This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Bypass a subroutine wrapper
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
a0ed51b3 35
a0c21aa1 36#ifndef EBCDIC
970ea3cb 37/* Separate prototypes needed because in ASCII systems these are
a0c21aa1
JH
38 * usually macros but they still are compiled as code, too. */
39PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
6dd9dce9 40PERL_CALLCONV UV Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen);
a0c21aa1
JH
41PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
42#endif
43
27da23d5
JH
44static const char unees[] =
45 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 46
48ef279e 47/*
ccfc67b7 48=head1 Unicode Support
a0ed51b3 49
166f8a29
DM
50This file contains various utility functions for manipulating UTF8-encoded
51strings. For the uninitiated, this is a method of representing arbitrary
61296642 52Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
53characters in the ASCII range are unmodified, and a zero byte never appears
54within non-zero characters.
166f8a29 55
eaf7a4d2
CS
56=cut
57*/
58
59/*
60=for apidoc is_ascii_string
61
a1433954 62Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
63or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
64is, if they are invariant. On ASCII-ish machines, only ASCII characters
65fit this definition, hence the function's name.
eaf7a4d2 66
9f7e3d64
MH
67If C<len> is 0, it will be calculated using C<strlen(s)>.
68
a1433954 69See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
70
71=cut
72*/
73
74bool
668b6d8d 75Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
76{
77 const U8* const send = s + (len ? len : strlen((const char *)s));
78 const U8* x = s;
79
80 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
81
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
84 break;
85 }
86
87 return x == send;
88}
89
90/*
87cea99e 91=for apidoc uvuni_to_utf8_flags
eebe1485 92
6ee84de2
KW
93Adds the UTF-8 representation of the code point C<uv> to the end
94of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
eebe1485 95bytes available. The return value is the pointer to the byte after the
9041c2e3 96end of the new character. In other words,
eebe1485 97
b851fbc1
JH
98 d = uvuni_to_utf8_flags(d, uv, flags);
99
100or, in most cases,
101
9041c2e3 102 d = uvuni_to_utf8(d, uv);
eebe1485 103
b851fbc1
JH
104(which is equivalent to)
105
106 d = uvuni_to_utf8_flags(d, uv, 0);
107
949cf498 108This is the recommended Unicode-aware way of saying
eebe1485
SC
109
110 *(d++) = uv;
111
949cf498
KW
112This function will convert to UTF-8 (and not warn) even code points that aren't
113legal Unicode or are problematic, unless C<flags> contains one or more of the
a1433954
KW
114following flags:
115
949cf498
KW
116If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
117the function will raise a warning, provided UTF8 warnings are enabled. If instead
118UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
119If both flags are set, the function will both warn and return NULL.
120
121The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
122affect how the function handles a Unicode non-character. And, likewise for the
123UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
124above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
125even less portable) can be warned and/or disallowed even if other above-Unicode
126code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
127flags.
128
129And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
130above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
131DISALLOW flags.
132
133
eebe1485
SC
134=cut
135*/
136
dfe13c55 137U8 *
b851fbc1 138Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 139{
7918f24d
NC
140 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
141
979f77b6
KW
142 /* The first problematic code point is the first surrogate */
143 if (uv >= UNICODE_SURROGATE_FIRST
144 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
145 {
949cf498
KW
146 if (UNICODE_IS_SURROGATE(uv)) {
147 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 148 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
149 "UTF-16 surrogate U+%04"UVXf, uv);
150 }
151 if (flags & UNICODE_DISALLOW_SURROGATE) {
152 return NULL;
153 }
154 }
155 else if (UNICODE_IS_SUPER(uv)) {
156 if (flags & UNICODE_WARN_SUPER
157 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
158 {
8457b38f 159 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
160 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
161 }
162 if (flags & UNICODE_DISALLOW_SUPER
163 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
164 {
165 return NULL;
166 }
167 }
168 else if (UNICODE_IS_NONCHAR(uv)) {
169 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 170 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
171 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
172 uv);
173 }
174 if (flags & UNICODE_DISALLOW_NONCHAR) {
175 return NULL;
176 }
177 }
507b9800 178 }
c4d5f83a 179 if (UNI_IS_INVARIANT(uv)) {
eb160463 180 *d++ = (U8)UTF_TO_NATIVE(uv);
a0ed51b3
LW
181 return d;
182 }
2d331972 183#if defined(EBCDIC)
1d72bdf6
NIS
184 else {
185 STRLEN len = UNISKIP(uv);
186 U8 *p = d+len-1;
187 while (p > d) {
eb160463 188 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
189 uv >>= UTF_ACCUMULATION_SHIFT;
190 }
eb160463 191 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
192 return d+len;
193 }
194#else /* Non loop style */
a0ed51b3 195 if (uv < 0x800) {
eb160463
GS
196 *d++ = (U8)(( uv >> 6) | 0xc0);
197 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
198 return d;
199 }
200 if (uv < 0x10000) {
eb160463
GS
201 *d++ = (U8)(( uv >> 12) | 0xe0);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
204 return d;
205 }
206 if (uv < 0x200000) {
eb160463
GS
207 *d++ = (U8)(( uv >> 18) | 0xf0);
208 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
209 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
210 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
211 return d;
212 }
213 if (uv < 0x4000000) {
eb160463
GS
214 *d++ = (U8)(( uv >> 24) | 0xf8);
215 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
218 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
219 return d;
220 }
221 if (uv < 0x80000000) {
eb160463
GS
222 *d++ = (U8)(( uv >> 30) | 0xfc);
223 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
227 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
228 return d;
229 }
6b8eaf93 230#ifdef HAS_QUAD
d7578b48 231 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
232#endif
233 {
eb160463
GS
234 *d++ = 0xfe; /* Can't match U+FEFF! */
235 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
239 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
240 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
241 return d;
242 }
6b8eaf93 243#ifdef HAS_QUAD
a0ed51b3 244 {
eb160463
GS
245 *d++ = 0xff; /* Can't match U+FFFE! */
246 *d++ = 0x80; /* 6 Reserved bits */
247 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
248 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
256 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
257 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
258 return d;
259 }
260#endif
1d72bdf6 261#endif /* Loop style */
a0ed51b3 262}
9041c2e3 263
646ca15d
JH
264/*
265
f7d739d1 266Tests if the first C<len> bytes of string C<s> form a valid UTF-8
646ca15d 267character. Note that an INVARIANT (i.e. ASCII) character is a valid
f7d739d1 268UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
269will be returned if it is valid, otherwise 0.
270
271This is the "slow" version as opposed to the "fast" version which is
272the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
273difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
274or less you should use the IS_UTF8_CHAR(), for lengths of five or more
275you should use the _slow(). In practice this means that the _slow()
276will be used very rarely, since the maximum Unicode code point (as of
277Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
278the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
279five bytes or more.
280
281=cut */
c053b435 282STATIC STRLEN
5f66b61c 283S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 284{
cd7e6c88 285 dTHX; /* The function called below requires thread context */
646ca15d 286
cd7e6c88 287 STRLEN actual_len;
646ca15d 288
cd7e6c88 289 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 290
cd7e6c88 291 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 292
cd7e6c88 293 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 294}
9041c2e3
NIS
295
296/*
492a624f
KW
297=for apidoc is_utf8_char_buf
298
299Returns the number of bytes that comprise the first UTF-8 encoded character in
300buffer C<buf>. C<buf_end> should point to one position beyond the end of the
301buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
302encoded character.
303
304Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
305machines) is a valid UTF-8 character.
306
307=cut */
308
309STRLEN
310Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
311{
312
313 STRLEN len;
314
315 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
316
317 if (buf_end <= buf) {
318 return 0;
319 }
320
321 len = buf_end - buf;
322 if (len > UTF8SKIP(buf)) {
323 len = UTF8SKIP(buf);
324 }
325
326#ifdef IS_UTF8_CHAR
327 if (IS_UTF8_CHAR_FAST(len))
328 return IS_UTF8_CHAR(buf, len) ? len : 0;
329#endif /* #ifdef IS_UTF8_CHAR */
330 return is_utf8_char_slow(buf, len);
331}
332
333/*
87cea99e 334=for apidoc is_utf8_char
eebe1485 335
76848387
KW
336DEPRECATED!
337
5da9da9e 338Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
339character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
340character is a valid UTF-8 character. The actual number of bytes in the UTF-8
341character will be returned if it is valid, otherwise 0.
9041c2e3 342
76848387 343This function is deprecated due to the possibility that malformed input could
a1433954 344cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 345instead.
e0328548 346
82686b01 347=cut */
76848387 348
067a85ef 349STRLEN
668b6d8d 350Perl_is_utf8_char(const U8 *s)
386d01d6 351{
7918f24d 352 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 353
76848387 354 /* Assumes we have enough space, which is why this is deprecated */
492a624f 355 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
356}
357
eaf7a4d2 358
6662521e 359/*
87cea99e 360=for apidoc is_utf8_string
6662521e 361
a1433954 362Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 363UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
364using C<strlen(s)> (which means if you use this option, that C<s> has to have a
365terminating NUL byte). Note that all characters being ASCII constitute 'a
366valid UTF-8 string'.
6662521e 367
a1433954 368See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 369
6662521e
GS
370=cut
371*/
372
8e84507e 373bool
668b6d8d 374Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 375{
35da51f7 376 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 377 const U8* x = s;
067a85ef 378
7918f24d 379 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 380
6662521e 381 while (x < send) {
1acdb0da 382 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
383 if (UTF8_IS_INVARIANT(*x)) {
384 x++;
385 }
1acdb0da 386 else if (!UTF8_IS_START(*x))
e0328548 387 return FALSE;
1acdb0da
JH
388 else {
389 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
390 const STRLEN c = UTF8SKIP(x);
391 const U8* const next_char_ptr = x + c;
392
393 if (next_char_ptr > send) {
394 return FALSE;
395 }
396
768c67ee
JH
397 if (IS_UTF8_CHAR_FAST(c)) {
398 if (!IS_UTF8_CHAR(x, c))
e0328548 399 return FALSE;
3c614e38 400 }
e0328548
KW
401 else if (! is_utf8_char_slow(x, c)) {
402 return FALSE;
403 }
404 x = next_char_ptr;
1acdb0da 405 }
6662521e 406 }
768c67ee 407
067a85ef 408 return TRUE;
6662521e
GS
409}
410
67e989fb 411/*
814fafa7
NC
412Implemented as a macro in utf8.h
413
87cea99e 414=for apidoc is_utf8_string_loc
814fafa7 415
a1433954
KW
416Like L</is_utf8_string> but stores the location of the failure (in the
417case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
418"utf8ness success") in the C<ep>.
419
a1433954 420See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 421
87cea99e 422=for apidoc is_utf8_string_loclen
81cd54e3 423
a1433954
KW
424Like L</is_utf8_string>() but stores the location of the failure (in the
425case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
426"utf8ness success") in the C<ep>, and the number of UTF-8
427encoded characters in the C<el>.
428
a1433954 429See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
430
431=cut
432*/
433
434bool
668b6d8d 435Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 436{
35da51f7 437 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 438 const U8* x = s;
81cd54e3 439 STRLEN c;
3ebfea28 440 STRLEN outlen = 0;
7918f24d
NC
441
442 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 443
81cd54e3 444 while (x < send) {
e0328548
KW
445 const U8* next_char_ptr;
446
81cd54e3
JH
447 /* Inline the easy bits of is_utf8_char() here for speed... */
448 if (UTF8_IS_INVARIANT(*x))
e0328548 449 next_char_ptr = x + 1;
768c67ee
JH
450 else if (!UTF8_IS_START(*x))
451 goto out;
81cd54e3 452 else {
768c67ee 453 /* ... and call is_utf8_char() only if really needed. */
768c67ee 454 c = UTF8SKIP(x);
e0328548
KW
455 next_char_ptr = c + x;
456 if (next_char_ptr > send) {
457 goto out;
458 }
768c67ee
JH
459 if (IS_UTF8_CHAR_FAST(c)) {
460 if (!IS_UTF8_CHAR(x, c))
461 c = 0;
462 } else
463 c = is_utf8_char_slow(x, c);
768c67ee
JH
464 if (!c)
465 goto out;
81cd54e3 466 }
e0328548 467 x = next_char_ptr;
3ebfea28 468 outlen++;
81cd54e3 469 }
768c67ee
JH
470
471 out:
3ebfea28
AL
472 if (el)
473 *el = outlen;
474
768c67ee
JH
475 if (ep)
476 *ep = x;
3ebfea28 477 return (x == send);
81cd54e3
JH
478}
479
480/*
768c67ee 481
87cea99e 482=for apidoc utf8n_to_uvuni
67e989fb 483
9041c2e3 484Bottom level UTF-8 decode routine.
746afd53
KW
485Returns the code point value of the first character in the string C<s>,
486which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
487C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
488the length, in bytes, of that character.
949cf498
KW
489
490The value of C<flags> determines the behavior when C<s> does not point to a
491well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
492zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
493next possible position in C<s> that could begin a non-malformed character.
494Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
495
496Various ALLOW flags can be set in C<flags> to allow (and not warn on)
497individual types of malformations, such as the sequence being overlong (that
498is, when there is a shorter sequence that can express the same code point;
499overlong sequences are expressly forbidden in the UTF-8 standard due to
500potential security issues). Another malformation example is the first byte of
501a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
502flags. For allowed 0 length strings, this function returns 0; for allowed
503overlong sequences, the computed code point is returned; for all other allowed
504malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
505determinable reasonable value.
949cf498
KW
506
507The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
508flags) malformation is found. If this flag is set, the routine assumes that
509the caller will raise a warning, and this function will silently just set
510C<retlen> to C<-1> and return zero.
511
512Certain code points are considered problematic. These are Unicode surrogates,
746afd53 513Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 514By default these are considered regular code points, but certain situations
5eafe189 515warrant special handling for them. If C<flags> contains
949cf498
KW
516UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
517malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
518UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
519maximum) can be set to disallow these categories individually.
520
521The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
522UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
523for their respective categories, but otherwise the code points are considered
524valid (not malformations). To get a category to both be treated as a
525malformation and raise a warning, specify both the WARN and DISALLOW flags.
526(But note that warnings are not raised if lexically disabled nor if
527UTF8_CHECK_ONLY is also specified.)
528
529Very large code points (above 0x7FFF_FFFF) are considered more problematic than
530the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
531reasons: they requre at least 32 bits to represent them on ASCII platforms, are
532not representable at all on EBCDIC platforms, and the original UTF-8
533specification never went above this number (the current 0x10FFFF limit was
534imposed later). (The smaller ones, those that fit into 32 bits, are
535representable by a UV on ASCII platforms, but not by an IV, which means that
536the number of operations that can be performed on them is quite restricted.)
