundary, disallow the
2382 * change, and return the original code point. See L<perlfunc/lc> for why;
2383 *
a1433954
KW
2384 * p points to the original string whose case was changed; assumed
2385 * by this routine to be well-formed
051a06d4
KW
2386 * result the code point of the first character in the changed-case string
2387 * ustrp points to the changed-case string (<result> represents its first char)
2388 * lenp points to the length of <ustrp> */
2389
2390 UV original; /* To store the first code point of <p> */
2391
2392 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2393
2394 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2395
2396 /* We know immediately if the first character in the string crosses the
2397 * boundary, so can skip */
2398 if (result > 255) {
2399
2400 /* Look at every character in the result; if any cross the
2401 * boundary, the whole thing is disallowed */
2402 U8* s = ustrp + UTF8SKIP(ustrp);
2403 U8* e = ustrp + *lenp;
2404 while (s < e) {
2405 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2406 {
2407 goto bad_crossing;
2408 }
2409 s += UTF8SKIP(s);
2410 }
2411
2412 /* Here, no characters crossed, result is ok as-is */
2413 return result;
2414 }
2415
2416bad_crossing:
2417
2418 /* Failed, have to return the original */
4b88fb76 2419 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2420 Copy(p, ustrp, *lenp, char);
2421 return original;
2422}
2423
d3e79532 2424/*
87cea99e 2425=for apidoc to_utf8_upper
d3e79532 2426
a1433954
KW
2427Convert the UTF-8 encoded character at C<p> to its uppercase version and
2428store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
89ebb4a3
JH
2429that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2430the uppercase version may be longer than the original character.
d3e79532
JH
2431
2432The first character of the uppercased version is returned
2433(but note, as explained above, that there may be more.)
2434
a1433954
KW
2435The character at C<p> is assumed by this routine to be well-formed.
2436
d3e79532
JH
2437=cut */
2438
051a06d4
KW
2439/* Not currently externally documented, and subject to change:
2440 * <flags> is set iff locale semantics are to be used for code points < 256
2441 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2442 * were used in the calculation; otherwise unchanged. */
2443
2104c8d9 2444UV
051a06d4 2445Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2446{
97aff369 2447 dVAR;
7918f24d 2448
051a06d4
KW
2449 UV result;
2450
2451 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2452
3a4c58c9 2453 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2454 if (flags) {
2455 result = toUPPER_LC(*p);
2456 }
2457 else {
81c6c7ce 2458 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2459 }
3a4c58c9
KW
2460 }
2461 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2462 if (flags) {
2463 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2464 }
2465 else {
81c6c7ce
KW
2466 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2467 ustrp, lenp, 'S');
051a06d4
KW
2468 }
2469 }
2470 else { /* utf8, ord above 255 */
2471 result = CALL_UPPER_CASE(p, ustrp, lenp);
2472
2473 if (flags) {
2474 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2475 }
2476 return result;
2477 }
2478
2479 /* Here, used locale rules. Convert back to utf8 */
2480 if (UTF8_IS_INVARIANT(result)) {
2481 *ustrp = (U8) result;
2482 *lenp = 1;
2483 }
2484 else {
2485 *ustrp = UTF8_EIGHT_BIT_HI(result);
2486 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2487 *lenp = 2;
3a4c58c9
KW
2488 }
2489
051a06d4
KW
2490 if (tainted_ptr) {
2491 *tainted_ptr = TRUE;
2492 }
2493 return result;
983ffd37 2494}
a0ed51b3 2495
d3e79532 2496/*
87cea99e 2497=for apidoc to_utf8_title
d3e79532 2498
a1433954
KW
2499Convert the UTF-8 encoded character at C<p> to its titlecase version and
2500store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2501that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2502titlecase version may be longer than the original character.
d3e79532
JH
2503
2504The first character of the titlecased version is returned
2505(but note, as explained above, that there may be more.)
2506
a1433954
KW
2507The character at C<p> is assumed by this routine to be well-formed.
2508
d3e79532
JH
2509=cut */
2510
051a06d4
KW
2511/* Not currently externally documented, and subject to change:
2512 * <flags> is set iff locale semantics are to be used for code points < 256
2513 * Since titlecase is not defined in POSIX, uppercase is used instead
2514 * for these/
2515 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2516 * were used in the calculation; otherwise unchanged. */
2517
983ffd37 2518UV
051a06d4 2519Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2520{
97aff369 2521 dVAR;
7918f24d 2522
051a06d4
KW
2523 UV result;
2524
2525 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2526
3a4c58c9 2527 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2528 if (flags) {
2529 result = toUPPER_LC(*p);
2530 }
2531 else {
81c6c7ce 2532 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2533 }
3a4c58c9
KW
2534 }
2535 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2536 if (flags) {
2537 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2538 }
2539 else {
81c6c7ce
KW
2540 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2541 ustrp, lenp, 's');
051a06d4
KW
2542 }
2543 }
2544 else { /* utf8, ord above 255 */
2545 result = CALL_TITLE_CASE(p, ustrp, lenp);
2546
2547 if (flags) {
2548 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2549 }
2550 return result;
2551 }
2552
2553 /* Here, used locale rules. Convert back to utf8 */
2554 if (UTF8_IS_INVARIANT(result)) {
2555 *ustrp = (U8) result;
2556 *lenp = 1;
2557 }
2558 else {
2559 *ustrp = UTF8_EIGHT_BIT_HI(result);
2560 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2561 *lenp = 2;
3a4c58c9
KW
2562 }
2563
051a06d4
KW
2564 if (tainted_ptr) {
2565 *tainted_ptr = TRUE;
2566 }
2567 return result;
a0ed51b3
LW
2568}
2569
d3e79532 2570/*
87cea99e 2571=for apidoc to_utf8_lower
d3e79532 2572
a1433954
KW
2573Convert the UTF-8 encoded character at C<p> to its lowercase version and
2574store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2575that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2576lowercase version may be longer than the original character.
d3e79532
JH
2577
2578The first character of the lowercased version is returned
2579(but note, as explained above, that there may be more.)
