This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
handy.h: Change documentation for perlapi
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
a0ed51b3 35
a0c21aa1 36#ifndef EBCDIC
970ea3cb 37/* Separate prototypes needed because in ASCII systems these are
a0c21aa1
JH
38 * usually macros but they still are compiled as code, too. */
39PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
6dd9dce9 40PERL_CALLCONV UV Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen);
a0c21aa1
JH
41PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
42#endif
43
27da23d5
JH
44static const char unees[] =
45 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 46
48ef279e 47/*
ccfc67b7 48=head1 Unicode Support
a0ed51b3 49
166f8a29
DM
50This file contains various utility functions for manipulating UTF8-encoded
51strings. For the uninitiated, this is a method of representing arbitrary
61296642 52Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
53characters in the ASCII range are unmodified, and a zero byte never appears
54within non-zero characters.
166f8a29 55
eaf7a4d2
CS
56=cut
57*/
58
59/*
60=for apidoc is_ascii_string
61
a1433954 62Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
63or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
64is, if they are invariant. On ASCII-ish machines, only ASCII characters
65fit this definition, hence the function's name.
eaf7a4d2 66
9f7e3d64
MH
67If C<len> is 0, it will be calculated using C<strlen(s)>.
68
a1433954 69See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
70
71=cut
72*/
73
74bool
668b6d8d 75Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
76{
77 const U8* const send = s + (len ? len : strlen((const char *)s));
78 const U8* x = s;
79
80 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
81
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
84 break;
85 }
86
87 return x == send;
88}
89
90/*
87cea99e 91=for apidoc uvuni_to_utf8_flags
eebe1485 92
6ee84de2
KW
93Adds the UTF-8 representation of the code point C<uv> to the end
94of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
eebe1485 95bytes available. The return value is the pointer to the byte after the
9041c2e3 96end of the new character. In other words,
eebe1485 97
b851fbc1
JH
98 d = uvuni_to_utf8_flags(d, uv, flags);
99
100or, in most cases,
101
9041c2e3 102 d = uvuni_to_utf8(d, uv);
eebe1485 103
b851fbc1
JH
104(which is equivalent to)
105
106 d = uvuni_to_utf8_flags(d, uv, 0);
107
949cf498 108This is the recommended Unicode-aware way of saying
eebe1485
SC
109
110 *(d++) = uv;
111
949cf498
KW
112This function will convert to UTF-8 (and not warn) even code points that aren't
113legal Unicode or are problematic, unless C<flags> contains one or more of the
a1433954
KW
114following flags:
115
949cf498
KW
116If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
117the function will raise a warning, provided UTF8 warnings are enabled. If instead
118UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
119If both flags are set, the function will both warn and return NULL.
120
121The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
122affect how the function handles a Unicode non-character. And, likewise for the
123UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
124above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
125even less portable) can be warned and/or disallowed even if other above-Unicode
126code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
127flags.
128
129And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
130above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
131DISALLOW flags.
132
133
eebe1485
SC
134=cut
135*/
136
dfe13c55 137U8 *
b851fbc1 138Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 139{
7918f24d
NC
140 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
141
979f77b6
KW
142 /* The first problematic code point is the first surrogate */
143 if (uv >= UNICODE_SURROGATE_FIRST
144 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
145 {
949cf498
KW
146 if (UNICODE_IS_SURROGATE(uv)) {
147 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 148 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
149 "UTF-16 surrogate U+%04"UVXf, uv);
150 }
151 if (flags & UNICODE_DISALLOW_SURROGATE) {
152 return NULL;
153 }
154 }
155 else if (UNICODE_IS_SUPER(uv)) {
156 if (flags & UNICODE_WARN_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
158 {
8457b38f 159 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
160 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
161 }
162 if (flags & UNICODE_DISALLOW_SUPER
163 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
164 {
165 return NULL;
166 }
167 }
168 else if (UNICODE_IS_NONCHAR(uv)) {
169 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 170 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
171 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
172 uv);
173 }
174 if (flags & UNICODE_DISALLOW_NONCHAR) {
175 return NULL;
176 }
177 }
507b9800 178 }
c4d5f83a 179 if (UNI_IS_INVARIANT(uv)) {
eb160463 180 *d++ = (U8)UTF_TO_NATIVE(uv);
a0ed51b3
LW
181 return d;
182 }
2d331972 183#if defined(EBCDIC)
1d72bdf6
NIS
184 else {
185 STRLEN len = UNISKIP(uv);
186 U8 *p = d+len-1;
187 while (p > d) {
eb160463 188 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
189 uv >>= UTF_ACCUMULATION_SHIFT;
190 }
eb160463 191 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
192 return d+len;
193 }
194#else /* Non loop style */
a0ed51b3 195 if (uv < 0x800) {
eb160463
GS
196 *d++ = (U8)(( uv >> 6) | 0xc0);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
198 return d;
199 }
200 if (uv < 0x10000) {
eb160463
GS
201 *d++ = (U8)(( uv >> 12) | 0xe0);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
204 return d;
205 }
206 if (uv < 0x200000) {
eb160463
GS
207 *d++ = (U8)(( uv >> 18) | 0xf0);
208 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
209 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
210 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
211 return d;
212 }
213 if (uv < 0x4000000) {
eb160463
GS
214 *d++ = (U8)(( uv >> 24) | 0xf8);
215 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
218 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
219 return d;
220 }
221 if (uv < 0x80000000) {
eb160463
GS
222 *d++ = (U8)(( uv >> 30) | 0xfc);
223 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
227 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
228 return d;
229 }
6b8eaf93 230#ifdef HAS_QUAD
d7578b48 231 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
232#endif
233 {
eb160463
GS
234 *d++ = 0xfe; /* Can't match U+FEFF! */
235 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
239 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
240 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
241 return d;
242 }
6b8eaf93 243#ifdef HAS_QUAD
a0ed51b3 244 {
eb160463
GS
245 *d++ = 0xff; /* Can't match U+FFFE! */
246 *d++ = 0x80; /* 6 Reserved bits */
247 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
248 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
257 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
258 return d;
259 }
260#endif
1d72bdf6 261#endif /* Loop style */
a0ed51b3 262}
9041c2e3 263
646ca15d
JH
264/*
265
f7d739d1 266Tests if the first C<len> bytes of string C<s> form a valid UTF-8
646ca15d 267character. Note that an INVARIANT (i.e. ASCII) character is a valid
f7d739d1 268UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
269will be returned if it is valid, otherwise 0.
270
271This is the "slow" version as opposed to the "fast" version which is
272the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
273difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
274or less you should use the IS_UTF8_CHAR(), for lengths of five or more
275you should use the _slow(). In practice this means that the _slow()
276will be used very rarely, since the maximum Unicode code point (as of
277Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
278the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
279five bytes or more.
280
281=cut */
7af276bc 282PERL_STATIC_INLINE STRLEN
5f66b61c 283S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 284{
cd7e6c88 285 dTHX; /* The function called below requires thread context */
646ca15d 286
cd7e6c88 287 STRLEN actual_len;
646ca15d 288
cd7e6c88 289 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 290
cd7e6c88 291 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 292
cd7e6c88 293 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 294}
9041c2e3
NIS
295
296/*
492a624f
KW
297=for apidoc is_utf8_char_buf
298
299Returns the number of bytes that comprise the first UTF-8 encoded character in
300buffer C<buf>. C<buf_end> should point to one position beyond the end of the
301buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
302encoded character.
303
304Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
305machines) is a valid UTF-8 character.
306
307=cut */
308
309STRLEN
310Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
311{
312
313 STRLEN len;
314
315 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
316
317 if (buf_end <= buf) {
318 return 0;
319 }
320
321 len = buf_end - buf;
322 if (len > UTF8SKIP(buf)) {
323 len = UTF8SKIP(buf);
324 }
325
326#ifdef IS_UTF8_CHAR
327 if (IS_UTF8_CHAR_FAST(len))
328 return IS_UTF8_CHAR(buf, len) ? len : 0;
329#endif /* #ifdef IS_UTF8_CHAR */
330 return is_utf8_char_slow(buf, len);
331}
332
333/*
87cea99e 334=for apidoc is_utf8_char
eebe1485 335
76848387
KW
336DEPRECATED!
337
5da9da9e 338Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
339character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
340character is a valid UTF-8 character. The actual number of bytes in the UTF-8
341character will be returned if it is valid, otherwise 0.
9041c2e3 342
76848387 343This function is deprecated due to the possibility that malformed input could
a1433954 344cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 345instead.
e0328548 346
82686b01 347=cut */
76848387 348
067a85ef 349STRLEN
668b6d8d 350Perl_is_utf8_char(const U8 *s)
386d01d6 351{
7918f24d 352 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 353
76848387 354 /* Assumes we have enough space, which is why this is deprecated */
492a624f 355 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
356}
357
eaf7a4d2 358
6662521e 359/*
87cea99e 360=for apidoc is_utf8_string
6662521e 361
a1433954 362Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 363UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
364using C<strlen(s)> (which means if you use this option, that C<s> has to have a
365terminating NUL byte). Note that all characters being ASCII constitute 'a
366valid UTF-8 string'.
6662521e 367
a1433954 368See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 369
6662521e
GS
370=cut
371*/
372
8e84507e 373bool
668b6d8d 374Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 375{
35da51f7 376 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 377 const U8* x = s;
067a85ef 378
7918f24d 379 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 380
6662521e 381 while (x < send) {
1acdb0da 382 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
383 if (UTF8_IS_INVARIANT(*x)) {
384 x++;
385 }
1acdb0da
JH
386 else {
387 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
388 const STRLEN c = UTF8SKIP(x);
389 const U8* const next_char_ptr = x + c;
390
391 if (next_char_ptr > send) {
392 return FALSE;
393 }
394
768c67ee
JH
395 if (IS_UTF8_CHAR_FAST(c)) {
396 if (!IS_UTF8_CHAR(x, c))
e0328548 397 return FALSE;
3c614e38 398 }
e0328548
KW
399 else if (! is_utf8_char_slow(x, c)) {
400 return FALSE;
401 }
402 x = next_char_ptr;
1acdb0da 403 }
6662521e 404 }
768c67ee 405
067a85ef 406 return TRUE;
6662521e
GS
407}
408
67e989fb 409/*
814fafa7
NC
410Implemented as a macro in utf8.h
411
87cea99e 412=for apidoc is_utf8_string_loc
814fafa7 413
a1433954
KW
414Like L</is_utf8_string> but stores the location of the failure (in the
415case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
416"utf8ness success") in the C<ep>.
417
a1433954 418See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 419
87cea99e 420=for apidoc is_utf8_string_loclen
81cd54e3 421
a1433954
KW
422Like L</is_utf8_string>() but stores the location of the failure (in the
423case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
424"utf8ness success") in the C<ep>, and the number of UTF-8
425encoded characters in the C<el>.
426
a1433954 427See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
428
429=cut
430*/
431
432bool
668b6d8d 433Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 434{
35da51f7 435 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 436 const U8* x = s;
81cd54e3 437 STRLEN c;
3ebfea28 438 STRLEN outlen = 0;
7918f24d
NC
439
440 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 441
81cd54e3 442 while (x < send) {
e0328548
KW
443 const U8* next_char_ptr;
444
81cd54e3
JH
445 /* Inline the easy bits of is_utf8_char() here for speed... */
446 if (UTF8_IS_INVARIANT(*x))
e0328548 447 next_char_ptr = x + 1;
81cd54e3 448 else {
768c67ee 449 /* ... and call is_utf8_char() only if really needed. */
768c67ee 450 c = UTF8SKIP(x);
e0328548
KW
451 next_char_ptr = c + x;
452 if (next_char_ptr > send) {
453 goto out;
454 }
768c67ee
JH
455 if (IS_UTF8_CHAR_FAST(c)) {
456 if (!IS_UTF8_CHAR(x, c))
457 c = 0;
458 } else
459 c = is_utf8_char_slow(x, c);
768c67ee
JH
460 if (!c)
461 goto out;
81cd54e3 462 }
e0328548 463 x = next_char_ptr;
3ebfea28 464 outlen++;
81cd54e3 465 }
768c67ee
JH
466
467 out:
3ebfea28
AL
468 if (el)
469 *el = outlen;
470
768c67ee
JH
471 if (ep)
472 *ep = x;
3ebfea28 473 return (x == send);
81cd54e3
JH
474}
475
476/*
768c67ee 477
87cea99e 478=for apidoc utf8n_to_uvuni
67e989fb 479
9041c2e3 480Bottom level UTF-8 decode routine.
746afd53
KW
481Returns the code point value of the first character in the string C<s>,
482which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
483C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
484the length, in bytes, of that character.
949cf498
KW
485
486The value of C<flags> determines the behavior when C<s> does not point to a
487well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
488zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
489next possible position in C<s> that could begin a non-malformed character.
490Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
491
492Various ALLOW flags can be set in C<flags> to allow (and not warn on)
493individual types of malformations, such as the sequence being overlong (that
494is, when there is a shorter sequence that can express the same code point;
495overlong sequences are expressly forbidden in the UTF-8 standard due to
496potential security issues). Another malformation example is the first byte of
497a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
498flags. For allowed 0 length strings, this function returns 0; for allowed
499overlong sequences, the computed code point is returned; for all other allowed
500malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
501determinable reasonable value.
949cf498
KW
502
503The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
504flags) malformation is found. If this flag is set, the routine assumes that
505the caller will raise a warning, and this function will silently just set
d088425d
KW
506C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
507
508Note that this API requires disambiguation between successful decoding a NUL
509character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
510in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
511first byte of C<s> is 0 as well. If so, the input was a NUL; if not, the input
512had an error.
949cf498
KW
513
514Certain code points are considered problematic. These are Unicode surrogates,
746afd53 515Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 516By default these are considered regular code points, but certain situations
5eafe189 517warrant special handling for them. If C<flags> contains
949cf498
KW
518UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
519malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
520UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
521maximum) can be set to disallow these categories individually.
522
523The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
524UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
525for their respective categories, but otherwise the code points are considered
526valid (not malformations). To get a category to both be treated as a
527malformation and raise a warning, specify both the WARN and DISALLOW flags.
528(But note that warnings are not raised if lexically disabled nor if
529UTF8_CHECK_ONLY is also specified.)