537The UTF-8 encoding on ASCII platforms for these large code points begins with a
538byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
539be treated as malformations, while allowing smaller above-Unicode code points.
540(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
541including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
542the other WARN flags, but applies just to these code points.
949cf498
KW
543
544All other code points corresponding to Unicode characters, including private
545use and those yet to be assigned, are never considered malformed and never
546warn.
67e989fb 547
ec5f19d0 548Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
9041c2e3 549
37607a96
PK
550=cut
551*/
67e989fb 552
a0ed51b3 553UV
7fc63493 554Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 555{
97aff369 556 dVAR;
d4c19fe8 557 const U8 * const s0 = s;
eb83ed87 558 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 559 U8 * send;
eb83ed87
KW
560 UV uv = *s;
561 STRLEN expectlen;
949cf498 562 SV* sv = NULL;
eb83ed87
KW
563 UV outlier_ret = 0; /* return value when input is in error or problematic
564 */
565 UV pack_warn = 0; /* Save result of packWARN() for later */
566 bool unexpected_non_continuation = FALSE;
567 bool overflowed = FALSE;
2f8f112e 568 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 569
eb83ed87 570 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 571
eb83ed87 572 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
a0dbb045 573
eb83ed87
KW
574 /* The order of malformation tests here is important. We should consume as
575 * few bytes as possible in order to not skip any valid character. This is
576 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
577 * http://unicode.org/reports/tr36 for more discussion as to why. For
578 * example, once we've done a UTF8SKIP, we can tell the expected number of
579 * bytes, and could fail right off the bat if the input parameters indicate
580 * that there are too few available. But it could be that just that first
581 * byte is garbled, and the intended character occupies fewer bytes. If we
582 * blindly assumed that the first byte is correct, and skipped based on
583 * that number, we could skip over a valid input character. So instead, we
584 * always examine the sequence byte-by-byte.
585 *
586 * We also should not consume too few bytes, otherwise someone could inject
587 * things. For example, an input could be deliberately designed to
588 * overflow, and if this code bailed out immediately upon discovering that,
589 * returning to the caller *retlen pointing to the very next byte (one
590 * which is actually part of of the overflowing sequence), that could look
591 * legitimate to the caller, which could discard the initial partial
592 * sequence and process the rest, inappropriately */
593
594 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 595 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
596 if (retlen) {
597 *retlen = 0;
598 }
a0dbb045 599
eb83ed87
KW
600 if (flags & UTF8_ALLOW_EMPTY) {
601 return 0;
602 }
603 if (! (flags & UTF8_CHECK_ONLY)) {
604 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
605 }
0c443dc2
JH
606 goto malformed;
607 }
608
eb83ed87
KW
609 expectlen = UTF8SKIP(s);
610
611 /* A well-formed UTF-8 character, as the vast majority of calls to this
612 * function will be for, has this expected length. For efficiency, set
613 * things up here to return it. It will be overriden only in those rare
614 * cases where a malformation is found */
615 if (retlen) {
616 *retlen = expectlen;
617 }
618
619 /* An invariant is trivially well-formed */
1d72bdf6 620 if (UTF8_IS_INVARIANT(uv)) {
c4d5f83a 621 return (UV) (NATIVE_TO_UTF(*s));
a0ed51b3 622 }
67e989fb 623
eb83ed87 624 /* A continuation character can't start a valid sequence */
b5b9af04 625 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
626 if (flags & UTF8_ALLOW_CONTINUATION) {
627 if (retlen) {
628 *retlen = 1;
629 }
630 return UNICODE_REPLACEMENT;
631 }
ba210ebe 632
eb83ed87
KW
633 if (! (flags & UTF8_CHECK_ONLY)) {
634 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
635 }
636 curlen = 1;
ba210ebe
JH
637 goto malformed;
638 }
9041c2e3 639
1d72bdf6 640#ifdef EBCDIC
75383841 641 uv = NATIVE_TO_UTF(uv);
1d72bdf6
NIS
642#endif
643
eb83ed87
KW
644 /* Here is not a continuation byte, nor an invariant. The only thing left
645 * is a start byte (possibly for an overlong) */
ba210ebe 646
eb83ed87
KW
647 /* Remove the leading bits that indicate the number of bytes in the
648 * character's whole UTF-8 sequence, leaving just the bits that are part of
649 * the value */
650 uv &= UTF_START_MASK(expectlen);
ba210ebe 651
eb83ed87
KW
652 /* Now, loop through the remaining bytes in the character's sequence,
653 * accumulating each into the working value as we go. Be sure to not look
654 * past the end of the input string */
0b8d30e8
KW
655 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
656
eb83ed87 657 for (s = s0 + 1; s < send; s++) {
b5b9af04 658 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
659#ifndef EBCDIC /* Can't overflow in EBCDIC */
660 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
661
662 /* The original implementors viewed this malformation as more
663 * serious than the others (though I, khw, don't understand
664 * why, since other malformations also give very very wrong
665 * results), so there is no way to turn off checking for it.
666 * Set a flag, but keep going in the loop, so that we absorb
667 * the rest of the bytes that comprise the character. */
668 overflowed = TRUE;
669 overflow_byte = *s; /* Save for warning message's use */
670 }
671#endif
8850bf83 672 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
673 }
674 else {
675 /* Here, found a non-continuation before processing all expected
676 * bytes. This byte begins a new character, so quit, even if
677 * allowing this malformation. */
678 unexpected_non_continuation = TRUE;
679 break;
680 }
681 } /* End of loop through the character's bytes */
682
683 /* Save how many bytes were actually in the character */
684 curlen = s - s0;
685
686 /* The loop above finds two types of malformations: non-continuation and/or
687 * overflow. The non-continuation malformation is really a too-short
688 * malformation, as it means that the current character ended before it was
689 * expected to (being terminated prematurely by the beginning of the next
690 * character, whereas in the too-short malformation there just are too few
691 * bytes available to hold the character. In both cases, the check below
692 * that we have found the expected number of bytes would fail if executed.)
693 * Thus the non-continuation malformation is really unnecessary, being a
694 * subset of the too-short malformation. But there may be existing
695 * applications that are expecting the non-continuation type, so we retain
696 * it, and return it in preference to the too-short malformation. (If this
697 * code were being written from scratch, the two types might be collapsed
698 * into one.) I, khw, am also giving priority to returning the
699 * non-continuation and too-short malformations over overflow when multiple
700 * ones are present. I don't know of any real reason to prefer one over
701 * the other, except that it seems to me that multiple-byte errors trumps
702 * errors from a single byte */
b5b9af04 703 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
704 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
705 if (! (flags & UTF8_CHECK_ONLY)) {
706 if (curlen == 1) {
707 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
708 }
709 else {
710 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, %d bytes after start byte 0x%02x, expected %d bytes)", malformed_text, *s, (int) curlen, *s0, (int)expectlen));
a0dbb045
JH
711 }
712 }
eb83ed87
KW
713 goto malformed;
714 }
715 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
716
717 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
718 * as what the original expectations were. */
719 do_overlong_test = FALSE;
eb83ed87
KW
720 if (retlen) {
721 *retlen = curlen;
722 }
723 }
b5b9af04 724 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
725 if (! (flags & UTF8_ALLOW_SHORT)) {
726 if (! (flags & UTF8_CHECK_ONLY)) {
727 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, *s0));
a0dbb045 728 }
eb83ed87
KW
729 goto malformed;
730 }
731 uv = UNICODE_REPLACEMENT;
2f8f112e 732 do_overlong_test = FALSE;
eb83ed87
KW
733 if (retlen) {
734 *retlen = curlen;
735 }
736 }
737
738#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
741 {
742 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
743 * generation of the sv, since no warnings are raised under CHECK */
744 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
745 && ckWARN_d(WARN_UTF8))
746 {
42303544
KW
747 /* This message is deliberately not of the same syntax as the other
748 * messages for malformations, for backwards compatibility in the
749 * unlikely event that code is relying on its precise earlier text
750 */
eb83ed87
KW
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
752 pack_warn = packWARN(WARN_UTF8);
753 }
754 if (flags & UTF8_DISALLOW_FE_FF) {
755 goto malformed;
ba210ebe 756 }
ba210ebe 757 }
b5b9af04 758 if (UNLIKELY(overflowed)) {
ba210ebe 759
eb83ed87
KW
760 /* If the first byte is FF, it will overflow a 32-bit word. If the
761 * first byte is FE, it will overflow a signed 32-bit word. The
762 * above preserves backward compatibility, since its message was used
763 * in earlier versions of this code in preference to overflow */
764 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 765 goto malformed;
eb83ed87
KW
766 }
767#endif
768
2f8f112e
KW
769 if (do_overlong_test
770 && expectlen > (STRLEN)UNISKIP(uv)
771 && ! (flags & UTF8_ALLOW_LONG))
772 {
eb83ed87
KW
773 /* The overlong malformation has lower precedence than the others.
774 * Note that if this malformation is allowed, we return the actual
775 * value, instead of the replacement character. This is because this
776 * value is actually well-defined. */
777 if (! (flags & UTF8_CHECK_ONLY)) {
778 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), *s0));
779 }
780 goto malformed;
781 }
782
783 /* Here, the input is considered to be well-formed , but could be a
784 * problematic code point that is not allowed by the input parameters. */
785 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
786 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
787 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
788 {
949cf498 789 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
790 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
791 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
792 {
111d382d 793 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 794 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
795 }
796 if (flags & UTF8_DISALLOW_SURROGATE) {
797 goto disallowed;
798 }
799 }
949cf498 800 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
801 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
802 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
803 {
111d382d 804 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 805 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
806 }
807 if (flags & UTF8_DISALLOW_SUPER) {
808 goto disallowed;
809 }
810 }
4190d317
KW
811 else if (UNICODE_IS_NONCHAR(uv)) {
812 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
813 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
814 {
815 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
816 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
817 }
818 if (flags & UTF8_DISALLOW_NONCHAR) {
819 goto disallowed;
820 }
821 }
949cf498 822
eb83ed87
KW
823 if (sv) {
824 outlier_ret = uv;
825 goto do_warn;
826 }
827
949cf498
KW
828 /* Here, this is not considered a malformed character, so drop through
829 * to return it */
a0ed51b3 830 }
ba210ebe 831
a0ed51b3 832 return uv;
ba210ebe 833
eb83ed87
KW
834 /* There are three cases which get to beyond this point. In all 3 cases:
835 * <sv> if not null points to a string to print as a warning.
836 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
837 * set.
838 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
839 * This is done by initializing it to 0, and changing it only
840 * for case 1).
841 * The 3 cases are:
842 * 1) The input is valid but problematic, and to be warned about. The
843 * return value is the resultant code point; <*retlen> is set to
844 * <curlen>, the number of bytes that comprise the code point.
845 * <pack_warn> contains the result of packWARN() for the warning
846 * types. The entry point for this case is the label <do_warn>;
847 * 2) The input is a valid code point but disallowed by the parameters to
848 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
849 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
850 * comprise the code point. <pack_warn> contains the result of
851 * packWARN() for the warning types. The entry point for this case is
852 * the label <disallowed>.
853 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
854 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
855 * bytes that comprise the malformation. All such malformations are
856 * assumed to be warning type <utf8>. The entry point for this case
857 * is the label <malformed>.
858 */
949cf498 859
ba210ebe
JH
860malformed:
861
eb83ed87
KW
862 if (sv && ckWARN_d(WARN_UTF8)) {
863 pack_warn = packWARN(WARN_UTF8);
864 }
865
866disallowed:
867
fcc8fcf6 868 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 869 if (retlen)
10edeb5d 870 *retlen = ((STRLEN) -1);
ba210ebe
JH
871 return 0;
872 }
873
eb83ed87 874do_warn:
5b311467 875
eb83ed87
KW
876 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
877 if warnings are to be raised. */
f555bc63 878 const char * const string = SvPVX_const(sv);
a0dbb045 879
f555bc63
KW
880 if (PL_op)
881 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
882 else
883 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
884 }
885
eb83ed87
KW
886 if (retlen) {
887 *retlen = curlen;
888 }
ba210ebe 889
eb83ed87 890 return outlier_ret;
a0ed51b3
LW
891}
892
8e84507e 893/*
ec5f19d0
KW
894=for apidoc utf8_to_uvchr_buf
895
896Returns the native code point of the first character in the string C<s> which
897is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 898C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 899
524080c4
KW
900If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
901enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
902NULL) to -1. If those warnings are off, the computed value if well-defined (or
903the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
904is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
905next possible position in C<s> that could begin a non-malformed character.
906See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
907
908=cut
909*/
910
911
912UV
913Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
914{
915 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
916
917 assert(s < send);
918
919 return utf8n_to_uvchr(s, send - s, retlen,
920 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
921}
922
27d6c58a 923/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 924 * there are no malformations in the input UTF-8 string C<s>. surrogates,
6dd9dce9
KW
925 * non-character code points, and non-Unicode code points are allowed. A macro
926 * in utf8.h is used to normally avoid this function wrapper */
27d6c58a
KW
927
928UV
929Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
930{
3986bb7c
KW
931 const UV uv = valid_utf8_to_uvuni(s, retlen);
932
27d6c58a
KW
933 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
934
3986bb7c 935 return UNI_TO_NATIVE(uv);
27d6c58a
KW
936}
937
ec5f19d0 938/*
87cea99e 939=for apidoc utf8_to_uvchr
9041c2e3 940
977c1d31
KW
941DEPRECATED!
942
6ee84de2 943Returns the native code point of the first character in the string C<s>
1e54db1a 944which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
945length, in bytes, of that character.
946
4b88fb76 947Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
948malformed input could cause reading beyond the end of the input buffer, which
949is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 950
524080c4
KW
951If C<s> points to one of the detected malformations, and UTF8 warnings are
952enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
953NULL) to -1. If those warnings are off, the computed value if well-defined (or
954the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
955is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
956next possible position in C<s> that could begin a non-malformed character.
957See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
958
959=cut
960*/
961
962UV
7fc63493 963Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 964{
7918f24d
NC
965 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
966
2ff6c191 967 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
968}
969
970/*
ec5f19d0
KW
971=for apidoc utf8_to_uvuni_buf
972
973Returns the Unicode code point of the first character in the string C<s> which
974is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
975C<retlen> will be set to the length, in bytes, of that character.
976
977This function should only be used when the returned UV is considered
978an index into the Unicode semantic tables (e.g. swashes).