2580
a1433954
KW
2581The character at C<p> is assumed by this routine to be well-formed.
2582
d3e79532
JH
2583=cut */
2584
051a06d4
KW
2585/* Not currently externally documented, and subject to change:
2586 * <flags> is set iff locale semantics are to be used for code points < 256
2587 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2588 * were used in the calculation; otherwise unchanged. */
2589
2104c8d9 2590UV
051a06d4 2591Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2592{
051a06d4
KW
2593 UV result;
2594
97aff369 2595 dVAR;
7918f24d 2596
051a06d4 2597 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2598
968c5e6a 2599 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2600 if (flags) {
2601 result = toLOWER_LC(*p);
2602 }
2603 else {
81c6c7ce 2604 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2605 }
968c5e6a
KW
2606 }
2607 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2608 if (flags) {
2609 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2610 }
2611 else {
81c6c7ce
KW
2612 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2613 ustrp, lenp);
051a06d4 2614 }
968c5e6a 2615 }
051a06d4
KW
2616 else { /* utf8, ord above 255 */
2617 result = CALL_LOWER_CASE(p, ustrp, lenp);
2618
2619 if (flags) {
2620 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2621 }
968c5e6a 2622
051a06d4
KW
2623 return result;
2624 }
2625
2626 /* Here, used locale rules. Convert back to utf8 */
2627 if (UTF8_IS_INVARIANT(result)) {
2628 *ustrp = (U8) result;
2629 *lenp = 1;
2630 }
2631 else {
2632 *ustrp = UTF8_EIGHT_BIT_HI(result);
2633 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2634 *lenp = 2;
2635 }
2636
2637 if (tainted_ptr) {
2638 *tainted_ptr = TRUE;
2639 }
2640 return result;
b4e400f9
JH
2641}
2642
d3e79532 2643/*
87cea99e 2644=for apidoc to_utf8_fold
d3e79532 2645
a1433954
KW
2646Convert the UTF-8 encoded character at C<p> to its foldcase version and
2647store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2648that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
d3e79532
JH
2649foldcase version may be longer than the original character (up to
2650three characters).
2651
2652The first character of the foldcased version is returned
2653(but note, as explained above, that there may be more.)
2654
a1433954
KW
2655The character at C<p> is assumed by this routine to be well-formed.
2656
d3e79532
JH
2657=cut */
2658
051a06d4
KW
2659/* Not currently externally documented, and subject to change,
2660 * in <flags>
2661 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2662 * points < 256. Since foldcase is not defined in
2663 * POSIX, lowercase is used instead
2664 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2665 * otherwise simple folds
a0270393
KW
2666 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2667 * prohibited
051a06d4
KW
2668 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2669 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2670
b4e400f9 2671UV
051a06d4 2672Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2673{
97aff369 2674 dVAR;
7918f24d 2675
051a06d4
KW
2676 UV result;
2677
36bb2ab6 2678 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2679
a0270393
KW
2680 /* These are mutually exclusive */
2681 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2682
50ba90ff
KW
2683 assert(p != ustrp); /* Otherwise overwrites */
2684
a1dde8de 2685 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2686 if (flags & FOLD_FLAGS_LOCALE) {
2687 result = toLOWER_LC(*p);
2688 }
2689 else {
81c6c7ce
KW
2690 return _to_fold_latin1(*p, ustrp, lenp,
2691 cBOOL(flags & FOLD_FLAGS_FULL));
051a06d4 2692 }
a1dde8de
KW
2693 }
2694 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2695 if (flags & FOLD_FLAGS_LOCALE) {
2696 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2697 }
2698 else {
81c6c7ce 2699 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
a0270393
KW
2700 ustrp, lenp,
2701 cBOOL((flags & FOLD_FLAGS_FULL
2702 /* If ASCII safe, don't allow full
2703 * folding, as that could include SHARP
2704 * S => ss; otherwise there is no
2705 * crossing of ascii/non-ascii in the
2706 * latin1 range */
2707 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
051a06d4 2708 }
a1dde8de 2709 }
051a06d4 2710 else { /* utf8, ord above 255 */
a0270393 2711 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2712
051a06d4 2713 if ((flags & FOLD_FLAGS_LOCALE)) {
a0270393 2714 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2715 }
a0270393
KW
2716 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2717 return result;
2718 }
2719 else {
2720 /* This is called when changing the case of a utf8-encoded
2721 * character above the Latin1 range, and the result should not
2722 * contain an ASCII character. */
2723
2724 UV original; /* To store the first code point of <p> */
2725
2726 /* Look at every character in the result; if any cross the
2727 * boundary, the whole thing is disallowed */
2728 U8* s = ustrp;
2729 U8* e = ustrp + *lenp;
2730 while (s < e) {
2731 if (isASCII(*s)) {
2732 /* Crossed, have to return the original */
2733 original = valid_utf8_to_uvchr(p, lenp);
2734 Copy(p, ustrp, *lenp, char);
2735 return original;
2736 }
2737 s += UTF8SKIP(s);
2738 }
051a06d4 2739
a0270393
KW
2740 /* Here, no characters crossed, result is ok as-is */
2741 return result;
2742 }
051a06d4
KW
2743 }
2744
2745 /* Here, used locale rules. Convert back to utf8 */
2746 if (UTF8_IS_INVARIANT(result)) {
2747 *ustrp = (U8) result;
2748 *lenp = 1;
2749 }
2750 else {
2751 *ustrp = UTF8_EIGHT_BIT_HI(result);
2752 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2753 *lenp = 2;
2754 }
2755
2756 if (tainted_ptr) {
2757 *tainted_ptr = TRUE;
2758 }
2759 return result;
a0ed51b3
LW
2760}
2761
711a919c 2762/* Note:
f90a9a02 2763 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2764 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2765 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2766 */
c4a5db0c 2767
a0ed51b3 2768SV*
7fc63493 2769Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2770{
c4a5db0c
KW
2771 PERL_ARGS_ASSERT_SWASH_INIT;
2772
2773 /* Returns a copy of a swash initiated by the called function. This is the
2774 * public interface, and returning a copy prevents others from doing
2775 * mischief on the original */
2776
5d3d13d1 2777 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2778}
2779
2780SV*
5d3d13d1 2781Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2782{
2783 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2784 * by calling utf8_heavy.pl in the general case. The returned value may be
2785 * the swash's inversion list instead if the input parameters allow it.
