This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Move functions to inline_invlist.c
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
a0ed51b3 35
a0c21aa1 36#ifndef EBCDIC
970ea3cb 37/* Separate prototypes needed because in ASCII systems these are
a0c21aa1
JH
38 * usually macros but they still are compiled as code, too. */
39PERL_CALLCONV UV Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags);
40PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
41#endif
42
27da23d5
JH
43static const char unees[] =
44 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 45
48ef279e 46/*
ccfc67b7 47=head1 Unicode Support
a0ed51b3 48
166f8a29
DM
49This file contains various utility functions for manipulating UTF8-encoded
50strings. For the uninitiated, this is a method of representing arbitrary
61296642 51Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
52characters in the ASCII range are unmodified, and a zero byte never appears
53within non-zero characters.
166f8a29 54
eaf7a4d2
CS
55=cut
56*/
57
58/*
59=for apidoc is_ascii_string
60
a1433954 61Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
62or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
63is, if they are invariant. On ASCII-ish machines, only ASCII characters
64fit this definition, hence the function's name.
eaf7a4d2 65
9f7e3d64
MH
66If C<len> is 0, it will be calculated using C<strlen(s)>.
67
a1433954 68See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
69
70=cut
71*/
72
73bool
668b6d8d 74Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
75{
76 const U8* const send = s + (len ? len : strlen((const char *)s));
77 const U8* x = s;
78
79 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
80
81 for (; x < send; ++x) {
82 if (!UTF8_IS_INVARIANT(*x))
83 break;
84 }
85
86 return x == send;
87}
88
89/*
87cea99e 90=for apidoc uvuni_to_utf8_flags
eebe1485 91
6ee84de2
KW
92Adds the UTF-8 representation of the code point C<uv> to the end
93of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
eebe1485 94bytes available. The return value is the pointer to the byte after the
9041c2e3 95end of the new character. In other words,
eebe1485 96
b851fbc1
JH
97 d = uvuni_to_utf8_flags(d, uv, flags);
98
99or, in most cases,
100
9041c2e3 101 d = uvuni_to_utf8(d, uv);
eebe1485 102
b851fbc1
JH
103(which is equivalent to)
104
105 d = uvuni_to_utf8_flags(d, uv, 0);
106
949cf498 107This is the recommended Unicode-aware way of saying
eebe1485
SC
108
109 *(d++) = uv;
110
949cf498
KW
111This function will convert to UTF-8 (and not warn) even code points that aren't
112legal Unicode or are problematic, unless C<flags> contains one or more of the
a1433954
KW
113following flags:
114
949cf498
KW
115If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
116the function will raise a warning, provided UTF8 warnings are enabled. If instead
117UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
118If both flags are set, the function will both warn and return NULL.
119
120The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
121affect how the function handles a Unicode non-character. And, likewise for the
122UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, and code points that are
123above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
124even less portable) can be warned and/or disallowed even if other above-Unicode
125code points are accepted by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
126flags.
127
128And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
129above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
130DISALLOW flags.
131
132
eebe1485
SC
133=cut
134*/
135
dfe13c55 136U8 *
b851fbc1 137Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 138{
7918f24d
NC
139 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
140
979f77b6
KW
141 /* The first problematic code point is the first surrogate */
142 if (uv >= UNICODE_SURROGATE_FIRST
143 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
144 {
949cf498
KW
145 if (UNICODE_IS_SURROGATE(uv)) {
146 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 147 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
148 "UTF-16 surrogate U+%04"UVXf, uv);
149 }
150 if (flags & UNICODE_DISALLOW_SURROGATE) {
151 return NULL;
152 }
153 }
154 else if (UNICODE_IS_SUPER(uv)) {
155 if (flags & UNICODE_WARN_SUPER
156 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
157 {
8457b38f 158 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
159 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
160 }
161 if (flags & UNICODE_DISALLOW_SUPER
162 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
163 {
164 return NULL;
165 }
166 }
167 else if (UNICODE_IS_NONCHAR(uv)) {
168 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 169 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
170 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
171 uv);
172 }
173 if (flags & UNICODE_DISALLOW_NONCHAR) {
174 return NULL;
175 }
176 }
507b9800 177 }
c4d5f83a 178 if (UNI_IS_INVARIANT(uv)) {
eb160463 179 *d++ = (U8)UTF_TO_NATIVE(uv);
a0ed51b3
LW
180 return d;
181 }
2d331972 182#if defined(EBCDIC)
1d72bdf6
NIS
183 else {
184 STRLEN len = UNISKIP(uv);
185 U8 *p = d+len-1;
186 while (p > d) {
eb160463 187 *p-- = (U8)UTF_TO_NATIVE((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
188 uv >>= UTF_ACCUMULATION_SHIFT;
189 }
eb160463 190 *p = (U8)UTF_TO_NATIVE((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
191 return d+len;
192 }
193#else /* Non loop style */
a0ed51b3 194 if (uv < 0x800) {
eb160463
GS
195 *d++ = (U8)(( uv >> 6) | 0xc0);
196 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
197 return d;
198 }
199 if (uv < 0x10000) {
eb160463
GS
200 *d++ = (U8)(( uv >> 12) | 0xe0);
201 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
202 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
203 return d;
204 }
205 if (uv < 0x200000) {
eb160463
GS
206 *d++ = (U8)(( uv >> 18) | 0xf0);
207 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
208 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
209 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
210 return d;
211 }
212 if (uv < 0x4000000) {
eb160463
GS
213 *d++ = (U8)(( uv >> 24) | 0xf8);
214 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
217 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
218 return d;
219 }
220 if (uv < 0x80000000) {
eb160463
GS
221 *d++ = (U8)(( uv >> 30) | 0xfc);
222 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
223 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
226 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
227 return d;
228 }
6b8eaf93 229#ifdef HAS_QUAD
d7578b48 230 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
231#endif
232 {
eb160463
GS
233 *d++ = 0xfe; /* Can't match U+FEFF! */
234 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
235 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
236 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
237 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
238 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
239 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
240 return d;
241 }
6b8eaf93 242#ifdef HAS_QUAD
a0ed51b3 243 {
eb160463
GS
244 *d++ = 0xff; /* Can't match U+FFFE! */
245 *d++ = 0x80; /* 6 Reserved bits */
246 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
247 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
248 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
249 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
250 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
251 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
252 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
253 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
254 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
255 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
256 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
257 return d;
258 }
259#endif
1d72bdf6 260#endif /* Loop style */
a0ed51b3 261}
9041c2e3 262
646ca15d
JH
263/*
264
f7d739d1 265Tests if the first C<len> bytes of string C<s> form a valid UTF-8
646ca15d 266character. Note that an INVARIANT (i.e. ASCII) character is a valid
f7d739d1 267UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
268will be returned if it is valid, otherwise 0.
269
270This is the "slow" version as opposed to the "fast" version which is
271the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
272difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
273or less you should use the IS_UTF8_CHAR(), for lengths of five or more
274you should use the _slow(). In practice this means that the _slow()
275will be used very rarely, since the maximum Unicode code point (as of
276Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
277the "Perl extended UTF-8" (the infamous 'v-strings') will encode into
278five bytes or more.
279
280=cut */
c053b435 281STATIC STRLEN
5f66b61c 282S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 283{
cd7e6c88 284 dTHX; /* The function called below requires thread context */
646ca15d 285
cd7e6c88 286 STRLEN actual_len;
646ca15d 287
cd7e6c88 288 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 289
cd7e6c88 290 utf8n_to_uvuni(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 291
cd7e6c88 292 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 293}
9041c2e3
NIS
294
295/*
492a624f
KW
296=for apidoc is_utf8_char_buf
297
298Returns the number of bytes that comprise the first UTF-8 encoded character in
299buffer C<buf>. C<buf_end> should point to one position beyond the end of the
300buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
301encoded character.
302
303Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
304machines) is a valid UTF-8 character.
305
306=cut */
307
308STRLEN
309Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
310{
311
312 STRLEN len;
313
314 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
315
316 if (buf_end <= buf) {
317 return 0;
318 }
319
320 len = buf_end - buf;
321 if (len > UTF8SKIP(buf)) {
322 len = UTF8SKIP(buf);
323 }
324
325#ifdef IS_UTF8_CHAR
326 if (IS_UTF8_CHAR_FAST(len))
327 return IS_UTF8_CHAR(buf, len) ? len : 0;
328#endif /* #ifdef IS_UTF8_CHAR */
329 return is_utf8_char_slow(buf, len);
330}
331
332/*
87cea99e 333=for apidoc is_utf8_char
eebe1485 334
76848387
KW
335DEPRECATED!
336
5da9da9e 337Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
338character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
339character is a valid UTF-8 character. The actual number of bytes in the UTF-8
340character will be returned if it is valid, otherwise 0.
9041c2e3 341
76848387 342This function is deprecated due to the possibility that malformed input could
a1433954 343cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 344instead.
e0328548 345
82686b01 346=cut */
76848387 347
067a85ef 348STRLEN
668b6d8d 349Perl_is_utf8_char(const U8 *s)
386d01d6 350{
7918f24d 351 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 352
76848387 353 /* Assumes we have enough space, which is why this is deprecated */
492a624f 354 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
355}
356
eaf7a4d2 357
6662521e 358/*
87cea99e 359=for apidoc is_utf8_string
6662521e 360
a1433954 361Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 362UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
363using C<strlen(s)> (which means if you use this option, that C<s> has to have a
364terminating NUL byte). Note that all characters being ASCII constitute 'a
365valid UTF-8 string'.
6662521e 366
a1433954 367See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 368
6662521e
GS
369=cut
370*/
371
8e84507e 372bool
668b6d8d 373Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 374{
35da51f7 375 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 376 const U8* x = s;
067a85ef 377
7918f24d 378 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 379
6662521e 380 while (x < send) {
1acdb0da 381 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
382 if (UTF8_IS_INVARIANT(*x)) {
383 x++;
384 }
1acdb0da 385 else if (!UTF8_IS_START(*x))
e0328548 386 return FALSE;
1acdb0da
JH
387 else {
388 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
389 const STRLEN c = UTF8SKIP(x);
390 const U8* const next_char_ptr = x + c;
391
392 if (next_char_ptr > send) {
393 return FALSE;
394 }
395
768c67ee
JH
396 if (IS_UTF8_CHAR_FAST(c)) {
397 if (!IS_UTF8_CHAR(x, c))
e0328548 398 return FALSE;
3c614e38 399 }
e0328548
KW
400 else if (! is_utf8_char_slow(x, c)) {
401 return FALSE;
402 }
403 x = next_char_ptr;
1acdb0da 404 }
6662521e 405 }
768c67ee 406
067a85ef 407 return TRUE;
6662521e
GS
408}
409
67e989fb 410/*
814fafa7
NC
411Implemented as a macro in utf8.h
412
87cea99e 413=for apidoc is_utf8_string_loc
814fafa7 414
a1433954
KW
415Like L</is_utf8_string> but stores the location of the failure (in the
416case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
417"utf8ness success") in the C<ep>.
