This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
embed.fnc: Mark internal function as "may change"
[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);
40PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
41#endif
42
27da23d5
JH
43static const char unees[] =
44 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 45
48ef279e 46/*
ccfc67b7 47=head1 Unicode Support
a0ed51b3 48
166f8a29
DM
49This file contains various utility functions for manipulating UTF8-encoded
50strings. For the uninitiated, this is a method of representing arbitrary
61296642 51Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
52characters in the ASCII range are unmodified, and a zero byte never appears
53within non-zero characters.
166f8a29 54
eaf7a4d2
CS
55=cut
56*/
57
58/*
59=for apidoc is_ascii_string
60
a1433954 61Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
62or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
63is, if they are invariant. On ASCII-ish machines, only ASCII characters
64fit this definition, hence the function's name.
eaf7a4d2 65
9f7e3d64
MH
66If C<len> is 0, it will be calculated using C<strlen(s)>.
67
a1433954 68See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
69
70=cut
71*/
72
73bool
668b6d8d 74Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
75{
76 const U8* const send = s + (len ? len : strlen((const char *)s));
77 const U8* x = s;
78
79 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
80
81 for (; x < send; ++x) {
82 if (!UTF8_IS_INVARIANT(*x))
83 break;
84 }
85
86 return x == send;
87}
88
89/*
87cea99e 90=for apidoc uvuni_to_utf8_flags
eebe1485 91
6ee84de2
KW
92Adds the UTF-8 representation of the code point C<uv> to the end
93of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
eebe1485 94bytes available. The return value is the pointer to the byte after the
9041c2e3 95end of the new character. In other words,
eebe1485 96
b851fbc1
JH
97 d = uvuni_to_utf8_flags(d, uv, flags);
98
99or, in most cases,
100
9041c2e3 101 d = uvuni_to_utf8(d, uv);
eebe1485 102
b851fbc1
JH
103(which is equivalent to)
104
105 d = uvuni_to_utf8_flags(d, uv, 0);
106
949cf498 107This is the recommended Unicode-aware way of saying
eebe1485
SC
108
109 *(d++) = uv;
110
949cf498
KW
111This function will convert to UTF-8 (and not warn) even code points that aren't
112legal Unicode or are problematic, unless C<flags> contains one or more of the
a1433954
KW
113following flags:
114
949cf498
KW
115If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
116the function will raise a warning, provided UTF8 warnings are enabled. If instead
117UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
118If both flags are set, the function will both warn and return NULL.
119
120The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
121affect how the function handles a Unicode non-character. And, likewise for the
122UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
123above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
124even less portable) can be warned and/or disallowed even if other above-Unicode
125code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
126flags.
127
128And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
129above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
130DISALLOW flags.
131
132
eebe1485
SC
133=cut
134*/
135
dfe13c55 136U8 *
b851fbc1 137Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 138{
7918f24d
NC
139 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
140
979f77b6
KW
141 /* The first problematic code point is the first surrogate */
142 if (uv >= UNICODE_SURROGATE_FIRST
143 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
144 {
949cf498
KW
145 if (UNICODE_IS_SURROGATE(uv)) {
146 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 147 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
148 "UTF-16 surrogate U+%04"UVXf, uv);
149 }
150 if (flags & UNICODE_DISALLOW_SURROGATE) {
151 return NULL;
152 }
153 }
154 else if (UNICODE_IS_SUPER(uv)) {
155 if (flags & UNICODE_WARN_SUPER
156 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
157 {
8457b38f 158 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
159 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
160 }
161 if (flags & UNICODE_DISALLOW_SUPER
162 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
163 {
164 return NULL;
165 }
166 }
167 else if (UNICODE_IS_NONCHAR(uv)) {
168 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 169 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
170 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
171 uv);
172 }
173 if (flags & UNICODE_DISALLOW_NONCHAR) {
174 return NULL;
175 }
176 }
507b9800 177 }
c4d5f83a 178 if (UNI_IS_INVARIANT(uv)) {
eb160463 179 *d++ = (U8)UTF_TO_NATIVE(uv);
a0ed51b3
LW
180 return d;
181 }
2d331972 182#if defined(EBCDIC)
1d72bdf6
NIS
183 else {
184 STRLEN len = UNISKIP(uv);
185 U8 *p = d+len-1;
186 while (p > d) {
eb160463 187 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
188 uv >>= UTF_ACCUMULATION_SHIFT;
189 }
eb160463 190 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
191 return d+len;
192 }
193#else /* Non loop style */
a0ed51b3 194 if (uv < 0x800) {
eb160463
GS
195 *d++ = (U8)(( uv >> 6) | 0xc0);
196 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
197 return d;
198 }
199 if (uv < 0x10000) {
eb160463
GS
200 *d++ = (U8)(( uv >> 12) | 0xe0);
201 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
202 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
203 return d;
204 }
205 if (uv < 0x200000) {
eb160463
GS
206 *d++ = (U8)(( uv >> 18) | 0xf0);
207 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
208 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
209 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
210 return d;
211 }
212 if (uv < 0x4000000) {
eb160463
GS
213 *d++ = (U8)(( uv >> 24) | 0xf8);
214 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
217 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
218 return d;
219 }
220 if (uv < 0x80000000) {
eb160463
GS
221 *d++ = (U8)(( uv >> 30) | 0xfc);
222 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
223 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
226 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
227 return d;
228 }
6b8eaf93 229#ifdef HAS_QUAD
d7578b48 230 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
231#endif
232 {
eb160463
GS
233 *d++ = 0xfe; /* Can't match U+FEFF! */
234 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
235 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
239 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
240 return d;
241 }
6b8eaf93 242#ifdef HAS_QUAD
a0ed51b3 243 {
eb160463
GS
244 *d++ = 0xff; /* Can't match U+FFFE! */
245 *d++ = 0x80; /* 6 Reserved bits */
246 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
247 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
248 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
256 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
257 return d;
258 }
259#endif
1d72bdf6 260#endif /* Loop style */
a0ed51b3 261}
9041c2e3 262
646ca15d
JH
263/*
264
f7d739d1 265Tests if the first C<len> bytes of string C<s> form a valid UTF-8
646ca15d 266character. Note that an INVARIANT (i.e. ASCII) character is a valid
f7d739d1 267UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
268will be returned if it is valid, otherwise 0.
269
270This is the "slow" version as opposed to the "fast" version which is
271the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
272difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
273or less you should use the IS_UTF8_CHAR(), for lengths of five or more
274you should use the _slow(). In practice this means that the _slow()
275will be used very rarely, since the maximum Unicode code point (as of
276Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
277the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
278five bytes or more.
279
280=cut */
c053b435 281STATIC STRLEN
5f66b61c 282S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 283{
cd7e6c88 284 dTHX; /* The function called below requires thread context */
646ca15d 285
cd7e6c88 286 STRLEN actual_len;
646ca15d 287
cd7e6c88 288 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 289
cd7e6c88 290 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 291
cd7e6c88 292 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 293}
9041c2e3
NIS
294
295/*
492a624f
KW
296=for apidoc is_utf8_char_buf
297
298Returns the number of bytes that comprise the first UTF-8 encoded character in
299buffer C<buf>. C<buf_end> should point to one position beyond the end of the
300buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
301encoded character.
302
303Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
304machines) is a valid UTF-8 character.
305
306=cut */
307
308STRLEN
309Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
310{
311
312 STRLEN len;
313
314 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
315
316 if (buf_end <= buf) {
317 return 0;
318 }
319
320 len = buf_end - buf;
321 if (len > UTF8SKIP(buf)) {
322 len = UTF8SKIP(buf);
323 }
324
325#ifdef IS_UTF8_CHAR
326 if (IS_UTF8_CHAR_FAST(len))
327 return IS_UTF8_CHAR(buf, len) ? len : 0;
328#endif /* #ifdef IS_UTF8_CHAR */
329 return is_utf8_char_slow(buf, len);
330}
331
332/*
87cea99e 333=for apidoc is_utf8_char
eebe1485 334
76848387
KW
335DEPRECATED!
336
5da9da9e 337Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
338character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
339character is a valid UTF-8 character. The actual number of bytes in the UTF-8
340character will be returned if it is valid, otherwise 0.
9041c2e3 341
76848387 342This function is deprecated due to the possibility that malformed input could
a1433954 343cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 344instead.
e0328548 345
82686b01 346=cut */
76848387 347
067a85ef 348STRLEN
668b6d8d 349Perl_is_utf8_char(const U8 *s)
386d01d6 350{
7918f24d 351 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 352
76848387 353 /* Assumes we have enough space, which is why this is deprecated */
492a624f 354 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
355}
356
eaf7a4d2 357
6662521e 358/*
87cea99e 359=for apidoc is_utf8_string
6662521e 360
a1433954 361Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 362UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
363using C<strlen(s)> (which means if you use this option, that C<s> has to have a
364terminating NUL byte). Note that all characters being ASCII constitute 'a
365valid UTF-8 string'.
6662521e 366
a1433954 367See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 368
6662521e
GS
369=cut
370*/
371
8e84507e 372bool
668b6d8d 373Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 374{
35da51f7 375 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 376 const U8* x = s;
067a85ef 377
7918f24d 378 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 379
6662521e 380 while (x < send) {
1acdb0da 381 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
382 if (UTF8_IS_INVARIANT(*x)) {
383 x++;
384 }
1acdb0da 385 else if (!UTF8_IS_START(*x))
e0328548 386 return FALSE;
1acdb0da
JH
387 else {
388 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
389 const STRLEN c = UTF8SKIP(x);
390 const U8* const next_char_ptr = x + c;
391
392 if (next_char_ptr > send) {
393 return FALSE;
394 }
395
768c67ee
JH
396 if (IS_UTF8_CHAR_FAST(c)) {
397 if (!IS_UTF8_CHAR(x, c))
e0328548 398 return FALSE;
3c614e38 399 }
e0328548
KW
400 else if (! is_utf8_char_slow(x, c)) {
401 return FALSE;
402 }
403 x = next_char_ptr;
1acdb0da 404 }
6662521e 405 }
768c67ee 406
067a85ef 407 return TRUE;
6662521e
GS
408}
409
67e989fb 410/*
814fafa7
NC
411Implemented as a macro in utf8.h
412
87cea99e 413=for apidoc is_utf8_string_loc
814fafa7 414
a1433954
KW
415Like L</is_utf8_string> but stores the location of the failure (in the
416case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
417"utf8ness success") in the C<ep>.
418
a1433954 419See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 420
87cea99e 421=for apidoc is_utf8_string_loclen
81cd54e3 422
a1433954
KW
423Like L</is_utf8_string>() but stores the location of the failure (in the
424case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
425"utf8ness success") in the C<ep>, and the number of UTF-8
426encoded characters in the C<el>.