2786 * Which is returned should be immaterial to callers, as the only
2787 * operations permitted on a swash, swash_fetch() and
2788 * _get_swash_invlist(), handle both these transparently.
c4a5db0c
KW
2789 *
2790 * This interface should only be used by functions that won't destroy or
2791 * adversely change the swash, as doing so affects all other uses of the
2792 * swash in the program; the general public should use 'Perl_swash_init'
2793 * instead.
2794 *
2795 * pkg is the name of the package that <name> should be in.
2796 * name is the name of the swash to find. Typically it is a Unicode
2797 * property name, including user-defined ones
2798 * listsv is a string to initialize the swash with. It must be of the form
2799 * documented as the subroutine return value in
2800 * L<perlunicode/User-Defined Character Properties>
2801 * minbits is the number of bits required to represent each data element.
2802 * It is '1' for binary properties.
2803 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2804 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2805 * flags_p if non-NULL is the address of various input and output flag bits
2806 * to the routine, as follows: ('I' means is input to the routine;
2807 * 'O' means output from the routine. Only flags marked O are
2808 * meaningful on return.)
2809 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2810 * came from a user-defined property. (I O)
5d3d13d1
KW
2811 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2812 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2813 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2814 * return of an inversion list instead of a swash hash if this routine
2815 * thinks that would result in faster execution of swash_fetch() later
2816 * on. (I)
9a53f6cf
KW
2817 *
2818 * Thus there are three possible inputs to find the swash: <name>,
2819 * <listsv>, and <invlist>. At least one must be specified. The result
2820 * will be the union of the specified ones, although <listsv>'s various
2821 * actions can intersect, etc. what <name> gives.
2822 *
2823 * <invlist> is only valid for binary properties */
c4a5db0c 2824
27da23d5 2825 dVAR;
c4a5db0c 2826 SV* retval = &PL_sv_undef;
83199d38 2827 HV* swash_hv = NULL;
87367d5f
KW
2828 const int invlist_swash_boundary =
2829 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2830 ? 512 /* Based on some benchmarking, but not extensive, see commit
2831 message */
2832 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2833
2834 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2835 assert(! invlist || minbits == 1);
2836
2837 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2838 * so */
2839 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2840 dSP;
2841 const size_t pkg_len = strlen(pkg);
2842 const size_t name_len = strlen(name);
2843 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2844 SV* errsv_save;
2845 GV *method;
2846
2847 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2848
2849 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2850 ENTER;
69794297
KW
2851 SAVEHINTS();
2852 save_re_context();
650f067c
JL
2853 /* We might get here via a subroutine signature which uses a utf8
2854 * parameter name, at which point PL_subname will have been set
2855 * but not yet used. */
2856 save_item(PL_subname);
69794297
KW
2857 if (PL_parser && PL_parser->error_count)
2858 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2859 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2860 if (!method) { /* demand load utf8 */
2861 ENTER;
2862 errsv_save = newSVsv(ERRSV);
36048493 2863 SAVEFREESV(errsv_save);
69794297
KW
2864 /* It is assumed that callers of this routine are not passing in
2865 * any user derived data. */
2866 /* Need to do this after save_re_context() as it will set
2867 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2868 * in Perl_magic_get). Even line to create errsv_save can turn on
2869 * PL_tainted. */
284167a5
SM
2870#ifndef NO_TAINT_SUPPORT
2871 SAVEBOOL(TAINT_get);
2872 TAINT_NOT;
2873#endif
69794297
KW
2874 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2875 NULL);
2876 if (!SvTRUE(ERRSV))
2877 sv_setsv(ERRSV, errsv_save);
69794297
KW
2878 LEAVE;
2879 }
2880 SPAGAIN;
2881 PUSHMARK(SP);
2882 EXTEND(SP,5);
2883 mPUSHp(pkg, pkg_len);
2884 mPUSHp(name, name_len);
2885 PUSHs(listsv);
2886 mPUSHi(minbits);
2887 mPUSHi(none);
2888 PUTBACK;
f8be5cf0 2889 errsv_save = newSVsv(ERRSV);
36048493 2890 SAVEFREESV(errsv_save);
69794297
KW
2891 /* If we already have a pointer to the method, no need to use
2892 * call_method() to repeat the lookup. */
2893 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
2894 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2895 {
2896 retval = *PL_stack_sp--;
2897 SvREFCNT_inc(retval);
2898 }
f8be5cf0
JH
2899 if (!SvTRUE(ERRSV))
2900 sv_setsv(ERRSV, errsv_save);
ce3b816e 2901 LEAVE;
69794297
KW
2902 POPSTACK;
2903 if (IN_PERL_COMPILETIME) {
2904 CopHINTS_set(PL_curcop, PL_hints);
2905 }
2906 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2907 if (SvPOK(retval))
2908
2909 /* If caller wants to handle missing properties, let them */
5d3d13d1 2910 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
2911 return NULL;
2912 }
2913 Perl_croak(aTHX_
2914 "Can't find Unicode property definition \"%"SVf"\"",
2915 SVfARG(retval));
2916 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2917 }
9a53f6cf 2918 } /* End of calling the module to find the swash */
36eb48b4 2919
83199d38
KW
2920 /* If this operation fetched a swash, and we will need it later, get it */