979
524080c4
KW
980If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
981enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
982NULL) to -1. If those warnings are off, the computed value if well-defined (or
983the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
984is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
985next possible position in C<s> that could begin a non-malformed character.
986See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
987
988=cut
989*/
990
991UV
992Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
993{
994 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
995
996 assert(send > s);
997
998 /* Call the low level routine asking for checks */
999 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
1000 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1001}
1002
27d6c58a 1003/* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1004 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1005 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1006
1007UV
1008Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1009{
3986bb7c
KW
1010 UV expectlen = UTF8SKIP(s);
1011 const U8* send = s + expectlen;
1012 UV uv = NATIVE_TO_UTF(*s);
1013
27d6c58a
KW
1014 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1015
3986bb7c
KW
1016 if (retlen) {
1017 *retlen = expectlen;
1018 }
1019
1020 /* An invariant is trivially returned */
1021 if (expectlen == 1) {
1022 return uv;
1023 }
1024
1025 /* Remove the leading bits that indicate the number of bytes, leaving just
1026 * the bits that are part of the value */
1027 uv &= UTF_START_MASK(expectlen);
1028
1029 /* Now, loop through the remaining bytes, accumulating each into the
08bc774e
KW
1030 * working total as we go. (I khw tried unrolling the loop for up to 4
1031 * bytes, but there was no performance improvement) */
3986bb7c
KW
1032 for (++s; s < send; s++) {
1033 uv = UTF8_ACCUMULATE(uv, *s);
1034 }
1035
1036 return uv;
27d6c58a
KW
1037}
1038
ec5f19d0 1039/*
87cea99e 1040=for apidoc utf8_to_uvuni
9041c2e3 1041
977c1d31
KW
1042DEPRECATED!
1043
9041c2e3 1044Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1045which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1046length, in bytes, of that character.
1047
39e518fd
KW
1048This function should only be used when the returned UV is considered
1049an index into the Unicode semantic tables (e.g. swashes).
1050
4b88fb76 1051Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1052malformed input could cause reading beyond the end of the input buffer, which
1053is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
9041c2e3 1054
524080c4
KW
1055If C<s> points to one of the detected malformations, and UTF8 warnings are
1056enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1057NULL) to -1. If those warnings are off, the computed value if well-defined (or
1058the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1059is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1060next possible position in C<s> that could begin a non-malformed character.
1061See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1062
1063=cut
1064*/
1065
1066UV
7fc63493 1067Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1068{
7918f24d
NC
1069 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1070
4b88fb76 1071 return valid_utf8_to_uvuni(s, retlen);
8e84507e
NIS
1072}
1073
b76347f2 1074/*
87cea99e 1075=for apidoc utf8_length
b76347f2
JH
1076
1077Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1078Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1079up past C<e>, croaks.
b76347f2
JH
1080
1081=cut
1082*/
1083
1084STRLEN
35a4481c 1085Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1086{
97aff369 1087 dVAR;
b76347f2
JH
1088 STRLEN len = 0;
1089
7918f24d
NC
1090 PERL_ARGS_ASSERT_UTF8_LENGTH;
1091
8850bf83
JH
1092 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1093 * the bitops (especially ~) can create illegal UTF-8.
1094 * In other words: in Perl UTF-8 is not just for Unicode. */
1095
a3b680e6
AL
1096 if (e < s)
1097 goto warn_and_return;
b76347f2 1098 while (s < e) {
8e91ec7f
AV
1099 if (!UTF8_IS_INVARIANT(*s))
1100 s += UTF8SKIP(s);
1101 else
1102 s++;
1103 len++;
1104 }
1105
1106 if (e != s) {
1107 len--;
1108 warn_and_return:
9b387841
NC
1109 if (PL_op)
1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1111 "%s in %s", unees, OP_DESC(PL_op));
1112 else
61a12c31 1113 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1114 }
1115
1116 return len;
1117}
1118
b06226ff 1119/*
87cea99e 1120=for apidoc utf8_distance
b06226ff 1121
1e54db1a 1122Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1123and C<b>.
1124
1125WARNING: use only if you *know* that the pointers point inside the
1126same UTF-8 buffer.
1127
37607a96
PK
1128=cut
1129*/
a0ed51b3 1130
02eb7b47 1131IV
35a4481c 1132Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1133{
7918f24d
NC
1134 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1135
bf1665bc 1136 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1137}
1138
b06226ff 1139/*
87cea99e 1140=for apidoc utf8_hop
b06226ff 1141
8850bf83
JH
1142Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1143forward or backward.
b06226ff
JH
1144
1145WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1146the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1147on the first byte of character or just after the last byte of a character.
b06226ff 1148
37607a96
PK
1149=cut
1150*/
a0ed51b3
LW
1151
1152U8 *
4373e329 1153Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1154{
7918f24d
NC
1155 PERL_ARGS_ASSERT_UTF8_HOP;
1156
96a5add6 1157 PERL_UNUSED_CONTEXT;
8850bf83
JH
1158 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1159 * the bitops (especially ~) can create illegal UTF-8.
1160 * In other words: in Perl UTF-8 is not just for Unicode. */
1161
a0ed51b3
LW
1162 if (off >= 0) {
1163 while (off--)
1164 s += UTF8SKIP(s);
1165 }
1166 else {
1167 while (off++) {
1168 s--;
8850bf83
JH
1169 while (UTF8_IS_CONTINUATION(*s))
1170 s--;
a0ed51b3
LW
1171 }
1172 }
4373e329 1173 return (U8 *)s;
a0ed51b3
LW
1174}
1175
6940069f 1176/*
fed3ba5d
NC
1177=for apidoc bytes_cmp_utf8
1178
a1433954
KW
1179Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1180sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1181equal, -1 or -2 if the first string is less than the second string, +1 or +2
1182if the first string is greater than the second string.
1183
1184-1 or +1 is returned if the shorter string was identical to the start of the
1185longer string. -2 or +2 is returned if the was a difference between characters
1186within the strings.
1187
1188=cut
1189*/
1190
1191int
1192Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1193{
1194 const U8 *const bend = b + blen;
1195 const U8 *const uend = u + ulen;
1196
1197 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1198
1199 PERL_UNUSED_CONTEXT;
1200
1201 while (b < bend && u < uend) {
1202 U8 c = *u++;
1203 if (!UTF8_IS_INVARIANT(c)) {
1204 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1205 if (u < uend) {
1206 U8 c1 = *u++;
1207 if (UTF8_IS_CONTINUATION(c1)) {
356979f4 1208 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
fed3ba5d
NC
1209 } else {
1210 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1211 "Malformed UTF-8 character "
1212 "(unexpected non-continuation byte 0x%02x"
1213 ", immediately after start byte 0x%02x)"
1214 /* Dear diag.t, it's in the pod. */
1215 "%s%s", c1, c,
1216 PL_op ? " in " : "",
1217 PL_op ? OP_DESC(PL_op) : "");
1218 return -2;
1219 }
1220 } else {
1221 if (PL_op)
1222 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1223 "%s in %s", unees, OP_DESC(PL_op));
1224 else
61a12c31 1225 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1226 return -2; /* Really want to return undef :-) */
1227 }
1228 } else {
1229 return -2;
1230 }
1231 }
1232 if (*b != c) {
1233 return *b < c ? -2 : +2;
1234 }
1235 ++b;
1236 }
1237
1238 if (b == bend && u == uend)
1239 return 0;
1240
1241 return b < bend ? +1 : -1;
1242}
1243
1244/*
87cea99e 1245=for apidoc utf8_to_bytes
6940069f 1246
2bbc8d55 1247Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1248Unlike L</bytes_to_utf8>, this over-writes the original string, and
1249updates C<len> to contain the new length.
67e989fb 1250Returns zero on failure, setting C<len> to -1.
6940069f 1251
a1433954 1252If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1253
6940069f
GS
1254=cut
1255*/
1256
1257U8 *
37607a96 1258Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1259{
d4c19fe8
AL
1260 U8 * const save = s;
1261 U8 * const send = s + *len;
6940069f 1262 U8 *d;
246fae53 1263
7918f24d
NC
1264 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1265
1e54db1a 1266 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1267 while (s < send) {
dcad2880
JH
1268 U8 c = *s++;
1269
1d72bdf6
NIS
1270 if (!UTF8_IS_INVARIANT(c) &&
1271 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1272 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
10edeb5d 1273 *len = ((STRLEN) -1);
dcad2880
JH
1274 return 0;
1275 }
246fae53 1276 }
dcad2880
JH
1277
1278 d = s = save;
6940069f 1279 while (s < send) {
ed646e6e 1280 STRLEN ulen;
4b88fb76 1281 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
ed646e6e 1282 s += ulen;
6940069f
GS
1283 }
1284 *d = '\0';
246fae53 1285 *len = d - save;
6940069f
GS
1286 return save;
1287}
1288
1289/*
87cea99e 1290=for apidoc bytes_from_utf8
f9a63242 1291
2bbc8d55 1292Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1293Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1294the newly-created string, and updates C<len> to contain the new
1295length. Returns the original string if no conversion occurs, C<len>
1296is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
12970 if C<s> is converted or consisted entirely of characters that are invariant
1298in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1299
37607a96
PK
1300=cut
1301*/
f9a63242
JH
1302
1303U8 *
e1ec3a88 1304Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1305{
f9a63242 1306 U8 *d;
e1ec3a88
AL
1307 const U8 *start = s;
1308 const U8 *send;
f9a63242
JH
1309 I32 count = 0;
1310
7918f24d
NC
1311 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1312
96a5add6 1313 PERL_UNUSED_CONTEXT;
f9a63242 1314 if (!*is_utf8)
73d840c0 1315 return (U8 *)start;
f9a63242 1316
1e54db1a 1317 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1318 for (send = s + *len; s < send;) {
e1ec3a88 1319 U8 c = *s++;
1d72bdf6 1320 if (!UTF8_IS_INVARIANT(c)) {
db42d148
NIS
1321 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1322 (c = *s++) && UTF8_IS_CONTINUATION(c))
1323 count++;
1324 else
73d840c0 1325 return (U8 *)start;
db42d148 1326 }
f9a63242
JH
1327 }
1328
35da51f7 1329 *is_utf8 = FALSE;
f9a63242 1330
212542aa 1331 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1332 s = start; start = d;
f9a63242
JH
1333 while (s < send) {
1334 U8 c = *s++;
c4d5f83a
NIS
1335 if (!UTF8_IS_INVARIANT(c)) {
1336 /* Then it is two-byte encoded */
356979f4 1337 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
c4d5f83a
NIS
1338 }
1339 *d++ = c;
f9a63242
JH
1340 }
1341 *d = '\0';
1342 *len = d - start;
73d840c0 1343 return (U8 *)start;
f9a63242
JH
1344}
1345
1346/*
87cea99e 1347=for apidoc bytes_to_utf8
6940069f 1348
ff97e5cf
KW
1349Converts a string C<s> of length C<len> bytes from the native encoding into
1350UTF-8.
6662521e 1351Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1352reflect the new length in bytes.
6940069f 1353
2bbc8d55
SP
1354A NUL character will be written after the end of the string.
1355
1356If you want to convert to UTF-8 from encodings other than
1357the native (Latin1 or EBCDIC),
a1433954 1358see L</sv_recode_to_utf8>().
c9ada85f 1359
497711e7 1360=cut
6940069f
GS
1361*/
1362
c682ebef
FC
1363/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1364 likewise need duplication. */
1365
6940069f 1366U8*
35a4481c 1367Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1368{
35a4481c 1369 const U8 * const send = s + (*len);
6940069f
GS
1370 U8 *d;
1371 U8 *dst;
7918f24d
NC
1372
1373 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1374 PERL_UNUSED_CONTEXT;
6940069f 1375
212542aa 1376 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1377 dst = d;
1378
1379 while (s < send) {
35a4481c 1380 const UV uv = NATIVE_TO_ASCII(*s++);
c4d5f83a 1381 if (UNI_IS_INVARIANT(uv))
eb160463 1382 *d++ = (U8)UTF_TO_NATIVE(uv);
6940069f 1383 else {
eb160463
GS
1384 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1385 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
6940069f
GS
1386 }
1387 }
1388 *d = '\0';
6662521e 1389 *len = d-dst;
6940069f
GS
1390 return dst;
1391}
1392
a0ed51b3 1393/*
dea0fc0b 1394 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1395 *
1396 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1397 * We optimize for native, for obvious reasons. */
1398
1399U8*
dea0fc0b 1400Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1401{
dea0fc0b
JH
1402 U8* pend;
1403 U8* dstart = d;
1404
7918f24d
NC
1405 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1406
dea0fc0b 1407 if (bytelen & 1)
f5992bc4 1408 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1409
1410 pend = p + bytelen;
1411
a0ed51b3 1412 while (p < pend) {
dea0fc0b
JH
1413 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1414 p += 2;
a0ed51b3 1415 if (uv < 0x80) {
e294cc5d
JH
1416#ifdef EBCDIC
1417 *d++ = UNI_TO_NATIVE(uv);
1418#else
eb160463 1419 *d++ = (U8)uv;
e294cc5d 1420#endif
a0ed51b3
LW
1421 continue;
1422 }
1423 if (uv < 0x800) {
eb160463
GS
1424 *d++ = (U8)(( uv >> 6) | 0xc0);
1425 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1426 continue;
1427 }
52b9aa85 1428 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
01ea242b 1429 if (p >= pend) {
dea0fc0b 1430 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1431 } else {
1432 UV low = (p[0] << 8) + p[1];
1433 p += 2;
52b9aa85 1434 if (low < 0xdc00 || low > 0xdfff)
01ea242b
NC
1435 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1436 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1437 }
dbde1951
NC
1438 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1439 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3
LW
1440 }
1441 if (uv < 0x10000) {
eb160463
GS
1442 *d++ = (U8)(( uv >> 12) | 0xe0);
1443 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1444 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1445 continue;
1446 }
1447 else {
eb160463
GS
1448 *d++ = (U8)(( uv >> 18) | 0xf0);
1449 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1450 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1451 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1452 continue;
1453 }
1454 }
dea0fc0b 1455 *newlen = d - dstart;
a0ed51b3
LW
1456 return d;
1457}
1458
1459/* Note: this one is slightly destructive of the source. */
1460
1461U8*
dea0fc0b 1462Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1463{
1464 U8* s = (U8*)p;
d4c19fe8 1465 U8* const send = s + bytelen;
7918f24d
NC
1466
1467 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1468
e0ea5e2d
NC
1469 if (bytelen & 1)
1470 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1471 (UV)bytelen);
1472
a0ed51b3 1473 while (s < send) {
d4c19fe8 1474 const U8 tmp = s[0];
a0ed51b3
LW
1475 s[0] = s[1];
1476 s[1] = tmp;
1477 s += 2;
1478 }
dea0fc0b 1479 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1480}
1481
c3fd2246
KW
1482/* for now these are all defined (inefficiently) in terms of the utf8 versions.