530
531Very large code points (above 0x7FFF_FFFF) are considered more problematic than
532the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
533reasons: they requre at least 32 bits to represent them on ASCII platforms, are
534not representable at all on EBCDIC platforms, and the original UTF-8
535specification never went above this number (the current 0x10FFFF limit was
536imposed later). (The smaller ones, those that fit into 32 bits, are
537representable by a UV on ASCII platforms, but not by an IV, which means that
538the number of operations that can be performed on them is quite restricted.)
539The UTF-8 encoding on ASCII platforms for these large code points begins with a
540byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
541be treated as malformations, while allowing smaller above-Unicode code points.
542(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
543including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
544the other WARN flags, but applies just to these code points.
949cf498
KW
545
546All other code points corresponding to Unicode characters, including private
547use and those yet to be assigned, are never considered malformed and never
548warn.
67e989fb 549
ec5f19d0 550Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
9041c2e3 551
37607a96
PK
552=cut
553*/
67e989fb 554
a0ed51b3 555UV
7fc63493 556Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 557{
97aff369 558 dVAR;
d4c19fe8 559 const U8 * const s0 = s;
eb83ed87 560 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 561 U8 * send;
eb83ed87
KW
562 UV uv = *s;
563 STRLEN expectlen;
949cf498 564 SV* sv = NULL;
eb83ed87
KW
565 UV outlier_ret = 0; /* return value when input is in error or problematic
566 */
567 UV pack_warn = 0; /* Save result of packWARN() for later */
568 bool unexpected_non_continuation = FALSE;
569 bool overflowed = FALSE;
2f8f112e 570 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 571
eb83ed87 572 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 573
eb83ed87 574 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
a0dbb045 575
eb83ed87
KW
576 /* The order of malformation tests here is important. We should consume as
577 * few bytes as possible in order to not skip any valid character. This is
578 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
579 * http://unicode.org/reports/tr36 for more discussion as to why. For
580 * example, once we've done a UTF8SKIP, we can tell the expected number of
581 * bytes, and could fail right off the bat if the input parameters indicate
582 * that there are too few available. But it could be that just that first
583 * byte is garbled, and the intended character occupies fewer bytes. If we
584 * blindly assumed that the first byte is correct, and skipped based on
585 * that number, we could skip over a valid input character. So instead, we
586 * always examine the sequence byte-by-byte.
587 *
588 * We also should not consume too few bytes, otherwise someone could inject
589 * things. For example, an input could be deliberately designed to
590 * overflow, and if this code bailed out immediately upon discovering that,
591 * returning to the caller *retlen pointing to the very next byte (one
592 * which is actually part of of the overflowing sequence), that could look
593 * legitimate to the caller, which could discard the initial partial
594 * sequence and process the rest, inappropriately */
595
596 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 597 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
598 if (retlen) {
599 *retlen = 0;
600 }
a0dbb045 601
eb83ed87
KW
602 if (flags & UTF8_ALLOW_EMPTY) {
603 return 0;
604 }
605 if (! (flags & UTF8_CHECK_ONLY)) {
606 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
607 }
0c443dc2
JH
608 goto malformed;
609 }
610
eb83ed87
KW
611 expectlen = UTF8SKIP(s);
612
613 /* A well-formed UTF-8 character, as the vast majority of calls to this
614 * function will be for, has this expected length. For efficiency, set
615 * things up here to return it. It will be overriden only in those rare
616 * cases where a malformation is found */
617 if (retlen) {
618 *retlen = expectlen;
619 }
620
621 /* An invariant is trivially well-formed */
1d72bdf6 622 if (UTF8_IS_INVARIANT(uv)) {
c4d5f83a 623 return (UV) (NATIVE_TO_UTF(*s));
a0ed51b3 624 }
67e989fb 625
eb83ed87 626 /* A continuation character can't start a valid sequence */
b5b9af04 627 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
628 if (flags & UTF8_ALLOW_CONTINUATION) {
629 if (retlen) {
630 *retlen = 1;
631 }
632 return UNICODE_REPLACEMENT;
633 }
ba210ebe 634
eb83ed87
KW
635 if (! (flags & UTF8_CHECK_ONLY)) {
636 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
637 }
638 curlen = 1;
ba210ebe
JH
639 goto malformed;
640 }
9041c2e3 641
1d72bdf6 642#ifdef EBCDIC
75383841 643 uv = NATIVE_TO_UTF(uv);
1d72bdf6
NIS
644#endif
645
eb83ed87
KW
646 /* Here is not a continuation byte, nor an invariant. The only thing left
647 * is a start byte (possibly for an overlong) */
ba210ebe 648
eb83ed87
KW
649 /* Remove the leading bits that indicate the number of bytes in the
650 * character's whole UTF-8 sequence, leaving just the bits that are part of
651 * the value */
652 uv &= UTF_START_MASK(expectlen);
ba210ebe 653
eb83ed87
KW
654 /* Now, loop through the remaining bytes in the character's sequence,
655 * accumulating each into the working value as we go. Be sure to not look
656 * past the end of the input string */
0b8d30e8
KW
657 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
658
eb83ed87 659 for (s = s0 + 1; s < send; s++) {
b5b9af04 660 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
661#ifndef EBCDIC /* Can't overflow in EBCDIC */
662 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
663
664 /* The original implementors viewed this malformation as more
665 * serious than the others (though I, khw, don't understand
666 * why, since other malformations also give very very wrong
667 * results), so there is no way to turn off checking for it.
668 * Set a flag, but keep going in the loop, so that we absorb
669 * the rest of the bytes that comprise the character. */
670 overflowed = TRUE;
671 overflow_byte = *s; /* Save for warning message's use */
672 }
673#endif
8850bf83 674 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
675 }
676 else {
677 /* Here, found a non-continuation before processing all expected
678 * bytes. This byte begins a new character, so quit, even if
679 * allowing this malformation. */
680 unexpected_non_continuation = TRUE;
681 break;
682 }
683 } /* End of loop through the character's bytes */
684
685 /* Save how many bytes were actually in the character */
686 curlen = s - s0;
687
688 /* The loop above finds two types of malformations: non-continuation and/or
689 * overflow. The non-continuation malformation is really a too-short
690 * malformation, as it means that the current character ended before it was
691 * expected to (being terminated prematurely by the beginning of the next
692 * character, whereas in the too-short malformation there just are too few
693 * bytes available to hold the character. In both cases, the check below
694 * that we have found the expected number of bytes would fail if executed.)
695 * Thus the non-continuation malformation is really unnecessary, being a
696 * subset of the too-short malformation. But there may be existing
697 * applications that are expecting the non-continuation type, so we retain
698 * it, and return it in preference to the too-short malformation. (If this
699 * code were being written from scratch, the two types might be collapsed
700 * into one.) I, khw, am also giving priority to returning the
701 * non-continuation and too-short malformations over overflow when multiple
702 * ones are present. I don't know of any real reason to prefer one over
703 * the other, except that it seems to me that multiple-byte errors trumps
704 * errors from a single byte */
b5b9af04 705 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
706 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
707 if (! (flags & UTF8_CHECK_ONLY)) {
708 if (curlen == 1) {
709 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
710 }
711 else {
712 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
713 }
714 }
eb83ed87
KW
715 goto malformed;
716 }
717 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
718
719 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
720 * as what the original expectations were. */
721 do_overlong_test = FALSE;
eb83ed87
KW
722 if (retlen) {
723 *retlen = curlen;
724 }
725 }
b5b9af04 726 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
727 if (! (flags & UTF8_ALLOW_SHORT)) {
728 if (! (flags & UTF8_CHECK_ONLY)) {
729 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 730 }
eb83ed87
KW
731 goto malformed;
732 }
733 uv = UNICODE_REPLACEMENT;
2f8f112e 734 do_overlong_test = FALSE;
eb83ed87
KW
735 if (retlen) {
736 *retlen = curlen;
737 }
738 }
739
740#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 741 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
742 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
743 {
744 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
745 * generation of the sv, since no warnings are raised under CHECK */
746 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
747 && ckWARN_d(WARN_UTF8))
748 {
42303544
KW
749 /* This message is deliberately not of the same syntax as the other
750 * messages for malformations, for backwards compatibility in the
751 * unlikely event that code is relying on its precise earlier text
752 */
eb83ed87
KW
753 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
754 pack_warn = packWARN(WARN_UTF8);
755 }
756 if (flags & UTF8_DISALLOW_FE_FF) {
757 goto malformed;
ba210ebe 758 }
ba210ebe 759 }
b5b9af04 760 if (UNLIKELY(overflowed)) {
ba210ebe 761
eb83ed87
KW
762 /* If the first byte is FF, it will overflow a 32-bit word. If the
763 * first byte is FE, it will overflow a signed 32-bit word. The
764 * above preserves backward compatibility, since its message was used
765 * in earlier versions of this code in preference to overflow */
766 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 767 goto malformed;
eb83ed87
KW
768 }
769#endif
770
2f8f112e
KW
771 if (do_overlong_test
772 && expectlen > (STRLEN)UNISKIP(uv)
773 && ! (flags & UTF8_ALLOW_LONG))
774 {
eb83ed87
KW
775 /* The overlong malformation has lower precedence than the others.
776 * Note that if this malformation is allowed, we return the actual
777 * value, instead of the replacement character. This is because this
778 * value is actually well-defined. */
779 if (! (flags & UTF8_CHECK_ONLY)) {
780 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));
781 }
782 goto malformed;
783 }
784
785 /* Here, the input is considered to be well-formed , but could be a
786 * problematic code point that is not allowed by the input parameters. */
787 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
788 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
789 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
790 {
949cf498 791 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
792 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
793 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
794 {
111d382d 795 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 796 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
797 }
798 if (flags & UTF8_DISALLOW_SURROGATE) {
799 goto disallowed;
800 }
801 }
949cf498 802 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
803 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
804 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
805 {
111d382d 806 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 807 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
808 }
809 if (flags & UTF8_DISALLOW_SUPER) {
810 goto disallowed;
811 }
812 }
4190d317
KW
813 else if (UNICODE_IS_NONCHAR(uv)) {
814 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
815 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
816 {
817 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
818 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
819 }
820 if (flags & UTF8_DISALLOW_NONCHAR) {
821 goto disallowed;
822 }
823 }
949cf498 824
eb83ed87
KW
825 if (sv) {
826 outlier_ret = uv;
827 goto do_warn;
828 }
829
949cf498
KW
830 /* Here, this is not considered a malformed character, so drop through
831 * to return it */
a0ed51b3 832 }
ba210ebe 833
a0ed51b3 834 return uv;
ba210ebe 835
eb83ed87
KW
836 /* There are three cases which get to beyond this point. In all 3 cases:
837 * <sv> if not null points to a string to print as a warning.
838 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
839 * set.
840 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
841 * This is done by initializing it to 0, and changing it only
842 * for case 1).
843 * The 3 cases are:
844 * 1) The input is valid but problematic, and to be warned about. The
845 * return value is the resultant code point; <*retlen> is set to
846 * <curlen>, the number of bytes that comprise the code point.
847 * <pack_warn> contains the result of packWARN() for the warning
848 * types. The entry point for this case is the label <do_warn>;
849 * 2) The input is a valid code point but disallowed by the parameters to
850 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
851 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
852 * comprise the code point. <pack_warn> contains the result of
853 * packWARN() for the warning types. The entry point for this case is
854 * the label <disallowed>.
855 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
856 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
857 * bytes that comprise the malformation. All such malformations are
858 * assumed to be warning type <utf8>. The entry point for this case
859 * is the label <malformed>.
860 */
949cf498 861
ba210ebe
JH
862malformed:
863
eb83ed87
KW
864 if (sv && ckWARN_d(WARN_UTF8)) {
865 pack_warn = packWARN(WARN_UTF8);
866 }
867
868disallowed:
869
fcc8fcf6 870 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 871 if (retlen)
10edeb5d 872 *retlen = ((STRLEN) -1);
ba210ebe
JH
873 return 0;
874 }
875
eb83ed87 876do_warn:
5b311467 877
eb83ed87
KW
878 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
879 if warnings are to be raised. */
f555bc63 880 const char * const string = SvPVX_const(sv);
a0dbb045 881
f555bc63
KW
882 if (PL_op)
883 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
884 else
885 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
886 }
887
eb83ed87
KW
888 if (retlen) {
889 *retlen = curlen;
890 }
ba210ebe 891
eb83ed87 892 return outlier_ret;
a0ed51b3
LW
893}
894
8e84507e 895/*
ec5f19d0
KW
896=for apidoc utf8_to_uvchr_buf
897
898Returns the native code point of the first character in the string C<s> which
899is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 900C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 901
524080c4
KW
902If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
903enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
904NULL) to -1. If those warnings are off, the computed value if well-defined (or
905the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
906is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
907next possible position in C<s> that could begin a non-malformed character.
908See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
909
910=cut
911*/
912
913
914UV
915Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
916{
917 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
918
919 assert(s < send);
920
921 return utf8n_to_uvchr(s, send - s, retlen,
922 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
923}
924
27d6c58a 925/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 926 * there are no malformations in the input UTF-8 string C<s>. surrogates,
6dd9dce9
KW
927 * non-character code points, and non-Unicode code points are allowed. A macro
928 * in utf8.h is used to normally avoid this function wrapper */
27d6c58a
KW
929
930UV
931Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
932{
3986bb7c
KW
933 const UV uv = valid_utf8_to_uvuni(s, retlen);
934
27d6c58a
KW
935 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
936
3986bb7c 937 return UNI_TO_NATIVE(uv);
27d6c58a
KW
938}
939
ec5f19d0 940/*
87cea99e 941=for apidoc utf8_to_uvchr
9041c2e3 942
977c1d31
KW
943DEPRECATED!
944
6ee84de2 945Returns the native code point of the first character in the string C<s>
1e54db1a 946which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
947length, in bytes, of that character.
948
4b88fb76 949Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
950malformed input could cause reading beyond the end of the input buffer, which
951is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 952
524080c4
KW
953If C<s> points to one of the detected malformations, and UTF8 warnings are
954enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
955NULL) to -1. If those warnings are off, the computed value if well-defined (or
956the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
957is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
958next possible position in C<s> that could begin a non-malformed character.
959See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
960
961=cut
962*/
963
964UV
7fc63493 965Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 966{
7918f24d
NC
967 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
968
2ff6c191 969 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
970}
971
972/*
ec5f19d0
KW
973=for apidoc utf8_to_uvuni_buf
974
975Returns the Unicode code point of the first character in the string C<s> which
976is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
977C<retlen> will be set to the length, in bytes, of that character.