418
a1433954 419See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 420
87cea99e 421=for apidoc is_utf8_string_loclen
81cd54e3 422
a1433954
KW
423Like L</is_utf8_string>() but stores the location of the failure (in the
424case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
425"utf8ness success") in the C<ep>, and the number of UTF-8
426encoded characters in the C<el>.
427
a1433954 428See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
429
430=cut
431*/
432
433bool
668b6d8d 434Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 435{
35da51f7 436 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 437 const U8* x = s;
81cd54e3 438 STRLEN c;
3ebfea28 439 STRLEN outlen = 0;
7918f24d
NC
440
441 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 442
81cd54e3 443 while (x < send) {
e0328548
KW
444 const U8* next_char_ptr;
445
81cd54e3
JH
446 /* Inline the easy bits of is_utf8_char() here for speed... */
447 if (UTF8_IS_INVARIANT(*x))
e0328548 448 next_char_ptr = x + 1;
768c67ee
JH
449 else if (!UTF8_IS_START(*x))
450 goto out;
81cd54e3 451 else {
768c67ee 452 /* ... and call is_utf8_char() only if really needed. */
768c67ee 453 c = UTF8SKIP(x);
e0328548
KW
454 next_char_ptr = c + x;
455 if (next_char_ptr > send) {
456 goto out;
457 }
768c67ee
JH
458 if (IS_UTF8_CHAR_FAST(c)) {
459 if (!IS_UTF8_CHAR(x, c))
460 c = 0;
461 } else
462 c = is_utf8_char_slow(x, c);
768c67ee
JH
463 if (!c)
464 goto out;
81cd54e3 465 }
e0328548 466 x = next_char_ptr;
3ebfea28 467 outlen++;
81cd54e3 468 }
768c67ee
JH
469
470 out:
3ebfea28
AL
471 if (el)
472 *el = outlen;
473
768c67ee
JH
474 if (ep)
475 *ep = x;
3ebfea28 476 return (x == send);
81cd54e3
JH
477}
478
479/*
768c67ee 480
87cea99e 481=for apidoc utf8n_to_uvuni
67e989fb 482
9041c2e3 483Bottom level UTF-8 decode routine.
746afd53
KW
484Returns the code point value of the first character in the string C<s>,
485which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
486C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
487the length, in bytes, of that character.
949cf498
KW
488
489The value of C<flags> determines the behavior when C<s> does not point to a
490well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
491zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
492next possible position in C<s> that could begin a non-malformed character.
493Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
494
495Various ALLOW flags can be set in C<flags> to allow (and not warn on)
496individual types of malformations, such as the sequence being overlong (that
497is, when there is a shorter sequence that can express the same code point;
498overlong sequences are expressly forbidden in the UTF-8 standard due to
499potential security issues). Another malformation example is the first byte of
500a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
501flags. For allowed 0 length strings, this function returns 0; for allowed
502overlong sequences, the computed code point is returned; for all other allowed
503malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
504determinable reasonable value.
949cf498
KW
505
506The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
507flags) malformation is found. If this flag is set, the routine assumes that
508the caller will raise a warning, and this function will silently just set
509C<retlen> to C<-1> and return zero.
510
511Certain code points are considered problematic. These are Unicode surrogates,
746afd53 512Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 513By default these are considered regular code points, but certain situations
5eafe189 514warrant special handling for them. If C<flags> contains
949cf498
KW
515UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
516malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
517UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
518maximum) can be set to disallow these categories individually.
519
520The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
521UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
522for their respective categories, but otherwise the code points are considered
523valid (not malformations). To get a category to both be treated as a
524malformation and raise a warning, specify both the WARN and DISALLOW flags.
525(But note that warnings are not raised if lexically disabled nor if
526UTF8_CHECK_ONLY is also specified.)
527
528Very large code points (above 0x7FFF_FFFF) are considered more problematic than
529the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
530reasons: they requre at least 32 bits to represent them on ASCII platforms, are
531not representable at all on EBCDIC platforms, and the original UTF-8
532specification never went above this number (the current 0x10FFFF limit was
533imposed later). (The smaller ones, those that fit into 32 bits, are
534representable by a UV on ASCII platforms, but not by an IV, which means that
535the number of operations that can be performed on them is quite restricted.)
536The UTF-8 encoding on ASCII platforms for these large code points begins with a
537byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
538be treated as malformations, while allowing smaller above-Unicode code points.
539(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
540including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
541the other WARN flags, but applies just to these code points.
949cf498
KW
542
543All other code points corresponding to Unicode characters, including private
544use and those yet to be assigned, are never considered malformed and never
545warn.
67e989fb 546
ec5f19d0 547Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
9041c2e3 548
37607a96
PK
549=cut
550*/
67e989fb 551
a0ed51b3 552UV
7fc63493 553Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 554{
97aff369 555 dVAR;
d4c19fe8 556 const U8 * const s0 = s;
eb83ed87 557 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 558 U8 * send;
eb83ed87
KW
559 UV uv = *s;
560 STRLEN expectlen;
949cf498 561 SV* sv = NULL;
eb83ed87
KW
562 UV outlier_ret = 0; /* return value when input is in error or problematic
563 */
564 UV pack_warn = 0; /* Save result of packWARN() for later */
565 bool unexpected_non_continuation = FALSE;
566 bool overflowed = FALSE;
2f8f112e 567 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 568
eb83ed87 569 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 570
eb83ed87 571 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
a0dbb045 572
eb83ed87
KW
573 /* The order of malformation tests here is important. We should consume as
574 * few bytes as possible in order to not skip any valid character. This is
575 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
576 * http://unicode.org/reports/tr36 for more discussion as to why. For
577 * example, once we've done a UTF8SKIP, we can tell the expected number of
578 * bytes, and could fail right off the bat if the input parameters indicate
579 * that there are too few available. But it could be that just that first
580 * byte is garbled, and the intended character occupies fewer bytes. If we
581 * blindly assumed that the first byte is correct, and skipped based on
582 * that number, we could skip over a valid input character. So instead, we
583 * always examine the sequence byte-by-byte.
584 *
585 * We also should not consume too few bytes, otherwise someone could inject
586 * things. For example, an input could be deliberately designed to
587 * overflow, and if this code bailed out immediately upon discovering that,
588 * returning to the caller *retlen pointing to the very next byte (one
589 * which is actually part of of the overflowing sequence), that could look
590 * legitimate to the caller, which could discard the initial partial
591 * sequence and process the rest, inappropriately */
592
593 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 594 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
595 if (retlen) {
596 *retlen = 0;
597 }
a0dbb045 598
eb83ed87
KW
599 if (flags & UTF8_ALLOW_EMPTY) {
600 return 0;
601 }
602 if (! (flags & UTF8_CHECK_ONLY)) {
603 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
604 }
0c443dc2
JH
605 goto malformed;
606 }
607
eb83ed87
KW
608 expectlen = UTF8SKIP(s);
609
610 /* A well-formed UTF-8 character, as the vast majority of calls to this
611 * function will be for, has this expected length. For efficiency, set
612 * things up here to return it. It will be overriden only in those rare
613 * cases where a malformation is found */
614 if (retlen) {
615 *retlen = expectlen;
616 }
617
618 /* An invariant is trivially well-formed */
1d72bdf6 619 if (UTF8_IS_INVARIANT(uv)) {
c4d5f83a 620 return (UV) (NATIVE_TO_UTF(*s));
a0ed51b3 621 }
67e989fb 622
eb83ed87 623 /* A continuation character can't start a valid sequence */
b5b9af04 624 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
625 if (flags & UTF8_ALLOW_CONTINUATION) {
626 if (retlen) {
627 *retlen = 1;
628 }
629 return UNICODE_REPLACEMENT;
630 }
ba210ebe 631
eb83ed87
KW
632 if (! (flags & UTF8_CHECK_ONLY)) {
633 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
634 }
635 curlen = 1;
ba210ebe
JH
636 goto malformed;
637 }
9041c2e3 638
1d72bdf6 639#ifdef EBCDIC
75383841 640 uv = NATIVE_TO_UTF(uv);
1d72bdf6
NIS
641#endif
642
eb83ed87
KW
643 /* Here is not a continuation byte, nor an invariant. The only thing left
644 * is a start byte (possibly for an overlong) */
ba210ebe 645
eb83ed87
KW
646 /* Remove the leading bits that indicate the number of bytes in the
647 * character's whole UTF-8 sequence, leaving just the bits that are part of
648 * the value */
649 uv &= UTF_START_MASK(expectlen);
ba210ebe 650
eb83ed87
KW
651 /* Now, loop through the remaining bytes in the character's sequence,
652 * accumulating each into the working value as we go. Be sure to not look
653 * past the end of the input string */
0b8d30e8
KW
654 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
655
eb83ed87 656 for (s = s0 + 1; s < send; s++) {
b5b9af04 657 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
658#ifndef EBCDIC /* Can't overflow in EBCDIC */
659 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
660
661 /* The original implementors viewed this malformation as more
662 * serious than the others (though I, khw, don't understand
663 * why, since other malformations also give very very wrong
664 * results), so there is no way to turn off checking for it.
665 * Set a flag, but keep going in the loop, so that we absorb
666 * the rest of the bytes that comprise the character. */
667 overflowed = TRUE;
668 overflow_byte = *s; /* Save for warning message's use */
669 }
670#endif
8850bf83 671 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
672 }
673 else {
674 /* Here, found a non-continuation before processing all expected
675 * bytes. This byte begins a new character, so quit, even if
676 * allowing this malformation. */
677 unexpected_non_continuation = TRUE;
678 break;
679 }
680 } /* End of loop through the character's bytes */
681
682 /* Save how many bytes were actually in the character */
683 curlen = s - s0;
684
685 /* The loop above finds two types of malformations: non-continuation and/or
686 * overflow. The non-continuation malformation is really a too-short
687 * malformation, as it means that the current character ended before it was
688 * expected to (being terminated prematurely by the beginning of the next
689 * character, whereas in the too-short malformation there just are too few
690 * bytes available to hold the character. In both cases, the check below
691 * that we have found the expected number of bytes would fail if executed.)