2921 if (retval != &PL_sv_undef
2922 && (minbits == 1 || (flags_p
2923 && ! (*flags_p
2924 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2925 {
2926 swash_hv = MUTABLE_HV(SvRV(retval));
2927
2928 /* If we don't already know that there is a user-defined component to
2929 * this swash, and the user has indicated they wish to know if there is
2930 * one (by passing <flags_p>), find out */
2931 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2932 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2933 if (user_defined && SvUV(*user_defined)) {
2934 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2935 }
2936 }
2937 }
2938
36eb48b4
KW
2939 /* Make sure there is an inversion list for binary properties */
2940 if (minbits == 1) {
2941 SV** swash_invlistsvp = NULL;
2942 SV* swash_invlist = NULL;
9a53f6cf 2943 bool invlist_in_swash_is_valid = FALSE;
36eb48b4 2944
9a53f6cf 2945 /* If this operation fetched a swash, get its already existing
83199d38 2946 * inversion list, or create one for it */
36eb48b4 2947
83199d38 2948 if (swash_hv) {
5c9f4bd2 2949 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2950 if (swash_invlistsvp) {
2951 swash_invlist = *swash_invlistsvp;
2952 invlist_in_swash_is_valid = TRUE;
2953 }
2954 else {
36eb48b4 2955 swash_invlist = _swash_to_invlist(retval);
9a53f6cf
KW
2956 }
2957 }
2958
2959 /* If an inversion list was passed in, have to include it */
2960 if (invlist) {
2961
2962 /* Any fetched swash will by now have an inversion list in it;
2963 * otherwise <swash_invlist> will be NULL, indicating that we
2964 * didn't fetch a swash */
2965 if (swash_invlist) {
2966
2967 /* Add the passed-in inversion list, which invalidates the one
2968 * already stored in the swash */
2969 invlist_in_swash_is_valid = FALSE;
2970 _invlist_union(invlist, swash_invlist, &swash_invlist);
2971 }
2972 else {
2973
87367d5f
KW
2974 /* Here, there is no swash already. Set up a minimal one, if
2975 * we are going to return a swash */
2976 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f
KW
2977 swash_hv = newHV();
2978 retval = newRV_inc(MUTABLE_SV(swash_hv));
87367d5f 2979 }
9a53f6cf
KW
2980 swash_invlist = invlist;
2981 }
9a53f6cf
KW
2982 }
2983
2984 /* Here, we have computed the union of all the passed-in data. It may
2985 * be that there was an inversion list in the swash which didn't get
2986 * touched; otherwise save the one computed one */
87367d5f
KW
2987 if (! invlist_in_swash_is_valid
2988 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2989 {
5c9f4bd2 2990 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2991 {
2992 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2993 }
9a53f6cf 2994 }
87367d5f
KW
2995
2996 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2997 SvREFCNT_dec(retval);
2998 retval = newRV_inc(swash_invlist);
2999 }
36eb48b4
KW
3000 }
3001
a0ed51b3
LW
3002 return retval;
3003}
3004
035d37be
JH
3005
3006/* This API is wrong for special case conversions since we may need to
3007 * return several Unicode characters for a single Unicode character
3008 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3009 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
3010 * multiple values. --jhi
3011 * For those, you should use to_utf8_case() instead */
b0e3252e 3012/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3013
3014/* Note:
3015 * Returns the value of property/mapping C<swash> for the first character
3016 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3017 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3018 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3019 *
3020 * A "swash" is a hash which contains initially the keys/values set up by
3021 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3022 * property for all possible code points. Things are stored in a compact form
3023 * (see utf8_heavy.pl) so that calculation is required to find the actual
3024 * property value for a given code point. As code points are looked up, new
3025 * key/value pairs are added to the hash, so that the calculation doesn't have
3026 * to ever be re-done. Further, each calculation is done, not just for the
3027 * desired one, but for a whole block of code points adjacent to that one.
3028 * For binary properties on ASCII machines, the block is usually for 64 code
3029 * points, starting with a code point evenly divisible by 64. Thus if the
3030 * property value for code point 257 is requested, the code goes out and
3031 * calculates the property values for all 64 code points between 256 and 319,
3032 * and stores these as a single 64-bit long bit vector, called a "swatch",
3033 * under the key for code point 256. The key is the UTF-8 encoding for code
3034 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3035 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3036 * for code point 258 is then requested, this code realizes that it would be
3037 * stored under the key for 256, and would find that value and extract the
3038 * relevant bit, offset from 256.
3039 *
3040 * Non-binary properties are stored in as many bits as necessary to represent
3041 * their values (32 currently, though the code is more general than that), not
3042 * as single bits, but the principal is the same: the value for each key is a
3043 * vector that encompasses the property values for all code points whose UTF-8
3044 * representations are represented by the key. That is, for all code points
3045 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3046 * bytes of that.