978
979This function should only be used when the returned UV is considered
980an index into the Unicode semantic tables (e.g. swashes).
981
524080c4
KW
982If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
983enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
984NULL) to -1. If those warnings are off, the computed value if well-defined (or
985the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
986is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
987next possible position in C<s> that could begin a non-malformed character.
988See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
989
990=cut
991*/
992
993UV
994Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
995{
996 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
997
998 assert(send > s);
999
1000 /* Call the low level routine asking for checks */
1001 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1002 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1003}
1004
27d6c58a 1005/* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1006 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1007 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1008
1009UV
1010Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1011{
3986bb7c
KW
1012 UV expectlen = UTF8SKIP(s);
1013 const U8* send = s + expectlen;
1014 UV uv = NATIVE_TO_UTF(*s);
1015
27d6c58a
KW
1016 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1017
3986bb7c
KW
1018 if (retlen) {
1019 *retlen = expectlen;
1020 }
1021
1022 /* An invariant is trivially returned */
1023 if (expectlen == 1) {
1024 return uv;
1025 }
1026
1027 /* Remove the leading bits that indicate the number of bytes, leaving just
1028 * the bits that are part of the value */
1029 uv &= UTF_START_MASK(expectlen);
1030
1031 /* Now, loop through the remaining bytes, accumulating each into the
08bc774e
KW
1032 * working total as we go. (I khw tried unrolling the loop for up to 4
1033 * bytes, but there was no performance improvement) */
3986bb7c
KW
1034 for (++s; s < send; s++) {
1035 uv = UTF8_ACCUMULATE(uv, *s);
1036 }
1037
1038 return uv;
27d6c58a
KW
1039}
1040
ec5f19d0 1041/*
87cea99e 1042=for apidoc utf8_to_uvuni
9041c2e3 1043
977c1d31
KW
1044DEPRECATED!
1045
9041c2e3 1046Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1047which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1048length, in bytes, of that character.
1049
39e518fd
KW
1050This function should only be used when the returned UV is considered
1051an index into the Unicode semantic tables (e.g. swashes).
1052
4b88fb76 1053Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1054malformed input could cause reading beyond the end of the input buffer, which
1055is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
9041c2e3 1056
524080c4
KW
1057If C<s> points to one of the detected malformations, and UTF8 warnings are
1058enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1059NULL) to -1. If those warnings are off, the computed value if well-defined (or
1060the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1061is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1062next possible position in C<s> that could begin a non-malformed character.
1063See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1064
1065=cut
1066*/
1067
1068UV
7fc63493 1069Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1070{
7918f24d
NC
1071 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1072
4b88fb76 1073 return valid_utf8_to_uvuni(s, retlen);
8e84507e
NIS
1074}
1075
b76347f2 1076/*
87cea99e 1077=for apidoc utf8_length
b76347f2
JH
1078
1079Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1080Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1081up past C<e>, croaks.
b76347f2
JH
1082
1083=cut
1084*/
1085
1086STRLEN
35a4481c 1087Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1088{
97aff369 1089 dVAR;
b76347f2
JH
1090 STRLEN len = 0;
1091
7918f24d
NC
1092 PERL_ARGS_ASSERT_UTF8_LENGTH;
1093
8850bf83
JH
1094 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1095 * the bitops (especially ~) can create illegal UTF-8.
1096 * In other words: in Perl UTF-8 is not just for Unicode. */
1097
a3b680e6
AL
1098 if (e < s)
1099 goto warn_and_return;
b76347f2 1100 while (s < e) {
4cbf4130 1101 s += UTF8SKIP(s);
8e91ec7f
AV
1102 len++;
1103 }
1104
1105 if (e != s) {
1106 len--;
1107 warn_and_return:
9b387841
NC
1108 if (PL_op)
1109 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1110 "%s in %s", unees, OP_DESC(PL_op));
1111 else
61a12c31 1112 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1113 }
1114
1115 return len;
1116}
1117
b06226ff 1118/*
87cea99e 1119=for apidoc utf8_distance
b06226ff 1120
1e54db1a 1121Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1122and C<b>.
1123
1124WARNING: use only if you *know* that the pointers point inside the
1125same UTF-8 buffer.
1126
37607a96
PK
1127=cut
1128*/
a0ed51b3 1129
02eb7b47 1130IV
35a4481c 1131Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1132{
7918f24d
NC
1133 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1134
bf1665bc 1135 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1136}
1137
b06226ff 1138/*
87cea99e 1139=for apidoc utf8_hop
b06226ff 1140
8850bf83
JH
1141Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1142forward or backward.
b06226ff
JH
1143
1144WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1145the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1146on the first byte of character or just after the last byte of a character.
b06226ff 1147
37607a96
PK
1148=cut
1149*/
a0ed51b3
LW
1150
1151U8 *
4373e329 1152Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1153{
7918f24d
NC
1154 PERL_ARGS_ASSERT_UTF8_HOP;
1155
96a5add6 1156 PERL_UNUSED_CONTEXT;
8850bf83
JH
1157 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1158 * the bitops (especially ~) can create illegal UTF-8.
1159 * In other words: in Perl UTF-8 is not just for Unicode. */
1160
a0ed51b3
LW
1161 if (off >= 0) {
1162 while (off--)
1163 s += UTF8SKIP(s);
1164 }
1165 else {
1166 while (off++) {
1167 s--;
8850bf83
JH
1168 while (UTF8_IS_CONTINUATION(*s))
1169 s--;
a0ed51b3
LW
1170 }
1171 }
4373e329 1172 return (U8 *)s;
a0ed51b3
LW
1173}
1174
6940069f 1175/*
fed3ba5d
NC
1176=for apidoc bytes_cmp_utf8
1177
a1433954
KW
1178Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1179sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1180equal, -1 or -2 if the first string is less than the second string, +1 or +2
1181if the first string is greater than the second string.
1182
1183-1 or +1 is returned if the shorter string was identical to the start of the
1184longer string. -2 or +2 is returned if the was a difference between characters
1185within the strings.
1186
1187=cut
1188*/
1189
1190int
1191Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1192{
1193 const U8 *const bend = b + blen;
1194 const U8 *const uend = u + ulen;
1195
1196 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1197
1198 PERL_UNUSED_CONTEXT;
1199
1200 while (b < bend && u < uend) {
1201 U8 c = *u++;
1202 if (!UTF8_IS_INVARIANT(c)) {
1203 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1204 if (u < uend) {
1205 U8 c1 = *u++;
1206 if (UTF8_IS_CONTINUATION(c1)) {
356979f4 1207 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
fed3ba5d
NC
1208 } else {
1209 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1210 "Malformed UTF-8 character "
1211 "(unexpected non-continuation byte 0x%02x"
1212 ", immediately after start byte 0x%02x)"
1213 /* Dear diag.t, it's in the pod. */
1214 "%s%s", c1, c,
1215 PL_op ? " in " : "",
1216 PL_op ? OP_DESC(PL_op) : "");
1217 return -2;
1218 }
1219 } else {
1220 if (PL_op)
1221 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1222 "%s in %s", unees, OP_DESC(PL_op));
1223 else
61a12c31 1224 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1225 return -2; /* Really want to return undef :-) */
1226 }
1227 } else {
1228 return -2;
1229 }
1230 }
1231 if (*b != c) {
1232 return *b < c ? -2 : +2;
1233 }
1234 ++b;
1235 }
1236
1237 if (b == bend && u == uend)
1238 return 0;
1239
1240 return b < bend ? +1 : -1;
1241}
1242
1243/*
87cea99e 1244=for apidoc utf8_to_bytes
6940069f 1245
2bbc8d55 1246Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1247Unlike L</bytes_to_utf8>, this over-writes the original string, and
1248updates C<len> to contain the new length.
67e989fb 1249Returns zero on failure, setting C<len> to -1.
6940069f 1250
a1433954 1251If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1252
6940069f
GS
1253=cut
1254*/
1255
1256U8 *
37607a96 1257Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1258{
d4c19fe8
AL
1259 U8 * const save = s;
1260 U8 * const send = s + *len;
6940069f 1261 U8 *d;
246fae53 1262
7918f24d
NC
1263 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1264
1e54db1a 1265 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1266 while (s < send) {
dcad2880
JH
1267 U8 c = *s++;
1268
1d72bdf6
NIS
1269 if (!UTF8_IS_INVARIANT(c) &&
1270 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1271 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
10edeb5d 1272 *len = ((STRLEN) -1);
dcad2880
JH
1273 return 0;
1274 }
246fae53 1275 }
dcad2880
JH
1276
1277 d = s = save;
6940069f 1278 while (s < send) {
ed646e6e 1279 STRLEN ulen;
4b88fb76 1280 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
ed646e6e 1281 s += ulen;
6940069f
GS
1282 }
1283 *d = '\0';
246fae53 1284 *len = d - save;
6940069f
GS
1285 return save;
1286}
1287
1288/*
87cea99e 1289=for apidoc bytes_from_utf8
f9a63242 1290
2bbc8d55 1291Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1292Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1293the newly-created string, and updates C<len> to contain the new
1294length. Returns the original string if no conversion occurs, C<len>
1295is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
12960 if C<s> is converted or consisted entirely of characters that are invariant
1297in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1298
37607a96
PK
1299=cut
1300*/
f9a63242
JH
1301
1302U8 *
e1ec3a88 1303Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1304{
f9a63242 1305 U8 *d;
e1ec3a88
AL
1306 const U8 *start = s;
1307 const U8 *send;
f9a63242
JH
1308 I32 count = 0;
1309
7918f24d
NC
1310 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1311
96a5add6 1312 PERL_UNUSED_CONTEXT;
f9a63242 1313 if (!*is_utf8)
73d840c0 1314 return (U8 *)start;
f9a63242 1315
1e54db1a 1316 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1317 for (send = s + *len; s < send;) {
e1ec3a88 1318 U8 c = *s++;
1d72bdf6 1319 if (!UTF8_IS_INVARIANT(c)) {
db42d148
NIS
1320 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1321 (c = *s++) && UTF8_IS_CONTINUATION(c))
1322 count++;
1323 else
73d840c0 1324 return (U8 *)start;
db42d148 1325 }
f9a63242
JH
1326 }
1327
35da51f7 1328 *is_utf8 = FALSE;
f9a63242 1329
212542aa 1330 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1331 s = start; start = d;
f9a63242
JH
1332 while (s < send) {
1333 U8 c = *s++;
c4d5f83a
NIS
1334 if (!UTF8_IS_INVARIANT(c)) {
1335 /* Then it is two-byte encoded */
356979f4 1336 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
c4d5f83a
NIS
1337 }
1338 *d++ = c;
f9a63242
JH
1339 }
1340 *d = '\0';
1341 *len = d - start;
73d840c0 1342 return (U8 *)start;
f9a63242
JH
1343}
1344
1345/*
87cea99e 1346=for apidoc bytes_to_utf8
6940069f 1347
ff97e5cf
KW
1348Converts a string C<s> of length C<len> bytes from the native encoding into
1349UTF-8.
6662521e 1350Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1351reflect the new length in bytes.
6940069f 1352
2bbc8d55
SP
1353A NUL character will be written after the end of the string.
1354
1355If you want to convert to UTF-8 from encodings other than
1356the native (Latin1 or EBCDIC),
a1433954 1357see L</sv_recode_to_utf8>().
c9ada85f 1358
497711e7 1359=cut
6940069f
GS
1360*/
1361
c682ebef
FC
1362/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1363 likewise need duplication. */
1364
6940069f 1365U8*
35a4481c 1366Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1367{
35a4481c 1368 const U8 * const send = s + (*len);
6940069f
GS
1369 U8 *d;
1370 U8 *dst;
7918f24d
NC
1371
1372 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1373 PERL_UNUSED_CONTEXT;
6940069f 1374
212542aa 1375 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1376 dst = d;
1377
1378 while (s < send) {
35a4481c 1379 const UV uv = NATIVE_TO_ASCII(*s++);
c4d5f83a 1380 if (UNI_IS_INVARIANT(uv))
eb160463 1381 *d++ = (U8)UTF_TO_NATIVE(uv);
6940069f 1382 else {
eb160463
GS
1383 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1384 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
6940069f
GS
1385 }
1386 }
1387 *d = '\0';
6662521e 1388 *len = d-dst;
6940069f
GS
1389 return dst;
1390}
1391
a0ed51b3 1392/*
dea0fc0b 1393 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1394 *
1395 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1396 * We optimize for native, for obvious reasons. */
1397
1398U8*
dea0fc0b 1399Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1400{
dea0fc0b
JH
1401 U8* pend;
1402 U8* dstart = d;
1403
7918f24d
NC
1404 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1405
dea0fc0b 1406 if (bytelen & 1)
f5992bc4 1407 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1408
1409 pend = p + bytelen;
1410
a0ed51b3 1411 while (p < pend) {
dea0fc0b
JH
1412 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1413 p += 2;
a0ed51b3 1414 if (uv < 0x80) {
e294cc5d
JH
1415#ifdef EBCDIC
1416 *d++ = UNI_TO_NATIVE(uv);
1417#else
eb160463 1418 *d++ = (U8)uv;
e294cc5d 1419#endif
a0ed51b3
LW
1420 continue;
1421 }
1422 if (uv < 0x800) {
eb160463
GS
1423 *d++ = (U8)(( uv >> 6) | 0xc0);
1424 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1425 continue;
1426 }
52b9aa85 1427 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
01ea242b 1428 if (p >= pend) {
dea0fc0b 1429 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1430 } else {
1431 UV low = (p[0] << 8) + p[1];
1432 p += 2;
52b9aa85 1433 if (low < 0xdc00 || low > 0xdfff)
01ea242b
NC
1434 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1435 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1436 }
dbde1951
NC
1437 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1438 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3
LW
1439 }
1440 if (uv < 0x10000) {
eb160463
GS
1441 *d++ = (U8)(( uv >> 12) | 0xe0);
1442 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1443 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1444 continue;
1445 }
1446 else {
eb160463
GS
1447 *d++ = (U8)(( uv >> 18) | 0xf0);
1448 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1449 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1450 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1451 continue;
1452 }
1453 }
dea0fc0b 1454 *newlen = d - dstart;
a0ed51b3
LW
1455 return d;
1456}
1457
1458/* Note: this one is slightly destructive of the source. */
1459
1460U8*
dea0fc0b 1461Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1462{
1463 U8* s = (U8*)p;
d4c19fe8 1464 U8* const send = s + bytelen;
7918f24d
NC
1465
1466 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1467
e0ea5e2d
NC
1468 if (bytelen & 1)
1469 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1470 (UV)bytelen);
1471
a0ed51b3 1472 while (s < send) {
d4c19fe8 1473 const U8 tmp = s[0];
a0ed51b3
LW
1474 s[0] = s[1];
1475 s[1] = tmp;
1476 s += 2;
1477 }
dea0fc0b 1478 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1479}
1480
c3fd2246
KW
1481/* for now these are all defined (inefficiently) in terms of the utf8 versions.