1483 * Note that the macros in handy.h that call these short-circuit calling them
1484 * for Latin-1 range inputs */
a0ed51b3
LW
1485
1486bool
84afefe6 1487Perl_is_uni_alnum(pTHX_ UV c)
a0ed51b3 1488{
89ebb4a3 1489 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1490 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1491 return is_utf8_alnum(tmpbuf);
1492}
1493
1494bool
84afefe6 1495Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1496{
89ebb4a3 1497 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1498 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1499 return is_utf8_idfirst(tmpbuf);
1500}
1501
1502bool
84afefe6 1503Perl_is_uni_alpha(pTHX_ UV c)
a0ed51b3 1504{
89ebb4a3 1505 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1506 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1507 return is_utf8_alpha(tmpbuf);
1508}
1509
1510bool
84afefe6 1511Perl_is_uni_ascii(pTHX_ UV c)
4d61ec05 1512{
bc39fe24 1513 return isASCII(c);
4d61ec05
GS
1514}
1515
1516bool
bdd8600f
KW
1517Perl_is_uni_blank(pTHX_ UV c)
1518{
1519 U8 tmpbuf[UTF8_MAXBYTES+1];
1520 uvchr_to_utf8(tmpbuf, c);
1521 return is_utf8_blank(tmpbuf);
1522}
1523
1524bool
84afefe6 1525Perl_is_uni_space(pTHX_ UV c)
a0ed51b3 1526{
89ebb4a3 1527 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1528 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1529 return is_utf8_space(tmpbuf);
1530}
1531
1532bool
84afefe6 1533Perl_is_uni_digit(pTHX_ UV c)
a0ed51b3 1534{
89ebb4a3 1535 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1536 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1537 return is_utf8_digit(tmpbuf);
1538}
1539
1540bool
84afefe6 1541Perl_is_uni_upper(pTHX_ UV c)
a0ed51b3 1542{
89ebb4a3 1543 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1544 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1545 return is_utf8_upper(tmpbuf);
1546}
1547
1548bool
84afefe6 1549Perl_is_uni_lower(pTHX_ UV c)
a0ed51b3 1550{
89ebb4a3 1551 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1552 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1553 return is_utf8_lower(tmpbuf);
1554}
1555
1556bool
84afefe6 1557Perl_is_uni_cntrl(pTHX_ UV c)
b8c5462f 1558{
7b952154 1559 return isCNTRL_L1(c);
b8c5462f
JH
1560}
1561
1562bool
84afefe6 1563Perl_is_uni_graph(pTHX_ UV c)
b8c5462f 1564{
89ebb4a3 1565 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1566 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1567 return is_utf8_graph(tmpbuf);
1568}
1569
1570bool
84afefe6 1571Perl_is_uni_print(pTHX_ UV c)
a0ed51b3 1572{
89ebb4a3 1573 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1574 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1575 return is_utf8_print(tmpbuf);
1576}
1577
b8c5462f 1578bool
84afefe6 1579Perl_is_uni_punct(pTHX_ UV c)
b8c5462f 1580{
89ebb4a3 1581 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1582 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1583 return is_utf8_punct(tmpbuf);
1584}
1585
4d61ec05 1586bool
84afefe6 1587Perl_is_uni_xdigit(pTHX_ UV c)
4d61ec05 1588{
89ebb4a3 1589 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
230880c1 1590 uvchr_to_utf8(tmpbuf, c);
4d61ec05
GS
1591 return is_utf8_xdigit(tmpbuf);
1592}
1593
3a4c58c9
KW
1594UV
1595Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1596{
1597 /* We have the latin1-range values compiled into the core, so just use
1598 * those, converting the result to utf8. The only difference between upper
1599 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1600 * either "SS" or "Ss". Which one to use is passed into the routine in
1601 * 'S_or_s' to avoid a test */
1602
1603 UV converted = toUPPER_LATIN1_MOD(c);
1604
1605 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1606
1607 assert(S_or_s == 'S' || S_or_s == 's');
1608
1609 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1610 characters in this range */
1611 *p = (U8) converted;
1612 *lenp = 1;
1613 return converted;
1614 }
1615
1616 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1617 * which it maps to one of them, so as to only have to have one check for
1618 * it in the main case */
1619 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1620 switch (c) {
1621 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1622 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1623 break;
1624 case MICRO_SIGN:
1625 converted = GREEK_CAPITAL_LETTER_MU;
1626 break;
1627 case LATIN_SMALL_LETTER_SHARP_S:
1628 *(p)++ = 'S';
1629 *p = S_or_s;
1630 *lenp = 2;
1631 return 'S';
1632 default:
1633 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1634 assert(0); /* NOTREACHED */
3a4c58c9
KW
1635 }
1636 }
1637
1638 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1639 *p = UTF8_TWO_BYTE_LO(converted);
1640 *lenp = 2;
1641
1642 return converted;
1643}
1644
50bda2c3
KW
1645/* Call the function to convert a UTF-8 encoded character to the specified case.
1646 * Note that there may be more than one character in the result.
1647 * INP is a pointer to the first byte of the input character
1648 * OUTP will be set to the first byte of the string of changed characters. It
1649 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1650 * LENP will be set to the length in bytes of the string of changed characters
1651 *
1652 * The functions return the ordinal of the first character in the string of OUTP */
f90a9a02
KW
1653#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1654#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1655#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
50bda2c3
KW
1656
1657/* This additionally has the input parameter SPECIALS, which if non-zero will
1658 * cause this to use the SPECIALS hash for folding (meaning get full case
1659 * folding); otherwise, when zero, this implies a simple case fold */
f90a9a02 1660#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 1661
84afefe6
JH
1662UV
1663Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1664{
3a4c58c9
KW
1665 dVAR;
1666
a1433954
KW
1667 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1668 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1669 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1670 * the changed version may be longer than the original character.
1671 *
1672 * The ordinal of the first character of the changed version is returned
1673 * (but note, as explained above, that there may be more.) */
1674
7918f24d
NC
1675 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1676
3a4c58c9
KW
1677 if (c < 256) {
1678 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1679 }
1680
0ebc6274 1681 uvchr_to_utf8(p, c);
3a4c58c9 1682 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1683}
1684
84afefe6
JH
1685UV
1686Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1687{
3a4c58c9
KW
1688 dVAR;
1689
7918f24d
NC
1690 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1691
3a4c58c9
KW
1692 if (c < 256) {
1693 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1694 }
1695
0ebc6274 1696 uvchr_to_utf8(p, c);
3a4c58c9 1697 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1698}
1699
afc16117
KW
1700STATIC U8
1701S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1702{
1703 /* We have the latin1-range values compiled into the core, so just use
1704 * those, converting the result to utf8. Since the result is always just
a1433954 1705 * one character, we allow <p> to be NULL */
afc16117
KW
1706
1707 U8 converted = toLOWER_LATIN1(c);
1708
1709 if (p != NULL) {
1710 if (UNI_IS_INVARIANT(converted)) {
1711 *p = converted;
1712 *lenp = 1;
1713 }
1714 else {
1715 *p = UTF8_TWO_BYTE_HI(converted);
1716 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1717 *lenp = 2;
1718 }
1719 }
1720 return converted;
1721}
1722
84afefe6
JH
1723UV
1724Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1725{
968c5e6a
KW
1726 dVAR;
1727
7918f24d
NC
1728 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1729
afc16117
KW
1730 if (c < 256) {
1731 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1732 }
1733
afc16117 1734 uvchr_to_utf8(p, c);
968c5e6a 1735 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1736}
1737
84afefe6 1738UV
f673fad4 1739Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
a1dde8de 1740{
a1433954 1741 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
f673fad4
KW
1742 * folding */
1743
a1dde8de
KW
1744 UV converted;
1745
1746 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1747
1748 if (c == MICRO_SIGN) {
1749 converted = GREEK_SMALL_LETTER_MU;
1750 }
1751 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1752 *(p)++ = 's';
1753 *p = 's';
1754 *lenp = 2;
1755 return 's';
1756 }
1757 else { /* In this range the fold of all other characters is their lower
1758 case */
1759 converted = toLOWER_LATIN1(c);
1760 }
1761
1762 if (UNI_IS_INVARIANT(converted)) {
1763 *p = (U8) converted;
1764 *lenp = 1;
1765 }
1766 else {
1767 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1768 *p = UTF8_TWO_BYTE_LO(converted);
1769 *lenp = 2;
1770 }
1771
1772 return converted;
1773}
1774
1775UV
a0270393 1776Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
84afefe6 1777{
4b593389 1778
a0270393
KW
1779 /* Not currently externally documented, and subject to change
1780 * <flags> bits meanings:
1781 * FOLD_FLAGS_FULL iff full folding is to be used;
1782 * FOLD_FLAGS_LOCALE iff in locale
1783 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1784 */
4b593389 1785
36bb2ab6 1786 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1787
a1dde8de 1788 if (c < 256) {
a0270393
KW
1789 UV result = _to_fold_latin1((U8) c, p, lenp,
1790 cBOOL(((flags & FOLD_FLAGS_FULL)
1791 /* If ASCII-safe, don't allow full folding,
1792 * as that could include SHARP S => ss;
1793 * otherwise there is no crossing of
1794 * ascii/non-ascii in the latin1 range */
1795 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1796 /* It is illegal for the fold to cross the 255/256 boundary under
1797 * locale; in this case return the original */
1798 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1799 ? c
1800 : result;
a1dde8de
KW
1801 }
1802
a0270393
KW
1803 /* If no special needs, just use the macro */
1804 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1805 uvchr_to_utf8(p, c);
1806 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1807 }
1808 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1809 the special flags. */
1810 U8 utf8_c[UTF8_MAXBYTES + 1];
1811 uvchr_to_utf8(utf8_c, c);
1812 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1813 }
84afefe6
JH
1814}
1815
ea317ccb
KW
1816/* for now these all assume no locale info available for Unicode > 255; and
1817 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1818 * called instead, so that these don't get called for < 255 */
a0ed51b3
LW
1819
1820bool
84afefe6 1821Perl_is_uni_alnum_lc(pTHX_ UV c)
a0ed51b3
LW
1822{
1823 return is_uni_alnum(c); /* XXX no locale support yet */
1824}
1825
1826bool
84afefe6 1827Perl_is_uni_idfirst_lc(pTHX_ UV c)
a0ed51b3
LW
1828{
1829 return is_uni_idfirst(c); /* XXX no locale support yet */
1830}
1831
1832bool
84afefe6 1833Perl_is_uni_alpha_lc(pTHX_ UV c)
a0ed51b3
LW
1834{
1835 return is_uni_alpha(c); /* XXX no locale support yet */
1836}
1837
1838bool
84afefe6 1839Perl_is_uni_ascii_lc(pTHX_ UV c)
4d61ec05
GS
1840{
1841 return is_uni_ascii(c); /* XXX no locale support yet */
1842}
1843
1844bool
bdd8600f
KW
1845Perl_is_uni_blank_lc(pTHX_ UV c)
1846{
1847 return is_uni_blank(c); /* XXX no locale support yet */
1848}
1849
1850bool
84afefe6 1851Perl_is_uni_space_lc(pTHX_ UV c)
a0ed51b3
LW
1852{
1853 return is_uni_space(c); /* XXX no locale support yet */
1854}
1855
1856bool
84afefe6 1857Perl_is_uni_digit_lc(pTHX_ UV c)
a0ed51b3
LW
1858{
1859 return is_uni_digit(c); /* XXX no locale support yet */
1860}
1861
1862bool
84afefe6 1863Perl_is_uni_upper_lc(pTHX_ UV c)
a0ed51b3
LW
1864{
1865 return is_uni_upper(c); /* XXX no locale support yet */
1866}
1867
1868bool
84afefe6 1869Perl_is_uni_lower_lc(pTHX_ UV c)
a0ed51b3
LW
1870{
1871 return is_uni_lower(c); /* XXX no locale support yet */
1872}
1873
1874bool
84afefe6 1875Perl_is_uni_cntrl_lc(pTHX_ UV c)
b8c5462f
JH
1876{
1877 return is_uni_cntrl(c); /* XXX no locale support yet */
1878}
1879
1880bool
84afefe6 1881Perl_is_uni_graph_lc(pTHX_ UV c)
b8c5462f
JH
1882{
1883 return is_uni_graph(c); /* XXX no locale support yet */
1884}
1885
1886bool
84afefe6 1887Perl_is_uni_print_lc(pTHX_ UV c)
a0ed51b3
LW
1888{
1889 return is_uni_print(c); /* XXX no locale support yet */
1890}
1891
b8c5462f 1892bool
84afefe6 1893Perl_is_uni_punct_lc(pTHX_ UV c)
b8c5462f
JH
1894{
1895 return is_uni_punct(c); /* XXX no locale support yet */
1896}
1897
4d61ec05 1898bool
84afefe6 1899Perl_is_uni_xdigit_lc(pTHX_ UV c)
4d61ec05
GS
1900{
1901 return is_uni_xdigit(c); /* XXX no locale support yet */
1902}
1903
b7ac61fa
JH
1904U32
1905Perl_to_uni_upper_lc(pTHX_ U32 c)
1906{
ee099d14
JH
1907 /* XXX returns only the first character -- do not use XXX */
1908 /* XXX no locale support yet */
1909 STRLEN len;
89ebb4a3 1910 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1911 return (U32)to_uni_upper(c, tmpbuf, &len);
b7ac61fa
JH
1912}
1913
1914U32
1915Perl_to_uni_title_lc(pTHX_ U32 c)
1916{
ee099d14
JH
1917 /* XXX returns only the first character XXX -- do not use XXX */
1918 /* XXX no locale support yet */
1919 STRLEN len;
89ebb4a3 1920 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1921 return (U32)to_uni_title(c, tmpbuf, &len);
b7ac61fa
JH
1922}
1923
1924U32
1925Perl_to_uni_lower_lc(pTHX_ U32 c)
1926{
ee099d14
JH
1927 /* XXX returns only the first character -- do not use XXX */
1928 /* XXX no locale support yet */
1929 STRLEN len;
89ebb4a3 1930 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1931 return (U32)to_uni_lower(c, tmpbuf, &len);