680c470c 3047 */
a0ed51b3 3048UV
680c470c 3049Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3050{
27da23d5 3051 dVAR;
ef8f7699 3052 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3053 U32 klen;
3054 U32 off;
a0ed51b3 3055 STRLEN slen;
7d85a32c 3056 STRLEN needents;
cfd0369c 3057 const U8 *tmps = NULL;
a0ed51b3 3058 U32 bit;
979f2922 3059 SV *swatch;
3568d838 3060 U8 tmputf8[2];
35da51f7 3061 const UV c = NATIVE_TO_ASCII(*ptr);
3568d838 3062
7918f24d
NC
3063 PERL_ARGS_ASSERT_SWASH_FETCH;
3064
87367d5f
KW
3065 /* If it really isn't a hash, it isn't really swash; must be an inversion
3066 * list */
3067 if (SvTYPE(hv) != SVt_PVHV) {
3068 return _invlist_contains_cp((SV*)hv,
3069 (do_utf8)
3070 ? valid_utf8_to_uvchr(ptr, NULL)
3071 : c);
3072 }
3073
dbe7a391 3074 /* Convert to utf8 if not already */
3568d838 3075 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
979f2922
ST
3076 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3077 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3078 ptr = tmputf8;
3568d838
JH
3079 }
3080 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
37e2e78e 3081 * then the "swatch" is a vec() for all the chars which start
3568d838
JH
3082 * with 0xAA..0xYY
3083 * So the key in the hash (klen) is length of encoded char -1
3084 */
3085 klen = UTF8SKIP(ptr) - 1;
3086 off = ptr[klen];
a0ed51b3 3087
979f2922 3088 if (klen == 0) {
37e2e78e 3089 /* If char is invariant then swatch is for all the invariant chars
1e54db1a 3090 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
7d85a32c 3091 */
979f2922
ST
3092 needents = UTF_CONTINUATION_MARK;
3093 off = NATIVE_TO_UTF(ptr[klen]);
3094 }
3095 else {
7d85a32c 3096 /* If char is encoded then swatch is for the prefix */
979f2922
ST
3097 needents = (1 << UTF_ACCUMULATION_SHIFT);
3098 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3099 }
7d85a32c 3100
a0ed51b3
LW
3101 /*
3102 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3103 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3104 * it's nothing to sniff at.) Pity we usually come through at least
3105 * two function calls to get here...
3106 *
3107 * NB: this code assumes that swatches are never modified, once generated!
3108 */
3109
3568d838 3110 if (hv == PL_last_swash_hv &&
a0ed51b3 3111 klen == PL_last_swash_klen &&
27da23d5 3112 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3113 {
3114 tmps = PL_last_swash_tmps;
3115 slen = PL_last_swash_slen;
3116 }
3117 else {
3118 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3119 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3120
b0e3252e 3121 /* If not cached, generate it via swatch_get */
979f2922
ST
3122 if (!svp || !SvPOK(*svp)
3123 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2b9d42f0
NIS
3124 /* We use utf8n_to_uvuni() as we want an index into
3125 Unicode tables, not a native character number.
3126 */
aec46f14 3127 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
872c91ae
JH
3128 ckWARN(WARN_UTF8) ?
3129 0 : UTF8_ALLOW_ANY);
b0e3252e 3130 swatch = swatch_get(swash,
979f2922 3131 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
361ee0fe 3132 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
979f2922
ST
3133 needents);
3134
923e4eb5 3135 if (IN_PERL_COMPILETIME)
623e6609 3136 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3137
979f2922 3138 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3139
979f2922
ST
3140 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3141 || (slen << 3) < needents)
5637ef5b
NC
3142 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3143 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3144 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3145 }
3146
3147 PL_last_swash_hv = hv;
16d8f38a 3148 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3149 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3150 /* FIXME change interpvar.h? */
3151 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3152 PL_last_swash_slen = slen;
3153 if (klen)
3154 Copy(ptr, PL_last_swash_key, klen, U8);
3155 }
3156
9faf8d75 3157 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3158 case 1:
3159 bit = 1 << (off & 7);
3160 off >>= 3;
3161 return (tmps[off] & bit) != 0;
3162 case 8:
3163 return tmps[off];
3164 case 16:
3165 off <<= 1;
3166 return (tmps[off] << 8) + tmps[off + 1] ;
3167 case 32:
3168 off <<= 2;
3169 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3170 }
5637ef5b
NC
3171 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3172 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3173 NORETURN_FUNCTION_END;
a0ed51b3 3174}
2b9d42f0 3175
319009ee
KW
3176/* Read a single line of the main body of the swash input text. These are of
3177 * the form:
3178 * 0053 0056 0073
3179 * where each number is hex. The first two numbers form the minimum and
3180 * maximum of a range, and the third is the value associated with the range.
3181 * Not all swashes should have a third number
3182 *
3183 * On input: l points to the beginning of the line to be examined; it points
3184 * to somewhere in the string of the whole input text, and is
3185 * terminated by a \n or the null string terminator.
3186 * lend points to the null terminator of that string
3187 * wants_value is non-zero if the swash expects a third number
3188 * typestr is the name of the swash's mapping, like 'ToLower'
3189 * On output: *min, *max, and *val are set to the values read from the line.