1482 * Note that the macros in handy.h that call these short-circuit calling them
1483 * for Latin-1 range inputs */
a0ed51b3
LW
1484
1485bool
84afefe6 1486Perl_is_uni_alnum(pTHX_ UV c)
a0ed51b3 1487{
89ebb4a3 1488 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1489 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1490 return is_utf8_alnum(tmpbuf);
1491}
1492
1493bool
5092f92a
KW
1494Perl_is_uni_alnumc(pTHX_ UV c)
1495{
1496 U8 tmpbuf[UTF8_MAXBYTES+1];
1497 uvchr_to_utf8(tmpbuf, c);
1498 return is_utf8_alnumc(tmpbuf);
1499}
1500
1501bool
84afefe6 1502Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1503{
89ebb4a3 1504 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1505 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1506 return is_utf8_idfirst(tmpbuf);
1507}
1508
1509bool
f91dcd13
KW
1510Perl__is_uni_perl_idstart(pTHX_ UV c)
1511{
1512 U8 tmpbuf[UTF8_MAXBYTES+1];
1513 uvchr_to_utf8(tmpbuf, c);
1514 return _is_utf8_perl_idstart(tmpbuf);
1515}
1516
1517bool
84afefe6 1518Perl_is_uni_alpha(pTHX_ UV c)
a0ed51b3 1519{
89ebb4a3 1520 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1521 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1522 return is_utf8_alpha(tmpbuf);
1523}
1524
1525bool
84afefe6 1526Perl_is_uni_ascii(pTHX_ UV c)
4d61ec05 1527{
bc39fe24 1528 return isASCII(c);
4d61ec05
GS
1529}
1530
1531bool
bdd8600f
KW
1532Perl_is_uni_blank(pTHX_ UV c)
1533{
2cafb56b 1534 return isBLANK_uni(c);
bdd8600f
KW
1535}
1536
1537bool
84afefe6 1538Perl_is_uni_space(pTHX_ UV c)
a0ed51b3 1539{
add4123a 1540 return isSPACE_uni(c);
a0ed51b3
LW
1541}
1542
1543bool
84afefe6 1544Perl_is_uni_digit(pTHX_ UV c)
a0ed51b3 1545{
89ebb4a3 1546 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1547 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1548 return is_utf8_digit(tmpbuf);
1549}
1550
1551bool
84afefe6 1552Perl_is_uni_upper(pTHX_ UV c)
a0ed51b3 1553{
89ebb4a3 1554 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1555 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1556 return is_utf8_upper(tmpbuf);
1557}
1558
1559bool
84afefe6 1560Perl_is_uni_lower(pTHX_ UV c)
a0ed51b3 1561{
89ebb4a3 1562 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1563 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1564 return is_utf8_lower(tmpbuf);
1565}
1566
1567bool
84afefe6 1568Perl_is_uni_cntrl(pTHX_ UV c)
b8c5462f 1569{
7b952154 1570 return isCNTRL_L1(c);
b8c5462f
JH
1571}
1572
1573bool
84afefe6 1574Perl_is_uni_graph(pTHX_ UV c)
b8c5462f 1575{
89ebb4a3 1576 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1577 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1578 return is_utf8_graph(tmpbuf);
1579}
1580
1581bool
84afefe6 1582Perl_is_uni_print(pTHX_ UV c)
a0ed51b3 1583{
89ebb4a3 1584 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1585 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1586 return is_utf8_print(tmpbuf);
1587}
1588
b8c5462f 1589bool
84afefe6 1590Perl_is_uni_punct(pTHX_ UV c)
b8c5462f 1591{
89ebb4a3 1592 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1593 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1594 return is_utf8_punct(tmpbuf);
1595}
1596
4d61ec05 1597bool
84afefe6 1598Perl_is_uni_xdigit(pTHX_ UV c)
4d61ec05 1599{
4ac6419d 1600 return isXDIGIT_uni(c);
4d61ec05
GS
1601}
1602
3a4c58c9
KW
1603UV
1604Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1605{
1606 /* We have the latin1-range values compiled into the core, so just use
1607 * those, converting the result to utf8. The only difference between upper
1608 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1609 * either "SS" or "Ss". Which one to use is passed into the routine in
1610 * 'S_or_s' to avoid a test */
1611
1612 UV converted = toUPPER_LATIN1_MOD(c);
1613
1614 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1615
1616 assert(S_or_s == 'S' || S_or_s == 's');
1617
1618 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1619 characters in this range */
1620 *p = (U8) converted;
1621 *lenp = 1;
1622 return converted;
1623 }
1624
1625 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1626 * which it maps to one of them, so as to only have to have one check for
1627 * it in the main case */
1628 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1629 switch (c) {
1630 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1631 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1632 break;
1633 case MICRO_SIGN:
1634 converted = GREEK_CAPITAL_LETTER_MU;
1635 break;
1636 case LATIN_SMALL_LETTER_SHARP_S:
1637 *(p)++ = 'S';
1638 *p = S_or_s;
1639 *lenp = 2;
1640 return 'S';
1641 default:
1642 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1643 assert(0); /* NOTREACHED */
3a4c58c9
KW
1644 }
1645 }
1646
1647 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1648 *p = UTF8_TWO_BYTE_LO(converted);
1649 *lenp = 2;
1650
1651 return converted;
1652}
1653
50bda2c3
KW
1654/* Call the function to convert a UTF-8 encoded character to the specified case.
1655 * Note that there may be more than one character in the result.
1656 * INP is a pointer to the first byte of the input character
1657 * OUTP will be set to the first byte of the string of changed characters. It
1658 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1659 * LENP will be set to the length in bytes of the string of changed characters
1660 *
1661 * The functions return the ordinal of the first character in the string of OUTP */
f90a9a02
KW
1662#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1663#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1664#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
50bda2c3
KW
1665
1666/* This additionally has the input parameter SPECIALS, which if non-zero will
1667 * cause this to use the SPECIALS hash for folding (meaning get full case
1668 * folding); otherwise, when zero, this implies a simple case fold */
f90a9a02 1669#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 1670
84afefe6
JH
1671UV
1672Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1673{
3a4c58c9
KW
1674 dVAR;
1675
a1433954
KW
1676 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1677 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1678 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1679 * the changed version may be longer than the original character.
1680 *
1681 * The ordinal of the first character of the changed version is returned
1682 * (but note, as explained above, that there may be more.) */
1683
7918f24d
NC
1684 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1685
3a4c58c9
KW
1686 if (c < 256) {
1687 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1688 }
1689
0ebc6274 1690 uvchr_to_utf8(p, c);
3a4c58c9 1691 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1692}
1693
84afefe6
JH
1694UV
1695Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1696{
3a4c58c9
KW
1697 dVAR;
1698
7918f24d
NC
1699 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1700
3a4c58c9
KW
1701 if (c < 256) {
1702 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1703 }
1704
0ebc6274 1705 uvchr_to_utf8(p, c);
3a4c58c9 1706 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1707}
1708
afc16117
KW
1709STATIC U8
1710S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1711{
1712 /* We have the latin1-range values compiled into the core, so just use
1713 * those, converting the result to utf8. Since the result is always just
a1433954 1714 * one character, we allow <p> to be NULL */
afc16117
KW
1715
1716 U8 converted = toLOWER_LATIN1(c);
1717
1718 if (p != NULL) {
1719 if (UNI_IS_INVARIANT(converted)) {
1720 *p = converted;
1721 *lenp = 1;
1722 }
1723 else {
1724 *p = UTF8_TWO_BYTE_HI(converted);
1725 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1726 *lenp = 2;
1727 }
1728 }
1729 return converted;
1730}
1731
84afefe6
JH
1732UV
1733Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1734{
968c5e6a
KW
1735 dVAR;
1736
7918f24d
NC
1737 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1738
afc16117
KW
1739 if (c < 256) {
1740 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1741 }
1742
afc16117 1743 uvchr_to_utf8(p, c);
968c5e6a 1744 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1745}
1746
84afefe6 1747UV
f673fad4 1748Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
a1dde8de 1749{
a1433954 1750 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
f673fad4
KW
1751 * folding */
1752
a1dde8de
KW
1753 UV converted;
1754
1755 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1756
1757 if (c == MICRO_SIGN) {
1758 converted = GREEK_SMALL_LETTER_MU;
1759 }
1760 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1761 *(p)++ = 's';
1762 *p = 's';
1763 *lenp = 2;
1764 return 's';
1765 }
1766 else { /* In this range the fold of all other characters is their lower
1767 case */
1768 converted = toLOWER_LATIN1(c);
1769 }
1770
1771 if (UNI_IS_INVARIANT(converted)) {
1772 *p = (U8) converted;
1773 *lenp = 1;
1774 }
1775 else {
1776 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1777 *p = UTF8_TWO_BYTE_LO(converted);
1778 *lenp = 2;
1779 }
1780
1781 return converted;
1782}
1783
1784UV
a0270393 1785Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
84afefe6 1786{
4b593389 1787
a0270393
KW
1788 /* Not currently externally documented, and subject to change
1789 * <flags> bits meanings:
1790 * FOLD_FLAGS_FULL iff full folding is to be used;
1791 * FOLD_FLAGS_LOCALE iff in locale
1792 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1793 */
4b593389 1794
36bb2ab6 1795 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1796
a1dde8de 1797 if (c < 256) {
a0270393
KW
1798 UV result = _to_fold_latin1((U8) c, p, lenp,
1799 cBOOL(((flags & FOLD_FLAGS_FULL)
1800 /* If ASCII-safe, don't allow full folding,
1801 * as that could include SHARP S => ss;
1802 * otherwise there is no crossing of
1803 * ascii/non-ascii in the latin1 range */
1804 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1805 /* It is illegal for the fold to cross the 255/256 boundary under
1806 * locale; in this case return the original */
1807 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1808 ? c
1809 : result;
a1dde8de
KW
1810 }
1811
a0270393
KW
1812 /* If no special needs, just use the macro */
1813 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1814 uvchr_to_utf8(p, c);
1815 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1816 }
1817 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1818 the special flags. */
1819 U8 utf8_c[UTF8_MAXBYTES + 1];
1820 uvchr_to_utf8(utf8_c, c);
1821 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1822 }
84afefe6
JH
1823}
1824
ea317ccb
KW
1825/* for now these all assume no locale info available for Unicode > 255; and
1826 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1827 * called instead, so that these don't get called for < 255 */
a0ed51b3
LW
1828
1829bool
84afefe6 1830Perl_is_uni_alnum_lc(pTHX_ UV c)
a0ed51b3
LW
1831{
1832 return is_uni_alnum(c); /* XXX no locale support yet */
1833}
1834
1835bool
5092f92a
KW
1836Perl_is_uni_alnumc_lc(pTHX_ UV c)
1837{
1838 if (c < 256) {
1839 return isALNUMC_LC(UNI_TO_NATIVE(c));
1840 }
1841 return is_uni_alnumc(c);
1842}
1843
1844bool
84afefe6 1845Perl_is_uni_idfirst_lc(pTHX_ UV c)
a0ed51b3 1846{
e85ea4f9 1847 return _is_uni_perl_idstart(c); /* XXX no locale support yet */
a0ed51b3
LW
1848}
1849
1850bool
84afefe6 1851Perl_is_uni_alpha_lc(pTHX_ UV c)
a0ed51b3
LW
1852{
1853 return is_uni_alpha(c); /* XXX no locale support yet */
1854}
1855
1856bool
84afefe6 1857Perl_is_uni_ascii_lc(pTHX_ UV c)
4d61ec05
GS
1858{
1859 return is_uni_ascii(c); /* XXX no locale support yet */
1860}
1861
1862bool
bdd8600f
KW
1863Perl_is_uni_blank_lc(pTHX_ UV c)
1864{
1865 return is_uni_blank(c); /* XXX no locale support yet */
1866}
1867
1868bool
84afefe6 1869Perl_is_uni_space_lc(pTHX_ UV c)
a0ed51b3
LW
1870{
1871 return is_uni_space(c); /* XXX no locale support yet */
1872}
1873
1874bool
84afefe6 1875Perl_is_uni_digit_lc(pTHX_ UV c)
a0ed51b3
LW
1876{
1877 return is_uni_digit(c); /* XXX no locale support yet */
1878}
1879
1880bool
84afefe6 1881Perl_is_uni_upper_lc(pTHX_ UV c)
a0ed51b3
LW
1882{
1883 return is_uni_upper(c); /* XXX no locale support yet */
1884}
1885
1886bool
84afefe6 1887Perl_is_uni_lower_lc(pTHX_ UV c)
a0ed51b3
LW
1888{
1889 return is_uni_lower(c); /* XXX no locale support yet */
1890}
1891
1892bool
84afefe6 1893Perl_is_uni_cntrl_lc(pTHX_ UV c)
b8c5462f
JH
1894{
1895 return is_uni_cntrl(c); /* XXX no locale support yet */
1896}
1897
1898bool
84afefe6 1899Perl_is_uni_graph_lc(pTHX_ UV c)
b8c5462f
JH
1900{
1901 return is_uni_graph(c); /* XXX no locale support yet */
1902}
1903
1904bool
84afefe6 1905Perl_is_uni_print_lc(pTHX_ UV c)
a0ed51b3
LW
1906{
1907 return is_uni_print(c); /* XXX no locale support yet */
1908}
1909
b8c5462f 1910bool
84afefe6 1911Perl_is_uni_punct_lc(pTHX_ UV c)
b8c5462f
JH
1912{
1913 return is_uni_punct(c); /* XXX no locale support yet */
1914}
1915
4d61ec05 1916bool
84afefe6 1917Perl_is_uni_xdigit_lc(pTHX_ UV c)
4d61ec05
GS
1918{
1919 return is_uni_xdigit(c); /* XXX no locale support yet */
1920}
1921
b7ac61fa
JH
1922U32
1923Perl_to_uni_upper_lc(pTHX_ U32 c)
1924{
ee099d14
JH
1925 /* XXX returns only the first character -- do not use XXX */
1926 /* XXX no locale support yet */
1927 STRLEN len;
89ebb4a3 1928 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1929 return (U32)to_uni_upper(c, tmpbuf, &len);
b7ac61fa
JH
1930}
1931
1932U32
1933Perl_to_uni_title_lc(pTHX_ U32 c)
1934{
ee099d14
JH
1935 /* XXX returns only the first character XXX -- do not use XXX */
1936 /* XXX no locale support yet */
1937 STRLEN len;
89ebb4a3 1938 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1939 return (U32)to_uni_title(c, tmpbuf, &len);
b7ac61fa
JH
1940}
1941
1942U32
1943Perl_to_uni_lower_lc(pTHX_ U32 c)
1944{
ee099d14
JH
1945 /* XXX returns only the first character -- do not use XXX */
1946 /* XXX no locale support yet */
1947 STRLEN len;
89ebb4a3 1948 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1949 return (U32)to_uni_lower(c, tmpbuf, &len);
b7ac61fa
JH
1950}
1951
26483009 1952PERL_STATIC_INLINE bool
5141f98e 1953S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
1954 const char *const swashname)
1955{
ea317ccb
KW
1956 /* returns a boolean giving whether or not the UTF8-encoded character that
1957 * starts at <p> is in the swash indicated by <swashname>. <swash>
1958 * contains a pointer to where the swash indicated by <swashname>
1959 * is to be stored; which this routine will do, so that future calls will
1960 * look at <*swash> and only generate a swash if it is not null
1961 *
1962 * Note that it is assumed that the buffer length of <p> is enough to
1963 * contain all the bytes that comprise the character. Thus, <*p> should
1964 * have been checked before this call for mal-formedness enough to assure
1965 * that. */
1966
97aff369 1967 dVAR;
7918f24d
NC
1968
1969 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1970
492a624f
KW