692 * Thus the non-continuation malformation is really unnecessary, being a
693 * subset of the too-short malformation. But there may be existing
694 * applications that are expecting the non-continuation type, so we retain
695 * it, and return it in preference to the too-short malformation. (If this
696 * code were being written from scratch, the two types might be collapsed
697 * into one.) I, khw, am also giving priority to returning the
698 * non-continuation and too-short malformations over overflow when multiple
699 * ones are present. I don't know of any real reason to prefer one over
700 * the other, except that it seems to me that multiple-byte errors trumps
701 * errors from a single byte */
b5b9af04 702 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
703 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
704 if (! (flags & UTF8_CHECK_ONLY)) {
705 if (curlen == 1) {
706 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
707 }
708 else {
709 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, %d bytes after start byte 0x%02x, expected %d bytes)", malformed_text, *s, (int) curlen, *s0, (int)expectlen));
a0dbb045
JH
710 }
711 }
eb83ed87
KW
712 goto malformed;
713 }
714 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
715
716 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
717 * as what the original expectations were. */
718 do_overlong_test = FALSE;
eb83ed87
KW
719 if (retlen) {
720 *retlen = curlen;
721 }
722 }
b5b9af04 723 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
724 if (! (flags & UTF8_ALLOW_SHORT)) {
725 if (! (flags & UTF8_CHECK_ONLY)) {
726 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)curlen, curlen == 1 ? "" : "s", (int)expectlen, *s0));
a0dbb045 727 }
eb83ed87
KW
728 goto malformed;
729 }
730 uv = UNICODE_REPLACEMENT;
2f8f112e 731 do_overlong_test = FALSE;
eb83ed87
KW
732 if (retlen) {
733 *retlen = curlen;
734 }
735 }
736
737#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 738 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
739 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
740 {
741 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
742 * generation of the sv, since no warnings are raised under CHECK */
743 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
744 && ckWARN_d(WARN_UTF8))
745 {
42303544
KW
746 /* This message is deliberately not of the same syntax as the other
747 * messages for malformations, for backwards compatibility in the
748 * unlikely event that code is relying on its precise earlier text
749 */
eb83ed87
KW
750 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
751 pack_warn = packWARN(WARN_UTF8);
752 }
753 if (flags & UTF8_DISALLOW_FE_FF) {
754 goto malformed;
ba210ebe 755 }
ba210ebe 756 }
b5b9af04 757 if (UNLIKELY(overflowed)) {
ba210ebe 758
eb83ed87
KW
759 /* If the first byte is FF, it will overflow a 32-bit word. If the
760 * first byte is FE, it will overflow a signed 32-bit word. The
761 * above preserves backward compatibility, since its message was used
762 * in earlier versions of this code in preference to overflow */
763 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 764 goto malformed;
eb83ed87
KW
765 }
766#endif
767
2f8f112e
KW
768 if (do_overlong_test
769 && expectlen > (STRLEN)UNISKIP(uv)
770 && ! (flags & UTF8_ALLOW_LONG))
771 {
eb83ed87
KW
772 /* The overlong malformation has lower precedence than the others.
773 * Note that if this malformation is allowed, we return the actual
774 * value, instead of the replacement character. This is because this
775 * value is actually well-defined. */
776 if (! (flags & UTF8_CHECK_ONLY)) {
777 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", UNISKIP(uv), *s0));
778 }
779 goto malformed;
780 }
781
782 /* Here, the input is considered to be well-formed , but could be a
783 * problematic code point that is not allowed by the input parameters. */
784 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
785 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
786 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
787 {
949cf498 788 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
789 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
790 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
791 {
111d382d 792 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 793 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
794 }
795 if (flags & UTF8_DISALLOW_SURROGATE) {
796 goto disallowed;
797 }
798 }
949cf498 799 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
800 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
801 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
802 {
111d382d 803 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 804 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
805 }
806 if (flags & UTF8_DISALLOW_SUPER) {
807 goto disallowed;
808 }
809 }
4190d317
KW
810 else if (UNICODE_IS_NONCHAR(uv)) {
811 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
812 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
813 {
814 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
815 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
816 }
817 if (flags & UTF8_DISALLOW_NONCHAR) {
818 goto disallowed;
819 }
820 }
949cf498 821
eb83ed87
KW
822 if (sv) {
823 outlier_ret = uv;
824 goto do_warn;
825 }
826
949cf498
KW
827 /* Here, this is not considered a malformed character, so drop through
828 * to return it */
a0ed51b3 829 }
ba210ebe 830
a0ed51b3 831 return uv;
ba210ebe 832
eb83ed87
KW
833 /* There are three cases which get to beyond this point. In all 3 cases:
834 * <sv> if not null points to a string to print as a warning.
835 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
836 * set.
837 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
838 * This is done by initializing it to 0, and changing it only
839 * for case 1).
840 * The 3 cases are:
841 * 1) The input is valid but problematic, and to be warned about. The
842 * return value is the resultant code point; <*retlen> is set to
843 * <curlen>, the number of bytes that comprise the code point.
844 * <pack_warn> contains the result of packWARN() for the warning
845 * types. The entry point for this case is the label <do_warn>;
846 * 2) The input is a valid code point but disallowed by the parameters to
847 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
848 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
849 * comprise the code point. <pack_warn> contains the result of
850 * packWARN() for the warning types. The entry point for this case is
851 * the label <disallowed>.
852 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
853 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
854 * bytes that comprise the malformation. All such malformations are
855 * assumed to be warning type <utf8>. The entry point for this case
856 * is the label <malformed>.
857 */
949cf498 858
ba210ebe
JH
859malformed:
860
eb83ed87
KW
861 if (sv && ckWARN_d(WARN_UTF8)) {
862 pack_warn = packWARN(WARN_UTF8);
863 }
864
865disallowed:
866
fcc8fcf6 867 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 868 if (retlen)
10edeb5d 869 *retlen = ((STRLEN) -1);
ba210ebe
JH
870 return 0;
871 }
872
eb83ed87 873do_warn:
5b311467 874
eb83ed87
KW
875 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
876 if warnings are to be raised. */
f555bc63 877 const char * const string = SvPVX_const(sv);
a0dbb045 878
f555bc63
KW
879 if (PL_op)
880 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
881 else
882 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
883 }
884
eb83ed87
KW
885 if (retlen) {
886 *retlen = curlen;
887 }
ba210ebe 888
eb83ed87 889 return outlier_ret;
a0ed51b3
LW
890}
891
8e84507e 892/*
ec5f19d0
KW
893=for apidoc utf8_to_uvchr_buf
894
895Returns the native code point of the first character in the string C<s> which
896is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 897C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 898
524080c4
KW
899If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
900enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
901NULL) to -1. If those warnings are off, the computed value if well-defined (or
902the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
903is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
904next possible position in C<s> that could begin a non-malformed character.
905See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
906
907=cut
908*/
909
910
911UV
912Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
913{
914 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
915
916 assert(s < send);
917
918 return utf8n_to_uvchr(s, send - s, retlen,
919 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
920}
921
27d6c58a 922/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c
KW
923 * there are no malformations in the input UTF-8 string C<s>. surrogates,
924 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
925
926UV
927Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
928{
3986bb7c
KW
929 const UV uv = valid_utf8_to_uvuni(s, retlen);
930
27d6c58a
KW
931 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
932
3986bb7c 933 return UNI_TO_NATIVE(uv);
27d6c58a
KW
934}
935
ec5f19d0 936/*
87cea99e 937=for apidoc utf8_to_uvchr
9041c2e3 938
977c1d31
KW
939DEPRECATED!
940
6ee84de2 941Returns the native code point of the first character in the string C<s>
1e54db1a 942which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
943length, in bytes, of that character.
944
4b88fb76 945Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
946malformed input could cause reading beyond the end of the input buffer, which
947is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 948
524080c4
KW
949If C<s> points to one of the detected malformations, and UTF8 warnings are
950enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
951NULL) to -1. If those warnings are off, the computed value if well-defined (or
952the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
953is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
954next possible position in C<s> that could begin a non-malformed character.
955See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
956
957=cut
958*/
959
960UV
7fc63493 961Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 962{
7918f24d
NC
963 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
964
2ff6c191 965 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
966}
967
968/*
ec5f19d0
KW
969=for apidoc utf8_to_uvuni_buf
970
971Returns the Unicode code point of the first character in the string C<s> which
972is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
973C<retlen> will be set to the length, in bytes, of that character.
974
975This function should only be used when the returned UV is considered
976an index into the Unicode semantic tables (e.g. swashes).
977
524080c4
KW
978If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
979enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
980NULL) to -1. If those warnings are off, the computed value if well-defined (or
981the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
982is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
983next possible position in C<s> that could begin a non-malformed character.
984See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
985
986=cut
987*/
988
989UV
990Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
991{
992 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
993
994 assert(send > s);
995
996 /* Call the low level routine asking for checks */
997 return Perl_utf8n_to_uvuni(aTHX_ s, send -s, retlen,
998 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
999}
1000
27d6c58a 1001/* Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1002 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1003 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1004
1005UV
1006Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1007{
3986bb7c
KW
1008 UV expectlen = UTF8SKIP(s);
1009 const U8* send = s + expectlen;
1010 UV uv = NATIVE_TO_UTF(*s);
1011
27d6c58a
KW
1012 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1013
3986bb7c
KW
1014 if (retlen) {
1015 *retlen = expectlen;
1016 }
1017
1018 /* An invariant is trivially returned */
1019 if (expectlen == 1) {
1020 return uv;
1021 }
1022
1023 /* Remove the leading bits that indicate the number of bytes, leaving just
1024 * the bits that are part of the value */
1025 uv &= UTF_START_MASK(expectlen);
1026
1027 /* Now, loop through the remaining bytes, accumulating each into the
1028 * working total as we go */
1029 for (++s; s < send; s++) {
1030 uv = UTF8_ACCUMULATE(uv, *s);
1031 }
1032
1033 return uv;
27d6c58a
KW
1034}
1035
ec5f19d0 1036/*
87cea99e 1037=for apidoc utf8_to_uvuni
9041c2e3 1038
977c1d31
KW
1039DEPRECATED!
1040
9041c2e3 1041Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1042which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1043length, in bytes, of that character.
1044
39e518fd
KW
1045This function should only be used when the returned UV is considered
1046an index into the Unicode semantic tables (e.g. swashes).
1047
4b88fb76 1048Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1049malformed input could cause reading beyond the end of the input buffer, which
1050is why this function is deprecated. Use L</utf8_to_uvuni_buf> instead.