427
a1433954 428See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
429
430=cut
431*/
432
433bool
668b6d8d 434Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 435{
35da51f7 436 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 437 const U8* x = s;
81cd54e3 438 STRLEN c;
3ebfea28 439 STRLEN outlen = 0;
7918f24d
NC
440
441 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 442
81cd54e3 443 while (x < send) {
e0328548
KW
444 const U8* next_char_ptr;
445
81cd54e3
JH
446 /* Inline the easy bits of is_utf8_char() here for speed... */
447 if (UTF8_IS_INVARIANT(*x))
e0328548 448 next_char_ptr = x + 1;
768c67ee
JH
449 else if (!UTF8_IS_START(*x))
450 goto out;
81cd54e3 451 else {
768c67ee 452 /* ... and call is_utf8_char() only if really needed. */
768c67ee 453 c = UTF8SKIP(x);
e0328548
KW
454 next_char_ptr = c + x;
455 if (next_char_ptr > send) {
456 goto out;
457 }
768c67ee
JH
458 if (IS_UTF8_CHAR_FAST(c)) {
459 if (!IS_UTF8_CHAR(x, c))
460 c = 0;
461 } else
462 c = is_utf8_char_slow(x, c);
768c67ee
JH
463 if (!c)
464 goto out;
81cd54e3 465 }
e0328548 466 x = next_char_ptr;
3ebfea28 467 outlen++;
81cd54e3 468 }
768c67ee
JH
469
470 out:
3ebfea28
AL
471 if (el)
472 *el = outlen;
473
768c67ee
JH
474 if (ep)
475 *ep = x;
3ebfea28 476 return (x == send);
81cd54e3
JH
477}
478
479/*
768c67ee 480
87cea99e 481=for apidoc utf8n_to_uvuni
67e989fb 482
9041c2e3 483Bottom level UTF-8 decode routine.
746afd53
KW
484Returns the code point value of the first character in the string C<s>,
485which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
486C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
487the length, in bytes, of that character.
949cf498
KW
488
489The value of C<flags> determines the behavior when C<s> does not point to a
490well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
491zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
492next possible position in C<s> that could begin a non-malformed character.
493Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
494
495Various ALLOW flags can be set in C<flags> to allow (and not warn on)
496individual types of malformations, such as the sequence being overlong (that
497is, when there is a shorter sequence that can express the same code point;
498overlong sequences are expressly forbidden in the UTF-8 standard due to
499potential security issues). Another malformation example is the first byte of
500a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
501flags. For allowed 0 length strings, this function returns 0; for allowed
502overlong sequences, the computed code point is returned; for all other allowed
503malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
504determinable reasonable value.
949cf498
KW
505
506The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
507flags) malformation is found. If this flag is set, the routine assumes that
508the caller will raise a warning, and this function will silently just set
509C<retlen> to C<-1> and return zero.
510
511Certain code points are considered problematic. These are Unicode surrogates,
746afd53 512Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 513By default these are considered regular code points, but certain situations
5eafe189 514warrant special handling for them. If C<flags> contains
949cf498
KW
515UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
516malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
517UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
518maximum) can be set to disallow these categories individually.
519
520The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
521UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
522for their respective categories, but otherwise the code points are considered
523valid (not malformations). To get a category to both be treated as a
524malformation and raise a warning, specify both the WARN and DISALLOW flags.
525(But note that warnings are not raised if lexically disabled nor if
526UTF8_CHECK_ONLY is also specified.)
527
528Very large code points (above 0x7FFF_FFFF) are considered more problematic than
529the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
530reasons: they requre at least 32 bits to represent them on ASCII platforms, are
531not representable at all on EBCDIC platforms, and the original UTF-8
532specification never went above this number (the current 0x10FFFF limit was
533imposed later). (The smaller ones, those that fit into 32 bits, are
534representable by a UV on ASCII platforms, but not by an IV, which means that
535the number of operations that can be performed on them is quite restricted.)
536The UTF-8 encoding on ASCII platforms for these large code points begins with a
537byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
538be treated as malformations, while allowing smaller above-Unicode code points.
539(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
540including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
541the other WARN flags, but applies just to these code points.
949cf498
KW
542
543All other code points corresponding to Unicode characters, including private
544use and those yet to be assigned, are never considered malformed and never
545warn.
67e989fb 546
ec5f19d0 547Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
9041c2e3 548
37607a96
PK
549=cut
550*/
67e989fb 551
a0ed51b3 552UV
7fc63493 553Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 554{
97aff369 555 dVAR;
d4c19fe8 556 const U8 * const s0 = s;
eb83ed87 557 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 558 U8 * send;
eb83ed87
KW
559 UV uv = *s;
560 STRLEN expectlen;
949cf498 561 SV* sv = NULL;
eb83ed87
KW
562 UV outlier_ret = 0; /* return value when input is in error or problematic
563 */
564 UV pack_warn = 0; /* Save result of packWARN() for later */
565 bool unexpected_non_continuation = FALSE;
566 bool overflowed = FALSE;
2f8f112e 567 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 568
eb83ed87 569 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 570
eb83ed87 571 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
a0dbb045 572
eb83ed87
KW
573 /* The order of malformation tests here is important. We should consume as
574 * few bytes as possible in order to not skip any valid character. This is
575 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
576 * http://unicode.org/reports/tr36 for more discussion as to why. For
577 * example, once we've done a UTF8SKIP, we can tell the expected number of
578 * bytes, and could fail right off the bat if the input parameters indicate
579 * that there are too few available. But it could be that just that first
580 * byte is garbled, and the intended character occupies fewer bytes. If we
581 * blindly assumed that the first byte is correct, and skipped based on
582 * that number, we could skip over a valid input character. So instead, we
583 * always examine the sequence byte-by-byte.
584 *
585 * We also should not consume too few bytes, otherwise someone could inject
586 * things. For example, an input could be deliberately designed to
587 * overflow, and if this code bailed out immediately upon discovering that,
588 * returning to the caller *retlen pointing to the very next byte (one
589 * which is actually part of of the overflowing sequence), that could look
590 * legitimate to the caller, which could discard the initial partial
591 * sequence and process the rest, inappropriately */
592
593 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 594 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
595 if (retlen) {
596 *retlen = 0;
597 }
a0dbb045 598
eb83ed87
KW
599 if (flags & UTF8_ALLOW_EMPTY) {
600 return 0;
601 }
602 if (! (flags & UTF8_CHECK_ONLY)) {
603 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
604 }
0c443dc2
JH
605 goto malformed;
606 }
607
eb83ed87
KW
608 expectlen = UTF8SKIP(s);
609
610 /* A well-formed UTF-8 character, as the vast majority of calls to this
611 * function will be for, has this expected length. For efficiency, set
612 * things up here to return it. It will be overriden only in those rare
613 * cases where a malformation is found */
614 if (retlen) {
615 *retlen = expectlen;
616 }
617
618 /* An invariant is trivially well-formed */
1d72bdf6 619 if (UTF8_IS_INVARIANT(uv)) {
c4d5f83a 620 return (UV) (NATIVE_TO_UTF(*s));
a0ed51b3 621 }
67e989fb 622
eb83ed87 623 /* A continuation character can't start a valid sequence */
b5b9af04 624 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
625 if (flags & UTF8_ALLOW_CONTINUATION) {
626 if (retlen) {
627 *retlen = 1;
628 }
629 return UNICODE_REPLACEMENT;
630 }
ba210ebe 631
eb83ed87
KW
632 if (! (flags & UTF8_CHECK_ONLY)) {
633 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
634 }
635 curlen = 1;
ba210ebe
JH
636 goto malformed;
637 }
9041c2e3 638
1d72bdf6 639#ifdef EBCDIC
75383841 640 uv = NATIVE_TO_UTF(uv);
1d72bdf6
NIS
641#endif
642
eb83ed87
KW
643 /* Here is not a continuation byte, nor an invariant. The only thing left
644 * is a start byte (possibly for an overlong) */
ba210ebe 645
eb83ed87
KW
646 /* Remove the leading bits that indicate the number of bytes in the
647 * character's whole UTF-8 sequence, leaving just the bits that are part of
648 * the value */
649 uv &= UTF_START_MASK(expectlen);
ba210ebe 650
eb83ed87
KW
651 /* Now, loop through the remaining bytes in the character's sequence,
652 * accumulating each into the working value as we go. Be sure to not look
653 * past the end of the input string */
0b8d30e8
KW
654 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
655
eb83ed87 656 for (s = s0 + 1; s < send; s++) {
b5b9af04 657 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
658#ifndef EBCDIC /* Can't overflow in EBCDIC */
659 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
660
661 /* The original implementors viewed this malformation as more
662 * serious than the others (though I, khw, don't understand
663 * why, since other malformations also give very very wrong
664 * results), so there is no way to turn off checking for it.
665 * Set a flag, but keep going in the loop, so that we absorb
666 * the rest of the bytes that comprise the character. */
667 overflowed = TRUE;
668 overflow_byte = *s; /* Save for warning message's use */
669 }
670#endif
8850bf83 671 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
672 }
673 else {
674 /* Here, found a non-continuation before processing all expected
675 * bytes. This byte begins a new character, so quit, even if
676 * allowing this malformation. */
677 unexpected_non_continuation = TRUE;
678 break;
679 }
680 } /* End of loop through the character's bytes */
681
682 /* Save how many bytes were actually in the character */
683 curlen = s - s0;
684
685 /* The loop above finds two types of malformations: non-continuation and/or
686 * overflow. The non-continuation malformation is really a too-short
687 * malformation, as it means that the current character ended before it was
688 * expected to (being terminated prematurely by the beginning of the next
689 * character, whereas in the too-short malformation there just are too few
690 * bytes available to hold the character. In both cases, the check below
691 * that we have found the expected number of bytes would fail if executed.)
692 * Thus the non-continuation malformation is really unnecessary, being a
693 * subset of the too-short malformation. But there may be existing
694 * applications that are expecting the non-continuation type, so we retain
695 * it, and return it in preference to the too-short malformation. (If this
696 * code were being written from scratch, the two types might be collapsed
697 * into one.) I, khw, am also giving priority to returning the
698 * non-continuation and too-short malformations over overflow when multiple
699 * ones are present. I don't know of any real reason to prefer one over
700 * the other, except that it seems to me that multiple-byte errors trumps
701 * errors from a single byte */
b5b9af04 702 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
703 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
704 if (! (flags & UTF8_CHECK_ONLY)) {
705 if (curlen == 1) {
706 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
707 }
708 else {
709 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
710 }
711 }
eb83ed87
KW
712 goto malformed;
713 }
714 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
715
716 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
717 * as what the original expectations were. */
718 do_overlong_test = FALSE;
eb83ed87
KW
719 if (retlen) {
720 *retlen = curlen;
721 }
722 }
b5b9af04 723 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
724 if (! (flags & UTF8_ALLOW_SHORT)) {
725 if (! (flags & UTF8_CHECK_ONLY)) {
726 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 727 }
eb83ed87
KW
728 goto malformed;
729 }
730 uv = UNICODE_REPLACEMENT;
2f8f112e 731 do_overlong_test = FALSE;
eb83ed87
KW
732 if (retlen) {
733 *retlen = curlen;
734 }
735 }
736
737#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 738 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
739 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
740 {
741 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
742 * generation of the sv, since no warnings are raised under CHECK */
743 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
744 && ckWARN_d(WARN_UTF8))
745 {
42303544
KW
746 /* This message is deliberately not of the same syntax as the other
747 * messages for malformations, for backwards compatibility in the
748 * unlikely event that code is relying on its precise earlier text
749 */
eb83ed87
KW
750 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
751 pack_warn = packWARN(WARN_UTF8);
752 }
753 if (flags & UTF8_DISALLOW_FE_FF) {
754 goto malformed;
ba210ebe 755 }
ba210ebe 756 }
b5b9af04 757 if (UNLIKELY(overflowed)) {
ba210ebe 758
eb83ed87
KW
759 /* If the first byte is FF, it will overflow a 32-bit word. If the
760 * first byte is FE, it will overflow a signed 32-bit word. The
761 * above preserves backward compatibility, since its message was used
762 * in earlier versions of this code in preference to overflow */
763 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 764 goto malformed;
eb83ed87
KW
765 }
766#endif
767
2f8f112e
KW
768 if (do_overlong_test
769 && expectlen > (STRLEN)UNISKIP(uv)
770 && ! (flags & UTF8_ALLOW_LONG))
771 {
eb83ed87
KW
772 /* The overlong malformation has lower precedence than the others.