1971 /* The API should have included a length for the UTF-8 character in <p>,
1972 * but it doesn't. We therefor assume that p has been validated at least
1973 * as far as there being enough bytes available in it to accommodate the
1974 * character without reading beyond the end, and pass that number on to the
1975 * validating routine */
1976 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
bde6a22d 1977 return FALSE;
87367d5f
KW
1978 if (!*swash) {
1979 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1980 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
1981 }
bde6a22d
NC
1982 return swash_fetch(*swash, p, TRUE) != 0;
1983}
1984
1985bool
7fc63493 1986Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 1987{
97aff369 1988 dVAR;
7918f24d
NC
1989
1990 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1991
671c33bf
NC
1992 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1993 * descendant of isalnum(3), in other words, it doesn't
1994 * contain the '_'. --jhi */
d4c19fe8 1995 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
a0ed51b3
LW
1996}
1997
1998bool
5092f92a
KW
1999Perl_is_utf8_alnumc(pTHX_ const U8 *p)
2000{
2001 dVAR;
2002
2003 PERL_ARGS_ASSERT_IS_UTF8_ALNUMC;
2004
2005 return is_utf8_common(p, &PL_utf8_alnumc, "IsAlnum");
2006}
2007
2008bool
7fc63493 2009Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 2010{
97aff369 2011 dVAR;
7918f24d
NC
2012
2013 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
2014
82686b01
JH
2015 if (*p == '_')
2016 return TRUE;
bde6a22d 2017 /* is_utf8_idstart would be more logical. */
d4c19fe8 2018 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
82686b01
JH
2019}
2020
2021bool
c11ff943
KW
2022Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
2023{
2024 dVAR;
2025
2026 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
2027
2028 if (*p == '_')
2029 return TRUE;
2030 /* is_utf8_idstart would be more logical. */
2031 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2032}
2033
2034bool
d65654cb 2035Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
b6912c02
KW
2036{
2037 dVAR;
2038
d65654cb 2039 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
b6912c02
KW
2040
2041 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2042}
2043
2044bool
7fc63493 2045Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2046{
97aff369 2047 dVAR;
7918f24d
NC
2048
2049 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2050
d4c19fe8 2051 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
2052}
2053
2054bool
c11ff943
KW
2055Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2056{
2057 dVAR;
2058
2059 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2060
c11ff943
KW
2061 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2062}
2063
2064bool
7fc63493 2065Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 2066{
97aff369 2067 dVAR;
7918f24d
NC
2068
2069 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2070
d4c19fe8 2071 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
a0ed51b3
LW
2072}
2073
2074bool
7fc63493 2075Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2076{
97aff369 2077 dVAR;
7918f24d
NC
2078
2079 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2080
bc39fe24
KW
2081 /* ASCII characters are the same whether in utf8 or not. So the macro
2082 * works on both utf8 and non-utf8 representations. */
2083 return isASCII(*p);
b8c5462f
JH
2084}
2085
2086bool
bdd8600f
KW
2087Perl_is_utf8_blank(pTHX_ const U8 *p)
2088{
2089 dVAR;
2090
2091 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2092
2cafb56b 2093 return isBLANK_utf8(p);
bdd8600f
KW
2094}
2095
2096bool
7fc63493 2097Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2098{
97aff369 2099 dVAR;
7918f24d
NC
2100
2101 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2102
add4123a 2103 return isSPACE_utf8(p);
a0ed51b3
LW
2104}
2105
2106bool
d1eb3177
YO
2107Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2108{
2109 dVAR;
2110
2111 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2112
c4428693
KW
2113 /* Only true if is an ASCII space-like character, and ASCII is invariant
2114 * under utf8, so can just use the macro */
2115 return isSPACE_A(*p);
d1eb3177
YO
2116}
2117
2118bool
2119Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2120{
2121 dVAR;
2122
2123 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2124
c4428693
KW
2125 /* Only true if is an ASCII word character, and ASCII is invariant
2126 * under utf8, so can just use the macro */
2127 return isWORDCHAR_A(*p);
d1eb3177
YO
2128}
2129
2130bool
7fc63493 2131Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2132{
97aff369 2133 dVAR;
7918f24d
NC
2134
2135 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2136
d4c19fe8 2137 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
a0ed51b3
LW
2138}
2139
2140bool
d1eb3177
YO
2141Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2142{
2143 dVAR;
2144
2145 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2146
c4428693
KW
2147 /* Only true if is an ASCII digit character, and ASCII is invariant
2148 * under utf8, so can just use the macro */
2149 return isDIGIT_A(*p);
d1eb3177
YO
2150}
2151
2152bool
7fc63493 2153Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2154{
97aff369 2155 dVAR;
7918f24d
NC
2156
2157 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2158
d4c19fe8 2159 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
a0ed51b3
LW
2160}
2161
2162bool
7fc63493 2163Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2164{
97aff369 2165 dVAR;
7918f24d
NC
2166
2167 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2168
d4c19fe8 2169 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
a0ed51b3
LW
2170}
2171
2172bool
7fc63493 2173Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2174{
97aff369 2175 dVAR;
7918f24d
NC
2176
2177 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2178
a35d759a 2179 return isCNTRL_utf8(p);
b8c5462f
JH
2180}
2181
2182bool
7fc63493 2183Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2184{
97aff369 2185 dVAR;
7918f24d
NC
2186
2187 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2188
d4c19fe8 2189 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
b8c5462f
JH
2190}
2191
2192bool
7fc63493 2193Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2194{
97aff369 2195 dVAR;
7918f24d
NC
2196
2197 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2198
d4c19fe8 2199 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
a0ed51b3
LW
2200}
2201
2202bool
7fc63493 2203Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2204{
97aff369 2205 dVAR;
7918f24d
NC
2206
2207 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2208
d4c19fe8 2209 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
b8c5462f
JH
2210}
2211
2212bool
7fc63493 2213Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2214{
97aff369 2215 dVAR;
7918f24d
NC
2216
2217 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2218
4ac6419d 2219 return is_XDIGIT_utf8(p);
b8c5462f
JH
2220}
2221
2222bool
7fc63493 2223Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2224{
97aff369 2225 dVAR;
7918f24d
NC
2226
2227 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2228
d4c19fe8 2229 return is_utf8_common(p, &PL_utf8_mark, "IsM");
a0ed51b3
LW
2230}
2231
37e2e78e 2232bool
27d4fc33 2233Perl_is_utf8_X_regular_begin(pTHX_ const U8 *p)
37e2e78e
KW
2234{
2235 dVAR;
2236
27d4fc33 2237 PERL_ARGS_ASSERT_IS_UTF8_X_REGULAR_BEGIN;
37e2e78e 2238
27d4fc33 2239 return is_utf8_common(p, &PL_utf8_X_regular_begin, "_X_Regular_Begin");
37e2e78e
KW
2240}
2241
2242bool
2243Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2244{
2245 dVAR;
2246
2247 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2248
2249 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2250}
2251
6b5c0936 2252/*
87cea99e 2253=for apidoc to_utf8_case
6b5c0936 2254
a1433954
KW
2255The C<p> contains the pointer to the UTF-8 string encoding
2256the character that is being converted. This routine assumes that the character
2257at C<p> is well-formed.
6b5c0936 2258
a1433954
KW
2259The C<ustrp> is a pointer to the character buffer to put the
2260conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2261of the result.
2262
a1433954 2263The C<swashp> is a pointer to the swash to use.
6b5c0936 2264
a1433954
KW
2265Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2266and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2267but not always, a multicharacter mapping), is tried first.
6b5c0936 2268
a1433954 2269The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2270hash %utf8::ToSpecLower. The access to the hash is through
2271Perl_to_utf8_case().
6b5c0936 2272
a1433954 2273The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2274%utf8::ToLower.
2275
2276=cut */
6b5c0936 2277
2104c8d9 2278UV
9a957fbc
AL
2279Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2280 SV **swashp, const char *normal, const char *special)
a0ed51b3 2281{
97aff369 2282 dVAR;
89ebb4a3 2283 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
0134edef 2284 STRLEN len = 0;
4b88fb76 2285 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
1feea2c7
JH
2286 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2287 * are necessary in EBCDIC, they are redundant no-ops
2288 * in ASCII-ish platforms, and hopefully optimized away. */
f54cb97a 2289 const UV uv1 = NATIVE_TO_UNI(uv0);
7918f24d
NC
2290
2291 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2292
9ae3ac1a
KW
2293 /* Note that swash_fetch() doesn't output warnings for these because it
2294 * assumes we will */
8457b38f 2295 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2296 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2297 if (ckWARN_d(WARN_SURROGATE)) {
2298 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2299 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2300 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2301 }
9ae3ac1a
KW
2302 }
2303 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2304 if (ckWARN_d(WARN_NON_UNICODE)) {
2305 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2306 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2307 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2308 }
9ae3ac1a
KW
2309 }
2310
2311 /* Note that non-characters are perfectly legal, so no warning should
2312 * be given */
2313 }
2314
1feea2c7 2315 uvuni_to_utf8(tmpbuf, uv1);
0134edef
JH
2316
2317 if (!*swashp) /* load on-demand */
5ab9d2ef 2318 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2319
a6f87d8c 2320 if (special) {
0134edef 2321 /* It might be "special" (sometimes, but not always,
2a37f04d 2322 * a multicharacter mapping) */
6673a63c 2323 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2324 SV **svp;
2325
35da51f7 2326 if (hv &&
b08cf34e
JH
2327 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2328 (*svp)) {
cfd0369c 2329 const char *s;
47654450 2330
cfd0369c 2331 s = SvPV_const(*svp, len);
47654450
JH
2332 if (len == 1)
2333 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2a37f04d 2334 else {
2f9475ad
JH
2335#ifdef EBCDIC
2336 /* If we have EBCDIC we need to remap the characters
2337 * since any characters in the low 256 are Unicode
2338 * code points, not EBCDIC. */
7cda7a3d 2339 U8 *t = (U8*)s, *tend = t + len, *d;
2f9475ad
JH
2340
2341 d = tmpbuf;
b08cf34e 2342 if (SvUTF8(*svp)) {
2f9475ad
JH
2343 STRLEN tlen = 0;
2344
2345 while (t < tend) {
6bd1c396 2346 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2f9475ad
JH
2347 if (tlen > 0) {
2348 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2349 t += tlen;
2350 }
2351 else
2352 break;
2353 }
2354 }
2355 else {
36fec512
JH
2356 while (t < tend) {
2357 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2358 t++;
2359 }
2f9475ad
JH
2360 }
2361 len = d - tmpbuf;
2362 Copy(tmpbuf, ustrp, len, U8);
2363#else
d2dcd0fb 2364 Copy(s, ustrp, len, U8);
2f9475ad 2365#endif
29e98929 2366 }
983ffd37 2367 }
0134edef
JH
2368 }
2369
2370 if (!len && *swashp) {
2114036c 2371 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
d4c19fe8 2372
0134edef
JH
2373 if (uv2) {
2374 /* It was "normal" (a single character mapping). */
d4c19fe8 2375 const UV uv3 = UNI_TO_NATIVE(uv2);
e9101d72 2376 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2a37f04d
JH
2377 }
2378 }
1feea2c7 2379
cbe07460
KW
2380 if (len) {
2381 if (lenp) {
2382 *lenp = len;
2383 }
2384 return valid_utf8_to_uvchr(ustrp, 0);
2385 }
2386
2387 /* Here, there was no mapping defined, which means that the code point maps
2388 * to itself. Return the inputs */
bfdf22ec 2389 len = UTF8SKIP(p);
ca9fab46
KW
2390 if (p != ustrp) { /* Don't copy onto itself */
2391 Copy(p, ustrp, len, U8);
2392 }
0134edef 2393
2a37f04d
JH
2394 if (lenp)
2395 *lenp = len;
2396
cbe07460
KW
2397 return uv0;
2398
a0ed51b3
LW
2399}
2400
051a06d4
KW
2401STATIC UV
2402S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2403{
2404 /* This is called when changing the case of a utf8-encoded character above
2405 * the Latin1 range, and the operation is in locale. If the result
2406 * contains a character that crosses the 255/256 boundary, disallow the
2407 * change, and return the original code point. See L<perlfunc/lc> for why;
2408 *
a1433954
KW
2409 * p points to the original string whose case was changed; assumed
2410 * by this routine to be well-formed
051a06d4
KW
2411 * result the code point of the first character in the changed-case string
2412 * ustrp points to the changed-case string (<result> represents its first char)
2413 * lenp points to the length of <ustrp> */
2414
2415 UV original; /* To store the first code point of <p> */
2416
2417 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2418
2419 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2420
2421 /* We know immediately if the first character in the string crosses the
2422 * boundary, so can skip */
2423 if (result > 255) {
2424
2425 /* Look at every character in the result; if any cross the
2426 * boundary, the whole thing is disallowed */
2427 U8* s = ustrp + UTF8SKIP(ustrp);
2428 U8* e = ustrp + *lenp;
2429 while (s < e) {
2430 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2431 {
2432 goto bad_crossing;
2433 }
2434 s += UTF8SKIP(s);
2435 }
2436
2437 /* Here, no characters crossed, result is ok as-is */
2438 return result;
2439 }
2440
2441bad_crossing:
2442
2443 /* Failed, have to return the original */
4b88fb76 2444 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2445 Copy(p, ustrp, *lenp, char);
2446 return original;
2447}
2448
d3e79532 2449/*
87cea99e 2450=for apidoc to_utf8_upper
d3e79532 2451
a1433954
KW
2452Convert the UTF-8 encoded character at C<p> to its uppercase version and
2453store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
89ebb4a3
JH
2454that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2455the uppercase version may be longer than the original character.
d3e79532
JH
2456
2457The first character of the uppercased version is returned
2458(but note, as explained above, that there may be more.)