3190 * returns a pointer just beyond the line examined. If there was no
3191 * valid min number on the line, returns lend+1
3192 */
3193
3194STATIC U8*
3195S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3196 const bool wants_value, const U8* const typestr)
3197{
3198 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3199 STRLEN numlen; /* Length of the number */
02470786
KW
3200 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3201 | PERL_SCAN_DISALLOW_PREFIX
3202 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3203
3204 /* nl points to the next \n in the scan */
3205 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3206
3207 /* Get the first number on the line: the range minimum */
3208 numlen = lend - l;
3209 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3210 if (numlen) /* If found a hex number, position past it */
3211 l += numlen;
3212 else if (nl) { /* Else, go handle next line, if any */
3213 return nl + 1; /* 1 is length of "\n" */
3214 }
3215 else { /* Else, no next line */
3216 return lend + 1; /* to LIST's end at which \n is not found */
3217 }
3218
3219 /* The max range value follows, separated by a BLANK */
3220 if (isBLANK(*l)) {
3221 ++l;
02470786
KW
3222 flags = PERL_SCAN_SILENT_ILLDIGIT
3223 | PERL_SCAN_DISALLOW_PREFIX
3224 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3225 numlen = lend - l;
3226 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3227 if (numlen)
3228 l += numlen;
3229 else /* If no value here, it is a single element range */
3230 *max = *min;
3231
3232 /* Non-binary tables have a third entry: what the first element of the
3233 * range maps to */
3234 if (wants_value) {
3235 if (isBLANK(*l)) {
3236 ++l;
f90a9a02
KW
3237
3238 /* The ToLc, etc table mappings are not in hex, and must be
3239 * corrected by adding the code point to them */
3240 if (typeto) {
3241 char *after_strtol = (char *) lend;
3242 *val = Strtol((char *)l, &after_strtol, 10);
3243 l = (U8 *) after_strtol;
f90a9a02
KW
3244 }
3245 else { /* Other tables are in hex, and are the correct result
3246 without tweaking */
a9d188b3
KW
3247 flags = PERL_SCAN_SILENT_ILLDIGIT
3248 | PERL_SCAN_DISALLOW_PREFIX
3249 | PERL_SCAN_SILENT_NON_PORTABLE;
3250 numlen = lend - l;
3251 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3252 if (numlen)
3253 l += numlen;
3254 else
3255 *val = 0;
f90a9a02 3256 }
319009ee
KW
3257 }
3258 else {
3259 *val = 0;
3260 if (typeto) {
dcbac5bb 3261 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3262 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3263 typestr, l);
3264 }
3265 }
3266 }
3267 else
3268 *val = 0; /* bits == 1, then any val should be ignored */
3269 }
3270 else { /* Nothing following range min, should be single element with no
3271 mapping expected */
3272 *max = *min;
3273 if (wants_value) {
3274 *val = 0;
3275 if (typeto) {
dcbac5bb 3276 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3277 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3278 }
3279 }
3280 else
3281 *val = 0; /* bits == 1, then val should be ignored */
3282 }
3283
3284 /* Position to next line if any, or EOF */
3285 if (nl)
3286 l = nl + 1;
3287 else
3288 l = lend;
3289
3290 return l;
3291}
3292
979f2922
ST
3293/* Note:
3294 * Returns a swatch (a bit vector string) for a code point sequence
3295 * that starts from the value C<start> and comprises the number C<span>.
3296 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3297 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3298 */
3299STATIC SV*
b0e3252e 3300S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3301{
3302 SV *swatch;
77f9f126 3303 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3304 STRLEN lcur, xcur, scur;
ef8f7699 3305 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3306 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3307
88d45d28
KW
3308 SV** listsvp = NULL; /* The string containing the main body of the table */
3309 SV** extssvp = NULL;
3310 SV** invert_it_svp = NULL;
3311 U8* typestr = NULL;
786861f5
KW
3312 STRLEN bits;
3313 STRLEN octets; /* if bits == 1, then octets == 0 */
3314 UV none;
3315 UV end = start + span;
972dd592 3316
36eb48b4 3317 if (invlistsvp == NULL) {
786861f5
KW
3318 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3319 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3320 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3321 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3322 listsvp = hv_fetchs(hv, "LIST", FALSE);
3323 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3324
3325 bits = SvUV(*bitssvp);
3326 none = SvUV(*nonesvp);
3327 typestr = (U8*)SvPV_nolen(*typesvp);
3328 }
36eb48b4
KW
3329 else {
3330 bits = 1;
3331 none = 0;
3332 }
786861f5 3333 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3334
b0e3252e 3335 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3336
979f2922 3337 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3338 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3339 (UV)bits);
979f2922
ST
3340 }
3341
84ea5ef6
KW
3342 /* If overflowed, use the max possible */
3343 if (end < start) {
3344 end = UV_MAX;
3345 span = end - start;
3346 }
3347
979f2922 3348 /* create and initialize $swatch */
979f2922 3349 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3350 swatch = newSV(scur);
3351 SvPOK_on(swatch);
979f2922
ST
3352 s = (U8*)SvPVX(swatch);
3353 if (octets && none) {
0bd48802 3354 const U8* const e = s + scur;
979f2922
ST
3355 while (s < e) {
3356 if (bits == 8)
3357 *s++ = (U8)(none & 0xff);
3358 else if (bits == 16) {
3359 *s++ = (U8)((none >> 8) & 0xff);
3360 *s++ = (U8)( none & 0xff);
3361 }
3362 else if (bits == 32) {
3363 *s++ = (U8)((none >> 24) & 0xff);
3364 *s++ = (U8)((none >> 16) & 0xff);
3365 *s++ = (U8)((none >> 8) & 0xff);
3366 *s++ = (U8)( none & 0xff);
3367 }
3368 }
3369 *s = '\0';
3370 }
3371 else {
3372 (void)memzero((U8*)s, scur + 1);