9041c2e3 1051
524080c4
KW
1052If C<s> points to one of the detected malformations, and UTF8 warnings are
1053enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1054NULL) to -1. If those warnings are off, the computed value if well-defined (or
1055the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1056is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1057next possible position in C<s> that could begin a non-malformed character.
1058See L</utf8n_to_uvuni> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1059
1060=cut
1061*/
1062
1063UV
7fc63493 1064Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1065{
7918f24d
NC
1066 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1067
4b88fb76 1068 return valid_utf8_to_uvuni(s, retlen);
8e84507e
NIS
1069}
1070
b76347f2 1071/*
87cea99e 1072=for apidoc utf8_length
b76347f2
JH
1073
1074Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1075Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1076up past C<e>, croaks.
b76347f2
JH
1077
1078=cut
1079*/
1080
1081STRLEN
35a4481c 1082Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1083{
97aff369 1084 dVAR;
b76347f2
JH
1085 STRLEN len = 0;
1086
7918f24d
NC
1087 PERL_ARGS_ASSERT_UTF8_LENGTH;
1088
8850bf83
JH
1089 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1090 * the bitops (especially ~) can create illegal UTF-8.
1091 * In other words: in Perl UTF-8 is not just for Unicode. */
1092
a3b680e6
AL
1093 if (e < s)
1094 goto warn_and_return;
b76347f2 1095 while (s < e) {
8e91ec7f
AV
1096 if (!UTF8_IS_INVARIANT(*s))
1097 s += UTF8SKIP(s);
1098 else
1099 s++;
1100 len++;
1101 }
1102
1103 if (e != s) {
1104 len--;
1105 warn_and_return:
9b387841
NC
1106 if (PL_op)
1107 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1108 "%s in %s", unees, OP_DESC(PL_op));
1109 else
61a12c31 1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1111 }
1112
1113 return len;
1114}
1115
b06226ff 1116/*
87cea99e 1117=for apidoc utf8_distance
b06226ff 1118
1e54db1a 1119Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1120and C<b>.
1121
1122WARNING: use only if you *know* that the pointers point inside the
1123same UTF-8 buffer.
1124
37607a96
PK
1125=cut
1126*/
a0ed51b3 1127
02eb7b47 1128IV
35a4481c 1129Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1130{
7918f24d
NC
1131 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1132
bf1665bc 1133 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1134}
1135
b06226ff 1136/*
87cea99e 1137=for apidoc utf8_hop
b06226ff 1138
8850bf83
JH
1139Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1140forward or backward.
b06226ff
JH
1141
1142WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1143the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1144on the first byte of character or just after the last byte of a character.
b06226ff 1145
37607a96
PK
1146=cut
1147*/
a0ed51b3
LW
1148
1149U8 *
4373e329 1150Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1151{
7918f24d
NC
1152 PERL_ARGS_ASSERT_UTF8_HOP;
1153
96a5add6 1154 PERL_UNUSED_CONTEXT;
8850bf83
JH
1155 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1156 * the bitops (especially ~) can create illegal UTF-8.
1157 * In other words: in Perl UTF-8 is not just for Unicode. */
1158
a0ed51b3
LW
1159 if (off >= 0) {
1160 while (off--)
1161 s += UTF8SKIP(s);
1162 }
1163 else {
1164 while (off++) {
1165 s--;
8850bf83
JH
1166 while (UTF8_IS_CONTINUATION(*s))
1167 s--;
a0ed51b3
LW
1168 }
1169 }
4373e329 1170 return (U8 *)s;
a0ed51b3
LW
1171}
1172
6940069f 1173/*
fed3ba5d
NC
1174=for apidoc bytes_cmp_utf8
1175
a1433954
KW
1176Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1177sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1178equal, -1 or -2 if the first string is less than the second string, +1 or +2
1179if the first string is greater than the second string.
1180
1181-1 or +1 is returned if the shorter string was identical to the start of the
1182longer string. -2 or +2 is returned if the was a difference between characters
1183within the strings.
1184
1185=cut
1186*/
1187
1188int
1189Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1190{
1191 const U8 *const bend = b + blen;
1192 const U8 *const uend = u + ulen;
1193
1194 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1195
1196 PERL_UNUSED_CONTEXT;
1197
1198 while (b < bend && u < uend) {
1199 U8 c = *u++;
1200 if (!UTF8_IS_INVARIANT(c)) {
1201 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1202 if (u < uend) {
1203 U8 c1 = *u++;
1204 if (UTF8_IS_CONTINUATION(c1)) {
356979f4 1205 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, c1));
fed3ba5d
NC
1206 } else {
1207 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1208 "Malformed UTF-8 character "
1209 "(unexpected non-continuation byte 0x%02x"
1210 ", immediately after start byte 0x%02x)"
1211 /* Dear diag.t, it's in the pod. */
1212 "%s%s", c1, c,
1213 PL_op ? " in " : "",
1214 PL_op ? OP_DESC(PL_op) : "");
1215 return -2;
1216 }
1217 } else {
1218 if (PL_op)
1219 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1220 "%s in %s", unees, OP_DESC(PL_op));
1221 else
61a12c31 1222 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1223 return -2; /* Really want to return undef :-) */
1224 }
1225 } else {
1226 return -2;
1227 }
1228 }
1229 if (*b != c) {
1230 return *b < c ? -2 : +2;
1231 }
1232 ++b;
1233 }
1234
1235 if (b == bend && u == uend)
1236 return 0;
1237
1238 return b < bend ? +1 : -1;
1239}
1240
1241/*
87cea99e 1242=for apidoc utf8_to_bytes
6940069f 1243
2bbc8d55 1244Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1245Unlike L</bytes_to_utf8>, this over-writes the original string, and
1246updates C<len> to contain the new length.
67e989fb 1247Returns zero on failure, setting C<len> to -1.
6940069f 1248
a1433954 1249If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1250
6940069f
GS
1251=cut
1252*/
1253
1254U8 *
37607a96 1255Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1256{
d4c19fe8
AL
1257 U8 * const save = s;
1258 U8 * const send = s + *len;
6940069f 1259 U8 *d;
246fae53 1260
7918f24d
NC
1261 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1262
1e54db1a 1263 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1264 while (s < send) {
dcad2880
JH
1265 U8 c = *s++;
1266
1d72bdf6
NIS
1267 if (!UTF8_IS_INVARIANT(c) &&
1268 (!UTF8_IS_DOWNGRADEABLE_START(c) || (s >= send)
1269 || !(c = *s++) || !UTF8_IS_CONTINUATION(c))) {
10edeb5d 1270 *len = ((STRLEN) -1);
dcad2880
JH
1271 return 0;
1272 }
246fae53 1273 }
dcad2880
JH
1274
1275 d = s = save;
6940069f 1276 while (s < send) {
ed646e6e 1277 STRLEN ulen;
4b88fb76 1278 *d++ = (U8)utf8_to_uvchr_buf(s, send, &ulen);
ed646e6e 1279 s += ulen;
6940069f
GS
1280 }
1281 *d = '\0';
246fae53 1282 *len = d - save;
6940069f
GS
1283 return save;
1284}
1285
1286/*
87cea99e 1287=for apidoc bytes_from_utf8
f9a63242 1288
2bbc8d55 1289Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1290Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1291the newly-created string, and updates C<len> to contain the new
1292length. Returns the original string if no conversion occurs, C<len>
1293is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
12940 if C<s> is converted or consisted entirely of characters that are invariant
1295in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1296
37607a96
PK
1297=cut
1298*/
f9a63242
JH
1299
1300U8 *
e1ec3a88 1301Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1302{
f9a63242 1303 U8 *d;
e1ec3a88
AL
1304 const U8 *start = s;
1305 const U8 *send;
f9a63242
JH
1306 I32 count = 0;
1307
7918f24d
NC
1308 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1309
96a5add6 1310 PERL_UNUSED_CONTEXT;
f9a63242 1311 if (!*is_utf8)
73d840c0 1312 return (U8 *)start;
f9a63242 1313
1e54db1a 1314 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1315 for (send = s + *len; s < send;) {
e1ec3a88 1316 U8 c = *s++;
1d72bdf6 1317 if (!UTF8_IS_INVARIANT(c)) {
db42d148
NIS
1318 if (UTF8_IS_DOWNGRADEABLE_START(c) && s < send &&
1319 (c = *s++) && UTF8_IS_CONTINUATION(c))
1320 count++;
1321 else
73d840c0 1322 return (U8 *)start;
db42d148 1323 }
f9a63242
JH
1324 }
1325
35da51f7 1326 *is_utf8 = FALSE;
f9a63242 1327
212542aa 1328 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1329 s = start; start = d;
f9a63242
JH
1330 while (s < send) {
1331 U8 c = *s++;
c4d5f83a
NIS
1332 if (!UTF8_IS_INVARIANT(c)) {
1333 /* Then it is two-byte encoded */
356979f4 1334 c = UNI_TO_NATIVE(TWO_BYTE_UTF8_TO_UNI(c, *s++));
c4d5f83a
NIS
1335 }
1336 *d++ = c;
f9a63242
JH
1337 }
1338 *d = '\0';
1339 *len = d - start;
73d840c0 1340 return (U8 *)start;
f9a63242
JH
1341}
1342
1343/*
87cea99e 1344=for apidoc bytes_to_utf8
6940069f 1345
ff97e5cf
KW
1346Converts a string C<s> of length C<len> bytes from the native encoding into
1347UTF-8.
6662521e 1348Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1349reflect the new length in bytes.
6940069f 1350
2bbc8d55
SP
1351A NUL character will be written after the end of the string.