2459
a1433954
KW
2460The character at C<p> is assumed by this routine to be well-formed.
2461
d3e79532
JH
2462=cut */
2463
051a06d4
KW
2464/* Not currently externally documented, and subject to change:
2465 * <flags> is set iff locale semantics are to be used for code points < 256
2466 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2467 * were used in the calculation; otherwise unchanged. */
2468
2104c8d9 2469UV
051a06d4 2470Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2471{
97aff369 2472 dVAR;
7918f24d 2473
051a06d4
KW
2474 UV result;
2475
2476 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2477
3a4c58c9 2478 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2479 if (flags) {
2480 result = toUPPER_LC(*p);
2481 }
2482 else {
81c6c7ce 2483 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2484 }
3a4c58c9
KW
2485 }
2486 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2487 if (flags) {
2488 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2489 }
2490 else {
81c6c7ce
KW
2491 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2492 ustrp, lenp, 'S');
051a06d4
KW
2493 }
2494 }
2495 else { /* utf8, ord above 255 */
2496 result = CALL_UPPER_CASE(p, ustrp, lenp);
2497
2498 if (flags) {
2499 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2500 }
2501 return result;
2502 }
2503
2504 /* Here, used locale rules. Convert back to utf8 */
2505 if (UTF8_IS_INVARIANT(result)) {
2506 *ustrp = (U8) result;
2507 *lenp = 1;
2508 }
2509 else {
2510 *ustrp = UTF8_EIGHT_BIT_HI(result);
2511 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2512 *lenp = 2;
3a4c58c9
KW
2513 }
2514
051a06d4
KW
2515 if (tainted_ptr) {
2516 *tainted_ptr = TRUE;
2517 }
2518 return result;
983ffd37 2519}
a0ed51b3 2520
d3e79532 2521/*
87cea99e 2522=for apidoc to_utf8_title
d3e79532 2523
a1433954
KW
2524Convert the UTF-8 encoded character at C<p> to its titlecase version and
2525store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2526that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2527titlecase version may be longer than the original character.
d3e79532
JH
2528
2529The first character of the titlecased version is returned
2530(but note, as explained above, that there may be more.)
2531
a1433954
KW
2532The character at C<p> is assumed by this routine to be well-formed.
2533
d3e79532
JH
2534=cut */
2535
051a06d4
KW
2536/* Not currently externally documented, and subject to change:
2537 * <flags> is set iff locale semantics are to be used for code points < 256
2538 * Since titlecase is not defined in POSIX, uppercase is used instead
2539 * for these/
2540 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2541 * were used in the calculation; otherwise unchanged. */
2542
983ffd37 2543UV
051a06d4 2544Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2545{
97aff369 2546 dVAR;
7918f24d 2547
051a06d4
KW
2548 UV result;
2549
2550 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2551
3a4c58c9 2552 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2553 if (flags) {
2554 result = toUPPER_LC(*p);
2555 }
2556 else {
81c6c7ce 2557 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2558 }
3a4c58c9
KW
2559 }
2560 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2561 if (flags) {
2562 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2563 }
2564 else {
81c6c7ce
KW
2565 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2566 ustrp, lenp, 's');
051a06d4
KW
2567 }
2568 }
2569 else { /* utf8, ord above 255 */
2570 result = CALL_TITLE_CASE(p, ustrp, lenp);
2571
2572 if (flags) {
2573 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2574 }
2575 return result;
2576 }
2577
2578 /* Here, used locale rules. Convert back to utf8 */
2579 if (UTF8_IS_INVARIANT(result)) {
2580 *ustrp = (U8) result;
2581 *lenp = 1;
2582 }
2583 else {
2584 *ustrp = UTF8_EIGHT_BIT_HI(result);
2585 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2586 *lenp = 2;
3a4c58c9
KW
2587 }
2588
051a06d4
KW
2589 if (tainted_ptr) {
2590 *tainted_ptr = TRUE;
2591 }
2592 return result;
a0ed51b3
LW
2593}
2594
d3e79532 2595/*
87cea99e 2596=for apidoc to_utf8_lower
d3e79532 2597
a1433954
KW
2598Convert the UTF-8 encoded character at C<p> to its lowercase version and
2599store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2600that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2601lowercase version may be longer than the original character.
d3e79532
JH
2602
2603The first character of the lowercased version is returned
2604(but note, as explained above, that there may be more.)
2605
a1433954
KW
2606The character at C<p> is assumed by this routine to be well-formed.
2607
d3e79532
JH
2608=cut */
2609
051a06d4
KW
2610/* Not currently externally documented, and subject to change:
2611 * <flags> is set iff locale semantics are to be used for code points < 256
2612 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2613 * were used in the calculation; otherwise unchanged. */
2614
2104c8d9 2615UV
051a06d4 2616Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2617{
051a06d4
KW
2618 UV result;
2619
97aff369 2620 dVAR;
7918f24d 2621
051a06d4 2622 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2623
968c5e6a 2624 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2625 if (flags) {
2626 result = toLOWER_LC(*p);
2627 }
2628 else {
81c6c7ce 2629 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2630 }
968c5e6a
KW
2631 }
2632 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2633 if (flags) {
2634 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2635 }
2636 else {
81c6c7ce
KW
2637 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2638 ustrp, lenp);
051a06d4 2639 }
968c5e6a 2640 }
051a06d4
KW
2641 else { /* utf8, ord above 255 */
2642 result = CALL_LOWER_CASE(p, ustrp, lenp);
2643
2644 if (flags) {
2645 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2646 }
968c5e6a 2647
051a06d4
KW
2648 return result;
2649 }
2650
2651 /* Here, used locale rules. Convert back to utf8 */
2652 if (UTF8_IS_INVARIANT(result)) {
2653 *ustrp = (U8) result;
2654 *lenp = 1;
2655 }
2656 else {
2657 *ustrp = UTF8_EIGHT_BIT_HI(result);
2658 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2659 *lenp = 2;
2660 }
2661
2662 if (tainted_ptr) {
2663 *tainted_ptr = TRUE;
2664 }
2665 return result;
b4e400f9
JH
2666}
2667
d3e79532 2668/*
87cea99e 2669=for apidoc to_utf8_fold
d3e79532 2670
a1433954
KW
2671Convert the UTF-8 encoded character at C<p> to its foldcase version and
2672store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2673that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
d3e79532
JH
2674foldcase version may be longer than the original character (up to
2675three characters).
2676
2677The first character of the foldcased version is returned
2678(but note, as explained above, that there may be more.)
2679
a1433954
KW
2680The character at C<p> is assumed by this routine to be well-formed.
2681
d3e79532
JH
2682=cut */
2683
051a06d4
KW
2684/* Not currently externally documented, and subject to change,
2685 * in <flags>
2686 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2687 * points < 256. Since foldcase is not defined in
2688 * POSIX, lowercase is used instead
2689 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2690 * otherwise simple folds
a0270393
KW
2691 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2692 * prohibited
051a06d4
KW
2693 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2694 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2695
b4e400f9 2696UV
051a06d4 2697Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2698{
97aff369 2699 dVAR;
7918f24d 2700
051a06d4
KW
2701 UV result;
2702
36bb2ab6 2703 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2704
a0270393
KW
2705 /* These are mutually exclusive */
2706 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2707
50ba90ff
KW
2708 assert(p != ustrp); /* Otherwise overwrites */
2709
a1dde8de 2710 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2711 if (flags & FOLD_FLAGS_LOCALE) {
2712 result = toLOWER_LC(*p);
2713 }
2714 else {
81c6c7ce
KW
2715 return _to_fold_latin1(*p, ustrp, lenp,
2716 cBOOL(flags & FOLD_FLAGS_FULL));
051a06d4 2717 }
a1dde8de
KW
2718 }
2719 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2720 if (flags & FOLD_FLAGS_LOCALE) {
2721 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2722 }
2723 else {
81c6c7ce 2724 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
a0270393
KW
2725 ustrp, lenp,
2726 cBOOL((flags & FOLD_FLAGS_FULL
2727 /* If ASCII safe, don't allow full
2728 * folding, as that could include SHARP
2729 * S => ss; otherwise there is no
2730 * crossing of ascii/non-ascii in the
2731 * latin1 range */
2732 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
051a06d4 2733 }
a1dde8de 2734 }
051a06d4 2735 else { /* utf8, ord above 255 */
a0270393 2736 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2737
051a06d4 2738 if ((flags & FOLD_FLAGS_LOCALE)) {
a0270393 2739 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2740 }
a0270393
KW
2741 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2742 return result;
2743 }
2744 else {
2745 /* This is called when changing the case of a utf8-encoded
2746 * character above the Latin1 range, and the result should not
2747 * contain an ASCII character. */
2748
2749 UV original; /* To store the first code point of <p> */
2750
2751 /* Look at every character in the result; if any cross the
2752 * boundary, the whole thing is disallowed */
2753 U8* s = ustrp;
2754 U8* e = ustrp + *lenp;
2755 while (s < e) {
2756 if (isASCII(*s)) {
2757 /* Crossed, have to return the original */
2758 original = valid_utf8_to_uvchr(p, lenp);
2759 Copy(p, ustrp, *lenp, char);
2760 return original;
2761 }
2762 s += UTF8SKIP(s);
2763 }
051a06d4 2764
a0270393
KW
2765 /* Here, no characters crossed, result is ok as-is */
2766 return result;
2767 }
051a06d4
KW
2768 }
2769
2770 /* Here, used locale rules. Convert back to utf8 */
2771 if (UTF8_IS_INVARIANT(result)) {
2772 *ustrp = (U8) result;
2773 *lenp = 1;
2774 }
2775 else {
2776 *ustrp = UTF8_EIGHT_BIT_HI(result);
2777 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2778 *lenp = 2;
2779 }
2780
2781 if (tainted_ptr) {
2782 *tainted_ptr = TRUE;
2783 }
2784 return result;
a0ed51b3
LW
2785}
2786
711a919c 2787/* Note:
f90a9a02 2788 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2789 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2790 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2791 */
c4a5db0c 2792
a0ed51b3 2793SV*
7fc63493 2794Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2795{
c4a5db0c
KW
2796 PERL_ARGS_ASSERT_SWASH_INIT;
2797
2798 /* Returns a copy of a swash initiated by the called function. This is the
2799 * public interface, and returning a copy prevents others from doing
2800 * mischief on the original */
2801
5d3d13d1 2802 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2803}
2804
2805SV*
5d3d13d1 2806Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2807{
2808 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2809 * by calling utf8_heavy.pl in the general case. The returned value may be
2810 * the swash's inversion list instead if the input parameters allow it.
2811 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2812 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2813 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2814 *
2815 * This interface should only be used by functions that won't destroy or
2816 * adversely change the swash, as doing so affects all other uses of the
2817 * swash in the program; the general public should use 'Perl_swash_init'
2818 * instead.
2819 *
2820 * pkg is the name of the package that <name> should be in.
2821 * name is the name of the swash to find. Typically it is a Unicode
2822 * property name, including user-defined ones
2823 * listsv is a string to initialize the swash with. It must be of the form
2824 * documented as the subroutine return value in
2825 * L<perlunicode/User-Defined Character Properties>
2826 * minbits is the number of bits required to represent each data element.
2827 * It is '1' for binary properties.
2828 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2829 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2830 * flags_p if non-NULL is the address of various input and output flag bits
2831 * to the routine, as follows: ('I' means is input to the routine;
2832 * 'O' means output from the routine. Only flags marked O are
2833 * meaningful on return.)
2834 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2835 * came from a user-defined property. (I O)
5d3d13d1
KW
2836 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2837 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2838 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2839 * return of an inversion list instead of a swash hash if this routine
2840 * thinks that would result in faster execution of swash_fetch() later
2841 * on. (I)
9a53f6cf
KW
2842 *
2843 * Thus there are three possible inputs to find the swash: <name>,
2844 * <listsv>, and <invlist>. At least one must be specified. The result
2845 * will be the union of the specified ones, although <listsv>'s various
2846 * actions can intersect, etc. what <name> gives.