773 * Note that if this malformation is allowed, we return the actual
774 * value, instead of the replacement character. This is because this
775 * value is actually well-defined. */
776 if (! (flags & UTF8_CHECK_ONLY)) {
777 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));
778 }
779 goto malformed;
780 }
781
782 /* Here, the input is considered to be well-formed , but could be a
783 * problematic code point that is not allowed by the input parameters. */
784 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
785 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
786 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
787 {
949cf498 788 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
789 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
790 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
791 {
111d382d 792 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 793 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
794 }
795 if (flags & UTF8_DISALLOW_SURROGATE) {
796 goto disallowed;
797 }
798 }
949cf498 799 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
800 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
801 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
802 {
111d382d 803 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 804 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
805 }
806 if (flags & UTF8_DISALLOW_SUPER) {
807 goto disallowed;
808 }
809 }
4190d317
KW
810 else if (UNICODE_IS_NONCHAR(uv)) {
811 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
812 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
813 {
814 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
815 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
816 }
817 if (flags & UTF8_DISALLOW_NONCHAR) {
818 goto disallowed;
819 }
820 }
949cf498 821
eb83ed87
KW
822 if (sv) {
823 outlier_ret = uv;
824 goto do_warn;
825 }
826
949cf498
KW
827 /* Here, this is not considered a malformed character, so drop through
828 * to return it */
a0ed51b3 829 }
ba210ebe 830
a0ed51b3 831 return uv;
ba210ebe 832
eb83ed87
KW
833 /* There are three cases which get to beyond this point. In all 3 cases:
834 * <sv> if not null points to a string to print as a warning.
835 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
836 * set.
837 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
838 * This is done by initializing it to 0, and changing it only
839 * for case 1).
840 * The 3 cases are:
841 * 1) The input is valid but problematic, and to be warned about. The
842 * return value is the resultant code point; <*retlen> is set to
843 * <curlen>, the number of bytes that comprise the code point.
844 * <pack_warn> contains the result of packWARN() for the warning
845 * types. The entry point for this case is the label <do_warn>;
846 * 2) The input is a valid code point but disallowed by the parameters to
847 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
848 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
849 * comprise the code point. <pack_warn> contains the result of
850 * packWARN() for the warning types. The entry point for this case is
851 * the label <disallowed>.
852 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
853 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
854 * bytes that comprise the malformation. All such malformations are
855 * assumed to be warning type <utf8>. The entry point for this case
856 * is the label <malformed>.
857 */
949cf498 858
ba210ebe
JH
859malformed:
860
eb83ed87
KW
861 if (sv && ckWARN_d(WARN_UTF8)) {
862 pack_warn = packWARN(WARN_UTF8);
863 }
864
865disallowed:
866
fcc8fcf6 867 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 868 if (retlen)
10edeb5d 869 *retlen = ((STRLEN) -1);
ba210ebe
JH
870 return 0;
871 }
872
eb83ed87 873do_warn:
5b311467 874
eb83ed87
KW
875 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
876 if warnings are to be raised. */
f555bc63 877 const char * const string = SvPVX_const(sv);
a0dbb045 878
f555bc63
KW
879 if (PL_op)
880 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
881 else
882 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
883 }
884
eb83ed87
KW
885 if (retlen) {
886 *retlen = curlen;
887 }
ba210ebe 888
eb83ed87 889 return outlier_ret;
a0ed51b3
LW
890}
891
8e84507e 892/*
ec5f19d0
KW
893=for apidoc utf8_to_uvchr_buf
894
895Returns the native code point of the first character in the string C<s> which
896is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 897C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 898
524080c4
KW
899If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
900enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
901NULL) to -1. If those warnings are off, the computed value if well-defined (or
902the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
903is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
904next possible position in C<s> that could begin a non-malformed character.
905See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
906
907=cut
908*/
909
910
911UV
912Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
913{
914 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
915
916 assert(s < send);
917
918 return utf8n_to_uvchr(s, send - s, retlen,
919 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
920}
921
27d6c58a 922/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c
KW
923 * there are no malformations in the input UTF-8 string C<s>. surrogates,
924 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
925
926UV
927Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
928{
3986bb7c
KW
929 const UV uv = valid_utf8_to_uvuni(s, retlen);
930
27d6c58a
KW
931 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
932
3986bb7c 933 return UNI_TO_NATIVE(uv);
27d6c58a
KW
934}
935
ec5f19d0 936/*
87cea99e 937=for apidoc utf8_to_uvchr
9041c2e3 938
977c1d31
KW
939DEPRECATED!
940
6ee84de2 941Returns the native code point of the first character in the string C<s>
1e54db1a 942which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
943length, in bytes, of that character.
944
4b88fb76 945Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
946malformed input could cause reading beyond the end of the input buffer, which
947is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 948
524080c4
KW
949If C<s> points to one of the detected malformations, and UTF8 warnings are
950enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
951NULL) to -1. If those warnings are off, the computed value if well-defined (or
952the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
953is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
954next possible position in C<s> that could begin a non-malformed character.
955See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
956
957=cut
958*/
959
960UV
7fc63493 961Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 962{
7918f24d
NC
963 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
964
2ff6c191 965 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
966}
967
968/*
ec5f19d0
KW
969=for apidoc utf8_to_uvuni_buf
970
971Returns the Unicode code point of the first character in the string C<s> which
972is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
973C<retlen> will be set to the length, in bytes, of that character.
974
975This function should only be used when the returned UV is considered
976an index into the Unicode semantic tables (e.g. swashes).
977
524080c4
KW
978If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
979enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
980NULL) to -1. If those warnings are off, the computed value if well-defined (or
981the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
982is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
983next possible position in C<s> that could begin a non-malformed character.
984See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
985
986=cut
987*/
988
989UV
990Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
991{
992 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
993
994 assert(send > s);
995
996 /* Call the low level routine asking for checks */
997 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
998 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
999}
1000
27d6c58a 1001/* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1002 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1003 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1004
1005UV
1006Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1007{
3986bb7c
KW
1008 UV expectlen = UTF8SKIP(s);
1009 const U8* send = s + expectlen;
1010 UV uv = NATIVE_TO_UTF(*s);
1011
27d6c58a
KW
1012 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1013
3986bb7c
KW
1014 if (retlen) {
1015 *retlen = expectlen;
1016 }
1017
1018 /* An invariant is trivially returned */
1019 if (expectlen == 1) {
1020 return uv;
1021 }
1022
1023 /* Remove the leading bits that indicate the number of bytes, leaving just
1024 * the bits that are part of the value */
1025 uv &= UTF_START_MASK(expectlen);
1026
1027 /* Now, loop through the remaining bytes, accumulating each into the
1028 * working total as we go */
1029 for (++s; s < send; s++) {
1030 uv = UTF8_ACCUMULATE(uv, *s);
1031 }
1032
1033 return uv;
27d6c58a
KW
1034}
1035
ec5f19d0 1036/*
87cea99e 1037=for apidoc utf8_to_uvuni
9041c2e3 1038
977c1d31
KW
1039DEPRECATED!
1040
9041c2e3 1041Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1042which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1043length, in bytes, of that character.
1044
39e518fd
KW
1045This function should only be used when the returned UV is considered
1046an index into the Unicode semantic tables (e.g. swashes).
1047
4b88fb76 1048Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1049malformed input could cause reading beyond the end of the input buffer, which
1050is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
9041c2e3 1051
524080c4
KW
1052If C<s> points to one of the detected malformations, and UTF8 warnings are
1053enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1054NULL) to -1. If those warnings are off, the computed value if well-defined (or
1055the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1056is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1057next possible position in C<s> that could begin a non-malformed character.
1058See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1059
1060=cut
1061*/
1062
1063UV
7fc63493 1064Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1065{
7918f24d
NC
1066 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1067
4b88fb76 1068 return valid_utf8_to_uvuni(s, retlen);
8e84507e
NIS
1069}
1070
b76347f2 1071/*
87cea99e 1072=for apidoc utf8_length
b76347f2
JH
1073
1074Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1075Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1076up past C<e>, croaks.
b76347f2
JH
1077
1078=cut
1079*/
1080
1081STRLEN
35a4481c 1082Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1083{
97aff369 1084 dVAR;
b76347f2
JH
1085 STRLEN len = 0;
1086
7918f24d
NC
1087 PERL_ARGS_ASSERT_UTF8_LENGTH;
1088
8850bf83
JH
1089 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1090 * the bitops (especially ~) can create illegal UTF-8.