1352
1353If you want to convert to UTF-8 from encodings other than
1354the native (Latin1 or EBCDIC),
a1433954 1355see L</sv_recode_to_utf8>().
c9ada85f 1356
497711e7 1357=cut
6940069f
GS
1358*/
1359
c682ebef
FC
1360/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1361 likewise need duplication. */
1362
6940069f 1363U8*
35a4481c 1364Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1365{
35a4481c 1366 const U8 * const send = s + (*len);
6940069f
GS
1367 U8 *d;
1368 U8 *dst;
7918f24d
NC
1369
1370 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1371 PERL_UNUSED_CONTEXT;
6940069f 1372
212542aa 1373 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1374 dst = d;
1375
1376 while (s < send) {
35a4481c 1377 const UV uv = NATIVE_TO_ASCII(*s++);
c4d5f83a 1378 if (UNI_IS_INVARIANT(uv))
eb160463 1379 *d++ = (U8)UTF_TO_NATIVE(uv);
6940069f 1380 else {
eb160463
GS
1381 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
1382 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
6940069f
GS
1383 }
1384 }
1385 *d = '\0';
6662521e 1386 *len = d-dst;
6940069f
GS
1387 return dst;
1388}
1389
a0ed51b3 1390/*
dea0fc0b 1391 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1392 *
1393 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1394 * We optimize for native, for obvious reasons. */
1395
1396U8*
dea0fc0b 1397Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1398{
dea0fc0b
JH
1399 U8* pend;
1400 U8* dstart = d;
1401
7918f24d
NC
1402 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1403
dea0fc0b 1404 if (bytelen & 1)
f5992bc4 1405 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1406
1407 pend = p + bytelen;
1408
a0ed51b3 1409 while (p < pend) {
dea0fc0b
JH
1410 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1411 p += 2;
a0ed51b3 1412 if (uv < 0x80) {
e294cc5d
JH
1413#ifdef EBCDIC
1414 *d++ = UNI_TO_NATIVE(uv);
1415#else
eb160463 1416 *d++ = (U8)uv;
e294cc5d 1417#endif
a0ed51b3
LW
1418 continue;
1419 }
1420 if (uv < 0x800) {
eb160463
GS
1421 *d++ = (U8)(( uv >> 6) | 0xc0);
1422 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1423 continue;
1424 }
52b9aa85 1425 if (uv >= 0xd800 && uv <= 0xdbff) { /* surrogates */
01ea242b 1426 if (p >= pend) {
dea0fc0b 1427 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1428 } else {
1429 UV low = (p[0] << 8) + p[1];
1430 p += 2;
52b9aa85 1431 if (low < 0xdc00 || low > 0xdfff)
01ea242b
NC
1432 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1433 uv = ((uv - 0xd800) << 10) + (low - 0xdc00) + 0x10000;
1434 }
dbde1951
NC
1435 } else if (uv >= 0xdc00 && uv <= 0xdfff) {
1436 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3
LW
1437 }
1438 if (uv < 0x10000) {
eb160463
GS
1439 *d++ = (U8)(( uv >> 12) | 0xe0);
1440 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1441 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1442 continue;
1443 }
1444 else {
eb160463
GS
1445 *d++ = (U8)(( uv >> 18) | 0xf0);
1446 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1447 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1448 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1449 continue;
1450 }
1451 }
dea0fc0b 1452 *newlen = d - dstart;
a0ed51b3
LW
1453 return d;
1454}
1455
1456/* Note: this one is slightly destructive of the source. */
1457
1458U8*
dea0fc0b 1459Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1460{
1461 U8* s = (U8*)p;
d4c19fe8 1462 U8* const send = s + bytelen;
7918f24d
NC
1463
1464 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1465
e0ea5e2d
NC
1466 if (bytelen & 1)
1467 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1468 (UV)bytelen);
1469
a0ed51b3 1470 while (s < send) {
d4c19fe8 1471 const U8 tmp = s[0];
a0ed51b3
LW
1472 s[0] = s[1];
1473 s[1] = tmp;
1474 s += 2;
1475 }
dea0fc0b 1476 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1477}
1478
c3fd2246
KW
1479/* for now these are all defined (inefficiently) in terms of the utf8 versions.
1480 * Note that the macros in handy.h that call these short-circuit calling them
1481 * for Latin-1 range inputs */
a0ed51b3
LW
1482
1483bool
84afefe6 1484Perl_is_uni_alnum(pTHX_ UV c)
a0ed51b3 1485{
89ebb4a3 1486 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1487 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1488 return is_utf8_alnum(tmpbuf);
1489}
1490
1491bool
84afefe6 1492Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1493{
89ebb4a3 1494 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1495 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1496 return is_utf8_idfirst(tmpbuf);
1497}
1498
1499bool
84afefe6 1500Perl_is_uni_alpha(pTHX_ UV c)
a0ed51b3 1501{
89ebb4a3 1502 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1503 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1504 return is_utf8_alpha(tmpbuf);
1505}
1506
1507bool
84afefe6 1508Perl_is_uni_ascii(pTHX_ UV c)
4d61ec05 1509{
bc39fe24 1510 return isASCII(c);
4d61ec05
GS
1511}
1512
1513bool
bdd8600f
KW
1514Perl_is_uni_blank(pTHX_ UV c)
1515{
1516 U8 tmpbuf[UTF8_MAXBYTES+1];
1517 uvchr_to_utf8(tmpbuf, c);
1518 return is_utf8_blank(tmpbuf);
1519}
1520
1521bool
84afefe6 1522Perl_is_uni_space(pTHX_ UV c)
a0ed51b3 1523{
89ebb4a3 1524 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1525 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1526 return is_utf8_space(tmpbuf);
1527}
1528
1529bool
84afefe6 1530Perl_is_uni_digit(pTHX_ UV c)
a0ed51b3 1531{
89ebb4a3 1532 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1533 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1534 return is_utf8_digit(tmpbuf);
1535}
1536
1537bool
84afefe6 1538Perl_is_uni_upper(pTHX_ UV c)
a0ed51b3 1539{
89ebb4a3 1540 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1541 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1542 return is_utf8_upper(tmpbuf);
1543}
1544
1545bool
84afefe6 1546Perl_is_uni_lower(pTHX_ UV c)
a0ed51b3 1547{
89ebb4a3 1548 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1549 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1550 return is_utf8_lower(tmpbuf);
1551}
1552
1553bool
84afefe6 1554Perl_is_uni_cntrl(pTHX_ UV c)
b8c5462f 1555{
7b952154 1556 return isCNTRL_L1(c);
b8c5462f
JH
1557}
1558
1559bool
84afefe6 1560Perl_is_uni_graph(pTHX_ UV c)
b8c5462f 1561{
89ebb4a3 1562 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1563 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1564 return is_utf8_graph(tmpbuf);
1565}
1566
1567bool
84afefe6 1568Perl_is_uni_print(pTHX_ UV c)
a0ed51b3 1569{
89ebb4a3 1570 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1571 uvchr_to_utf8(tmpbuf, c);
a0ed51b3
LW
1572 return is_utf8_print(tmpbuf);
1573}
1574
b8c5462f 1575bool
84afefe6 1576Perl_is_uni_punct(pTHX_ UV c)
b8c5462f 1577{
89ebb4a3 1578 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1579 uvchr_to_utf8(tmpbuf, c);
b8c5462f
JH
1580 return is_utf8_punct(tmpbuf);
1581}
1582
4d61ec05 1583bool
84afefe6 1584Perl_is_uni_xdigit(pTHX_ UV c)
4d61ec05 1585{
89ebb4a3 1586 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
230880c1 1587 uvchr_to_utf8(tmpbuf, c);
4d61ec05
GS
1588 return is_utf8_xdigit(tmpbuf);
1589}
1590
3a4c58c9
KW
1591UV
1592Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1593{
1594 /* We have the latin1-range values compiled into the core, so just use
1595 * those, converting the result to utf8. The only difference between upper
1596 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1597 * either "SS" or "Ss". Which one to use is passed into the routine in
1598 * 'S_or_s' to avoid a test */
1599
1600 UV converted = toUPPER_LATIN1_MOD(c);
1601
1602 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1603
1604 assert(S_or_s == 'S' || S_or_s == 's');
1605
1606 if (UNI_IS_INVARIANT(converted)) { /* No difference between the two for
1607 characters in this range */
1608 *p = (U8) converted;
1609 *lenp = 1;
1610 return converted;
1611 }
1612
1613 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1614 * which it maps to one of them, so as to only have to have one check for
1615 * it in the main case */
1616 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1617 switch (c) {
1618 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1619 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1620 break;
1621 case MICRO_SIGN:
1622 converted = GREEK_CAPITAL_LETTER_MU;
1623 break;
1624 case LATIN_SMALL_LETTER_SHARP_S:
1625 *(p)++ = 'S';
1626 *p = S_or_s;
1627 *lenp = 2;
1628 return 'S';
1629 default:
1630 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1631 assert(0); /* NOTREACHED */
3a4c58c9
KW
1632 }
1633 }
1634
1635 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1636 *p = UTF8_TWO_BYTE_LO(converted);
1637 *lenp = 2;
1638
1639 return converted;
1640}
1641
50bda2c3
KW
1642/* Call the function to convert a UTF-8 encoded character to the specified case.
1643 * Note that there may be more than one character in the result.
1644 * INP is a pointer to the first byte of the input character
1645 * OUTP will be set to the first byte of the string of changed characters. It
1646 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1647 * LENP will be set to the length in bytes of the string of changed characters
1648 *
1649 * The functions return the ordinal of the first character in the string of OUTP */
f90a9a02
KW
1650#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1651#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1652#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
50bda2c3
KW
1653
1654/* This additionally has the input parameter SPECIALS, which if non-zero will
1655 * cause this to use the SPECIALS hash for folding (meaning get full case
1656 * folding); otherwise, when zero, this implies a simple case fold */
f90a9a02 1657#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "utf8::ToSpecCf" : NULL)
c3fd2246 1658
84afefe6
JH
1659UV
1660Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1661{
3a4c58c9
KW
1662 dVAR;
1663
a1433954
KW
1664 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1665 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1666 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1667 * the changed version may be longer than the original character.