2847 *
2848 * <invlist> is only valid for binary properties */
c4a5db0c 2849
27da23d5 2850 dVAR;
c4a5db0c 2851 SV* retval = &PL_sv_undef;
83199d38 2852 HV* swash_hv = NULL;
87367d5f
KW
2853 const int invlist_swash_boundary =
2854 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2855 ? 512 /* Based on some benchmarking, but not extensive, see commit
2856 message */
2857 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2858
2859 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2860 assert(! invlist || minbits == 1);
2861
2862 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2863 * so */
2864 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2865 dSP;
2866 const size_t pkg_len = strlen(pkg);
2867 const size_t name_len = strlen(name);
2868 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2869 SV* errsv_save;
2870 GV *method;
2871
2872 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2873
2874 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2875 ENTER;
69794297
KW
2876 SAVEHINTS();
2877 save_re_context();
650f067c
JL
2878 /* We might get here via a subroutine signature which uses a utf8
2879 * parameter name, at which point PL_subname will have been set
2880 * but not yet used. */
2881 save_item(PL_subname);
69794297
KW
2882 if (PL_parser && PL_parser->error_count)
2883 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2884 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2885 if (!method) { /* demand load utf8 */
2886 ENTER;
db2c6cb3
FC
2887 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2888 GvSV(PL_errgv) = NULL;
69794297
KW
2889 /* It is assumed that callers of this routine are not passing in
2890 * any user derived data. */
2891 /* Need to do this after save_re_context() as it will set
2892 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2893 * in Perl_magic_get). Even line to create errsv_save can turn on
2894 * PL_tainted. */
284167a5
SM
2895#ifndef NO_TAINT_SUPPORT
2896 SAVEBOOL(TAINT_get);
2897 TAINT_NOT;
2898#endif
69794297
KW
2899 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2900 NULL);
eed484f9 2901 {
db2c6cb3
FC
2902 /* Not ERRSV, as there is no need to vivify a scalar we are
2903 about to discard. */
2904 SV * const errsv = GvSV(PL_errgv);
2905 if (!SvTRUE(errsv)) {
2906 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2907 SvREFCNT_dec(errsv);
2908 }
eed484f9 2909 }
69794297
KW
2910 LEAVE;
2911 }
2912 SPAGAIN;
2913 PUSHMARK(SP);
2914 EXTEND(SP,5);
2915 mPUSHp(pkg, pkg_len);
2916 mPUSHp(name, name_len);
2917 PUSHs(listsv);
2918 mPUSHi(minbits);
2919 mPUSHi(none);
2920 PUTBACK;
db2c6cb3
FC
2921 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2922 GvSV(PL_errgv) = NULL;
69794297
KW
2923 /* If we already have a pointer to the method, no need to use
2924 * call_method() to repeat the lookup. */
c41800a8
KW
2925 if (method
2926 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2927 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2928 {
2929 retval = *PL_stack_sp--;
2930 SvREFCNT_inc(retval);
2931 }
eed484f9 2932 {
db2c6cb3
FC
2933 /* Not ERRSV. See above. */
2934 SV * const errsv = GvSV(PL_errgv);
2935 if (!SvTRUE(errsv)) {
2936 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2937 SvREFCNT_dec(errsv);
2938 }
eed484f9 2939 }
ce3b816e 2940 LEAVE;
69794297
KW
2941 POPSTACK;
2942 if (IN_PERL_COMPILETIME) {
2943 CopHINTS_set(PL_curcop, PL_hints);
2944 }
2945 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2946 if (SvPOK(retval))
2947
2948 /* If caller wants to handle missing properties, let them */
5d3d13d1 2949 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
2950 return NULL;
2951 }
2952 Perl_croak(aTHX_
2953 "Can't find Unicode property definition \"%"SVf"\"",
2954 SVfARG(retval));
2955 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2956 }
9a53f6cf 2957 } /* End of calling the module to find the swash */
36eb48b4 2958
83199d38
KW
2959 /* If this operation fetched a swash, and we will need it later, get it */
2960 if (retval != &PL_sv_undef
2961 && (minbits == 1 || (flags_p
2962 && ! (*flags_p
2963 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2964 {
2965 swash_hv = MUTABLE_HV(SvRV(retval));
2966
2967 /* If we don't already know that there is a user-defined component to
2968 * this swash, and the user has indicated they wish to know if there is
2969 * one (by passing <flags_p>), find out */
2970 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2971 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2972 if (user_defined && SvUV(*user_defined)) {
2973 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2974 }
2975 }
2976 }
2977
36eb48b4
KW
2978 /* Make sure there is an inversion list for binary properties */
2979 if (minbits == 1) {
2980 SV** swash_invlistsvp = NULL;
2981 SV* swash_invlist = NULL;
9a53f6cf 2982 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2983 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2984 an unclaimed reference count */
36eb48b4 2985
9a53f6cf 2986 /* If this operation fetched a swash, get its already existing
83199d38 2987 * inversion list, or create one for it */
36eb48b4 2988
83199d38 2989 if (swash_hv) {
5c9f4bd2 2990 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2991 if (swash_invlistsvp) {
2992 swash_invlist = *swash_invlistsvp;
2993 invlist_in_swash_is_valid = TRUE;
2994 }
2995 else {
36eb48b4 2996 swash_invlist = _swash_to_invlist(retval);
02c85471 2997 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2998 }
2999 }
3000
3001 /* If an inversion list was passed in, have to include it */
3002 if (invlist) {
3003
3004 /* Any fetched swash will by now have an inversion list in it;
3005 * otherwise <swash_invlist> will be NULL, indicating that we
3006 * didn't fetch a swash */
3007 if (swash_invlist) {
3008
3009 /* Add the passed-in inversion list, which invalidates the one
3010 * already stored in the swash */
3011 invlist_in_swash_is_valid = FALSE;
3012 _invlist_union(invlist, swash_invlist, &swash_invlist);
3013 }
3014 else {
3015
87367d5f
KW
3016 /* Here, there is no swash already. Set up a minimal one, if
3017 * we are going to return a swash */
3018 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3019 swash_hv = newHV();
4aca0fe6 3020 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3021 }
9a53f6cf
KW
3022 swash_invlist = invlist;
3023 }
9a53f6cf
KW
3024 }
3025
3026 /* Here, we have computed the union of all the passed-in data. It may
3027 * be that there was an inversion list in the swash which didn't get
3028 * touched; otherwise save the one computed one */
87367d5f
KW
3029 if (! invlist_in_swash_is_valid
3030 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3031 {
5c9f4bd2 3032 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3033 {
3034 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3035 }
cc34d8c5
FC
3036 /* We just stole a reference count. */
3037 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3038 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3039 }
87367d5f 3040
c41800a8 3041 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3042 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3043 SvREFCNT_dec(retval);
02c85471
FC
3044 if (!swash_invlist_unclaimed)
3045 SvREFCNT_inc_simple_void_NN(swash_invlist);
3046 retval = newRV_noinc(swash_invlist);
87367d5f 3047 }
36eb48b4
KW
3048 }
3049
a0ed51b3
LW
3050 return retval;
3051}
3052
035d37be
JH
3053
3054/* This API is wrong for special case conversions since we may need to
3055 * return several Unicode characters for a single Unicode character
3056 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3057 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
3058 * multiple values. --jhi
3059 * For those, you should use to_utf8_case() instead */
b0e3252e 3060/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3061
3062/* Note:
3063 * Returns the value of property/mapping C<swash> for the first character
3064 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3065 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3066 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3067 *
3068 * A "swash" is a hash which contains initially the keys/values set up by
3069 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3070 * property for all possible code points. Things are stored in a compact form
3071 * (see utf8_heavy.pl) so that calculation is required to find the actual
3072 * property value for a given code point. As code points are looked up, new
3073 * key/value pairs are added to the hash, so that the calculation doesn't have
3074 * to ever be re-done. Further, each calculation is done, not just for the
3075 * desired one, but for a whole block of code points adjacent to that one.
3076 * For binary properties on ASCII machines, the block is usually for 64 code
3077 * points, starting with a code point evenly divisible by 64. Thus if the
3078 * property value for code point 257 is requested, the code goes out and
3079 * calculates the property values for all 64 code points between 256 and 319,
3080 * and stores these as a single 64-bit long bit vector, called a "swatch",
3081 * under the key for code point 256. The key is the UTF-8 encoding for code
3082 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3083 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3084 * for code point 258 is then requested, this code realizes that it would be
3085 * stored under the key for 256, and would find that value and extract the
3086 * relevant bit, offset from 256.
3087 *
3088 * Non-binary properties are stored in as many bits as necessary to represent
3089 * their values (32 currently, though the code is more general than that), not
3090 * as single bits, but the principal is the same: the value for each key is a
3091 * vector that encompasses the property values for all code points whose UTF-8
3092 * representations are represented by the key. That is, for all code points
3093 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3094 * bytes of that.
680c470c 3095 */
a0ed51b3 3096UV
680c470c 3097Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3098{
27da23d5 3099 dVAR;
ef8f7699 3100 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3101 U32 klen;
3102 U32 off;
a0ed51b3 3103 STRLEN slen;
7d85a32c 3104 STRLEN needents;
cfd0369c 3105 const U8 *tmps = NULL;
a0ed51b3 3106 U32 bit;
979f2922 3107 SV *swatch;
3568d838 3108 U8 tmputf8[2];
35da51f7 3109 const UV c = NATIVE_TO_ASCII(*ptr);
3568d838 3110
7918f24d
NC
3111 PERL_ARGS_ASSERT_SWASH_FETCH;
3112
87367d5f
KW
3113 /* If it really isn't a hash, it isn't really swash; must be an inversion
3114 * list */
3115 if (SvTYPE(hv) != SVt_PVHV) {
3116 return _invlist_contains_cp((SV*)hv,
3117 (do_utf8)
3118 ? valid_utf8_to_uvchr(ptr, NULL)
3119 : c);
3120 }
3121
dbe7a391 3122 /* Convert to utf8 if not already */
3568d838 3123 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
979f2922
ST
3124 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3125 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3126 ptr = tmputf8;
3568d838
JH
3127 }
3128 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
37e2e78e 3129 * then the "swatch" is a vec() for all the chars which start
3568d838
JH
3130 * with 0xAA..0xYY
3131 * So the key in the hash (klen) is length of encoded char -1
3132 */
3133 klen = UTF8SKIP(ptr) - 1;
3134 off = ptr[klen];
a0ed51b3 3135
979f2922 3136 if (klen == 0) {
37e2e78e 3137 /* If char is invariant then swatch is for all the invariant chars
1e54db1a 3138 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
7d85a32c 3139 */
979f2922
ST
3140 needents = UTF_CONTINUATION_MARK;
3141 off = NATIVE_TO_UTF(ptr[klen]);
3142 }
3143 else {
7d85a32c 3144 /* If char is encoded then swatch is for the prefix */
979f2922
ST
3145 needents = (1 << UTF_ACCUMULATION_SHIFT);
3146 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3147 }
7d85a32c 3148
a0ed51b3
LW
3149 /*
3150 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3151 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3152 * it's nothing to sniff at.) Pity we usually come through at least
3153 * two function calls to get here...
3154 *
3155 * NB: this code assumes that swatches are never modified, once generated!
3156 */
3157
3568d838 3158 if (hv == PL_last_swash_hv &&
a0ed51b3 3159 klen == PL_last_swash_klen &&
27da23d5 3160 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3161 {
3162 tmps = PL_last_swash_tmps;
3163 slen = PL_last_swash_slen;
3164 }
3165 else {
3166 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3167 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3168
b0e3252e 3169 /* If not cached, generate it via swatch_get */
979f2922
ST
3170 if (!svp || !SvPOK(*svp)
3171 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2b9d42f0
NIS
3172 /* We use utf8n_to_uvuni() as we want an index into
3173 Unicode tables, not a native character number.
3174 */
aec46f14 3175 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
872c91ae
JH
3176 ckWARN(WARN_UTF8) ?
3177 0 : UTF8_ALLOW_ANY);
b0e3252e 3178 swatch = swatch_get(swash,
979f2922 3179 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
361ee0fe 3180 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
979f2922
ST
3181 needents);
3182
923e4eb5 3183 if (IN_PERL_COMPILETIME)
623e6609 3184 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3185
979f2922 3186 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3187
979f2922
ST
3188 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3189 || (slen << 3) < needents)
5637ef5b
NC
3190 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3191 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3192 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3193 }
3194
3195 PL_last_swash_hv = hv;
16d8f38a 3196 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3197 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3198 /* FIXME change interpvar.h? */
3199 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3200 PL_last_swash_slen = slen;
3201 if (klen)
3202 Copy(ptr, PL_last_swash_key, klen, U8);
3203 }
3204
9faf8d75 3205 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3206 case 1:
3207 bit = 1 << (off & 7);
3208 off >>= 3;
3209 return (tmps[off] & bit) != 0;
3210 case 8:
3211 return tmps[off];
3212 case 16:
3213 off <<= 1;
3214 return (tmps[off] << 8) + tmps[off + 1] ;
3215 case 32:
3216 off <<= 2;
3217 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3218 }
5637ef5b
NC
3219 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3220 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3221 NORETURN_FUNCTION_END;
a0ed51b3 3222}
2b9d42f0 3223
319009ee
KW
3224/* Read a single line of the main body of the swash input text. These are of
3225 * the form:
3226 * 0053 0056 0073
3227 * where each number is hex. The first two numbers form the minimum and
3228 * maximum of a range, and the third is the value associated with the range.
3229 * Not all swashes should have a third number
3230 *
3231 * On input: l points to the beginning of the line to be examined; it points
3232 * to somewhere in the string of the whole input text, and is
3233 * terminated by a \n or the null string terminator.
3234 * lend points to the null terminator of that string
3235 * wants_value is non-zero if the swash expects a third number
3236 * typestr is the name of the swash's mapping, like 'ToLower'
3237 * On output: *min, *max, and *val are set to the values read from the line.