1091 * In other words: in Perl UTF-8 is not just for Unicode. */
1092
a3b680e6
AL
1093 if (e < s)
1094 goto warn_and_return;
b76347f2 1095 while (s < e) {
8e91ec7f
AV
1096 if (!UTF8_IS_INVARIANT(*s))
1097 s += UTF8SKIP(s);
1098 else
1099 s++;
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
NC
1956 return FALSE;
1957 if (!*swash)
711a919c 1958 *swash = swash_init("utf8", swashname, &PL_sv_undef, 1, 0);
bde6a22d
NC
1959 return swash_fetch(*swash, p, TRUE) != 0;
1960}
1961
1962bool
7fc63493 1963Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 1964{
97aff369 1965 dVAR;
7918f24d
NC
1966
1967 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1968
671c33bf
NC
1969 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1970 * descendant of isalnum(3), in other words, it doesn't
1971 * contain the '_'. --jhi */
d4c19fe8 1972 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
a0ed51b3
LW
1973}
1974
1975bool
7fc63493 1976Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1977{
97aff369 1978 dVAR;
7918f24d
NC
1979
1980 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1981
82686b01
JH
1982 if (*p == '_')
1983 return TRUE;
bde6a22d 1984 /* is_utf8_idstart would be more logical. */
d4c19fe8 1985 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
82686b01
JH
1986}
1987
1988bool
c11ff943
KW
1989Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1990{
1991 dVAR;
1992
1993 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1994
1995 if (*p == '_')
1996 return TRUE;
1997 /* is_utf8_idstart would be more logical. */
1998 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
1999}
2000
2001bool
b6912c02
KW
2002Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2003{
2004 dVAR;
2005
2006 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2007
2008 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2009}
2010
2011bool
7fc63493 2012Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2013{
97aff369 2014 dVAR;
7918f24d
NC
2015
2016 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2017
d4c19fe8 2018 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
2019}
2020
2021bool
c11ff943
KW
2022Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2023{
2024 dVAR;
2025
2026 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2027
c11ff943
KW
2028 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2029}
2030
2031bool
7fc63493 2032Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 2033{
97aff369 2034 dVAR;
7918f24d
NC
2035
2036 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2037
d4c19fe8 2038 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
a0ed51b3
LW
2039}
2040
2041bool
7fc63493 2042Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2043{
97aff369 2044 dVAR;
7918f24d
NC
2045
2046 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2047
bc39fe24
KW
2048 /* ASCII characters are the same whether in utf8 or not. So the macro
2049 * works on both utf8 and non-utf8 representations. */
2050 return isASCII(*p);
b8c5462f
JH
2051}
2052
2053bool
bdd8600f
KW
2054Perl_is_utf8_blank(pTHX_ const U8 *p)
2055{
2056 dVAR;
2057
2058 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2059
2060 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2061}
2062
2063bool
7fc63493 2064Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2065{
97aff369 2066 dVAR;
7918f24d
NC
2067
2068 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2069
a34094a9 2070 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
a0ed51b3
LW
2071}
2072
2073bool
d1eb3177
YO
2074Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2075{
2076 dVAR;
2077
2078 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2079
c4428693
KW
2080 /* Only true if is an ASCII space-like character, and ASCII is invariant
2081 * under utf8, so can just use the macro */
2082 return isSPACE_A(*p);
d1eb3177
YO
2083}
2084
2085bool
2086Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2087{
2088 dVAR;
2089
2090 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2091
c4428693
KW
2092 /* Only true if is an ASCII word character, and ASCII is invariant
2093 * under utf8, so can just use the macro */
2094 return isWORDCHAR_A(*p);
d1eb3177
YO
2095}
2096
2097bool
7fc63493 2098Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2099{
97aff369 2100 dVAR;
7918f24d
NC
2101
2102 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2103
d4c19fe8 2104 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
a0ed51b3
LW
2105}
2106
2107bool
d1eb3177
YO
2108Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2109{
2110 dVAR;
2111
2112 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2113
c4428693
KW
2114 /* Only true if is an ASCII digit character, and ASCII is invariant
2115 * under utf8, so can just use the macro */
2116 return isDIGIT_A(*p);
d1eb3177
YO
2117}
2118
2119bool
7fc63493 2120Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2121{
97aff369 2122 dVAR;
7918f24d
NC
2123
2124 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2125
d4c19fe8 2126 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
a0ed51b3
LW
2127}
2128
2129bool
7fc63493 2130Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2131{
97aff369 2132 dVAR;
7918f24d
NC
2133
2134 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2135
d4c19fe8 2136 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
a0ed51b3
LW
2137}
2138
2139bool
7fc63493 2140Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2141{
97aff369 2142 dVAR;
7918f24d
NC
2143
2144 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2145
7b952154
KW
2146 if (isASCII(*p)) {
2147 return isCNTRL_A(*p);
2148 }
2149
2150 /* All controls are in Latin1 */
2151 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2152 return 0;
2153 }
2154 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
b8c5462f
JH
2155}
2156
2157bool
7fc63493 2158Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2159{
97aff369 2160 dVAR;
7918f24d
NC
2161
2162 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2163
d4c19fe8 2164 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
b8c5462f
JH
2165}
2166
2167bool
7fc63493 2168Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2169{
97aff369 2170 dVAR;
7918f24d
NC
2171
2172 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2173
d4c19fe8 2174 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
a0ed51b3
LW
2175}
2176
2177bool
7fc63493 2178Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2179{
97aff369 2180 dVAR;
7918f24d
NC
2181
2182 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2183
d4c19fe8 2184 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
b8c5462f
JH
2185}
2186
2187bool
7fc63493 2188Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2189{
97aff369 2190 dVAR;
7918f24d
NC
2191
2192 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2193
d1eb3177 2194 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
b8c5462f
JH
2195}
2196
2197bool
7fc63493 2198Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2199{
97aff369 2200 dVAR;
7918f24d
NC
2201
2202 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2203
d4c19fe8 2204 return is_utf8_common(p, &PL_utf8_mark, "IsM");
a0ed51b3
LW
2205}
2206
37e2e78e
KW
2207bool
2208Perl_is_utf8_X_begin(pTHX_ const U8 *p)
2209{
2210 dVAR;
2211
2212 PERL_ARGS_ASSERT_IS_UTF8_X_BEGIN;
2213
2214 return is_utf8_common(p, &PL_utf8_X_begin, "_X_Begin");
2215}
2216
2217bool
2218Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2219{
2220 dVAR;
2221
2222 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2223
2224 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2225}
2226
2227bool
2228Perl_is_utf8_X_prepend(pTHX_ const U8 *p)
2229{
2230 dVAR;
2231
2232 PERL_ARGS_ASSERT_IS_UTF8_X_PREPEND;
2233
f957fa32 2234 return is_utf8_common(p, &PL_utf8_X_prepend, "_X_GCB_Prepend");
37e2e78e
KW
2235}
2236
2237bool
2238Perl_is_utf8_X_non_hangul(pTHX_ const U8 *p)
2239{
2240 dVAR;
2241
2242 PERL_ARGS_ASSERT_IS_UTF8_X_NON_HANGUL;
2243
f957fa32 2244 return is_utf8_common(p, &PL_utf8_X_non_hangul, "_X_HST_Not_Applicable");
37e2e78e
KW
2245}
2246
2247bool
2248Perl_is_utf8_X_L(pTHX_ const U8 *p)
2249{
2250 dVAR;
2251
2252 PERL_ARGS_ASSERT_IS_UTF8_X_L;
2253
f957fa32 2254 return is_utf8_common(p, &PL_utf8_X_L, "_X_GCB_L");
37e2e78e
KW
2255}
2256
41de4811
KW
2257/* These constants are for finding GCB=LV and GCB=LVT. These are for the
2258 * pre-composed Hangul syllables, which are all in a contiguous block and
2259 * arranged there in such a way so as to facilitate alorithmic determination of
2260 * their characteristics. As such, they don't need a swash, but can be
2261 * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
2262 * is a GCB=LV */
2263#define SBASE 0xAC00 /* Start of block */
2264#define SCount 11172 /* Length of block */
2265#define TCount 28
2266
218282df 2267#if 0 /* This routine is not currently used */
37e2e78e
KW
2268bool
2269Perl_is_utf8_X_LV(pTHX_ const U8 *p)
2270{
41de4811
KW
2271 /* Unlike most other similarly named routines here, this does not create a
2272 * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
2273
37e2e78e
KW
2274 dVAR;
2275
41de4811
KW
2276 UV cp = valid_utf8_to_uvchr(p, NULL);
2277
37e2e78e
KW
2278 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
2279
41de4811
KW
2280 /* The earliest Unicode releases did not have these precomposed Hangul
2281 * syllables. Set to point to undef in that case, so will return false on
2282 * every call */
2283 if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
2284 PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
2285 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
2286 SvREFCNT_dec(PL_utf8_X_LV);
2287 PL_utf8_X_LV = &PL_sv_undef;
2288 }
2289 }
2290
2291 return (PL_utf8_X_LV != &PL_sv_undef
2292 && cp >= SBASE && cp < SBASE + SCount
2293 && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
37e2e78e 2294}
218282df 2295#endif
37e2e78e
KW
2296
2297bool
2298Perl_is_utf8_X_LVT(pTHX_ const U8 *p)
2299{
41de4811
KW
2300 /* Unlike most other similarly named routines here, this does not create a
2301 * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
2302
37e2e78e
KW
2303 dVAR;
2304
41de4811
KW
2305 UV cp = valid_utf8_to_uvchr(p, NULL);
2306
37e2e78e
KW
2307 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
2308
41de4811
KW
2309 /* The earliest Unicode releases did not have these precomposed Hangul
2310 * syllables. Set to point to undef in that case, so will return false on
2311 * every call */
2312 if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
2313 PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
2314 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
2315 SvREFCNT_dec(PL_utf8_X_LVT);
2316 PL_utf8_X_LVT = &PL_sv_undef;
2317 }
2318 }
2319
2320 return (PL_utf8_X_LVT != &PL_sv_undef
2321 && cp >= SBASE && cp < SBASE + SCount
2322 && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
37e2e78e
KW
2323}
2324
2325bool
2326Perl_is_utf8_X_T(pTHX_ const U8 *p)
2327{
2328 dVAR;
2329
2330 PERL_ARGS_ASSERT_IS_UTF8_X_T;
2331
f957fa32 2332 return is_utf8_common(p, &PL_utf8_X_T, "_X_GCB_T");
37e2e78e
KW
2333}
2334
2335bool
2336Perl_is_utf8_X_V(pTHX_ const U8 *p)
2337{
2338 dVAR;
2339
2340 PERL_ARGS_ASSERT_IS_UTF8_X_V;
2341
f957fa32 2342 return is_utf8_common(p, &PL_utf8_X_V, "_X_GCB_V");
37e2e78e
KW
2343}
2344
2345bool
2346Perl_is_utf8_X_LV_LVT_V(pTHX_ const U8 *p)
2347{
2348 dVAR;
2349
2350 PERL_ARGS_ASSERT_IS_UTF8_X_LV_LVT_V;
2351
2352 return is_utf8_common(p, &PL_utf8_X_LV_LVT_V, "_X_LV_LVT_V");
2353}
2354
2e2b2571
KW
2355bool
2356Perl__is_utf8_quotemeta(pTHX_ const U8 *p)
2357{
2358 /* For exclusive use of pp_quotemeta() */
2359
2360 dVAR;
2361
2362 PERL_ARGS_ASSERT__IS_UTF8_QUOTEMETA;
2363
2364 return is_utf8_common(p, &PL_utf8_quotemeta, "_Perl_Quotemeta");
2365}
2366
6b5c0936 2367/*
87cea99e 2368=for apidoc to_utf8_case
6b5c0936 2369
a1433954
KW
2370The C<p> contains the pointer to the UTF-8 string encoding
2371the character that is being converted. This routine assumes that the character
2372at C<p> is well-formed.
6b5c0936 2373
a1433954
KW
2374The C<ustrp> is a pointer to the character buffer to put the
2375conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2376of the result.
2377
a1433954 2378The C<swashp> is a pointer to the swash to use.
6b5c0936 2379
a1433954
KW
2380Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2381and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2382but not always, a multicharacter mapping), is tried first.
6b5c0936 2383
a1433954 2384The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2385hash %utf8::ToSpecLower. The access to the hash is through
2386Perl_to_utf8_case().
6b5c0936 2387
a1433954 2388The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2389%utf8::ToLower.