1668 *
1669 * The ordinal of the first character of the changed version is returned
1670 * (but note, as explained above, that there may be more.) */
1671
7918f24d
NC
1672 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1673
3a4c58c9
KW
1674 if (c < 256) {
1675 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1676 }
1677
0ebc6274 1678 uvchr_to_utf8(p, c);
3a4c58c9 1679 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1680}
1681
84afefe6
JH
1682UV
1683Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1684{
3a4c58c9
KW
1685 dVAR;
1686
7918f24d
NC
1687 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1688
3a4c58c9
KW
1689 if (c < 256) {
1690 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1691 }
1692
0ebc6274 1693 uvchr_to_utf8(p, c);
3a4c58c9 1694 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1695}
1696
afc16117
KW
1697STATIC U8
1698S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1699{
1700 /* We have the latin1-range values compiled into the core, so just use
1701 * those, converting the result to utf8. Since the result is always just
a1433954 1702 * one character, we allow <p> to be NULL */
afc16117
KW
1703
1704 U8 converted = toLOWER_LATIN1(c);
1705
1706 if (p != NULL) {
1707 if (UNI_IS_INVARIANT(converted)) {
1708 *p = converted;
1709 *lenp = 1;
1710 }
1711 else {
1712 *p = UTF8_TWO_BYTE_HI(converted);
1713 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1714 *lenp = 2;
1715 }
1716 }
1717 return converted;
1718}
1719
84afefe6
JH
1720UV
1721Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1722{
968c5e6a
KW
1723 dVAR;
1724
7918f24d
NC
1725 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1726
afc16117
KW
1727 if (c < 256) {
1728 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1729 }
1730
afc16117 1731 uvchr_to_utf8(p, c);
968c5e6a 1732 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1733}
1734
84afefe6 1735UV
f673fad4 1736Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const bool flags)
a1dde8de 1737{
a1433954 1738 /* Corresponds to to_lower_latin1(), <flags> is TRUE if to use full case
f673fad4
KW
1739 * folding */
1740
a1dde8de
KW
1741 UV converted;
1742
1743 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1744
1745 if (c == MICRO_SIGN) {
1746 converted = GREEK_SMALL_LETTER_MU;
1747 }
1748 else if (flags && c == LATIN_SMALL_LETTER_SHARP_S) {
1749 *(p)++ = 's';
1750 *p = 's';
1751 *lenp = 2;
1752 return 's';
1753 }
1754 else { /* In this range the fold of all other characters is their lower
1755 case */
1756 converted = toLOWER_LATIN1(c);
1757 }
1758
1759 if (UNI_IS_INVARIANT(converted)) {
1760 *p = (U8) converted;
1761 *lenp = 1;
1762 }
1763 else {
1764 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1765 *p = UTF8_TWO_BYTE_LO(converted);
1766 *lenp = 2;
1767 }
1768
1769 return converted;
1770}
1771
1772UV
a0270393 1773Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
84afefe6 1774{
4b593389 1775
a0270393
KW
1776 /* Not currently externally documented, and subject to change
1777 * <flags> bits meanings:
1778 * FOLD_FLAGS_FULL iff full folding is to be used;
1779 * FOLD_FLAGS_LOCALE iff in locale
1780 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1781 */
4b593389 1782
36bb2ab6 1783 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1784
a1dde8de 1785 if (c < 256) {
a0270393
KW
1786 UV result = _to_fold_latin1((U8) c, p, lenp,
1787 cBOOL(((flags & FOLD_FLAGS_FULL)
1788 /* If ASCII-safe, don't allow full folding,
1789 * as that could include SHARP S => ss;
1790 * otherwise there is no crossing of
1791 * ascii/non-ascii in the latin1 range */
1792 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
1793 /* It is illegal for the fold to cross the 255/256 boundary under
1794 * locale; in this case return the original */
1795 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1796 ? c
1797 : result;
a1dde8de
KW
1798 }
1799
a0270393
KW
1800 /* If no special needs, just use the macro */
1801 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1802 uvchr_to_utf8(p, c);
1803 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1804 }
1805 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1806 the special flags. */
1807 U8 utf8_c[UTF8_MAXBYTES + 1];
1808 uvchr_to_utf8(utf8_c, c);
1809 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1810 }
84afefe6
JH
1811}
1812
ea317ccb
KW
1813/* for now these all assume no locale info available for Unicode > 255; and
1814 * the corresponding macros in handy.h (like isALNUM_LC_uvchr) should have been
1815 * called instead, so that these don't get called for < 255 */
a0ed51b3
LW
1816
1817bool
84afefe6 1818Perl_is_uni_alnum_lc(pTHX_ UV c)
a0ed51b3
LW
1819{
1820 return is_uni_alnum(c); /* XXX no locale support yet */
1821}
1822
1823bool
84afefe6 1824Perl_is_uni_idfirst_lc(pTHX_ UV c)
a0ed51b3
LW
1825{
1826 return is_uni_idfirst(c); /* XXX no locale support yet */
1827}
1828
1829bool
84afefe6 1830Perl_is_uni_alpha_lc(pTHX_ UV c)
a0ed51b3
LW
1831{
1832 return is_uni_alpha(c); /* XXX no locale support yet */
1833}
1834
1835bool
84afefe6 1836Perl_is_uni_ascii_lc(pTHX_ UV c)
4d61ec05
GS
1837{
1838 return is_uni_ascii(c); /* XXX no locale support yet */
1839}
1840
1841bool
bdd8600f
KW
1842Perl_is_uni_blank_lc(pTHX_ UV c)
1843{
1844 return is_uni_blank(c); /* XXX no locale support yet */
1845}
1846
1847bool
84afefe6 1848Perl_is_uni_space_lc(pTHX_ UV c)
a0ed51b3
LW
1849{
1850 return is_uni_space(c); /* XXX no locale support yet */
1851}
1852
1853bool
84afefe6 1854Perl_is_uni_digit_lc(pTHX_ UV c)
a0ed51b3
LW
1855{
1856 return is_uni_digit(c); /* XXX no locale support yet */
1857}
1858
1859bool
84afefe6 1860Perl_is_uni_upper_lc(pTHX_ UV c)
a0ed51b3
LW
1861{
1862 return is_uni_upper(c); /* XXX no locale support yet */
1863}
1864
1865bool
84afefe6 1866Perl_is_uni_lower_lc(pTHX_ UV c)
a0ed51b3
LW
1867{
1868 return is_uni_lower(c); /* XXX no locale support yet */
1869}
1870
1871bool
84afefe6 1872Perl_is_uni_cntrl_lc(pTHX_ UV c)
b8c5462f
JH
1873{
1874 return is_uni_cntrl(c); /* XXX no locale support yet */
1875}
1876
1877bool
84afefe6 1878Perl_is_uni_graph_lc(pTHX_ UV c)
b8c5462f
JH
1879{
1880 return is_uni_graph(c); /* XXX no locale support yet */
1881}
1882
1883bool
84afefe6 1884Perl_is_uni_print_lc(pTHX_ UV c)
a0ed51b3
LW
1885{
1886 return is_uni_print(c); /* XXX no locale support yet */
1887}
1888
b8c5462f 1889bool
84afefe6 1890Perl_is_uni_punct_lc(pTHX_ UV c)
b8c5462f
JH
1891{
1892 return is_uni_punct(c); /* XXX no locale support yet */
1893}
1894
4d61ec05 1895bool
84afefe6 1896Perl_is_uni_xdigit_lc(pTHX_ UV c)
4d61ec05
GS
1897{
1898 return is_uni_xdigit(c); /* XXX no locale support yet */
1899}
1900
b7ac61fa
JH
1901U32
1902Perl_to_uni_upper_lc(pTHX_ U32 c)
1903{
ee099d14
JH
1904 /* XXX returns only the first character -- do not use XXX */
1905 /* XXX no locale support yet */
1906 STRLEN len;
89ebb4a3 1907 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1908 return (U32)to_uni_upper(c, tmpbuf, &len);
b7ac61fa
JH
1909}
1910
1911U32
1912Perl_to_uni_title_lc(pTHX_ U32 c)
1913{
ee099d14
JH
1914 /* XXX returns only the first character XXX -- do not use XXX */
1915 /* XXX no locale support yet */
1916 STRLEN len;
89ebb4a3 1917 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1918 return (U32)to_uni_title(c, tmpbuf, &len);
b7ac61fa
JH
1919}
1920
1921U32
1922Perl_to_uni_lower_lc(pTHX_ U32 c)
1923{
ee099d14
JH
1924 /* XXX returns only the first character -- do not use XXX */
1925 /* XXX no locale support yet */
1926 STRLEN len;
89ebb4a3 1927 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 1928 return (U32)to_uni_lower(c, tmpbuf, &len);
b7ac61fa
JH
1929}
1930
7452cf6a 1931static bool
5141f98e 1932S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
1933 const char *const swashname)
1934{
ea317ccb
KW
1935 /* returns a boolean giving whether or not the UTF8-encoded character that
1936 * starts at <p> is in the swash indicated by <swashname>. <swash>
1937 * contains a pointer to where the swash indicated by <swashname>
1938 * is to be stored; which this routine will do, so that future calls will
1939 * look at <*swash> and only generate a swash if it is not null
1940 *
1941 * Note that it is assumed that the buffer length of <p> is enough to
1942 * contain all the bytes that comprise the character. Thus, <*p> should
1943 * have been checked before this call for mal-formedness enough to assure
1944 * that. */
1945
97aff369 1946 dVAR;
7918f24d
NC
1947
1948 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1949
492a624f
KW
1950 /* The API should have included a length for the UTF-8 character in <p>,
1951 * but it doesn't. We therefor assume that p has been validated at least
1952 * as far as there being enough bytes available in it to accommodate the
1953 * character without reading beyond the end, and pass that number on to the
1954 * validating routine */
1955 if (!is_utf8_char_buf(p, p + UTF8SKIP(p)))
bde6a22d
NC
1956 return FALSE;
1957 if (!*swash)
711a919c 1958 *swash = swash_init("utf8", swashname, &PL_sv_undef, 1, 0);
bde6a22d
NC
1959 return swash_fetch(*swash, p, TRUE) != 0;
1960}
1961
1962bool
7fc63493 1963Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 1964{
97aff369 1965 dVAR;
7918f24d
NC
1966
1967 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
1968
671c33bf
NC
1969 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
1970 * descendant of isalnum(3), in other words, it doesn't
1971 * contain the '_'. --jhi */
d4c19fe8 1972 return is_utf8_common(p, &PL_utf8_alnum, "IsWord");
a0ed51b3
LW
1973}
1974
1975bool