3238 * returns a pointer just beyond the line examined. If there was no
3239 * valid min number on the line, returns lend+1
3240 */
3241
3242STATIC U8*
3243S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3244 const bool wants_value, const U8* const typestr)
3245{
3246 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3247 STRLEN numlen; /* Length of the number */
02470786
KW
3248 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3249 | PERL_SCAN_DISALLOW_PREFIX
3250 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3251
3252 /* nl points to the next \n in the scan */
3253 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3254
3255 /* Get the first number on the line: the range minimum */
3256 numlen = lend - l;
3257 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3258 if (numlen) /* If found a hex number, position past it */
3259 l += numlen;
3260 else if (nl) { /* Else, go handle next line, if any */
3261 return nl + 1; /* 1 is length of "\n" */
3262 }
3263 else { /* Else, no next line */
3264 return lend + 1; /* to LIST's end at which \n is not found */
3265 }
3266
3267 /* The max range value follows, separated by a BLANK */
3268 if (isBLANK(*l)) {
3269 ++l;
02470786
KW
3270 flags = PERL_SCAN_SILENT_ILLDIGIT
3271 | PERL_SCAN_DISALLOW_PREFIX
3272 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3273 numlen = lend - l;
3274 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3275 if (numlen)
3276 l += numlen;
3277 else /* If no value here, it is a single element range */
3278 *max = *min;
3279
3280 /* Non-binary tables have a third entry: what the first element of the
3281 * range maps to */
3282 if (wants_value) {
3283 if (isBLANK(*l)) {
3284 ++l;
f90a9a02
KW
3285
3286 /* The ToLc, etc table mappings are not in hex, and must be
3287 * corrected by adding the code point to them */
3288 if (typeto) {
3289 char *after_strtol = (char *) lend;
3290 *val = Strtol((char *)l, &after_strtol, 10);
3291 l = (U8 *) after_strtol;
f90a9a02
KW
3292 }
3293 else { /* Other tables are in hex, and are the correct result
3294 without tweaking */
a9d188b3
KW
3295 flags = PERL_SCAN_SILENT_ILLDIGIT
3296 | PERL_SCAN_DISALLOW_PREFIX
3297 | PERL_SCAN_SILENT_NON_PORTABLE;
3298 numlen = lend - l;
3299 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3300 if (numlen)
3301 l += numlen;
3302 else
3303 *val = 0;
f90a9a02 3304 }
319009ee
KW
3305 }
3306 else {
3307 *val = 0;
3308 if (typeto) {
dcbac5bb 3309 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3310 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3311 typestr, l);
3312 }
3313 }
3314 }
3315 else
3316 *val = 0; /* bits == 1, then any val should be ignored */
3317 }
3318 else { /* Nothing following range min, should be single element with no
3319 mapping expected */
3320 *max = *min;
3321 if (wants_value) {
3322 *val = 0;
3323 if (typeto) {
dcbac5bb 3324 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3325 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3326 }
3327 }
3328 else
3329 *val = 0; /* bits == 1, then val should be ignored */
3330 }
3331
3332 /* Position to next line if any, or EOF */
3333 if (nl)
3334 l = nl + 1;
3335 else
3336 l = lend;
3337
3338 return l;
3339}
3340
979f2922
ST
3341/* Note:
3342 * Returns a swatch (a bit vector string) for a code point sequence
3343 * that starts from the value C<start> and comprises the number C<span>.
3344 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3345 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3346 */
3347STATIC SV*
b0e3252e 3348S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3349{
3350 SV *swatch;
77f9f126 3351 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3352 STRLEN lcur, xcur, scur;
ef8f7699 3353 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3354 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3355
88d45d28
KW
3356 SV** listsvp = NULL; /* The string containing the main body of the table */
3357 SV** extssvp = NULL;
3358 SV** invert_it_svp = NULL;
3359 U8* typestr = NULL;
786861f5
KW
3360 STRLEN bits;
3361 STRLEN octets; /* if bits == 1, then octets == 0 */
3362 UV none;
3363 UV end = start + span;
972dd592 3364
36eb48b4 3365 if (invlistsvp == NULL) {
786861f5
KW
3366 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3367 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3368 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3369 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3370 listsvp = hv_fetchs(hv, "LIST", FALSE);
3371 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3372
3373 bits = SvUV(*bitssvp);
3374 none = SvUV(*nonesvp);
3375 typestr = (U8*)SvPV_nolen(*typesvp);
3376 }
36eb48b4
KW
3377 else {
3378 bits = 1;
3379 none = 0;
3380 }
786861f5 3381 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3382
b0e3252e 3383 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3384
979f2922 3385 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3386 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3387 (UV)bits);
979f2922
ST
3388 }
3389
84ea5ef6
KW
3390 /* If overflowed, use the max possible */
3391 if (end < start) {
3392 end = UV_MAX;
3393 span = end - start;
3394 }
3395
979f2922 3396 /* create and initialize $swatch */
979f2922 3397 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3398 swatch = newSV(scur);
3399 SvPOK_on(swatch);
979f2922
ST
3400 s = (U8*)SvPVX(swatch);
3401 if (octets && none) {
0bd48802 3402 const U8* const e = s + scur;
979f2922
ST
3403 while (s < e) {
3404 if (bits == 8)
3405 *s++ = (U8)(none & 0xff);
3406 else if (bits == 16) {
3407 *s++ = (U8)((none >> 8) & 0xff);
3408 *s++ = (U8)( none & 0xff);
3409 }
3410 else if (bits == 32) {
3411 *s++ = (U8)((none >> 24) & 0xff);
3412 *s++ = (U8)((none >> 16) & 0xff);
3413 *s++ = (U8)((none >> 8) & 0xff);
3414 *s++ = (U8)( none & 0xff);
3415 }
3416 }
3417 *s = '\0';
3418 }
3419 else {
3420 (void)memzero((U8*)s, scur + 1);
3421 }
3422 SvCUR_set(swatch, scur);
3423 s = (U8*)SvPVX(swatch);
3424
36eb48b4
KW
3425 if (invlistsvp) { /* If has an inversion list set up use that */
3426 _invlist_populate_swatch(*invlistsvp, start, end, s);
3427 return swatch;
3428 }
3429
3430 /* read $swash->{LIST} */
979f2922
ST
3431 l = (U8*)SvPV(*listsvp, lcur);
3432 lend = l + lcur;
3433 while (l < lend) {
8ed25d53 3434 UV min, max, val, upper;
319009ee
KW
3435 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3436 cBOOL(octets), typestr);
3437 if (l > lend) {
979f2922
ST
3438 break;
3439 }
3440
972dd592 3441 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3442 if (max < start)
3443 continue;
3444
8ed25d53
KW
3445 /* <end> is generally 1 beyond where we want to set things, but at the
3446 * platform's infinity, where we can't go any higher, we want to
3447 * include the code point at <end> */
3448 upper = (max < end)
3449 ? max
3450 : (max != UV_MAX || end != UV_MAX)
3451 ? end - 1
3452 : end;
3453
979f2922 3454 if (octets) {
35da51f7 3455 UV key;
979f2922
ST
3456 if (min < start) {
3457 if (!none || val < none) {
3458 val += start - min;
3459 }
3460 min = start;
3461 }
8ed25d53 3462 for (key = min; key <= upper; key++) {
979f2922 3463 STRLEN offset;
979f2922
ST
3464 /* offset must be non-negative (start <= min <= key < end) */
3465 offset = octets * (key - start);
3466 if (bits == 8)
3467 s[offset] = (U8)(val & 0xff);
3468 else if (bits == 16) {
3469 s[offset ] = (U8)((val >> 8) & 0xff);
3470 s[offset + 1] = (U8)( val & 0xff);
3471 }
3472 else if (bits == 32) {
3473 s[offset ] = (U8)((val >> 24) & 0xff);
3474 s[offset + 1] = (U8)((val >> 16) & 0xff);
3475 s[offset + 2] = (U8)((val >> 8) & 0xff);
3476 s[offset + 3] = (U8)( val & 0xff);
3477 }
3478
3479 if (!none || val < none)
3480 ++val;
3481 }
3482 }
711a919c 3483 else { /* bits == 1, then val should be ignored */
35da51f7 3484 UV key;
979f2922
ST
3485 if (min < start)
3486 min = start;
6cb05c12 3487
8ed25d53 3488 for (key = min; key <= upper; key++) {
0bd48802 3489 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3490 s[offset >> 3] |= 1 << (offset & 7);
3491 }
3492 }
3493 } /* while */
979f2922 3494
9479a769 3495 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3496 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3497
3498 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3499 * be 0, and their inversion should also be 0, as we don't succeed any
3500 * Unicode property matches for non-Unicode code points */
3501 if (start <= PERL_UNICODE_MAX) {
3502
3503 /* The code below assumes that we never cross the
3504 * Unicode/above-Unicode boundary in a range, as otherwise we would
3505 * have to figure out where to stop flipping the bits. Since this
3506 * boundary is divisible by a large power of 2, and swatches comes
3507 * in small powers of 2, this should be a valid assumption */
3508 assert(start + span - 1 <= PERL_UNICODE_MAX);
3509
507a8485
KW
3510 send = s + scur;
3511 while (s < send) {
3512 *s = ~(*s);
3513 s++;
3514 }
0bda3001 3515 }
77f9f126
KW
3516 }
3517
d73c39c5
KW
3518 /* read $swash->{EXTRAS}
3519 * This code also copied to swash_to_invlist() below */
979f2922
ST
3520 x = (U8*)SvPV(*extssvp, xcur);
3521 xend = x + xcur;
3522 while (x < xend) {
3523 STRLEN namelen;
3524 U8 *namestr;
3525 SV** othersvp;
3526 HV* otherhv;
3527 STRLEN otherbits;
3528 SV **otherbitssvp, *other;
711a919c 3529 U8 *s, *o, *nl;
979f2922
ST
3530 STRLEN slen, olen;
3531
35da51f7 3532 const U8 opc = *x++;
979f2922
ST
3533 if (opc == '\n')
3534 continue;
3535
3536 nl = (U8*)memchr(x, '\n', xend - x);
3537
3538 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3539 if (nl) {
3540 x = nl + 1; /* 1 is length of "\n" */
3541 continue;
3542 }
3543 else {
3544 x = xend; /* to EXTRAS' end at which \n is not found */
3545 break;
3546 }
3547 }
3548
3549 namestr = x;
3550 if (nl) {
3551 namelen = nl - namestr;
3552 x = nl + 1;
3553 }
3554 else {
3555 namelen = xend - namestr;
3556 x = xend;
3557 }
3558
3559 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3560 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3561 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3562 otherbits = (STRLEN)SvUV(*otherbitssvp);
3563 if (bits < otherbits)
5637ef5b
NC
3564 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3565 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3566
3567 /* The "other" swatch must be destroyed after. */
b0e3252e 3568 other = swatch_get(*othersvp, start, span);
979f2922
ST
3569 o = (U8*)SvPV(other, olen);
3570
3571 if (!olen)
b0e3252e 3572 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3573
3574 s = (U8*)SvPV(swatch, slen);
3575 if (bits == 1 && otherbits == 1) {
3576 if (slen != olen)
5637ef5b
NC
3577 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3578 "mismatch, slen=%"UVuf", olen=%"UVuf,
3579 (UV)slen, (UV)olen);
979f2922
ST
3580
3581 switch (opc) {
3582 case '+':
3583 while (slen--)
3584 *s++ |= *o++;
3585 break;
3586 case '!':
3587 while (slen--)
3588 *s++ |= ~*o++;
3589 break;
3590 case '-':
3591 while (slen--)
3592 *s++ &= ~*o++;
3593 break;
3594 case '&':
3595 while (slen--)
3596 *s++ &= *o++;
3597 break;
3598 default:
3599 break;
3600 }
3601 }
711a919c 3602 else {
979f2922
ST
3603 STRLEN otheroctets = otherbits >> 3;
3604 STRLEN offset = 0;
35da51f7 3605 U8* const send = s + slen;
979f2922
ST
3606
3607 while (s < send) {
3608 UV otherval = 0;
3609
3610 if (otherbits == 1) {
3611 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3612 ++offset;
3613 }
3614 else {
3615 STRLEN vlen = otheroctets;
3616 otherval = *o++;
3617 while (--vlen) {
3618 otherval <<= 8;
3619 otherval |= *o++;
3620 }
3621 }
3622
711a919c 3623 if (opc == '+' && otherval)
6f207bd3 3624 NOOP; /* replace with otherval */
979f2922
ST
3625 else if (opc == '!' && !otherval)
3626 otherval = 1;
3627 else if (opc == '-' && otherval)
3628 otherval = 0;
3629 else if (opc == '&' && !otherval)
3630 otherval = 0;
3631 else {
711a919c 3632 s += octets; /* no replacement */
979f2922
ST
3633 continue;
3634 }
3635
3636 if (bits == 8)
3637 *s++ = (U8)( otherval & 0xff);
3638 else if (bits == 16) {
3639 *s++ = (U8)((otherval >> 8) & 0xff);
3640 *s++ = (U8)( otherval & 0xff);
3641 }
3642 else if (bits == 32) {
3643 *s++ = (U8)((otherval >> 24) & 0xff);
3644 *s++ = (U8)((otherval >> 16) & 0xff);
3645 *s++ = (U8)((otherval >> 8) & 0xff);
3646 *s++ = (U8)( otherval & 0xff);
3647 }
3648 }
3649 }
3650 sv_free(other); /* through with it! */
3651 } /* while */
3652 return swatch;
3653}
3654
064c021d 3655HV*
4c2e1131 3656Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3657{
3658
79a2a0e8 3659 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3660 * Can't be used on a property that is subject to user override, as it
3661 * relies on the value of SPECIALS in the swash which would be set by
3662 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3663 * for overridden properties
064c021d
KW
3664 *
3665 * Returns a hash which is the inversion and closure of a swash mapping.
3666 * For example, consider the input lines:
3667 * 004B 006B
3668 * 004C 006C
3669 * 212A 006B
3670 *
3671 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3672 * 006C. The value for each key is an array. For 006C, the array would
3673 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3674 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3675 *
3676 * Essentially, for any code point, it gives all the code points that map to
3677 * it, or the list of 'froms' for that point.
3678 *
5662e334
KW
3679 * Currently it ignores any additions or deletions from other swashes,
3680 * looking at just the main body of the swash, and if there are SPECIALS
3681 * in the swash, at that hash
3682 *
3683 * The specials hash can be extra code points, and most likely consists of
3684 * maps from single code points to multiple ones (each expressed as a string
3685 * of utf8 characters). This function currently returns only 1-1 mappings.
3686 * However consider this possible input in the specials hash:
3687 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3688 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3689 *
3690 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3691 * currently handle. But it also means that FB05 and FB06 are equivalent in
3692 * a 1-1 mapping which we should handle, and this relationship may not be in
3693 * the main table. Therefore this function examines all the multi-char
3694 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3695
3696 U8 *l, *lend;
3697 STRLEN lcur;
3698 HV *const hv = MUTABLE_HV(SvRV(swash));
3699
923b6d4e
KW
3700 /* The string containing the main body of the table. This will have its
3701 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3702 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3703
3704 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3705 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3706 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3707 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3708 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3709 const STRLEN bits = SvUV(*bitssvp);
3710 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3711 const UV none = SvUV(*nonesvp);
5662e334 3712 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3713
3714 HV* ret = newHV();
3715
3716 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3717
3718 /* Must have at least 8 bits to get the mappings */