2390
2391=cut */
6b5c0936 2392
2104c8d9 2393UV
9a957fbc
AL
2394Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2395 SV **swashp, const char *normal, const char *special)
a0ed51b3 2396{
97aff369 2397 dVAR;
89ebb4a3 2398 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
0134edef 2399 STRLEN len = 0;
4b88fb76 2400 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
1feea2c7
JH
2401 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2402 * are necessary in EBCDIC, they are redundant no-ops
2403 * in ASCII-ish platforms, and hopefully optimized away. */
f54cb97a 2404 const UV uv1 = NATIVE_TO_UNI(uv0);
7918f24d
NC
2405
2406 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2407
9ae3ac1a
KW
2408 /* Note that swash_fetch() doesn't output warnings for these because it
2409 * assumes we will */
8457b38f 2410 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2411 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2412 if (ckWARN_d(WARN_SURROGATE)) {
2413 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2414 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2415 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2416 }
9ae3ac1a
KW
2417 }
2418 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2419 if (ckWARN_d(WARN_NON_UNICODE)) {
2420 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2421 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2422 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2423 }
9ae3ac1a
KW
2424 }
2425
2426 /* Note that non-characters are perfectly legal, so no warning should
2427 * be given */
2428 }
2429
1feea2c7 2430 uvuni_to_utf8(tmpbuf, uv1);
0134edef
JH
2431
2432 if (!*swashp) /* load on-demand */
2433 *swashp = swash_init("utf8", normal, &PL_sv_undef, 4, 0);
2434
a6f87d8c 2435 if (special) {
0134edef 2436 /* It might be "special" (sometimes, but not always,
2a37f04d 2437 * a multicharacter mapping) */
6673a63c 2438 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2439 SV **svp;
2440
35da51f7 2441 if (hv &&
b08cf34e
JH
2442 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2443 (*svp)) {
cfd0369c 2444 const char *s;
47654450 2445
cfd0369c 2446 s = SvPV_const(*svp, len);
47654450
JH
2447 if (len == 1)
2448 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2a37f04d 2449 else {
2f9475ad
JH
2450#ifdef EBCDIC
2451 /* If we have EBCDIC we need to remap the characters
2452 * since any characters in the low 256 are Unicode
2453 * code points, not EBCDIC. */
7cda7a3d 2454 U8 *t = (U8*)s, *tend = t + len, *d;
2f9475ad
JH
2455
2456 d = tmpbuf;
b08cf34e 2457 if (SvUTF8(*svp)) {
2f9475ad
JH
2458 STRLEN tlen = 0;
2459
2460 while (t < tend) {
6bd1c396 2461 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2f9475ad
JH
2462 if (tlen > 0) {
2463 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2464 t += tlen;
2465 }
2466 else
2467 break;
2468 }
2469 }
2470 else {
36fec512
JH
2471 while (t < tend) {
2472 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2473 t++;
2474 }
2f9475ad
JH
2475 }
2476 len = d - tmpbuf;
2477 Copy(tmpbuf, ustrp, len, U8);
2478#else
d2dcd0fb 2479 Copy(s, ustrp, len, U8);
2f9475ad 2480#endif
29e98929 2481 }
983ffd37 2482 }
0134edef
JH
2483 }
2484
2485 if (!len && *swashp) {
2114036c 2486 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
d4c19fe8 2487
0134edef
JH
2488 if (uv2) {
2489 /* It was "normal" (a single character mapping). */
d4c19fe8 2490 const UV uv3 = UNI_TO_NATIVE(uv2);
e9101d72 2491 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2a37f04d
JH
2492 }
2493 }
1feea2c7 2494
cbe07460
KW
2495 if (len) {
2496 if (lenp) {
2497 *lenp = len;
2498 }
2499 return valid_utf8_to_uvchr(ustrp, 0);
2500 }
2501
2502 /* Here, there was no mapping defined, which means that the code point maps
2503 * to itself. Return the inputs */
bfdf22ec
KW
2504 len = UTF8SKIP(p);
2505 Copy(p, ustrp, len, U8);
0134edef 2506
2a37f04d
JH
2507 if (lenp)
2508 *lenp = len;
2509
cbe07460
KW
2510 return uv0;
2511
a0ed51b3
LW
2512}
2513
051a06d4
KW
2514STATIC UV
2515S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2516{
2517 /* This is called when changing the case of a utf8-encoded character above
2518 * the Latin1 range, and the operation is in locale. If the result
2519 * contains a character that crosses the 255/256 boundary, disallow the
2520 * change, and return the original code point. See L<perlfunc/lc> for why;
2521 *
a1433954
KW
2522 * p points to the original string whose case was changed; assumed
2523 * by this routine to be well-formed
051a06d4
KW
2524 * result the code point of the first character in the changed-case string
2525 * ustrp points to the changed-case string (<result> represents its first char)
2526 * lenp points to the length of <ustrp> */
2527
2528 UV original; /* To store the first code point of <p> */
2529
2530 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2531
2532 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2533
2534 /* We know immediately if the first character in the string crosses the
2535 * boundary, so can skip */
2536 if (result > 255) {
2537
2538 /* Look at every character in the result; if any cross the
2539 * boundary, the whole thing is disallowed */
2540 U8* s = ustrp + UTF8SKIP(ustrp);
2541 U8* e = ustrp + *lenp;
2542 while (s < e) {
2543 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2544 {
2545 goto bad_crossing;
2546 }
2547 s += UTF8SKIP(s);
2548 }
2549
2550 /* Here, no characters crossed, result is ok as-is */
2551 return result;
2552 }
2553
2554bad_crossing:
2555
2556 /* Failed, have to return the original */
4b88fb76 2557 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2558 Copy(p, ustrp, *lenp, char);
2559 return original;
2560}
2561
d3e79532 2562/*
87cea99e 2563=for apidoc to_utf8_upper
d3e79532 2564
a1433954
KW
2565Convert the UTF-8 encoded character at C<p> to its uppercase version and
2566store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
89ebb4a3
JH
2567that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2568the uppercase version may be longer than the original character.
d3e79532
JH
2569
2570The first character of the uppercased version is returned
2571(but note, as explained above, that there may be more.)
2572
a1433954
KW
2573The character at C<p> is assumed by this routine to be well-formed.
2574
d3e79532
JH
2575=cut */
2576
051a06d4
KW
2577/* Not currently externally documented, and subject to change:
2578 * <flags> is set iff locale semantics are to be used for code points < 256
2579 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2580 * were used in the calculation; otherwise unchanged. */
2581
2104c8d9 2582UV
051a06d4 2583Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2584{
97aff369 2585 dVAR;
7918f24d 2586
051a06d4
KW
2587 UV result;
2588
2589 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2590
3a4c58c9 2591 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2592 if (flags) {
2593 result = toUPPER_LC(*p);
2594 }
2595 else {
81c6c7ce 2596 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2597 }
3a4c58c9
KW
2598 }
2599 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2600 if (flags) {
2601 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2602 }
2603 else {
81c6c7ce
KW
2604 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2605 ustrp, lenp, 'S');
051a06d4
KW
2606 }
2607 }
2608 else { /* utf8, ord above 255 */
2609 result = CALL_UPPER_CASE(p, ustrp, lenp);
2610
2611 if (flags) {
2612 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2613 }
2614 return result;
2615 }
2616
2617 /* Here, used locale rules. Convert back to utf8 */
2618 if (UTF8_IS_INVARIANT(result)) {
2619 *ustrp = (U8) result;
2620 *lenp = 1;
2621 }
2622 else {
2623 *ustrp = UTF8_EIGHT_BIT_HI(result);
2624 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2625 *lenp = 2;
3a4c58c9
KW
2626 }
2627
051a06d4
KW
2628 if (tainted_ptr) {
2629 *tainted_ptr = TRUE;
2630 }
2631 return result;
983ffd37 2632}
a0ed51b3 2633
d3e79532 2634/*
87cea99e 2635=for apidoc to_utf8_title
d3e79532 2636
a1433954
KW
2637Convert the UTF-8 encoded character at C<p> to its titlecase version and
2638store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2639that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2640titlecase version may be longer than the original character.
d3e79532
JH
2641
2642The first character of the titlecased version is returned
2643(but note, as explained above, that there may be more.)
2644
a1433954
KW
2645The character at C<p> is assumed by this routine to be well-formed.
2646
d3e79532
JH
2647=cut */
2648
051a06d4
KW
2649/* Not currently externally documented, and subject to change:
2650 * <flags> is set iff locale semantics are to be used for code points < 256
2651 * Since titlecase is not defined in POSIX, uppercase is used instead
2652 * for these/
2653 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2654 * were used in the calculation; otherwise unchanged. */
2655
983ffd37 2656UV
051a06d4 2657Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2658{
97aff369 2659 dVAR;
7918f24d 2660
051a06d4
KW
2661 UV result;
2662
2663 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2664
3a4c58c9 2665 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2666 if (flags) {
2667 result = toUPPER_LC(*p);
2668 }
2669 else {
81c6c7ce 2670 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2671 }
3a4c58c9
KW
2672 }
2673 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2674 if (flags) {
2675 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2676 }
2677 else {
81c6c7ce
KW
2678 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2679 ustrp, lenp, 's');
051a06d4
KW
2680 }
2681 }
2682 else { /* utf8, ord above 255 */
2683 result = CALL_TITLE_CASE(p, ustrp, lenp);
2684
2685 if (flags) {
2686 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2687 }
2688 return result;
2689 }
2690
2691 /* Here, used locale rules. Convert back to utf8 */
2692 if (UTF8_IS_INVARIANT(result)) {
2693 *ustrp = (U8) result;
2694 *lenp = 1;
2695 }
2696 else {
2697 *ustrp = UTF8_EIGHT_BIT_HI(result);
2698 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2699 *lenp = 2;
3a4c58c9
KW
2700 }
2701
051a06d4
KW
2702 if (tainted_ptr) {
2703 *tainted_ptr = TRUE;
2704 }
2705 return result;
a0ed51b3
LW
2706}
2707
d3e79532 2708/*
87cea99e 2709=for apidoc to_utf8_lower
d3e79532 2710
a1433954
KW
2711Convert the UTF-8 encoded character at C<p> to its lowercase version and
2712store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2713that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2714lowercase version may be longer than the original character.
d3e79532
JH
2715
2716The first character of the lowercased version is returned
2717(but note, as explained above, that there may be more.)
2718
a1433954
KW
2719The character at C<p> is assumed by this routine to be well-formed.
2720
d3e79532
JH
2721=cut */
2722
051a06d4
KW
2723/* Not currently externally documented, and subject to change:
2724 * <flags> is set iff locale semantics are to be used for code points < 256
2725 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2726 * were used in the calculation; otherwise unchanged. */
2727
2104c8d9 2728UV
051a06d4 2729Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2730{
051a06d4
KW
2731 UV result;
2732
97aff369 2733 dVAR;
7918f24d 2734
051a06d4 2735 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2736
968c5e6a 2737 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2738 if (flags) {
2739 result = toLOWER_LC(*p);
2740 }
2741 else {
81c6c7ce 2742 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2743 }
968c5e6a
KW
2744 }
2745 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2746 if (flags) {
2747 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2748 }
2749 else {
81c6c7ce
KW
2750 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2751 ustrp, lenp);
051a06d4 2752 }
968c5e6a 2753 }
051a06d4
KW
2754 else { /* utf8, ord above 255 */
2755 result = CALL_LOWER_CASE(p, ustrp, lenp);
2756
2757 if (flags) {
2758 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2759 }
968c5e6a 2760
051a06d4
KW
2761 return result;
2762 }
2763
2764 /* Here, used locale rules. Convert back to utf8 */
2765 if (UTF8_IS_INVARIANT(result)) {
2766 *ustrp = (U8) result;
2767 *lenp = 1;
2768 }
2769 else {
2770 *ustrp = UTF8_EIGHT_BIT_HI(result);
2771 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2772 *lenp = 2;
2773 }
2774
2775 if (tainted_ptr) {
2776 *tainted_ptr = TRUE;
2777 }
2778 return result;
b4e400f9
JH
2779}
2780
d3e79532 2781/*
87cea99e 2782=for apidoc to_utf8_fold
d3e79532 2783
a1433954
KW
2784Convert the UTF-8 encoded character at C<p> to its foldcase version and
2785store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2786that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
d3e79532
JH
2787foldcase version may be longer than the original character (up to
2788three characters).