7fc63493 1976Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1977{
97aff369 1978 dVAR;
7918f24d
NC
1979
1980 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1981
82686b01
JH
1982 if (*p == '_')
1983 return TRUE;
bde6a22d 1984 /* is_utf8_idstart would be more logical. */
d4c19fe8 1985 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
82686b01
JH
1986}
1987
1988bool
c11ff943
KW
1989Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1990{
1991 dVAR;
1992
1993 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1994
1995 if (*p == '_')
1996 return TRUE;
1997 /* is_utf8_idstart would be more logical. */
1998 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
1999}
2000
2001bool
b6912c02
KW
2002Perl__is_utf8__perl_idstart(pTHX_ const U8 *p)
2003{
2004 dVAR;
2005
2006 PERL_ARGS_ASSERT__IS_UTF8__PERL_IDSTART;
2007
2008 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2009}
2010
2011bool
7fc63493 2012Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2013{
97aff369 2014 dVAR;
7918f24d
NC
2015
2016 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2017
d4c19fe8 2018 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
2019}
2020
2021bool
c11ff943
KW
2022Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2023{
2024 dVAR;
2025
2026 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2027
c11ff943
KW
2028 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2029}
2030
2031bool
7fc63493 2032Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 2033{
97aff369 2034 dVAR;
7918f24d
NC
2035
2036 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2037
d4c19fe8 2038 return is_utf8_common(p, &PL_utf8_alpha, "IsAlpha");
a0ed51b3
LW
2039}
2040
2041bool
7fc63493 2042Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2043{
97aff369 2044 dVAR;
7918f24d
NC
2045
2046 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2047
bc39fe24
KW
2048 /* ASCII characters are the same whether in utf8 or not. So the macro
2049 * works on both utf8 and non-utf8 representations. */
2050 return isASCII(*p);
b8c5462f
JH
2051}
2052
2053bool
bdd8600f
KW
2054Perl_is_utf8_blank(pTHX_ const U8 *p)
2055{
2056 dVAR;
2057
2058 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2059
2060 return is_utf8_common(p, &PL_utf8_blank, "XPosixBlank");
2061}
2062
2063bool
7fc63493 2064Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2065{
97aff369 2066 dVAR;
7918f24d
NC
2067
2068 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2069
a34094a9 2070 return is_utf8_common(p, &PL_utf8_space, "IsXPerlSpace");
a0ed51b3
LW
2071}
2072
2073bool
d1eb3177
YO
2074Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2075{
2076 dVAR;
2077
2078 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2079
c4428693
KW
2080 /* Only true if is an ASCII space-like character, and ASCII is invariant
2081 * under utf8, so can just use the macro */
2082 return isSPACE_A(*p);
d1eb3177
YO
2083}
2084
2085bool
2086Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2087{
2088 dVAR;
2089
2090 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2091
c4428693
KW
2092 /* Only true if is an ASCII word character, and ASCII is invariant
2093 * under utf8, so can just use the macro */
2094 return isWORDCHAR_A(*p);
d1eb3177
YO
2095}
2096
2097bool
7fc63493 2098Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2099{
97aff369 2100 dVAR;
7918f24d
NC
2101
2102 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2103
d4c19fe8 2104 return is_utf8_common(p, &PL_utf8_digit, "IsDigit");
a0ed51b3
LW
2105}
2106
2107bool
d1eb3177
YO
2108Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2109{
2110 dVAR;
2111
2112 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2113
c4428693
KW
2114 /* Only true if is an ASCII digit character, and ASCII is invariant
2115 * under utf8, so can just use the macro */
2116 return isDIGIT_A(*p);
d1eb3177
YO
2117}
2118
2119bool
7fc63493 2120Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2121{
97aff369 2122 dVAR;
7918f24d
NC
2123
2124 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2125
d4c19fe8 2126 return is_utf8_common(p, &PL_utf8_upper, "IsUppercase");
a0ed51b3
LW
2127}
2128
2129bool
7fc63493 2130Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2131{
97aff369 2132 dVAR;
7918f24d
NC
2133
2134 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2135
d4c19fe8 2136 return is_utf8_common(p, &PL_utf8_lower, "IsLowercase");
a0ed51b3
LW
2137}
2138
2139bool
7fc63493 2140Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2141{
97aff369 2142 dVAR;
7918f24d
NC
2143
2144 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2145
7b952154
KW
2146 if (isASCII(*p)) {
2147 return isCNTRL_A(*p);
2148 }
2149
2150 /* All controls are in Latin1 */
2151 if (! UTF8_IS_DOWNGRADEABLE_START(*p)) {
2152 return 0;
2153 }
2154 return isCNTRL_L1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
b8c5462f
JH
2155}
2156
2157bool
7fc63493 2158Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2159{
97aff369 2160 dVAR;
7918f24d
NC
2161
2162 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2163
d4c19fe8 2164 return is_utf8_common(p, &PL_utf8_graph, "IsGraph");
b8c5462f
JH
2165}
2166
2167bool
7fc63493 2168Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2169{
97aff369 2170 dVAR;
7918f24d
NC
2171
2172 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2173
d4c19fe8 2174 return is_utf8_common(p, &PL_utf8_print, "IsPrint");
a0ed51b3
LW
2175}
2176
2177bool
7fc63493 2178Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2179{
97aff369 2180 dVAR;
7918f24d
NC
2181
2182 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2183
d4c19fe8 2184 return is_utf8_common(p, &PL_utf8_punct, "IsPunct");
b8c5462f
JH
2185}
2186
2187bool
7fc63493 2188Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2189{
97aff369 2190 dVAR;
7918f24d
NC
2191
2192 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2193
d1eb3177 2194 return is_utf8_common(p, &PL_utf8_xdigit, "IsXDigit");
b8c5462f
JH
2195}
2196
2197bool
7fc63493 2198Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2199{
97aff369 2200 dVAR;
7918f24d
NC
2201
2202 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2203
d4c19fe8 2204 return is_utf8_common(p, &PL_utf8_mark, "IsM");
a0ed51b3
LW
2205}
2206
37e2e78e
KW
2207bool
2208Perl_is_utf8_X_begin(pTHX_ const U8 *p)
2209{
2210 dVAR;
2211
2212 PERL_ARGS_ASSERT_IS_UTF8_X_BEGIN;
2213
2214 return is_utf8_common(p, &PL_utf8_X_begin, "_X_Begin");
2215}
2216
2217bool
2218Perl_is_utf8_X_extend(pTHX_ const U8 *p)
2219{
2220 dVAR;
2221
2222 PERL_ARGS_ASSERT_IS_UTF8_X_EXTEND;
2223
2224 return is_utf8_common(p, &PL_utf8_X_extend, "_X_Extend");
2225}
2226
2227bool
2228Perl_is_utf8_X_prepend(pTHX_ const U8 *p)
2229{
2230 dVAR;
2231
2232 PERL_ARGS_ASSERT_IS_UTF8_X_PREPEND;
2233
f957fa32 2234 return is_utf8_common(p, &PL_utf8_X_prepend, "_X_GCB_Prepend");
37e2e78e
KW
2235}
2236
2237bool
2238Perl_is_utf8_X_non_hangul(pTHX_ const U8 *p)
2239{
2240 dVAR;
2241
2242 PERL_ARGS_ASSERT_IS_UTF8_X_NON_HANGUL;
2243
f957fa32 2244 return is_utf8_common(p, &PL_utf8_X_non_hangul, "_X_HST_Not_Applicable");
37e2e78e
KW
2245}
2246
2247bool
2248Perl_is_utf8_X_L(pTHX_ const U8 *p)
2249{
2250 dVAR;
2251
2252 PERL_ARGS_ASSERT_IS_UTF8_X_L;
2253
f957fa32 2254 return is_utf8_common(p, &PL_utf8_X_L, "_X_GCB_L");
37e2e78e
KW
2255}
2256
2257bool
2258Perl_is_utf8_X_LV(pTHX_ const U8 *p)
2259{
2260 dVAR;
2261
2262 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
2263
f957fa32 2264 return is_utf8_common(p, &PL_utf8_X_LV, "_X_GCB_LV");
37e2e78e
KW
2265}
2266
2267bool
2268Perl_is_utf8_X_LVT(pTHX_ const U8 *p)
2269{
2270 dVAR;
2271
2272 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
2273
f957fa32 2274 return is_utf8_common(p, &PL_utf8_X_LVT, "_X_GCB_LVT");
37e2e78e
KW
2275}
2276
2277bool
2278Perl_is_utf8_X_T(pTHX_ const U8 *p)
2279{
2280 dVAR;
2281
2282 PERL_ARGS_ASSERT_IS_UTF8_X_T;
2283
f957fa32 2284 return is_utf8_common(p, &PL_utf8_X_T, "_X_GCB_T");
37e2e78e
KW
2285}
2286
2287bool
2288Perl_is_utf8_X_V(pTHX_ const U8 *p)
2289{
2290 dVAR;
2291
2292 PERL_ARGS_ASSERT_IS_UTF8_X_V;
2293
f957fa32 2294 return is_utf8_common(p, &PL_utf8_X_V, "_X_GCB_V");
37e2e78e
KW
2295}
2296
2297bool
2298Perl_is_utf8_X_LV_LVT_V(pTHX_ const U8 *p)
2299{
2300 dVAR;
2301
2302 PERL_ARGS_ASSERT_IS_UTF8_X_LV_LVT_V;
2303
2304 return is_utf8_common(p, &PL_utf8_X_LV_LVT_V, "_X_LV_LVT_V");
2305}
2306
2e2b2571
KW
2307bool
2308Perl__is_utf8_quotemeta(pTHX_ const U8 *p)
2309{
2310 /* For exclusive use of pp_quotemeta() */
2311
2312 dVAR;
2313
2314 PERL_ARGS_ASSERT__IS_UTF8_QUOTEMETA;
2315
2316 return is_utf8_common(p, &PL_utf8_quotemeta, "_Perl_Quotemeta");
2317}
2318
6b5c0936 2319/*
87cea99e 2320=for apidoc to_utf8_case
6b5c0936 2321
a1433954
KW
2322The C<p> contains the pointer to the UTF-8 string encoding
2323the character that is being converted. This routine assumes that the character
2324at C<p> is well-formed.
6b5c0936 2325
a1433954
KW
2326The C<ustrp> is a pointer to the character buffer to put the
2327conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2328of the result.
2329
a1433954 2330The C<swashp> is a pointer to the swash to use.
6b5c0936 2331
a1433954
KW
2332Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2333and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2334but not always, a multicharacter mapping), is tried first.
6b5c0936 2335
a1433954 2336The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2337hash %utf8::ToSpecLower. The access to the hash is through
2338Perl_to_utf8_case().
6b5c0936 2339
a1433954 2340The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2341%utf8::ToLower.