b7ac61fa
JH
1932}
1933
7452cf6a 1934static bool
5141f98e 1935S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
1936 const char *const swashname)
1937{
ea317ccb
KW
1938 /* returns a boolean giving whether or not the UTF8-encoded character that
1939 * starts at <p> is in the swash indicated by <swashname>. <swash>
1940 * contains a pointer to where the swash indicated by <swashname>
1941 * is to be stored; which this routine will do, so that future calls will
1942 * look at <*swash> and only generate a swash if it is not null
1943 *
1944 * Note that it is assumed that the buffer length of <p> is enough to
1945 * contain all the bytes that comprise the character. Thus, <*p> should
1946 * have been checked before this call for mal-formedness enough to assure
1947 * that. */
1948
97aff369 1949 dVAR;
7918f24d
NC
1950
1951 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1952
492a624f
KW
1953 /* The API should have included a length for the UTF-8 character in <p>,
1954 * but it doesn't. We therefor assume that p has been validated at least
1955 * as far as there being enough bytes available in it to accommodate the
1956 * character without reading beyond the end, and pass that number on to the
1957 * validating routine */
1958 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
bde6a22d
NC
1959 return FALSE;
1960 if (!*swash)
711a919c 1961 *swash = swash_init("utf8", swashname, &PL_sv_undef, 1, 0);
bde6a22d
NC
1962 return swash_fetch(*swash, p, TRUE) != 0;
1963}
1964
1965bool
7fc63493 1966Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 1967{
97aff369 1968 dVAR;
7918f24d
NC
1969
1970 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1971
671c33bf
NC
1972 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1973 * descendant of isalnum(3), in other words, it doesn't
1974 * contain the '_'. --jhi */
d4c19fe8 1975 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
a0ed51b3
LW
1976}
1977
1978bool
7fc63493 1979Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1980{
97aff369 1981 dVAR;
7918f24d
NC
1982
1983 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1984
82686b01
JH
1985 if (*p == '_')
1986 return TRUE;
bde6a22d 1987 /* is_utf8_idstart would be more logical. */
d4c19fe8 1988 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
82686b01
JH
1989}
1990
1991bool
c11ff943
KW
1992Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1993{
1994 dVAR;
1995
1996 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1997
1998 if (*p == '_')
1999 return TRUE;
2000 /* is_utf8_idstart would be more logical. */
2001 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2002}
2003
2004bool
b6912c02
KW
2005Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2006{
2007 dVAR;
2008
2009 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2010
2011 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2012}
2013
2014bool
7fc63493 2015Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2016{
97aff369 2017 dVAR;
7918f24d
NC
2018
2019 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2020
d4c19fe8 2021 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
2022}
2023
2024bool
c11ff943
KW
2025Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2026{
2027 dVAR;
2028
2029 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2030
c11ff943
KW
2031 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2032}
2033
2034bool
7fc63493 2035Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 2036{
97aff369 2037 dVAR;
7918f24d
NC
2038
2039 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2040
d4c19fe8 2041 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
a0ed51b3
LW
2042}
2043
2044bool
7fc63493 2045Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2046{
97aff369 2047 dVAR;
7918f24d
NC
2048
2049 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2050
bc39fe24
KW
2051 /* ASCII characters are the same whether in utf8 or not. So the macro
2052 * works on both utf8 and non-utf8 representations. */
2053 return isASCII(*p);
b8c5462f
JH
2054}
2055
2056bool
bdd8600f
KW
2057Perl_is_utf8_blank(pTHX_ const U8 *p)
2058{
2059 dVAR;
2060
2061 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2062
2063 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2064}
2065
2066bool
7fc63493 2067Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2068{
97aff369 2069 dVAR;
7918f24d
NC
2070
2071 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2072
a34094a9 2073 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
a0ed51b3
LW
2074}
2075
2076bool
d1eb3177
YO
2077Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2078{
2079 dVAR;
2080
2081 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2082
c4428693
KW
2083 /* Only true if is an ASCII space-like character, and ASCII is invariant
2084 * under utf8, so can just use the macro */
2085 return isSPACE_A(*p);
d1eb3177
YO
2086}
2087
2088bool
2089Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2090{
2091 dVAR;
2092
2093 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2094
c4428693
KW
2095 /* Only true if is an ASCII word character, and ASCII is invariant
2096 * under utf8, so can just use the macro */
2097 return isWORDCHAR_A(*p);
d1eb3177
YO
2098}
2099
2100bool
7fc63493 2101Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2102{
97aff369 2103 dVAR;
7918f24d
NC
2104
2105 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2106
d4c19fe8 2107 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
a0ed51b3
LW
2108}
2109
2110bool
d1eb3177
YO
2111Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2112{
2113 dVAR;
2114
2115 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2116
c4428693
KW
2117 /* Only true if is an ASCII digit character, and ASCII is invariant
2118 * under utf8, so can just use the macro */
2119 return isDIGIT_A(*p);
d1eb3177
YO
2120}
2121
2122bool
7fc63493 2123Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2124{
97aff369 2125 dVAR;
7918f24d
NC
2126
2127 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2128
d4c19fe8 2129 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
a0ed51b3
LW
2130}
2131
2132bool
7fc63493 2133Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2134{
97aff369 2135 dVAR;
7918f24d
NC
2136
2137 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2138
d4c19fe8 2139 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
a0ed51b3
LW
2140}
2141
2142bool
7fc63493 2143Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2144{
97aff369 2145 dVAR;
7918f24d
NC
2146
2147 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2148
7b952154
KW
2149 if (isASCII(*p)) {
2150 return isCNTRL_A(*p);
2151 }
2152
2153 /* All controls are in Latin1 */
2154 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2155 return 0;
2156 }
2157 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
b8c5462f
JH
2158}
2159
2160bool
7fc63493 2161Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2162{
97aff369 2163 dVAR;
7918f24d
NC
2164
2165 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2166
d4c19fe8 2167 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
b8c5462f
JH
2168}
2169
2170bool
7fc63493 2171Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2172{
97aff369 2173 dVAR;
7918f24d
NC
2174
2175 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2176
d4c19fe8 2177 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
a0ed51b3
LW
2178}
2179
2180bool
7fc63493 2181Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2182{
97aff369 2183 dVAR;
7918f24d
NC
2184
2185 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2186
d4c19fe8 2187 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
b8c5462f
JH
2188}
2189
2190bool
7fc63493 2191Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2192{
97aff369 2193 dVAR;
7918f24d
NC
2194
2195 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2196
d1eb3177 2197 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
b8c5462f
JH
2198}
2199
2200bool
7fc63493 2201Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2202{
97aff369 2203 dVAR;
7918f24d
NC
2204
2205 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2206
d4c19fe8 2207 return is_utf8_common(p, &PL_utf8_mark, "IsM");
a0ed51b3
LW
2208}
2209
37e2e78e
KW
2210bool
2211Perl_is_utf8_X_begin(pTHX_ const U8 *p)
2212{
2213 dVAR;
2214
2215 PERL_ARGS_ASSERT_IS_UTF8_X_BEGIN;
2216
2217 return is_utf8_common(p, &PL_utf8_X_begin, "_X_Begin");
2218}
2219
2220bool
2221Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2222{
2223 dVAR;
2224
2225 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2226
2227 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2228}
2229
2230bool
2231Perl_is_utf8_X_prepend(pTHX_ const U8 *p)
2232{
2233 dVAR;
2234
2235 PERL_ARGS_ASSERT_IS_UTF8_X_PREPEND;
2236
f957fa32 2237 return is_utf8_common(p, &PL_utf8_X_prepend, "_X_GCB_Prepend");
37e2e78e
KW
2238}
2239
2240bool
2241Perl_is_utf8_X_non_hangul(pTHX_ const U8 *p)
2242{
2243 dVAR;
2244
2245 PERL_ARGS_ASSERT_IS_UTF8_X_NON_HANGUL;
2246
f957fa32 2247 return is_utf8_common(p, &PL_utf8_X_non_hangul, "_X_HST_Not_Applicable");
37e2e78e
KW
2248}
2249
2250bool
2251Perl_is_utf8_X_L(pTHX_ const U8 *p)
2252{
2253 dVAR;
2254
2255 PERL_ARGS_ASSERT_IS_UTF8_X_L;
2256
f957fa32 2257 return is_utf8_common(p, &PL_utf8_X_L, "_X_GCB_L");
37e2e78e
KW
2258}
2259
41de4811
KW
2260/* These constants are for finding GCB=LV and GCB=LVT. These are for the
2261 * pre-composed Hangul syllables, which are all in a contiguous block and
2262 * arranged there in such a way so as to facilitate alorithmic determination of
2263 * their characteristics. As such, they don't need a swash, but can be
2264 * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
2265 * is a GCB=LV */
2266#define SBASE 0xAC00 /* Start of block */
2267#define SCount 11172 /* Length of block */
2268#define TCount 28
2269
218282df 2270#if 0 /* This routine is not currently used */
37e2e78e
KW
2271bool
2272Perl_is_utf8_X_LV(pTHX_ const U8 *p)
2273{
41de4811
KW
2274 /* Unlike most other similarly named routines here, this does not create a
2275 * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
2276
37e2e78e
KW
2277 dVAR;
2278
41de4811
KW
2279 UV cp = valid_utf8_to_uvchr(p, NULL);
2280
37e2e78e
KW
2281 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
2282
41de4811
KW
2283 /* The earliest Unicode releases did not have these precomposed Hangul
2284 * syllables. Set to point to undef in that case, so will return false on
2285 * every call */
2286 if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
2287 PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
2288 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
2289 SvREFCNT_dec(PL_utf8_X_LV);
2290 PL_utf8_X_LV = &PL_sv_undef;
2291 }
2292 }
2293
2294 return (PL_utf8_X_LV != &PL_sv_undef
2295 && cp >= SBASE && cp < SBASE + SCount
2296 && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
37e2e78e 2297}
218282df 2298#endif
37e2e78e
KW
2299
2300bool
2301Perl_is_utf8_X_LVT(pTHX_ const U8 *p)
2302{
41de4811
KW
2303 /* Unlike most other similarly named routines here, this does not create a
2304 * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
2305
37e2e78e
KW
2306 dVAR;
2307
41de4811
KW
2308 UV cp = valid_utf8_to_uvchr(p, NULL);
2309
37e2e78e
KW
2310 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
2311
41de4811
KW
2312 /* The earliest Unicode releases did not have these precomposed Hangul
2313 * syllables. Set to point to undef in that case, so will return false on
2314 * every call */
2315 if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
2316 PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
2317 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
2318 SvREFCNT_dec(PL_utf8_X_LVT);
2319 PL_utf8_X_LVT = &PL_sv_undef;
2320 }
2321 }
2322
2323 return (PL_utf8_X_LVT != &PL_sv_undef
2324 && cp >= SBASE && cp < SBASE + SCount
2325 && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
37e2e78e
KW
2326}
2327
2328bool
2329Perl_is_utf8_X_T(pTHX_ const U8 *p)
2330{
2331 dVAR;
2332
2333 PERL_ARGS_ASSERT_IS_UTF8_X_T;
2334
f957fa32 2335 return is_utf8_common(p, &PL_utf8_X_T, "_X_GCB_T");
37e2e78e
KW
2336}
2337
2338bool
2339Perl_is_utf8_X_V(pTHX_ const U8 *p)
2340{
2341 dVAR;
2342
2343 PERL_ARGS_ASSERT_IS_UTF8_X_V;
2344
f957fa32 2345 return is_utf8_common(p, &PL_utf8_X_V, "_X_GCB_V");
37e2e78e
KW
2346}
2347
2348bool
2349Perl_is_utf8_X_LV_LVT_V(pTHX_ const U8 *p)
2350{
2351 dVAR;
2352
2353 PERL_ARGS_ASSERT_IS_UTF8_X_LV_LVT_V;
2354
2355 return is_utf8_common(p, &PL_utf8_X_LV_LVT_V, "_X_LV_LVT_V");
2356}
2357
2e2b2571
KW
2358bool
2359Perl__is_utf8_quotemeta(pTHX_ const U8 *p)
2360{
2361 /* For exclusive use of pp_quotemeta() */
2362
2363 dVAR;
2364
2365 PERL_ARGS_ASSERT__IS_UTF8_QUOTEMETA;
2366
2367 return is_utf8_common(p, &PL_utf8_quotemeta, "_Perl_Quotemeta");
2368}
2369
6b5c0936 2370/*
87cea99e 2371=for apidoc to_utf8_case
6b5c0936 2372
a1433954
KW
2373The C<p> contains the pointer to the UTF-8 string encoding
2374the character that is being converted. This routine assumes that the character
2375at C<p> is well-formed.
6b5c0936 2376
a1433954
KW
2377The C<ustrp> is a pointer to the character buffer to put the
2378conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2379of the result.
2380
a1433954 2381The C<swashp> is a pointer to the swash to use.
6b5c0936 2382
a1433954
KW
2383Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2384and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2385but not always, a multicharacter mapping), is tried first.
6b5c0936 2386
a1433954 2387The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2388hash %utf8::ToSpecLower. The access to the hash is through
2389Perl_to_utf8_case().
6b5c0936 2390
a1433954 2391The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2392%utf8::ToLower.