2789
2790The first character of the foldcased version is returned
2791(but note, as explained above, that there may be more.)
2792
a1433954
KW
2793The character at C<p> is assumed by this routine to be well-formed.
2794
d3e79532
JH
2795=cut */
2796
051a06d4
KW
2797/* Not currently externally documented, and subject to change,
2798 * in <flags>
2799 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2800 * points < 256. Since foldcase is not defined in
2801 * POSIX, lowercase is used instead
2802 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2803 * otherwise simple folds
a0270393
KW
2804 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2805 * prohibited
051a06d4
KW
2806 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2807 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2808
b4e400f9 2809UV
051a06d4 2810Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2811{
97aff369 2812 dVAR;
7918f24d 2813
051a06d4
KW
2814 UV result;
2815
36bb2ab6 2816 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2817
a0270393
KW
2818 /* These are mutually exclusive */
2819 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2820
50ba90ff
KW
2821 assert(p != ustrp); /* Otherwise overwrites */
2822
a1dde8de 2823 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2824 if (flags & FOLD_FLAGS_LOCALE) {
2825 result = toLOWER_LC(*p);
2826 }
2827 else {
81c6c7ce
KW
2828 return _to_fold_latin1(*p, ustrp, lenp,
2829 cBOOL(flags & FOLD_FLAGS_FULL));
051a06d4 2830 }
a1dde8de
KW
2831 }
2832 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2833 if (flags & FOLD_FLAGS_LOCALE) {
2834 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2835 }
2836 else {
81c6c7ce 2837 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
a0270393
KW
2838 ustrp, lenp,
2839 cBOOL((flags & FOLD_FLAGS_FULL
2840 /* If ASCII safe, don't allow full
2841 * folding, as that could include SHARP
2842 * S => ss; otherwise there is no
2843 * crossing of ascii/non-ascii in the
2844 * latin1 range */
2845 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
051a06d4 2846 }
a1dde8de 2847 }
051a06d4 2848 else { /* utf8, ord above 255 */
a0270393 2849 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2850
051a06d4 2851 if ((flags & FOLD_FLAGS_LOCALE)) {
a0270393 2852 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2853 }
a0270393
KW
2854 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2855 return result;
2856 }
2857 else {
2858 /* This is called when changing the case of a utf8-encoded
2859 * character above the Latin1 range, and the result should not
2860 * contain an ASCII character. */
2861
2862 UV original; /* To store the first code point of <p> */
2863
2864 /* Look at every character in the result; if any cross the
2865 * boundary, the whole thing is disallowed */
2866 U8* s = ustrp;
2867 U8* e = ustrp + *lenp;
2868 while (s < e) {
2869 if (isASCII(*s)) {
2870 /* Crossed, have to return the original */
2871 original = valid_utf8_to_uvchr(p, lenp);
2872 Copy(p, ustrp, *lenp, char);
2873 return original;
2874 }
2875 s += UTF8SKIP(s);
2876 }
051a06d4 2877
a0270393
KW
2878 /* Here, no characters crossed, result is ok as-is */
2879 return result;
2880 }
051a06d4
KW
2881 }
2882
2883 /* Here, used locale rules. Convert back to utf8 */
2884 if (UTF8_IS_INVARIANT(result)) {
2885 *ustrp = (U8) result;
2886 *lenp = 1;
2887 }
2888 else {
2889 *ustrp = UTF8_EIGHT_BIT_HI(result);
2890 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2891 *lenp = 2;
2892 }
2893
2894 if (tainted_ptr) {
2895 *tainted_ptr = TRUE;
2896 }
2897 return result;
a0ed51b3
LW
2898}
2899
711a919c 2900/* Note:
f90a9a02 2901 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2902 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2903 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2904 */
c4a5db0c 2905
a0ed51b3 2906SV*
7fc63493 2907Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2908{
c4a5db0c
KW
2909 PERL_ARGS_ASSERT_SWASH_INIT;
2910
2911 /* Returns a copy of a swash initiated by the called function. This is the
2912 * public interface, and returning a copy prevents others from doing
2913 * mischief on the original */
2914
9a53f6cf 2915 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, FALSE, NULL, FALSE));
c4a5db0c
KW
2916}
2917
2918SV*
9a53f6cf 2919Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, bool return_if_undef, SV* invlist, bool passed_in_invlist_has_user_defined_property)
c4a5db0c
KW
2920{
2921 /* Initialize and return a swash, creating it if necessary. It does this
9a53f6cf 2922 * by calling utf8_heavy.pl in the general case.
c4a5db0c
KW
2923 *
2924 * This interface should only be used by functions that won't destroy or
2925 * adversely change the swash, as doing so affects all other uses of the
2926 * swash in the program; the general public should use 'Perl_swash_init'
2927 * instead.
2928 *
2929 * pkg is the name of the package that <name> should be in.
2930 * name is the name of the swash to find. Typically it is a Unicode
2931 * property name, including user-defined ones
2932 * listsv is a string to initialize the swash with. It must be of the form
2933 * documented as the subroutine return value in
2934 * L<perlunicode/User-Defined Character Properties>
2935 * minbits is the number of bits required to represent each data element.
2936 * It is '1' for binary properties.
2937 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf
KW
2938 * return_if_undef is TRUE if the routine shouldn't croak if it can't find
2939 * the requested property
2940 * invlist is an inversion list to initialize the swash with (or NULL)
2941 * has_user_defined_property is TRUE if <invlist> has some component that
2942 * came from a user-defined property
2943 *
2944 * Thus there are three possible inputs to find the swash: <name>,
2945 * <listsv>, and <invlist>. At least one must be specified. The result
2946 * will be the union of the specified ones, although <listsv>'s various
2947 * actions can intersect, etc. what <name> gives.
2948 *
2949 * <invlist> is only valid for binary properties */
c4a5db0c 2950
27da23d5 2951 dVAR;
c4a5db0c 2952 SV* retval = &PL_sv_undef;
9a53f6cf
KW
2953
2954 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2955 assert(! invlist || minbits == 1);
2956
2957 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2958 * so */
2959 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2960 dSP;
2961 const size_t pkg_len = strlen(pkg);
2962 const size_t name_len = strlen(name);
2963 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2964 SV* errsv_save;
2965 GV *method;
2966
2967 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2968
2969 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2970 ENTER;
69794297
KW
2971 SAVEHINTS();
2972 save_re_context();
2973 if (PL_parser && PL_parser->error_count)
2974 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2975 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2976 if (!method) { /* demand load utf8 */
2977 ENTER;
2978 errsv_save = newSVsv(ERRSV);
2979 /* It is assumed that callers of this routine are not passing in
2980 * any user derived data. */
2981 /* Need to do this after save_re_context() as it will set
2982 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2983 * in Perl_magic_get). Even line to create errsv_save can turn on
2984 * PL_tainted. */
2985 SAVEBOOL(PL_tainted);
2986 PL_tainted = 0;
2987 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2988 NULL);
2989 if (!SvTRUE(ERRSV))
2990 sv_setsv(ERRSV, errsv_save);
2991 SvREFCNT_dec(errsv_save);
2992 LEAVE;
2993 }
2994 SPAGAIN;
2995 PUSHMARK(SP);
2996 EXTEND(SP,5);
2997 mPUSHp(pkg, pkg_len);
2998 mPUSHp(name, name_len);
2999 PUSHs(listsv);
3000 mPUSHi(minbits);
3001 mPUSHi(none);
3002 PUTBACK;
f8be5cf0 3003 errsv_save = newSVsv(ERRSV);
69794297
KW
3004 /* If we already have a pointer to the method, no need to use
3005 * call_method() to repeat the lookup. */
3006 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
3007 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3008 {
3009 retval = *PL_stack_sp--;
3010 SvREFCNT_inc(retval);
3011 }
f8be5cf0
JH
3012 if (!SvTRUE(ERRSV))
3013 sv_setsv(ERRSV, errsv_save);
3014 SvREFCNT_dec(errsv_save);
ce3b816e 3015 LEAVE;
69794297
KW
3016 POPSTACK;
3017 if (IN_PERL_COMPILETIME) {
3018 CopHINTS_set(PL_curcop, PL_hints);
3019 }
3020 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3021 if (SvPOK(retval))
3022
3023 /* If caller wants to handle missing properties, let them */
3024 if (return_if_undef) {
3025 return NULL;
3026 }
3027 Perl_croak(aTHX_
3028 "Can't find Unicode property definition \"%"SVf"\"",
3029 SVfARG(retval));
3030 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3031 }
9a53f6cf 3032 } /* End of calling the module to find the swash */
36eb48b4
KW
3033
3034 /* Make sure there is an inversion list for binary properties */
3035 if (minbits == 1) {
3036 SV** swash_invlistsvp = NULL;
3037 SV* swash_invlist = NULL;
9a53f6cf 3038 bool invlist_in_swash_is_valid = FALSE;
88d45d28 3039 HV* swash_hv = NULL;
36eb48b4 3040
9a53f6cf
KW
3041 /* If this operation fetched a swash, get its already existing
3042 * inversion list or create one for it */
3043 if (retval != &PL_sv_undef) {
36eb48b4
KW
3044 swash_hv = MUTABLE_HV(SvRV(retval));
3045
3046 swash_invlistsvp = hv_fetchs(swash_hv, "INVLIST", FALSE);
9a53f6cf
KW
3047 if (swash_invlistsvp) {
3048 swash_invlist = *swash_invlistsvp;
3049 invlist_in_swash_is_valid = TRUE;
3050 }
3051 else {
36eb48b4 3052 swash_invlist = _swash_to_invlist(retval);
9a53f6cf
KW
3053 }
3054 }
3055
3056 /* If an inversion list was passed in, have to include it */
3057 if (invlist) {
3058
3059 /* Any fetched swash will by now have an inversion list in it;
3060 * otherwise <swash_invlist> will be NULL, indicating that we
3061 * didn't fetch a swash */
3062 if (swash_invlist) {
3063
3064 /* Add the passed-in inversion list, which invalidates the one
3065 * already stored in the swash */
3066 invlist_in_swash_is_valid = FALSE;
3067 _invlist_union(invlist, swash_invlist, &swash_invlist);
3068 }
3069 else {
3070
3071 /* Here, there is no swash already. Set up a minimal one */
3072 swash_hv = newHV();
3073 retval = newRV_inc(MUTABLE_SV(swash_hv));
3074 swash_invlist = invlist;
3075 }
3076
3077 if (passed_in_invlist_has_user_defined_property) {
3078 if (! hv_stores(swash_hv, "USER_DEFINED", newSVuv(1))) {
3079 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3080 }
3081 }
3082 }
3083
3084 /* Here, we have computed the union of all the passed-in data. It may
3085 * be that there was an inversion list in the swash which didn't get
3086 * touched; otherwise save the one computed one */
3087 if (! invlist_in_swash_is_valid) {
69794297
KW
3088 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "INVLIST", swash_invlist))
3089 {
3090 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3091 }
9a53f6cf 3092 }
36eb48b4
KW
3093 }
3094
a0ed51b3
LW
3095 return retval;
3096}
3097
035d37be
JH
3098
3099/* This API is wrong for special case conversions since we may need to
3100 * return several Unicode characters for a single Unicode character
3101 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3102 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
3103 * multiple values. --jhi
3104 * For those, you should use to_utf8_case() instead */
b0e3252e 3105/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3106
3107/* Note:
3108 * Returns the value of property/mapping C<swash> for the first character
3109 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3110 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3111 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3112 *
3113 * A "swash" is a hash which contains initially the keys/values set up by
3114 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3115 * property for all possible code points. Things are stored in a compact form
3116 * (see utf8_heavy.pl) so that calculation is required to find the actual
3117 * property value for a given code point. As code points are looked up, new
3118 * key/value pairs are added to the hash, so that the calculation doesn't have
3119 * to ever be re-done. Further, each calculation is done, not just for the
3120 * desired one, but for a whole block of code points adjacent to that one.