3373 }
3374 SvCUR_set(swatch, scur);
3375 s = (U8*)SvPVX(swatch);
3376
36eb48b4
KW
3377 if (invlistsvp) { /* If has an inversion list set up use that */
3378 _invlist_populate_swatch(*invlistsvp, start, end, s);
3379 return swatch;
3380 }
3381
3382 /* read $swash->{LIST} */
979f2922
ST
3383 l = (U8*)SvPV(*listsvp, lcur);
3384 lend = l + lcur;
3385 while (l < lend) {
8ed25d53 3386 UV min, max, val, upper;
319009ee
KW
3387 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3388 cBOOL(octets), typestr);
3389 if (l > lend) {
979f2922
ST
3390 break;
3391 }
3392
972dd592 3393 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3394 if (max < start)
3395 continue;
3396
8ed25d53
KW
3397 /* <end> is generally 1 beyond where we want to set things, but at the
3398 * platform's infinity, where we can't go any higher, we want to
3399 * include the code point at <end> */
3400 upper = (max < end)
3401 ? max
3402 : (max != UV_MAX || end != UV_MAX)
3403 ? end - 1
3404 : end;
3405
979f2922 3406 if (octets) {
35da51f7 3407 UV key;
979f2922
ST
3408 if (min < start) {
3409 if (!none || val < none) {
3410 val += start - min;
3411 }
3412 min = start;
3413 }
8ed25d53 3414 for (key = min; key <= upper; key++) {
979f2922 3415 STRLEN offset;
979f2922
ST
3416 /* offset must be non-negative (start <= min <= key < end) */
3417 offset = octets * (key - start);
3418 if (bits == 8)
3419 s[offset] = (U8)(val & 0xff);
3420 else if (bits == 16) {
3421 s[offset ] = (U8)((val >> 8) & 0xff);
3422 s[offset + 1] = (U8)( val & 0xff);
3423 }
3424 else if (bits == 32) {
3425 s[offset ] = (U8)((val >> 24) & 0xff);
3426 s[offset + 1] = (U8)((val >> 16) & 0xff);
3427 s[offset + 2] = (U8)((val >> 8) & 0xff);
3428 s[offset + 3] = (U8)( val & 0xff);
3429 }
3430
3431 if (!none || val < none)
3432 ++val;
3433 }
3434 }
711a919c 3435 else { /* bits == 1, then val should be ignored */
35da51f7 3436 UV key;
979f2922
ST
3437 if (min < start)
3438 min = start;
6cb05c12 3439
8ed25d53 3440 for (key = min; key <= upper; key++) {
0bd48802 3441 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3442 s[offset >> 3] |= 1 << (offset & 7);
3443 }
3444 }
3445 } /* while */
979f2922 3446
9479a769 3447 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3448 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3449
3450 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3451 * be 0, and their inversion should also be 0, as we don't succeed any
3452 * Unicode property matches for non-Unicode code points */
3453 if (start <= PERL_UNICODE_MAX) {
3454
3455 /* The code below assumes that we never cross the
3456 * Unicode/above-Unicode boundary in a range, as otherwise we would
3457 * have to figure out where to stop flipping the bits. Since this
3458 * boundary is divisible by a large power of 2, and swatches comes
3459 * in small powers of 2, this should be a valid assumption */
3460 assert(start + span - 1 <= PERL_UNICODE_MAX);
3461
507a8485
KW
3462 send = s + scur;
3463 while (s < send) {
3464 *s = ~(*s);
3465 s++;
3466 }
0bda3001 3467 }
77f9f126
KW
3468 }
3469
d73c39c5
KW
3470 /* read $swash->{EXTRAS}
3471 * This code also copied to swash_to_invlist() below */
979f2922
ST
3472 x = (U8*)SvPV(*extssvp, xcur);
3473 xend = x + xcur;
3474 while (x < xend) {
3475 STRLEN namelen;
3476 U8 *namestr;
3477 SV** othersvp;
3478 HV* otherhv;
3479 STRLEN otherbits;
3480 SV **otherbitssvp, *other;
711a919c 3481 U8 *s, *o, *nl;
979f2922
ST
3482 STRLEN slen, olen;
3483
35da51f7 3484 const U8 opc = *x++;
979f2922
ST
3485 if (opc == '\n')
3486 continue;
3487
3488 nl = (U8*)memchr(x, '\n', xend - x);
3489
3490 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3491 if (nl) {
3492 x = nl + 1; /* 1 is length of "\n" */
3493 continue;
3494 }
3495 else {
3496 x = xend; /* to EXTRAS' end at which \n is not found */
3497 break;
3498 }
3499 }
3500
3501 namestr = x;
3502 if (nl) {
3503 namelen = nl - namestr;
3504 x = nl + 1;
3505 }
3506 else {
3507 namelen = xend - namestr;
3508 x = xend;
3509 }
3510
3511 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3512 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3513 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3514 otherbits = (STRLEN)SvUV(*otherbitssvp);
3515 if (bits < otherbits)
5637ef5b
NC
3516 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3517 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3518
3519 /* The "other" swatch must be destroyed after. */
b0e3252e 3520 other = swatch_get(*othersvp, start, span);
979f2922
ST
3521 o = (U8*)SvPV(other, olen);
3522
3523 if (!olen)
b0e3252e 3524 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3525
3526 s = (U8*)SvPV(swatch, slen);
3527 if (bits == 1 && otherbits == 1) {
3528 if (slen != olen)
5637ef5b
NC
3529 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3530 "mismatch, slen=%"UVuf", olen=%"UVuf,
3531 (UV)slen, (UV)olen);
979f2922
ST
3532
3533 switch (opc) {
3534 case '+':
3535 while (slen--)
3536 *s++ |= *o++;
3537 break;
3538 case '!':
3539 while (slen--)
3540 *s++ |= ~*o++;
3541 break;
3542 case '-':
3543 while (slen--)
3544 *s++ &= ~*o++;
3545 break;
3546 case '&':
3547 while (slen--)
3548 *s++ &= *o++;
3549 break;
3550 default:
3551 break;
3552 }
3553 }
711a919c 3554 else {
979f2922
ST
3555 STRLEN otheroctets = otherbits >> 3;
3556 STRLEN offset = 0;
35da51f7 3557 U8* const send = s + slen;
979f2922
ST
3558
3559 while (s < send) {
3560 UV otherval = 0;
3561
3562 if (otherbits == 1) {
3563 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3564 ++offset;
3565 }
3566 else {
3567 STRLEN vlen = otheroctets;
3568 otherval = *o++;
3569 while (--vlen) {
3570 otherval <<= 8;
3571 otherval |= *o++;
3572 }
3573 }
3574
711a919c 3575 if (opc == '+' && otherval)
6f207bd3 3576 NOOP; /* replace with otherval */
979f2922
ST
3577 else if (opc == '!' && !otherval)
3578 otherval = 1;
3579 else if (opc == '-' && otherval)
3580 otherval = 0;