2393
2394=cut */
6b5c0936 2395
2104c8d9 2396UV
9a957fbc
AL
2397Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2398 SV **swashp, const char *normal, const char *special)
a0ed51b3 2399{
97aff369 2400 dVAR;
89ebb4a3 2401 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
0134edef 2402 STRLEN len = 0;
4b88fb76 2403 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
1feea2c7
JH
2404 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2405 * are necessary in EBCDIC, they are redundant no-ops
2406 * in ASCII-ish platforms, and hopefully optimized away. */
f54cb97a 2407 const UV uv1 = NATIVE_TO_UNI(uv0);
7918f24d
NC
2408
2409 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2410
9ae3ac1a
KW
2411 /* Note that swash_fetch() doesn't output warnings for these because it
2412 * assumes we will */
8457b38f 2413 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2414 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2415 if (ckWARN_d(WARN_SURROGATE)) {
2416 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2417 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2418 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2419 }
9ae3ac1a
KW
2420 }
2421 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2422 if (ckWARN_d(WARN_NON_UNICODE)) {
2423 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2424 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2425 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2426 }
9ae3ac1a
KW
2427 }
2428
2429 /* Note that non-characters are perfectly legal, so no warning should
2430 * be given */
2431 }
2432
1feea2c7 2433 uvuni_to_utf8(tmpbuf, uv1);
0134edef
JH
2434
2435 if (!*swashp) /* load on-demand */
5ab9d2ef 2436 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2437
a6f87d8c 2438 if (special) {
0134edef 2439 /* It might be "special" (sometimes, but not always,
2a37f04d 2440 * a multicharacter mapping) */
6673a63c 2441 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2442 SV **svp;
2443
35da51f7 2444 if (hv &&
b08cf34e
JH
2445 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2446 (*svp)) {
cfd0369c 2447 const char *s;
47654450 2448
cfd0369c 2449 s = SvPV_const(*svp, len);
47654450
JH
2450 if (len == 1)
2451 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2a37f04d 2452 else {
2f9475ad
JH
2453#ifdef EBCDIC
2454 /* If we have EBCDIC we need to remap the characters
2455 * since any characters in the low 256 are Unicode
2456 * code points, not EBCDIC. */
7cda7a3d 2457 U8 *t = (U8*)s, *tend = t + len, *d;
2f9475ad
JH
2458
2459 d = tmpbuf;
b08cf34e 2460 if (SvUTF8(*svp)) {
2f9475ad
JH
2461 STRLEN tlen = 0;
2462
2463 while (t < tend) {
6bd1c396 2464 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2f9475ad
JH
2465 if (tlen > 0) {
2466 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2467 t += tlen;
2468 }
2469 else
2470 break;
2471 }
2472 }
2473 else {
36fec512
JH
2474 while (t < tend) {
2475 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2476 t++;
2477 }
2f9475ad
JH
2478 }
2479 len = d - tmpbuf;
2480 Copy(tmpbuf, ustrp, len, U8);
2481#else
d2dcd0fb 2482 Copy(s, ustrp, len, U8);
2f9475ad 2483#endif
29e98929 2484 }
983ffd37 2485 }
0134edef
JH
2486 }
2487
2488 if (!len && *swashp) {
2114036c 2489 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
d4c19fe8 2490
0134edef
JH
2491 if (uv2) {
2492 /* It was "normal" (a single character mapping). */
d4c19fe8 2493 const UV uv3 = UNI_TO_NATIVE(uv2);
e9101d72 2494 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2a37f04d
JH
2495 }
2496 }
1feea2c7 2497
cbe07460
KW
2498 if (len) {
2499 if (lenp) {
2500 *lenp = len;
2501 }
2502 return valid_utf8_to_uvchr(ustrp, 0);
2503 }
2504
2505 /* Here, there was no mapping defined, which means that the code point maps
2506 * to itself. Return the inputs */
bfdf22ec
KW
2507 len = UTF8SKIP(p);
2508 Copy(p, ustrp, len, U8);
0134edef 2509
2a37f04d
JH
2510 if (lenp)
2511 *lenp = len;
2512
cbe07460
KW
2513 return uv0;
2514
a0ed51b3
LW
2515}
2516
051a06d4
KW
2517STATIC UV
2518S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2519{
2520 /* This is called when changing the case of a utf8-encoded character above
2521 * the Latin1 range, and the operation is in locale. If the result
2522 * contains a character that crosses the 255/256 boundary, disallow the
2523 * change, and return the original code point. See L<perlfunc/lc> for why;
2524 *
a1433954
KW
2525 * p points to the original string whose case was changed; assumed
2526 * by this routine to be well-formed
051a06d4
KW
2527 * result the code point of the first character in the changed-case string
2528 * ustrp points to the changed-case string (<result> represents its first char)
2529 * lenp points to the length of <ustrp> */
2530
2531 UV original; /* To store the first code point of <p> */
2532
2533 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2534
2535 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2536
2537 /* We know immediately if the first character in the string crosses the
2538 * boundary, so can skip */
2539 if (result > 255) {
2540
2541 /* Look at every character in the result; if any cross the
2542 * boundary, the whole thing is disallowed */
2543 U8* s = ustrp + UTF8SKIP(ustrp);
2544 U8* e = ustrp + *lenp;
2545 while (s < e) {
2546 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2547 {
2548 goto bad_crossing;
2549 }
2550 s += UTF8SKIP(s);
2551 }
2552
2553 /* Here, no characters crossed, result is ok as-is */
2554 return result;
2555 }
2556
2557bad_crossing:
2558
2559 /* Failed, have to return the original */
4b88fb76 2560 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2561 Copy(p, ustrp, *lenp, char);
2562 return original;
2563}
2564
d3e79532 2565/*
87cea99e 2566=for apidoc to_utf8_upper
d3e79532 2567
a1433954
KW
2568Convert the UTF-8 encoded character at C<p> to its uppercase version and
2569store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
89ebb4a3
JH
2570that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2571the uppercase version may be longer than the original character.
d3e79532
JH
2572
2573The first character of the uppercased version is returned
2574(but note, as explained above, that there may be more.)
2575
a1433954
KW
2576The character at C<p> is assumed by this routine to be well-formed.
2577
d3e79532
JH
2578=cut */
2579
051a06d4
KW
2580/* Not currently externally documented, and subject to change:
2581 * <flags> is set iff locale semantics are to be used for code points < 256
2582 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2583 * were used in the calculation; otherwise unchanged. */
2584
2104c8d9 2585UV
051a06d4 2586Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2587{
97aff369 2588 dVAR;
7918f24d 2589
051a06d4
KW
2590 UV result;
2591
2592 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2593
3a4c58c9 2594 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2595 if (flags) {
2596 result = toUPPER_LC(*p);
2597 }
2598 else {
81c6c7ce 2599 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2600 }
3a4c58c9
KW
2601 }
2602 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2603 if (flags) {
2604 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2605 }
2606 else {
81c6c7ce
KW
2607 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2608 ustrp, lenp, 'S');
051a06d4
KW
2609 }
2610 }
2611 else { /* utf8, ord above 255 */
2612 result = CALL_UPPER_CASE(p, ustrp, lenp);
2613
2614 if (flags) {
2615 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2616 }
2617 return result;
2618 }
2619
2620 /* Here, used locale rules. Convert back to utf8 */
2621 if (UTF8_IS_INVARIANT(result)) {
2622 *ustrp = (U8) result;
2623 *lenp = 1;
2624 }
2625 else {
2626 *ustrp = UTF8_EIGHT_BIT_HI(result);
2627 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2628 *lenp = 2;
3a4c58c9
KW
2629 }
2630
051a06d4
KW
2631 if (tainted_ptr) {
2632 *tainted_ptr = TRUE;
2633 }
2634 return result;
983ffd37 2635}
a0ed51b3 2636
d3e79532 2637/*
87cea99e 2638=for apidoc to_utf8_title
d3e79532 2639
a1433954
KW
2640Convert the UTF-8 encoded character at C<p> to its titlecase version and
2641store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2642that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2643titlecase version may be longer than the original character.
d3e79532
JH
2644
2645The first character of the titlecased version is returned
2646(but note, as explained above, that there may be more.)
2647
a1433954
KW
2648The character at C<p> is assumed by this routine to be well-formed.
2649
d3e79532
JH
2650=cut */
2651
051a06d4
KW
2652/* Not currently externally documented, and subject to change:
2653 * <flags> is set iff locale semantics are to be used for code points < 256
2654 * Since titlecase is not defined in POSIX, uppercase is used instead
2655 * for these/
2656 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2657 * were used in the calculation; otherwise unchanged. */
2658
983ffd37 2659UV
051a06d4 2660Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2661{
97aff369 2662 dVAR;
7918f24d 2663
051a06d4
KW
2664 UV result;
2665
2666 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2667
3a4c58c9 2668 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2669 if (flags) {
2670 result = toUPPER_LC(*p);
2671 }
2672 else {
81c6c7ce 2673 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2674 }
3a4c58c9
KW
2675 }
2676 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2677 if (flags) {
2678 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2679 }
2680 else {
81c6c7ce
KW
2681 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2682 ustrp, lenp, 's');
051a06d4
KW
2683 }
2684 }
2685 else { /* utf8, ord above 255 */
2686 result = CALL_TITLE_CASE(p, ustrp, lenp);
2687
2688 if (flags) {
2689 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2690 }
2691 return result;
2692 }
2693
2694 /* Here, used locale rules. Convert back to utf8 */
2695 if (UTF8_IS_INVARIANT(result)) {
2696 *ustrp = (U8) result;
2697 *lenp = 1;
2698 }
2699 else {
2700 *ustrp = UTF8_EIGHT_BIT_HI(result);
2701 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2702 *lenp = 2;
3a4c58c9
KW
2703 }
2704
051a06d4
KW
2705 if (tainted_ptr) {
2706 *tainted_ptr = TRUE;
2707 }
2708 return result;
a0ed51b3
LW
2709}
2710
d3e79532 2711/*
87cea99e 2712=for apidoc to_utf8_lower
d3e79532 2713
a1433954
KW
2714Convert the UTF-8 encoded character at C<p> to its lowercase version and
2715store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2716that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2717lowercase version may be longer than the original character.
d3e79532
JH
2718
2719The first character of the lowercased version is returned
2720(but note, as explained above, that there may be more.)
2721
a1433954
KW
2722The character at C<p> is assumed by this routine to be well-formed.
2723
d3e79532
JH
2724=cut */
2725
051a06d4
KW
2726/* Not currently externally documented, and subject to change:
2727 * <flags> is set iff locale semantics are to be used for code points < 256
2728 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2729 * were used in the calculation; otherwise unchanged. */
2730
2104c8d9 2731UV
051a06d4 2732Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2733{
051a06d4
KW
2734 UV result;
2735
97aff369 2736 dVAR;
7918f24d 2737
051a06d4 2738 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2739
968c5e6a 2740 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2741 if (flags) {
2742 result = toLOWER_LC(*p);
2743 }
2744 else {
81c6c7ce 2745 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2746 }
968c5e6a
KW
2747 }
2748 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2749 if (flags) {
2750 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2751 }
2752 else {
81c6c7ce
KW
2753 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2754 ustrp, lenp);
051a06d4 2755 }
968c5e6a 2756 }
051a06d4
KW
2757 else { /* utf8, ord above 255 */
2758 result = CALL_LOWER_CASE(p, ustrp, lenp);
2759
2760 if (flags) {
2761 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2762 }
968c5e6a 2763
051a06d4
KW
2764 return result;
2765 }
2766
2767 /* Here, used locale rules. Convert back to utf8 */
2768 if (UTF8_IS_INVARIANT(result)) {
2769 *ustrp = (U8) result;
2770 *lenp = 1;
2771 }
2772 else {
2773 *ustrp = UTF8_EIGHT_BIT_HI(result);
2774 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2775 *lenp = 2;
2776 }
2777
2778 if (tainted_ptr) {
2779 *tainted_ptr = TRUE;
2780 }
2781 return result;
b4e400f9
JH
2782}
2783
d3e79532 2784/*
87cea99e 2785=for apidoc to_utf8_fold
d3e79532 2786
a1433954
KW
2787Convert the UTF-8 encoded character at C<p> to its foldcase version and
2788store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2789that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
d3e79532
JH
2790foldcase version may be longer than the original character (up to
2791three characters).
2792
2793The first character of the foldcased version is returned
2794(but note, as explained above, that there may be more.)
2795
a1433954
KW
2796The character at C<p> is assumed by this routine to be well-formed.
2797
d3e79532
JH
2798=cut */
2799
051a06d4
KW
2800/* Not currently externally documented, and subject to change,
2801 * in <flags>
2802 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2803 * points < 256. Since foldcase is not defined in
2804 * POSIX, lowercase is used instead
2805 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2806 * otherwise simple folds
a0270393
KW
2807 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2808 * prohibited
051a06d4
KW
2809 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2810 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2811
b4e400f9 2812UV
051a06d4 2813Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2814{
97aff369 2815 dVAR;
7918f24d 2816
051a06d4
KW
2817 UV result;
2818
36bb2ab6 2819 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2820
a0270393
KW
2821 /* These are mutually exclusive */
2822 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2823
50ba90ff
KW
2824 assert(p != ustrp); /* Otherwise overwrites */
2825
a1dde8de 2826 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2827 if (flags & FOLD_FLAGS_LOCALE) {
2828 result = toLOWER_LC(*p);
2829 }
2830 else {
81c6c7ce
KW
2831 return _to_fold_latin1(*p, ustrp, lenp,
2832 cBOOL(flags & FOLD_FLAGS_FULL));
051a06d4 2833 }
a1dde8de
KW
2834 }
2835 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2836 if (flags & FOLD_FLAGS_LOCALE) {
2837 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2838 }
2839 else {
81c6c7ce 2840 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
a0270393
KW
2841 ustrp, lenp,
2842 cBOOL((flags & FOLD_FLAGS_FULL
2843 /* If ASCII safe, don't allow full
2844 * folding, as that could include SHARP
2845 * S => ss; otherwise there is no
2846 * crossing of ascii/non-ascii in the
2847 * latin1 range */
2848 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
051a06d4 2849 }
a1dde8de 2850 }
051a06d4 2851 else { /* utf8, ord above 255 */
a0270393 2852 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2853
051a06d4 2854 if ((flags & FOLD_FLAGS_LOCALE)) {
a0270393 2855 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2856 }
a0270393
KW
2857 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2858 return result;
2859 }
2860 else {
2861 /* This is called when changing the case of a utf8-encoded
2862 * character above the Latin1 range, and the result should not
2863 * contain an ASCII character. */
2864
2865 UV original; /* To store the first code point of <p> */
2866
2867 /* Look at every character in the result; if any cross the
2868 * boundary, the whole thing is disallowed */
2869 U8* s = ustrp;
2870 U8* e = ustrp + *lenp;
2871 while (s < e) {
2872 if (isASCII(*s)) {
2873 /* Crossed, have to return the original */
2874 original = valid_utf8_to_uvchr(p, lenp);
2875 Copy(p, ustrp, *lenp, char);
2876 return original;
2877 }
2878 s += UTF8SKIP(s);
2879 }
051a06d4 2880
a0270393
KW
2881 /* Here, no characters crossed, result is ok as-is */
2882 return result;
2883 }
051a06d4
KW
2884 }
2885
2886 /* Here, used locale rules. Convert back to utf8 */
2887 if (UTF8_IS_INVARIANT(result)) {
2888 *ustrp = (U8) result;
2889 *lenp = 1;
2890 }
2891 else {
2892 *ustrp = UTF8_EIGHT_BIT_HI(result);
2893 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2894 *lenp = 2;
2895 }
2896
2897 if (tainted_ptr) {
2898 *tainted_ptr = TRUE;
2899 }
2900 return result;
a0ed51b3
LW
2901}
2902
711a919c 2903/* Note:
f90a9a02 2904 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2905 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2906 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2907 */
c4a5db0c 2908
a0ed51b3 2909SV*
7fc63493 2910Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2911{
c4a5db0c
KW
2912 PERL_ARGS_ASSERT_SWASH_INIT;
2913
2914 /* Returns a copy of a swash initiated by the called function. This is the
2915 * public interface, and returning a copy prevents others from doing
2916 * mischief on the original */
2917
5d3d13d1 2918 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2919}
2920
2921SV*
5d3d13d1 2922Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2923{
2924 /* Initialize and return a swash, creating it if necessary. It does this
9a53f6cf 2925 * by calling utf8_heavy.pl in the general case.
c4a5db0c
KW
2926 *
2927 * This interface should only be used by functions that won't destroy or
2928 * adversely change the swash, as doing so affects all other uses of the
2929 * swash in the program; the general public should use 'Perl_swash_init'
2930 * instead.