3121 * For binary properties on ASCII machines, the block is usually for 64 code
3122 * points, starting with a code point evenly divisible by 64. Thus if the
3123 * property value for code point 257 is requested, the code goes out and
3124 * calculates the property values for all 64 code points between 256 and 319,
3125 * and stores these as a single 64-bit long bit vector, called a "swatch",
3126 * under the key for code point 256. The key is the UTF-8 encoding for code
3127 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3128 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3129 * for code point 258 is then requested, this code realizes that it would be
3130 * stored under the key for 256, and would find that value and extract the
3131 * relevant bit, offset from 256.
3132 *
3133 * Non-binary properties are stored in as many bits as necessary to represent
3134 * their values (32 currently, though the code is more general than that), not
3135 * as single bits, but the principal is the same: the value for each key is a
3136 * vector that encompasses the property values for all code points whose UTF-8
3137 * representations are represented by the key. That is, for all code points
3138 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3139 * bytes of that.
680c470c 3140 */
a0ed51b3 3141UV
680c470c 3142Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3143{
27da23d5 3144 dVAR;
ef8f7699 3145 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3146 U32 klen;
3147 U32 off;
a0ed51b3 3148 STRLEN slen;
7d85a32c 3149 STRLEN needents;
cfd0369c 3150 const U8 *tmps = NULL;
a0ed51b3 3151 U32 bit;
979f2922 3152 SV *swatch;
3568d838 3153 U8 tmputf8[2];
35da51f7 3154 const UV c = NATIVE_TO_ASCII(*ptr);
3568d838 3155
7918f24d
NC
3156 PERL_ARGS_ASSERT_SWASH_FETCH;
3157
dbe7a391 3158 /* Convert to utf8 if not already */
3568d838 3159 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
979f2922
ST
3160 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3161 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3162 ptr = tmputf8;
3568d838
JH
3163 }
3164 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
37e2e78e 3165 * then the "swatch" is a vec() for all the chars which start
3568d838
JH
3166 * with 0xAA..0xYY
3167 * So the key in the hash (klen) is length of encoded char -1
3168 */
3169 klen = UTF8SKIP(ptr) - 1;
3170 off = ptr[klen];
a0ed51b3 3171
979f2922 3172 if (klen == 0) {
37e2e78e 3173 /* If char is invariant then swatch is for all the invariant chars
1e54db1a 3174 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
7d85a32c 3175 */
979f2922
ST
3176 needents = UTF_CONTINUATION_MARK;
3177 off = NATIVE_TO_UTF(ptr[klen]);
3178 }
3179 else {
7d85a32c 3180 /* If char is encoded then swatch is for the prefix */
979f2922
ST
3181 needents = (1 << UTF_ACCUMULATION_SHIFT);
3182 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3183 }
7d85a32c 3184
a0ed51b3
LW
3185 /*
3186 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3187 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3188 * it's nothing to sniff at.) Pity we usually come through at least
3189 * two function calls to get here...
3190 *
3191 * NB: this code assumes that swatches are never modified, once generated!
3192 */
3193
3568d838 3194 if (hv == PL_last_swash_hv &&
a0ed51b3 3195 klen == PL_last_swash_klen &&
27da23d5 3196 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3197 {
3198 tmps = PL_last_swash_tmps;
3199 slen = PL_last_swash_slen;
3200 }
3201 else {
3202 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3203 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3204
b0e3252e 3205 /* If not cached, generate it via swatch_get */
979f2922
ST
3206 if (!svp || !SvPOK(*svp)
3207 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2b9d42f0
NIS
3208 /* We use utf8n_to_uvuni() as we want an index into
3209 Unicode tables, not a native character number.
3210 */
aec46f14 3211 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
872c91ae
JH
3212 ckWARN(WARN_UTF8) ?
3213 0 : UTF8_ALLOW_ANY);
b0e3252e 3214 swatch = swatch_get(swash,
979f2922 3215 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
361ee0fe 3216 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
979f2922
ST
3217 needents);
3218
923e4eb5 3219 if (IN_PERL_COMPILETIME)
623e6609 3220 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3221
979f2922 3222 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3223
979f2922
ST
3224 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3225 || (slen << 3) < needents)
5637ef5b
NC
3226 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3227 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3228 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3229 }
3230
3231 PL_last_swash_hv = hv;
16d8f38a 3232 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3233 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3234 /* FIXME change interpvar.h? */
3235 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3236 PL_last_swash_slen = slen;
3237 if (klen)
3238 Copy(ptr, PL_last_swash_key, klen, U8);
3239 }
3240
9faf8d75 3241 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3242 case 1:
3243 bit = 1 << (off & 7);
3244 off >>= 3;
3245 return (tmps[off] & bit) != 0;
3246 case 8:
3247 return tmps[off];
3248 case 16:
3249 off <<= 1;
3250 return (tmps[off] << 8) + tmps[off + 1] ;
3251 case 32:
3252 off <<= 2;
3253 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3254 }
5637ef5b
NC
3255 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3256 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3257 NORETURN_FUNCTION_END;
a0ed51b3 3258}
2b9d42f0 3259
319009ee
KW
3260/* Read a single line of the main body of the swash input text. These are of
3261 * the form:
3262 * 0053 0056 0073
3263 * where each number is hex. The first two numbers form the minimum and
3264 * maximum of a range, and the third is the value associated with the range.
3265 * Not all swashes should have a third number
3266 *
3267 * On input: l points to the beginning of the line to be examined; it points
3268 * to somewhere in the string of the whole input text, and is
3269 * terminated by a \n or the null string terminator.
3270 * lend points to the null terminator of that string
3271 * wants_value is non-zero if the swash expects a third number
3272 * typestr is the name of the swash's mapping, like 'ToLower'
3273 * On output: *min, *max, and *val are set to the values read from the line.
3274 * returns a pointer just beyond the line examined. If there was no
3275 * valid min number on the line, returns lend+1
3276 */
3277
3278STATIC U8*
3279S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3280 const bool wants_value, const U8* const typestr)
3281{
3282 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3283 STRLEN numlen; /* Length of the number */
02470786
KW
3284 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3285 | PERL_SCAN_DISALLOW_PREFIX
3286 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3287
3288 /* nl points to the next \n in the scan */
3289 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3290
3291 /* Get the first number on the line: the range minimum */
3292 numlen = lend - l;
3293 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3294 if (numlen) /* If found a hex number, position past it */
3295 l += numlen;
3296 else if (nl) { /* Else, go handle next line, if any */
3297 return nl + 1; /* 1 is length of "\n" */
3298 }
3299 else { /* Else, no next line */
3300 return lend + 1; /* to LIST's end at which \n is not found */
3301 }
3302
3303 /* The max range value follows, separated by a BLANK */
3304 if (isBLANK(*l)) {
3305 ++l;
02470786
KW
3306 flags = PERL_SCAN_SILENT_ILLDIGIT
3307 | PERL_SCAN_DISALLOW_PREFIX
3308 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3309 numlen = lend - l;
3310 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3311 if (numlen)
3312 l += numlen;
3313 else /* If no value here, it is a single element range */
3314 *max = *min;
3315
3316 /* Non-binary tables have a third entry: what the first element of the
3317 * range maps to */
3318 if (wants_value) {
3319 if (isBLANK(*l)) {
3320 ++l;
f90a9a02
KW
3321
3322 /* The ToLc, etc table mappings are not in hex, and must be
3323 * corrected by adding the code point to them */
3324 if (typeto) {
3325 char *after_strtol = (char *) lend;
3326 *val = Strtol((char *)l, &after_strtol, 10);
3327 l = (U8 *) after_strtol;
f90a9a02
KW
3328 }
3329 else { /* Other tables are in hex, and are the correct result
3330 without tweaking */
a9d188b3
KW
3331 flags = PERL_SCAN_SILENT_ILLDIGIT
3332 | PERL_SCAN_DISALLOW_PREFIX
3333 | PERL_SCAN_SILENT_NON_PORTABLE;
3334 numlen = lend - l;
3335 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3336 if (numlen)
3337 l += numlen;
3338 else
3339 *val = 0;
f90a9a02 3340 }
319009ee
KW
3341 }
3342 else {
3343 *val = 0;
3344 if (typeto) {
dcbac5bb 3345 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3346 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3347 typestr, l);
3348 }
3349 }
3350 }
3351 else
3352 *val = 0; /* bits == 1, then any val should be ignored */
3353 }
3354 else { /* Nothing following range min, should be single element with no
3355 mapping expected */
3356 *max = *min;
3357 if (wants_value) {
3358 *val = 0;
3359 if (typeto) {
dcbac5bb 3360 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3361 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3362 }
3363 }
3364 else
3365 *val = 0; /* bits == 1, then val should be ignored */
3366 }
3367
3368 /* Position to next line if any, or EOF */
3369 if (nl)
3370 l = nl + 1;
3371 else
3372 l = lend;
3373
3374 return l;
3375}
3376
979f2922
ST
3377/* Note:
3378 * Returns a swatch (a bit vector string) for a code point sequence
3379 * that starts from the value C<start> and comprises the number C<span>.