3581 else if (opc == '&' && !otherval)
3582 otherval = 0;
3583 else {
711a919c 3584 s += octets; /* no replacement */
979f2922
ST
3585 continue;
3586 }
3587
3588 if (bits == 8)
3589 *s++ = (U8)( otherval & 0xff);
3590 else if (bits == 16) {
3591 *s++ = (U8)((otherval >> 8) & 0xff);
3592 *s++ = (U8)( otherval & 0xff);
3593 }
3594 else if (bits == 32) {
3595 *s++ = (U8)((otherval >> 24) & 0xff);
3596 *s++ = (U8)((otherval >> 16) & 0xff);
3597 *s++ = (U8)((otherval >> 8) & 0xff);
3598 *s++ = (U8)( otherval & 0xff);
3599 }
3600 }
3601 }
3602 sv_free(other); /* through with it! */
3603 } /* while */
3604 return swatch;
3605}
3606
064c021d 3607HV*
4c2e1131 3608Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3609{
3610
79a2a0e8 3611 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3612 * Can't be used on a property that is subject to user override, as it
3613 * relies on the value of SPECIALS in the swash which would be set by
3614 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3615 * for overridden properties
064c021d
KW
3616 *
3617 * Returns a hash which is the inversion and closure of a swash mapping.
3618 * For example, consider the input lines:
3619 * 004B 006B
3620 * 004C 006C
3621 * 212A 006B
3622 *
3623 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3624 * 006C. The value for each key is an array. For 006C, the array would
3625 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3626 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3627 *
3628 * Essentially, for any code point, it gives all the code points that map to
3629 * it, or the list of 'froms' for that point.
3630 *
5662e334
KW
3631 * Currently it ignores any additions or deletions from other swashes,
3632 * looking at just the main body of the swash, and if there are SPECIALS
3633 * in the swash, at that hash
3634 *
3635 * The specials hash can be extra code points, and most likely consists of
3636 * maps from single code points to multiple ones (each expressed as a string
3637 * of utf8 characters). This function currently returns only 1-1 mappings.
3638 * However consider this possible input in the specials hash:
3639 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3640 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3641 *
3642 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3643 * currently handle. But it also means that FB05 and FB06 are equivalent in
3644 * a 1-1 mapping which we should handle, and this relationship may not be in
3645 * the main table. Therefore this function examines all the multi-char
3646 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3647
3648 U8 *l, *lend;
3649 STRLEN lcur;
3650 HV *const hv = MUTABLE_HV(SvRV(swash));
3651
3652 /* The string containing the main body of the table */
3653 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3654
3655 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3656 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3657 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3658 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3659 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3660 const STRLEN bits = SvUV(*bitssvp);
3661 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3662 const UV none = SvUV(*nonesvp);
5662e334 3663 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3664
3665 HV* ret = newHV();
3666
3667 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3668
3669 /* Must have at least 8 bits to get the mappings */
3670 if (bits != 8 && bits != 16 && bits != 32) {
3671 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3672 (UV)bits);
3673 }
3674
5662e334
KW
3675 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3676 mapping to more than one character */
3677
3678 /* Construct an inverse mapping hash for the specials */
3679 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3680 HV * specials_inverse = newHV();
3681 char *char_from; /* the lhs of the map */
3682 I32 from_len; /* its byte length */
3683 char *char_to; /* the rhs of the map */
3684 I32 to_len; /* its byte length */
3685 SV *sv_to; /* and in a sv */
3686 AV* from_list; /* list of things that map to each 'to' */
3687
3688 hv_iterinit(specials_hv);
3689
3690 /* The keys are the characters (in utf8) that map to the corresponding
3691 * utf8 string value. Iterate through the list creating the inverse
3692 * list. */
3693 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3694 SV** listp;
3695 if (! SvPOK(sv_to)) {
5637ef5b
NC
3696 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3697 "unexpectedly is not a string, flags=%lu",
3698 (unsigned long)SvFLAGS(sv_to));
5662e334 3699 }
4b88fb76 3700 /*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
3701
3702 /* Each key in the inverse list is a mapped-to value, and the key's
3703 * hash value is a list of the strings (each in utf8) that map to
3704 * it. Those strings are all one character long */
3705 if ((listp = hv_fetch(specials_inverse,
3706 SvPVX(sv_to),
3707 SvCUR(sv_to), 0)))
3708 {
3709 from_list = (AV*) *listp;
3710 }
3711 else { /* No entry yet for it: create one */
3712 from_list = newAV();
3713 if (! hv_store(specials_inverse,
3714 SvPVX(sv_to),
3715 SvCUR(sv_to),
3716 (SV*) from_list, 0))
3717 {
3718 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3719 }
3720 }
3721
3722 /* Here have the list associated with this 'to' (perhaps newly
3723 * created and empty). Just add to it. Note that we ASSUME that
3724 * the input is guaranteed to not have duplications, so we don't
3725 * check for that. Duplications just slow down execution time. */
3726 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3727 }
3728
3729 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3730 * it looking for cases like the FB05/FB06 examples above. There would
3731 * be an entry in the hash like
3732 * 'st' => [ FB05, FB06 ]
3733 * In this example we will create two lists that get stored in the