2931 *
2932 * pkg is the name of the package that <name> should be in.
2933 * name is the name of the swash to find. Typically it is a Unicode
2934 * property name, including user-defined ones
2935 * listsv is a string to initialize the swash with. It must be of the form
2936 * documented as the subroutine return value in
2937 * L<perlunicode/User-Defined Character Properties>
2938 * minbits is the number of bits required to represent each data element.
2939 * It is '1' for binary properties.
2940 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2941 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2942 * flags_p if non-NULL is the address of various input and output flag bits
2943 * to the routine, as follows: ('I' means is input to the routine;
2944 * 'O' means output from the routine. Only flags marked O are
2945 * meaningful on return.)
2946 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2947 * came from a user-defined property. (I O)
5d3d13d1
KW
2948 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2949 * when the swash cannot be located, to simply return NULL. (I)
9a53f6cf
KW
2950 *
2951 * Thus there are three possible inputs to find the swash: <name>,
2952 * <listsv>, and <invlist>. At least one must be specified. The result
2953 * will be the union of the specified ones, although <listsv>'s various
2954 * actions can intersect, etc. what <name> gives.
2955 *
2956 * <invlist> is only valid for binary properties */
c4a5db0c 2957
27da23d5 2958 dVAR;
c4a5db0c 2959 SV* retval = &PL_sv_undef;
83199d38 2960 HV* swash_hv = NULL;
9a53f6cf
KW
2961
2962 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2963 assert(! invlist || minbits == 1);
2964
2965 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2966 * so */
2967 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2968 dSP;
2969 const size_t pkg_len = strlen(pkg);
2970 const size_t name_len = strlen(name);
2971 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2972 SV* errsv_save;
2973 GV *method;
2974
2975 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2976
2977 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2978 ENTER;
69794297
KW
2979 SAVEHINTS();
2980 save_re_context();
2981 if (PL_parser && PL_parser->error_count)
2982 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2983 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2984 if (!method) { /* demand load utf8 */
2985 ENTER;
2986 errsv_save = newSVsv(ERRSV);
2987 /* It is assumed that callers of this routine are not passing in
2988 * any user derived data. */
2989 /* Need to do this after save_re_context() as it will set
2990 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2991 * in Perl_magic_get). Even line to create errsv_save can turn on
2992 * PL_tainted. */
2993 SAVEBOOL(PL_tainted);
2994 PL_tainted = 0;
2995 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2996 NULL);
2997 if (!SvTRUE(ERRSV))
2998 sv_setsv(ERRSV, errsv_save);
2999 SvREFCNT_dec(errsv_save);
3000 LEAVE;
3001 }
3002 SPAGAIN;
3003 PUSHMARK(SP);
3004 EXTEND(SP,5);
3005 mPUSHp(pkg, pkg_len);
3006 mPUSHp(name, name_len);
3007 PUSHs(listsv);
3008 mPUSHi(minbits);
3009 mPUSHi(none);
3010 PUTBACK;
f8be5cf0 3011 errsv_save = newSVsv(ERRSV);
69794297
KW
3012 /* If we already have a pointer to the method, no need to use
3013 * call_method() to repeat the lookup. */
3014 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
3015 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3016 {
3017 retval = *PL_stack_sp--;
3018 SvREFCNT_inc(retval);
3019 }
f8be5cf0
JH
3020 if (!SvTRUE(ERRSV))
3021 sv_setsv(ERRSV, errsv_save);
3022 SvREFCNT_dec(errsv_save);
ce3b816e 3023 LEAVE;
69794297
KW
3024 POPSTACK;
3025 if (IN_PERL_COMPILETIME) {
3026 CopHINTS_set(PL_curcop, PL_hints);
3027 }
3028 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3029 if (SvPOK(retval))
3030
3031 /* If caller wants to handle missing properties, let them */
5d3d13d1 3032 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
3033 return NULL;
3034 }
3035 Perl_croak(aTHX_
3036 "Can't find Unicode property definition \"%"SVf"\"",
3037 SVfARG(retval));
3038 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3039 }
9a53f6cf 3040 } /* End of calling the module to find the swash */
36eb48b4 3041
83199d38
KW
3042 /* If this operation fetched a swash, and we will need it later, get it */
3043 if (retval != &PL_sv_undef
3044 && (minbits == 1 || (flags_p
3045 && ! (*flags_p
3046 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3047 {
3048 swash_hv = MUTABLE_HV(SvRV(retval));
3049
3050 /* If we don't already know that there is a user-defined component to
3051 * this swash, and the user has indicated they wish to know if there is
3052 * one (by passing <flags_p>), find out */
3053 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3054 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3055 if (user_defined && SvUV(*user_defined)) {
3056 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3057 }
3058 }
3059 }
3060
36eb48b4
KW
3061 /* Make sure there is an inversion list for binary properties */
3062 if (minbits == 1) {
3063 SV** swash_invlistsvp = NULL;
3064 SV* swash_invlist = NULL;
9a53f6cf 3065 bool invlist_in_swash_is_valid = FALSE;
36eb48b4 3066
9a53f6cf 3067 /* If this operation fetched a swash, get its already existing
83199d38 3068 * inversion list, or create one for it */
36eb48b4 3069
83199d38 3070 if (swash_hv) {
5c9f4bd2 3071 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3072 if (swash_invlistsvp) {
3073 swash_invlist = *swash_invlistsvp;
3074 invlist_in_swash_is_valid = TRUE;
3075 }
3076 else {
36eb48b4 3077 swash_invlist = _swash_to_invlist(retval);
9a53f6cf
KW
3078 }
3079 }
3080
3081 /* If an inversion list was passed in, have to include it */
3082 if (invlist) {
3083
3084 /* Any fetched swash will by now have an inversion list in it;
3085 * otherwise <swash_invlist> will be NULL, indicating that we
3086 * didn't fetch a swash */
3087 if (swash_invlist) {
3088
3089 /* Add the passed-in inversion list, which invalidates the one
3090 * already stored in the swash */
3091 invlist_in_swash_is_valid = FALSE;
3092 _invlist_union(invlist, swash_invlist, &swash_invlist);
3093 }
3094 else {
3095
3096 /* Here, there is no swash already. Set up a minimal one */
3097 swash_hv = newHV();
3098 retval = newRV_inc(MUTABLE_SV(swash_hv));
3099 swash_invlist = invlist;
3100 }
9a53f6cf
KW
3101 }
3102
3103 /* Here, we have computed the union of all the passed-in data. It may
3104 * be that there was an inversion list in the swash which didn't get
3105 * touched; otherwise save the one computed one */
3106 if (! invlist_in_swash_is_valid) {
5c9f4bd2 3107 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3108 {
3109 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3110 }
9a53f6cf 3111 }
36eb48b4
KW
3112 }
3113
a0ed51b3
LW
3114 return retval;
3115}
3116
035d37be
JH
3117
3118/* This API is wrong for special case conversions since we may need to
3119 * return several Unicode characters for a single Unicode character
3120 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3121 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
3122 * multiple values. --jhi
3123 * For those, you should use to_utf8_case() instead */
b0e3252e 3124/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3125
3126/* Note:
3127 * Returns the value of property/mapping C<swash> for the first character
3128 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3129 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3130 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3131 *
3132 * A "swash" is a hash which contains initially the keys/values set up by
3133 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3134 * property for all possible code points. Things are stored in a compact form
3135 * (see utf8_heavy.pl) so that calculation is required to find the actual
3136 * property value for a given code point. As code points are looked up, new
3137 * key/value pairs are added to the hash, so that the calculation doesn't have
3138 * to ever be re-done. Further, each calculation is done, not just for the
3139 * desired one, but for a whole block of code points adjacent to that one.
3140 * For binary properties on ASCII machines, the block is usually for 64 code
3141 * points, starting with a code point evenly divisible by 64. Thus if the
3142 * property value for code point 257 is requested, the code goes out and
3143 * calculates the property values for all 64 code points between 256 and 319,
3144 * and stores these as a single 64-bit long bit vector, called a "swatch",
3145 * under the key for code point 256. The key is the UTF-8 encoding for code
3146 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3147 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3148 * for code point 258 is then requested, this code realizes that it would be
3149 * stored under the key for 256, and would find that value and extract the
3150 * relevant bit, offset from 256.
3151 *
3152 * Non-binary properties are stored in as many bits as necessary to represent
3153 * their values (32 currently, though the code is more general than that), not
3154 * as single bits, but the principal is the same: the value for each key is a
3155 * vector that encompasses the property values for all code points whose UTF-8
3156 * representations are represented by the key. That is, for all code points
3157 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3158 * bytes of that.
680c470c 3159 */
a0ed51b3 3160UV
680c470c 3161Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3162{
27da23d5 3163 dVAR;
ef8f7699 3164 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3165 U32 klen;
3166 U32 off;
a0ed51b3 3167 STRLEN slen;
7d85a32c 3168 STRLEN needents;
cfd0369c 3169 const U8 *tmps = NULL;
a0ed51b3 3170 U32 bit;
979f2922 3171 SV *swatch;
3568d838 3172 U8 tmputf8[2];
35da51f7 3173 const UV c = NATIVE_TO_ASCII(*ptr);
3568d838 3174
7918f24d
NC
3175 PERL_ARGS_ASSERT_SWASH_FETCH;
3176
dbe7a391 3177 /* Convert to utf8 if not already */
3568d838 3178 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
979f2922
ST
3179 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3180 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3181 ptr = tmputf8;
3568d838
JH
3182 }
3183 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
37e2e78e 3184 * then the "swatch" is a vec() for all the chars which start
3568d838
JH
3185 * with 0xAA..0xYY
3186 * So the key in the hash (klen) is length of encoded char -1
3187 */
3188 klen = UTF8SKIP(ptr) - 1;
3189 off = ptr[klen];
a0ed51b3 3190
979f2922 3191 if (klen == 0) {
37e2e78e 3192 /* If char is invariant then swatch is for all the invariant chars
1e54db1a 3193 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
7d85a32c 3194 */
979f2922
ST
3195 needents = UTF_CONTINUATION_MARK;
3196 off = NATIVE_TO_UTF(ptr[klen]);
3197 }
3198 else {
7d85a32c 3199 /* If char is encoded then swatch is for the prefix */
979f2922
ST
3200 needents = (1 << UTF_ACCUMULATION_SHIFT);
3201 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3202 }
7d85a32c 3203
a0ed51b3
LW
3204 /*
3205 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3206 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3207 * it's nothing to sniff at.) Pity we usually come through at least
3208 * two function calls to get here...
3209 *
3210 * NB: this code assumes that swatches are never modified, once generated!
3211 */
3212
3568d838 3213 if (hv == PL_last_swash_hv &&
a0ed51b3 3214 klen == PL_last_swash_klen &&
27da23d5 3215 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3216 {
3217 tmps = PL_last_swash_tmps;
3218 slen = PL_last_swash_slen;
3219 }
3220 else {
3221 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3222 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3223
b0e3252e 3224 /* If not cached, generate it via swatch_get */
979f2922
ST
3225 if (!svp || !SvPOK(*svp)
3226 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2b9d42f0
NIS
3227 /* We use utf8n_to_uvuni() as we want an index into
3228 Unicode tables, not a native character number.
3229 */
aec46f14 3230 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
872c91ae
JH
3231 ckWARN(WARN_UTF8) ?
3232 0 : UTF8_ALLOW_ANY);
b0e3252e 3233 swatch = swatch_get(swash,
979f2922 3234 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
361ee0fe 3235 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
979f2922
ST
3236 needents);
3237
923e4eb5 3238 if (IN_PERL_COMPILETIME)
623e6609 3239 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3240
979f2922 3241 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3242
979f2922
ST
3243 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3244 || (slen << 3) < needents)
5637ef5b
NC
3245 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3246 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3247 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3248 }
3249
3250 PL_last_swash_hv = hv;
16d8f38a 3251 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3252 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3253 /* FIXME change interpvar.h? */
3254 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3255 PL_last_swash_slen = slen;
3256 if (klen)
3257 Copy(ptr, PL_last_swash_key, klen, U8);
3258 }
3259
9faf8d75 3260 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3261 case 1:
3262 bit = 1 << (off & 7);
3263 off >>= 3;
3264 return (tmps[off] & bit) != 0;
3265 case 8:
3266 return tmps[off];
3267 case 16:
3268 off <<= 1;
3269 return (tmps[off] << 8) + tmps[off + 1] ;
3270 case 32:
3271 off <<= 2;
3272 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3273 }
5637ef5b
NC
3274 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3275 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3276 NORETURN_FUNCTION_END;
a0ed51b3 3277}
2b9d42f0 3278
319009ee
KW
3279/* Read a single line of the main body of the swash input text. These are of
3280 * the form:
3281 * 0053 0056 0073
3282 * where each number is hex. The first two numbers form the minimum and
3283 * maximum of a range, and the third is the value associated with the range.
3284 * Not all swashes should have a third number
3285 *
3286 * On input: l points to the beginning of the line to be examined; it points
3287 * to somewhere in the string of the whole input text, and is
3288 * terminated by a \n or the null string terminator.
3289 * lend points to the null terminator of that string
3290 * wants_value is non-zero if the swash expects a third number
3291 * typestr is the name of the swash's mapping, like 'ToLower'
3292 * On output: *min, *max, and *val are set to the values read from the line.