3380 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3381 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3382 */
3383STATIC SV*
b0e3252e 3384S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3385{
3386 SV *swatch;
77f9f126 3387 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3388 STRLEN lcur, xcur, scur;
ef8f7699 3389 HV *const hv = MUTABLE_HV(SvRV(swash));
36eb48b4
KW
3390 SV** const invlistsvp = hv_fetchs(hv, "INVLIST", FALSE);
3391
88d45d28
KW
3392 SV** listsvp = NULL; /* The string containing the main body of the table */
3393 SV** extssvp = NULL;
3394 SV** invert_it_svp = NULL;
3395 U8* typestr = NULL;
786861f5
KW
3396 STRLEN bits;
3397 STRLEN octets; /* if bits == 1, then octets == 0 */
3398 UV none;
3399 UV end = start + span;
972dd592 3400
36eb48b4 3401 if (invlistsvp == NULL) {
786861f5
KW
3402 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3403 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3404 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3405 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3406 listsvp = hv_fetchs(hv, "LIST", FALSE);
3407 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3408
3409 bits = SvUV(*bitssvp);
3410 none = SvUV(*nonesvp);
3411 typestr = (U8*)SvPV_nolen(*typesvp);
3412 }
36eb48b4
KW
3413 else {
3414 bits = 1;
3415 none = 0;
3416 }
786861f5 3417 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3418
b0e3252e 3419 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3420
979f2922 3421 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3422 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3423 (UV)bits);
979f2922
ST
3424 }
3425
84ea5ef6
KW
3426 /* If overflowed, use the max possible */
3427 if (end < start) {
3428 end = UV_MAX;
3429 span = end - start;
3430 }
3431
979f2922 3432 /* create and initialize $swatch */
979f2922 3433 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3434 swatch = newSV(scur);
3435 SvPOK_on(swatch);
979f2922
ST
3436 s = (U8*)SvPVX(swatch);
3437 if (octets && none) {
0bd48802 3438 const U8* const e = s + scur;
979f2922
ST
3439 while (s < e) {
3440 if (bits == 8)
3441 *s++ = (U8)(none & 0xff);
3442 else if (bits == 16) {
3443 *s++ = (U8)((none >> 8) & 0xff);
3444 *s++ = (U8)( none & 0xff);
3445 }
3446 else if (bits == 32) {
3447 *s++ = (U8)((none >> 24) & 0xff);
3448 *s++ = (U8)((none >> 16) & 0xff);
3449 *s++ = (U8)((none >> 8) & 0xff);
3450 *s++ = (U8)( none & 0xff);
3451 }
3452 }
3453 *s = '\0';
3454 }
3455 else {
3456 (void)memzero((U8*)s, scur + 1);
3457 }
3458 SvCUR_set(swatch, scur);
3459 s = (U8*)SvPVX(swatch);
3460
36eb48b4
KW
3461 if (invlistsvp) { /* If has an inversion list set up use that */
3462 _invlist_populate_swatch(*invlistsvp, start, end, s);
3463 return swatch;
3464 }
3465
3466 /* read $swash->{LIST} */
979f2922
ST
3467 l = (U8*)SvPV(*listsvp, lcur);
3468 lend = l + lcur;
3469 while (l < lend) {
8ed25d53 3470 UV min, max, val, upper;
319009ee
KW
3471 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3472 cBOOL(octets), typestr);
3473 if (l > lend) {
979f2922
ST
3474 break;
3475 }
3476
972dd592 3477 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3478 if (max < start)
3479 continue;
3480
8ed25d53
KW
3481 /* <end> is generally 1 beyond where we want to set things, but at the
3482 * platform's infinity, where we can't go any higher, we want to
3483 * include the code point at <end> */
3484 upper = (max < end)
3485 ? max
3486 : (max != UV_MAX || end != UV_MAX)
3487 ? end - 1
3488 : end;
3489
979f2922 3490 if (octets) {
35da51f7 3491 UV key;
979f2922
ST
3492 if (min < start) {
3493 if (!none || val < none) {
3494 val += start - min;
3495 }
3496 min = start;
3497 }
8ed25d53 3498 for (key = min; key <= upper; key++) {
979f2922 3499 STRLEN offset;
979f2922
ST
3500 /* offset must be non-negative (start <= min <= key < end) */
3501 offset = octets * (key - start);
3502 if (bits == 8)
3503 s[offset] = (U8)(val & 0xff);
3504 else if (bits == 16) {
3505 s[offset ] = (U8)((val >> 8) & 0xff);
3506 s[offset + 1] = (U8)( val & 0xff);
3507 }
3508 else if (bits == 32) {
3509 s[offset ] = (U8)((val >> 24) & 0xff);
3510 s[offset + 1] = (U8)((val >> 16) & 0xff);
3511 s[offset + 2] = (U8)((val >> 8) & 0xff);
3512 s[offset + 3] = (U8)( val & 0xff);
3513 }
3514
3515 if (!none || val < none)
3516 ++val;
3517 }
3518 }
711a919c 3519 else { /* bits == 1, then val should be ignored */
35da51f7 3520 UV key;
979f2922
ST
3521 if (min < start)
3522 min = start;
6cb05c12 3523
8ed25d53 3524 for (key = min; key <= upper; key++) {
0bd48802 3525 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3526 s[offset >> 3] |= 1 << (offset & 7);
3527 }
3528 }
3529 } /* while */
979f2922 3530
9479a769 3531 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3532 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3533
3534 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3535 * be 0, and their inversion should also be 0, as we don't succeed any
3536 * Unicode property matches for non-Unicode code points */
3537 if (start <= PERL_UNICODE_MAX) {
3538
3539 /* The code below assumes that we never cross the
3540 * Unicode/above-Unicode boundary in a range, as otherwise we would
3541 * have to figure out where to stop flipping the bits. Since this
3542 * boundary is divisible by a large power of 2, and swatches comes
3543 * in small powers of 2, this should be a valid assumption */
3544 assert(start + span - 1 <= PERL_UNICODE_MAX);
3545
507a8485
KW
3546 send = s + scur;
3547 while (s < send) {
3548 *s = ~(*s);
3549 s++;
3550 }
0bda3001 3551 }
77f9f126
KW
3552 }
3553
d73c39c5
KW
3554 /* read $swash->{EXTRAS}
3555 * This code also copied to swash_to_invlist() below */
979f2922
ST
3556 x = (U8*)SvPV(*extssvp, xcur);
3557 xend = x + xcur;
3558 while (x < xend) {
3559 STRLEN namelen;
3560 U8 *namestr;
3561 SV** othersvp;
3562 HV* otherhv;
3563 STRLEN otherbits;
3564 SV **otherbitssvp, *other;
711a919c 3565 U8 *s, *o, *nl;
979f2922
ST
3566 STRLEN slen, olen;
3567
35da51f7 3568 const U8 opc = *x++;
979f2922
ST
3569 if (opc == '\n')
3570 continue;
3571
3572 nl = (U8*)memchr(x, '\n', xend - x);
3573
3574 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3575 if (nl) {
3576 x = nl + 1; /* 1 is length of "\n" */
3577 continue;
3578 }
3579 else {
3580 x = xend; /* to EXTRAS' end at which \n is not found */
3581 break;
3582 }
3583 }
3584
3585 namestr = x;
3586 if (nl) {
3587 namelen = nl - namestr;
3588 x = nl + 1;
3589 }
3590 else {
3591 namelen = xend - namestr;
3592 x = xend;
3593 }
3594
3595 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3596 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3597 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3598 otherbits = (STRLEN)SvUV(*otherbitssvp);
3599 if (bits < otherbits)
5637ef5b
NC
3600 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3601 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3602
3603 /* The "other" swatch must be destroyed after. */
b0e3252e 3604 other = swatch_get(*othersvp, start, span);
979f2922
ST
3605 o = (U8*)SvPV(other, olen);
3606
3607 if (!olen)
b0e3252e 3608 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3609
3610 s = (U8*)SvPV(swatch, slen);
3611 if (bits == 1 && otherbits == 1) {
3612 if (slen != olen)
5637ef5b
NC
3613 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3614 "mismatch, slen=%"UVuf", olen=%"UVuf,
3615 (UV)slen, (UV)olen);
979f2922
ST
3616
3617 switch (opc) {
3618 case '+':
3619 while (slen--)
3620 *s++ |= *o++;
3621 break;
3622 case '!':
3623 while (slen--)
3624 *s++ |= ~*o++;
3625 break;
3626 case '-':
3627 while (slen--)
3628 *s++ &= ~*o++;
3629 break;
3630 case '&':
3631 while (slen--)
3632 *s++ &= *o++;
3633 break;
3634 default:
3635 break;
3636 }
3637 }
711a919c 3638 else {
979f2922
ST
3639 STRLEN otheroctets = otherbits >> 3;
3640 STRLEN offset = 0;
35da51f7 3641 U8* const send = s + slen;
979f2922
ST
3642
3643 while (s < send) {
3644 UV otherval = 0;
3645
3646 if (otherbits == 1) {
3647 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3648 ++offset;
3649 }
3650 else {
3651 STRLEN vlen = otheroctets;
3652 otherval = *o++;
3653 while (--vlen) {
3654 otherval <<= 8;
3655 otherval |= *o++;
3656 }
3657 }
3658
711a919c 3659 if (opc == '+' && otherval)
6f207bd3 3660 NOOP; /* replace with otherval */
979f2922
ST
3661 else if (opc == '!' && !otherval)
3662 otherval = 1;
3663 else if (opc == '-' && otherval)
3664 otherval = 0;
3665 else if (opc == '&' && !otherval)
3666 otherval = 0;
3667 else {
711a919c 3668 s += octets; /* no replacement */
979f2922
ST
3669 continue;
3670 }
3671
3672 if (bits == 8)
3673 *s++ = (U8)( otherval & 0xff);
3674 else if (bits == 16) {
3675 *s++ = (U8)((otherval >> 8) & 0xff);
3676 *s++ = (U8)( otherval & 0xff);
3677 }
3678 else if (bits == 32) {
3679 *s++ = (U8)((otherval >> 24) & 0xff);
3680 *s++ = (U8)((otherval >> 16) & 0xff);
3681 *s++ = (U8)((otherval >> 8) & 0xff);
3682 *s++ = (U8)( otherval & 0xff);
3683 }
3684 }
3685 }
3686 sv_free(other); /* through with it! */
3687 } /* while */
3688 return swatch;
3689}
3690
064c021d 3691HV*
4c2e1131 3692Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3693{
3694
5662e334
KW
3695 /* Subject to change or removal. For use only in one place in regcomp.c.
3696 * Can't be used on a property that is subject to user override, as it
3697 * relies on the value of SPECIALS in the swash which would be set by
3698 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3699 * for overridden properties
064c021d
KW
3700 *
3701 * Returns a hash which is the inversion and closure of a swash mapping.
3702 * For example, consider the input lines:
3703 * 004B 006B
3704 * 004C 006C
3705 * 212A 006B
3706 *
3707 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3708 * 006C. The value for each key is an array. For 006C, the array would
3709 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3710 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3711 *
3712 * Essentially, for any code point, it gives all the code points that map to
3713 * it, or the list of 'froms' for that point.
3714 *
5662e334
KW
3715 * Currently it ignores any additions or deletions from other swashes,
3716 * looking at just the main body of the swash, and if there are SPECIALS
3717 * in the swash, at that hash
3718 *
3719 * The specials hash can be extra code points, and most likely consists of
3720 * maps from single code points to multiple ones (each expressed as a string
3721 * of utf8 characters). This function currently returns only 1-1 mappings.
3722 * However consider this possible input in the specials hash:
3723 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3724 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3725 *
3726 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3727 * currently handle. But it also means that FB05 and FB06 are equivalent in
3728 * a 1-1 mapping which we should handle, and this relationship may not be in
3729 * the main table. Therefore this function examines all the multi-char
3730 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3731
3732 U8 *l, *lend;
3733 STRLEN lcur;
3734 HV *const hv = MUTABLE_HV(SvRV(swash));
3735
3736 /* The string containing the main body of the table */
3737 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3738
3739 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3740 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3741 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3742 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3743 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3744 const STRLEN bits = SvUV(*bitssvp);
3745 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3746 const UV none = SvUV(*nonesvp);
5662e334 3747 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3748
3749 HV* ret = newHV();
3750
3751 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3752
3753 /* Must have at least 8 bits to get the mappings */
3754 if (bits != 8 && bits != 16 && bits != 32) {
3755 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3756 (UV)bits);
3757 }
3758
5662e334
KW
3759 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3760 mapping t