3293 * returns a pointer just beyond the line examined. If there was no
3294 * valid min number on the line, returns lend+1
3295 */
3296
3297STATIC U8*
3298S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3299 const bool wants_value, const U8* const typestr)
3300{
3301 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3302 STRLEN numlen; /* Length of the number */
02470786
KW
3303 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3304 | PERL_SCAN_DISALLOW_PREFIX
3305 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3306
3307 /* nl points to the next \n in the scan */
3308 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3309
3310 /* Get the first number on the line: the range minimum */
3311 numlen = lend - l;
3312 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3313 if (numlen) /* If found a hex number, position past it */
3314 l += numlen;
3315 else if (nl) { /* Else, go handle next line, if any */
3316 return nl + 1; /* 1 is length of "\n" */
3317 }
3318 else { /* Else, no next line */
3319 return lend + 1; /* to LIST's end at which \n is not found */
3320 }
3321
3322 /* The max range value follows, separated by a BLANK */
3323 if (isBLANK(*l)) {
3324 ++l;
02470786
KW
3325 flags = PERL_SCAN_SILENT_ILLDIGIT
3326 | PERL_SCAN_DISALLOW_PREFIX
3327 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3328 numlen = lend - l;
3329 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3330 if (numlen)
3331 l += numlen;
3332 else /* If no value here, it is a single element range */
3333 *max = *min;
3334
3335 /* Non-binary tables have a third entry: what the first element of the
3336 * range maps to */
3337 if (wants_value) {
3338 if (isBLANK(*l)) {
3339 ++l;
f90a9a02
KW
3340
3341 /* The ToLc, etc table mappings are not in hex, and must be
3342 * corrected by adding the code point to them */
3343 if (typeto) {
3344 char *after_strtol = (char *) lend;
3345 *val = Strtol((char *)l, &after_strtol, 10);
3346 l = (U8 *) after_strtol;
f90a9a02
KW
3347 }
3348 else { /* Other tables are in hex, and are the correct result
3349 without tweaking */
a9d188b3
KW
3350 flags = PERL_SCAN_SILENT_ILLDIGIT
3351 | PERL_SCAN_DISALLOW_PREFIX
3352 | PERL_SCAN_SILENT_NON_PORTABLE;
3353 numlen = lend - l;
3354 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3355 if (numlen)
3356 l += numlen;
3357 else
3358 *val = 0;
f90a9a02 3359 }
319009ee
KW
3360 }
3361 else {
3362 *val = 0;
3363 if (typeto) {
dcbac5bb 3364 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3365 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3366 typestr, l);
3367 }
3368 }
3369 }
3370 else
3371 *val = 0; /* bits == 1, then any val should be ignored */
3372 }
3373 else { /* Nothing following range min, should be single element with no
3374 mapping expected */
3375 *max = *min;
3376 if (wants_value) {
3377 *val = 0;
3378 if (typeto) {
dcbac5bb 3379 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3380 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3381 }
3382 }
3383 else
3384 *val = 0; /* bits == 1, then val should be ignored */
3385 }
3386
3387 /* Position to next line if any, or EOF */
3388 if (nl)
3389 l = nl + 1;
3390 else
3391 l = lend;
3392
3393 return l;
3394}
3395
979f2922
ST
3396/* Note:
3397 * Returns a swatch (a bit vector string) for a code point sequence
3398 * that starts from the value C<start> and comprises the number C<span>.
3399 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3400 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3401 */
3402STATIC SV*
b0e3252e 3403S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3404{
3405 SV *swatch;
77f9f126 3406 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3407 STRLEN lcur, xcur, scur;
ef8f7699 3408 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3409 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3410
88d45d28
KW
3411 SV** listsvp = NULL; /* The string containing the main body of the table */
3412 SV** extssvp = NULL;
3413 SV** invert_it_svp = NULL;
3414 U8* typestr = NULL;
786861f5
KW
3415 STRLEN bits;
3416 STRLEN octets; /* if bits == 1, then octets == 0 */
3417 UV none;
3418 UV end = start + span;
972dd592 3419
36eb48b4 3420 if (invlistsvp == NULL) {
786861f5
KW
3421 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3422 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3423 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3424 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3425 listsvp = hv_fetchs(hv, "LIST", FALSE);
3426 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3427
3428 bits = SvUV(*bitssvp);
3429 none = SvUV(*nonesvp);
3430 typestr = (U8*)SvPV_nolen(*typesvp);
3431 }
36eb48b4
KW
3432 else {
3433 bits = 1;
3434 none = 0;
3435 }
786861f5 3436 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3437
b0e3252e 3438 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3439
979f2922 3440 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3441 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3442 (UV)bits);
979f2922
ST
3443 }
3444
84ea5ef6
KW
3445 /* If overflowed, use the max possible */
3446 if (end < start) {
3447 end = UV_MAX;
3448 span = end - start;
3449 }
3450
979f2922 3451 /* create and initialize $swatch */
979f2922 3452 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3453 swatch = newSV(scur);
3454 SvPOK_on(swatch);
979f2922
ST
3455 s = (U8*)SvPVX(swatch);
3456 if (octets && none) {
0bd48802 3457 const U8* const e = s + scur;
979f2922
ST
3458 while (s < e) {
3459 if (bits == 8)
3460 *s++ = (U8)(none & 0xff);
3461 else if (bits == 16) {
3462 *s++ = (U8)((none >> 8) & 0xff);
3463 *s++ = (U8)( none & 0xff);
3464 }
3465 else if (bits == 32) {
3466 *s++ = (U8)((none >> 24) & 0xff);
3467 *s++ = (U8)((none >> 16) & 0xff);
3468 *s++ = (U8)((none >> 8) & 0xff);
3469 *s++ = (U8)( none & 0xff);
3470 }
3471 }
3472 *s = '\0';
3473 }
3474 else {
3475 (void)memzero((U8*)s, scur + 1);
3476 }
3477 SvCUR_set(swatch, scur);
3478 s = (U8*)SvPVX(swatch);
3479
36eb48b4
KW
3480 if (invlistsvp) { /* If has an inversion list set up use that */
3481 _invlist_populate_swatch(*invlistsvp, start, end, s);
3482 return swatch;
3483 }
3484
3485 /* read $swash->{LIST} */
979f2922
ST
3486 l = (U8*)SvPV(*listsvp, lcur);
3487 lend = l + lcur;
3488 while (l < lend) {
8ed25d53 3489 UV min, max, val, upper;
319009ee
KW
3490 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3491 cBOOL(octets), typestr);
3492 if (l > lend) {
979f2922
ST
3493 break;
3494 }
3495
972dd592 3496 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3497 if (max < start)
3498 continue;
3499
8ed25d53
KW
3500 /* <end> is generally 1 beyond where we want to set things, but at the
3501 * platform's infinity, where we can't go any higher, we want to
3502 * include the code point at <end> */
3503 upper = (max < end)
3504 ? max
3505 : (max != UV_MAX || end != UV_MAX)
3506 ? end - 1
3507 : end;
3508
979f2922 3509 if (octets) {
35da51f7 3510 UV key;
979f2922
ST
3511 if (min < start) {
3512 if (!none || val < none) {
3513 val += start - min;
3514 }
3515 min = start;
3516 }
8ed25d53 3517 for (key = min; key <= upper; key++) {
979f2922 3518 STRLEN offset;
979f2922
ST
3519 /* offset must be non-negative (start <= min <= key < end) */
3520 offset = octets * (key - start);
3521 if (bits == 8)
3522 s[offset] = (U8)(val & 0xff);
3523 else if (bits == 16) {
3524 s[offset ] = (U8)((val >> 8) & 0xff);
3525 s[offset + 1] = (U8)( val & 0xff);
3526 }
3527 else if (bits == 32) {
3528 s[offset ] = (U8)((val >> 24) & 0xff);
3529 s[offset + 1] = (U8)((val >> 16) & 0xff);
3530 s[offset + 2] = (U8)((val >> 8) & 0xff);
3531 s[offset + 3] = (U8)( val & 0xff);
3532 }
3533
3534 if (!none || val < none)
3535 ++val;
3536 }
3537 }
711a919c 3538 else { /* bits == 1, then val should be ignored */
35da51f7 3539 UV key;
979f2922
ST
3540 if (min < start)
3541 min = start;
6cb05c12 3542
8ed25d53 3543 for (key = min; key <= upper; key++) {
0bd48802 3544 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3545 s[offset >> 3] |= 1 << (offset & 7);
3546 }
3547 }
3548 } /* while */
979f2922 3549
9479a769 3550 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3551 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3552
3553 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3554 * be 0, and their inversion should also be 0, as we don't succeed any
3555 * Unicode property matches for non-Unicode code points */
3556 if (start <= PERL_UNICODE_MAX) {
3557
3558 /* The code below assumes that we never cross the
3559 * Unicode/above-Unicode boundary in a range, as otherwise we would
3560 * have to figure out where to stop flipping the bits. Since this
3561 * boundary is divisible by a large power of 2, and swatches comes
3562 * in small powers of 2, this should be a valid assumption */
3563 assert(start + span - 1 <= PERL_UNICODE_MAX);
3564
507a8485
KW
3565 send = s + scur;
3566 while (s < send) {
3567 *s = ~(*s);
3568 s++;
3569 }
0bda3001 3570 }
77f9f126
KW
3571 }
3572
d73c39c5
KW
3573 /* read $swash->{EXTRAS}
3574 * This code also copied to swash_to_invlist() below */
979f2922
ST
3575 x = (U8*)SvPV(*extssvp, xcur);
3576 xend = x + xcur;
3577 while (x < xend) {
3578 STRLEN namelen;
3579 U8 *namestr;
3580 SV** othersvp;
3581 HV* otherhv;
3582 STRLEN otherbits;
3583 SV **otherbitssvp, *other;
711a919c 3584 U8 *s, *o, *nl;
979f2922
ST
3585 STRLEN slen, olen;
3586
35da51f7 3587 const U8 opc = *x++;
979f2922
ST
3588 if (opc == '\n')
3589 continue;
3590
3591 nl = (U8*)memchr(x, '\n', xend - x);
3592
3593 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3594 if (nl) {
3595 x = nl + 1; /* 1 is length of "\n" */
3596 continue;
3597 }
3598 else {
3599 x = xend; /* to EXTRAS' end at which \n is not found */
3600 break;
3601 }
3602 }
3603
3604 namestr = x;
3605 if (nl) {
3606 namelen = nl - namestr;
3607 x = nl + 1;
3608 }
3609 else {
3610 namelen = xend - namestr;
3611 x = xend;
3612 }
3613
3614 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3615 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3616 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3617 otherbits = (STRLEN)SvUV(*otherbitssvp);
3618 if (bits < otherbits)
5637ef5b
NC
3619 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3620 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3621
3622 /* The "other" swatch must be destroyed after. */
b0e3252e 3623 other = swatch_get(*othersvp, start, span);
979f2922
ST
3624 o = (U8*)SvPV(other, olen);
3625
3626 if (!olen)
b0e3252e 3627 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3628
3629 s = (U8*)SvPV(swatch, slen);
3630 if (bits == 1 && otherbits == 1) {
3631 if (slen != olen)
5637ef5b
NC
3632 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3633 "mismatch, slen=%"UVuf", olen=%"UVuf,
3634 (UV)slen, (UV)olen);
979f2922
ST
3635
3636 switch (opc) {
3637 case '+':
3638 while (slen--)
3639 *s++ |= *o++;
3640 break;
3641 case '!':
3642 while (slen--)
3643 *s++ |= ~*o++;
3644 break;
3645 case '-':
3646 while (slen--)
3647 *s++ &= ~*o++;
3648 break;
3649 case '&':
3650 while (slen--)
3651 *s++ &= *o++;
3652 break;
3653 default:
3654 break;
3655 }
3656 }
711a919c 3657 else {
979f2922
ST
3658 STRLEN otheroctets = otherbits >> 3;
3659 STRLEN offset = 0;
35da51f7 3660 U8* const send = s + slen;
979f2922
ST
3661
3662 while (s < send) {
3663 UV otherval = 0;
3664
3665 if (otherbits == 1) {
3666 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3667 ++offset;
3668 }
3669 else {
3670 STRLEN vlen = otheroctets;
3671 otherval = *o++;
3672 while (--vlen) {
3673 otherval <<= 8;
3674 otherval |= *o++;
3675 }
3676 }
3677
711a919c 3678 if (opc == '+' && otherval)
6f207bd3 3679 NOOP; /* replace with otherval */
979f2922
ST
3680 else if (opc == '!' && !otherval)
3681 otherval = 1;
3682 else if (opc == '-' && otherval)
3683 otherval = 0;
3684 else if (opc == '&' && !otherval)
3685 otherval = 0;
3686 else {
711a919c 3687 s += octets; /* no replacement */
979f2922
ST
3688 continue;
3689 }
3690
3691 if (bits == 8)
3692 *s++ = (U8)( otherval & 0xff);
3693 else if (bits == 16) {
3694 *s++ = (U8)((otherval >> 8) & 0xff);
3695 *s++ = (U8)( otherval & 0xff);
3696 }
3697 else if (bits == 32) {
3698 *s++ = (U8)((otherval >> 24) & 0xff);
3699 *s++ = (U8)((otherval >> 16) & 0xff);
3700 *s++ = (U8)((otherval >> 8) & 0xff);
3701 *s++ = (U8)( otherval & 0xff);
3702 }
3703 }
3704 }
3705 sv_free(other); /* through with it! */
3706 } /* while */
3707 return swatch;
3708}
3709
064c021d 3710HV*
4c2e1131 3711Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3712{
3713
5662e334
KW
3714 /* Subject to change or removal. For use only in one place in regcomp.c.
3715 * Can't be used on a property that is subject to user override, as it
3716 * relies on the value of SPECIALS in the swash which would be set by
3717 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3718 * for overridden properties
064c021d
KW
3719 *
3720 * Returns a hash which is the inversion and closure of a swash mapping.
3721 * For example, consider the input lines:
3722 * 004B 006B
3723 * 004C 006C
3724 * 212A 006B
3725 *
3726 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3727 * 006C. The value for each key is an array. For 006C, the array would
3728 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3729 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3730 *
3731 * Essentially, for any code point, it gives all the code points that map to
3732 * it, or the list of 'froms' for that point.
3733 *
5662e334
KW
3734 * Currently it ignores any additions or deletions from other swashes,
3735 * looking at just the main body of the swash, and if there are SPECIALS
3736 * in the swash, at that hash
3737 *
3738 * The specials hash can be extra code points, and most likely consists of
3739 * maps from single code points to multiple ones (each expressed as a string
3740 * of utf8 characters). This function currently returns only 1-1 mappings.
3741 * However consider this possible input in the specials hash:
3742 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3743 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3744 *
3745 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3746 * currently handle. But it also means that FB05 and FB06 are equivalent in
3747 * a 1-1 mapping which we should handle, and this relationship may not be in