2342
2343=cut */
6b5c0936 2344
2104c8d9 2345UV
9a957fbc
AL
2346Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2347 SV **swashp, const char *normal, const char *special)
a0ed51b3 2348{
97aff369 2349 dVAR;
89ebb4a3 2350 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
0134edef 2351 STRLEN len = 0;
4b88fb76 2352 const UV uv0 = valid_utf8_to_uvchr(p, NULL);
1feea2c7
JH
2353 /* The NATIVE_TO_UNI() and UNI_TO_NATIVE() mappings
2354 * are necessary in EBCDIC, they are redundant no-ops
2355 * in ASCII-ish platforms, and hopefully optimized away. */
f54cb97a 2356 const UV uv1 = NATIVE_TO_UNI(uv0);
7918f24d
NC
2357
2358 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2359
9ae3ac1a
KW
2360 /* Note that swash_fetch() doesn't output warnings for these because it
2361 * assumes we will */
8457b38f 2362 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2363 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2364 if (ckWARN_d(WARN_SURROGATE)) {
2365 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2366 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2367 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2368 }
9ae3ac1a
KW
2369 }
2370 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2371 if (ckWARN_d(WARN_NON_UNICODE)) {
2372 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2373 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2374 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2375 }
9ae3ac1a
KW
2376 }
2377
2378 /* Note that non-characters are perfectly legal, so no warning should
2379 * be given */
2380 }
2381
1feea2c7 2382 uvuni_to_utf8(tmpbuf, uv1);
0134edef
JH
2383
2384 if (!*swashp) /* load on-demand */
2385 *swashp = swash_init("utf8", normal, &PL_sv_undef, 4, 0);
2386
a6f87d8c 2387 if (special) {
0134edef 2388 /* It might be "special" (sometimes, but not always,
2a37f04d 2389 * a multicharacter mapping) */
6673a63c 2390 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2391 SV **svp;
2392
35da51f7 2393 if (hv &&
b08cf34e
JH
2394 (svp = hv_fetch(hv, (const char*)tmpbuf, UNISKIP(uv1), FALSE)) &&
2395 (*svp)) {
cfd0369c 2396 const char *s;
47654450 2397
cfd0369c 2398 s = SvPV_const(*svp, len);
47654450
JH
2399 if (len == 1)
2400 len = uvuni_to_utf8(ustrp, NATIVE_TO_UNI(*(U8*)s)) - ustrp;
2a37f04d 2401 else {
2f9475ad
JH
2402#ifdef EBCDIC
2403 /* If we have EBCDIC we need to remap the characters
2404 * since any characters in the low 256 are Unicode
2405 * code points, not EBCDIC. */
7cda7a3d 2406 U8 *t = (U8*)s, *tend = t + len, *d;
2f9475ad
JH
2407
2408 d = tmpbuf;
b08cf34e 2409 if (SvUTF8(*svp)) {
2f9475ad
JH
2410 STRLEN tlen = 0;
2411
2412 while (t < tend) {
6bd1c396 2413 const UV c = utf8_to_uvchr_buf(t, tend, &tlen);
2f9475ad
JH
2414 if (tlen > 0) {
2415 d = uvchr_to_utf8(d, UNI_TO_NATIVE(c));
2416 t += tlen;
2417 }
2418 else
2419 break;
2420 }
2421 }
2422 else {
36fec512
JH
2423 while (t < tend) {
2424 d = uvchr_to_utf8(d, UNI_TO_NATIVE(*t));
2425 t++;
2426 }
2f9475ad
JH
2427 }
2428 len = d - tmpbuf;
2429 Copy(tmpbuf, ustrp, len, U8);
2430#else
d2dcd0fb 2431 Copy(s, ustrp, len, U8);
2f9475ad 2432#endif
29e98929 2433 }
983ffd37 2434 }
0134edef
JH
2435 }
2436
2437 if (!len && *swashp) {
2114036c 2438 const UV uv2 = swash_fetch(*swashp, tmpbuf, TRUE /* => is utf8 */);
d4c19fe8 2439
0134edef
JH
2440 if (uv2) {
2441 /* It was "normal" (a single character mapping). */
d4c19fe8 2442 const UV uv3 = UNI_TO_NATIVE(uv2);
e9101d72 2443 len = uvchr_to_utf8(ustrp, uv3) - ustrp;
2a37f04d
JH
2444 }
2445 }
1feea2c7 2446
cbe07460
KW
2447 if (len) {
2448 if (lenp) {
2449 *lenp = len;
2450 }
2451 return valid_utf8_to_uvchr(ustrp, 0);
2452 }
2453
2454 /* Here, there was no mapping defined, which means that the code point maps
2455 * to itself. Return the inputs */
bfdf22ec
KW
2456 len = UTF8SKIP(p);
2457 Copy(p, ustrp, len, U8);
0134edef 2458
2a37f04d
JH
2459 if (lenp)
2460 *lenp = len;
2461
cbe07460
KW
2462 return uv0;
2463
a0ed51b3
LW
2464}
2465
051a06d4
KW
2466STATIC UV
2467S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2468{
2469 /* This is called when changing the case of a utf8-encoded character above
2470 * the Latin1 range, and the operation is in locale. If the result
2471 * contains a character that crosses the 255/256 boundary, disallow the
2472 * change, and return the original code point. See L<perlfunc/lc> for why;
2473 *
a1433954
KW
2474 * p points to the original string whose case was changed; assumed
2475 * by this routine to be well-formed
051a06d4
KW
2476 * result the code point of the first character in the changed-case string
2477 * ustrp points to the changed-case string (<result> represents its first char)
2478 * lenp points to the length of <ustrp> */
2479
2480 UV original; /* To store the first code point of <p> */
2481
2482 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2483
2484 assert(! UTF8_IS_INVARIANT(*p) && ! UTF8_IS_DOWNGRADEABLE_START(*p));
2485
2486 /* We know immediately if the first character in the string crosses the
2487 * boundary, so can skip */
2488 if (result > 255) {
2489
2490 /* Look at every character in the result; if any cross the
2491 * boundary, the whole thing is disallowed */
2492 U8* s = ustrp + UTF8SKIP(ustrp);
2493 U8* e = ustrp + *lenp;
2494 while (s < e) {
2495 if (UTF8_IS_INVARIANT(*s) || UTF8_IS_DOWNGRADEABLE_START(*s))
2496 {
2497 goto bad_crossing;
2498 }
2499 s += UTF8SKIP(s);
2500 }
2501
2502 /* Here, no characters crossed, result is ok as-is */
2503 return result;
2504 }
2505
2506bad_crossing:
2507
2508 /* Failed, have to return the original */
4b88fb76 2509 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2510 Copy(p, ustrp, *lenp, char);
2511 return original;
2512}
2513
d3e79532 2514/*
87cea99e 2515=for apidoc to_utf8_upper
d3e79532 2516
a1433954
KW
2517Convert the UTF-8 encoded character at C<p> to its uppercase version and
2518store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
89ebb4a3
JH
2519that the ustrp needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2520the uppercase version may be longer than the original character.
d3e79532
JH
2521
2522The first character of the uppercased version is returned
2523(but note, as explained above, that there may be more.)
2524
a1433954
KW
2525The character at C<p> is assumed by this routine to be well-formed.
2526
d3e79532
JH
2527=cut */
2528
051a06d4
KW
2529/* Not currently externally documented, and subject to change:
2530 * <flags> is set iff locale semantics are to be used for code points < 256
2531 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2532 * were used in the calculation; otherwise unchanged. */
2533
2104c8d9 2534UV
051a06d4 2535Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2536{
97aff369 2537 dVAR;
7918f24d 2538
051a06d4
KW
2539 UV result;
2540
2541 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2542
3a4c58c9 2543 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2544 if (flags) {
2545 result = toUPPER_LC(*p);
2546 }
2547 else {
81c6c7ce 2548 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2549 }
3a4c58c9
KW
2550 }
2551 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2552 if (flags) {
2553 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2554 }
2555 else {
81c6c7ce
KW
2556 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2557 ustrp, lenp, 'S');
051a06d4
KW
2558 }
2559 }
2560 else { /* utf8, ord above 255 */
2561 result = CALL_UPPER_CASE(p, ustrp, lenp);
2562
2563 if (flags) {
2564 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2565 }
2566 return result;
2567 }
2568
2569 /* Here, used locale rules. Convert back to utf8 */
2570 if (UTF8_IS_INVARIANT(result)) {
2571 *ustrp = (U8) result;
2572 *lenp = 1;
2573 }
2574 else {
2575 *ustrp = UTF8_EIGHT_BIT_HI(result);
2576 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2577 *lenp = 2;
3a4c58c9
KW
2578 }
2579
051a06d4
KW
2580 if (tainted_ptr) {
2581 *tainted_ptr = TRUE;
2582 }
2583 return result;
983ffd37 2584}
a0ed51b3 2585
d3e79532 2586/*
87cea99e 2587=for apidoc to_utf8_title
d3e79532 2588
a1433954
KW
2589Convert the UTF-8 encoded character at C<p> to its titlecase version and
2590store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2591that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2592titlecase version may be longer than the original character.
d3e79532
JH
2593
2594The first character of the titlecased version is returned
2595(but note, as explained above, that there may be more.)
2596
a1433954
KW
2597The character at C<p> is assumed by this routine to be well-formed.
2598
d3e79532
JH
2599=cut */
2600
051a06d4
KW
2601/* Not currently externally documented, and subject to change:
2602 * <flags> is set iff locale semantics are to be used for code points < 256
2603 * Since titlecase is not defined in POSIX, uppercase is used instead
2604 * for these/
2605 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2606 * were used in the calculation; otherwise unchanged. */
2607
983ffd37 2608UV
051a06d4 2609Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2610{
97aff369 2611 dVAR;
7918f24d 2612
051a06d4
KW
2613 UV result;
2614
2615 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2616
3a4c58c9 2617 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2618 if (flags) {
2619 result = toUPPER_LC(*p);
2620 }
2621 else {
81c6c7ce 2622 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2623 }
3a4c58c9
KW
2624 }
2625 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2626 if (flags) {
2627 result = toUPPER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2628 }
2629 else {
81c6c7ce
KW
2630 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2631 ustrp, lenp, 's');
051a06d4
KW
2632 }
2633 }
2634 else { /* utf8, ord above 255 */
2635 result = CALL_TITLE_CASE(p, ustrp, lenp);
2636
2637 if (flags) {
2638 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2639 }
2640 return result;
2641 }
2642
2643 /* Here, used locale rules. Convert back to utf8 */
2644 if (UTF8_IS_INVARIANT(result)) {
2645 *ustrp = (U8) result;
2646 *lenp = 1;
2647 }
2648 else {
2649 *ustrp = UTF8_EIGHT_BIT_HI(result);
2650 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2651 *lenp = 2;
3a4c58c9
KW
2652 }
2653
051a06d4
KW
2654 if (tainted_ptr) {
2655 *tainted_ptr = TRUE;
2656 }
2657 return result;
a0ed51b3
LW
2658}
2659
d3e79532 2660/*
87cea99e 2661=for apidoc to_utf8_lower
d3e79532 2662
a1433954
KW
2663Convert the UTF-8 encoded character at C<p> to its lowercase version and
2664store that in UTF-8 in ustrp and its length in bytes in C<lenp>. Note
2665that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
89ebb4a3 2666lowercase version may be longer than the original character.
d3e79532
JH
2667
2668The first character of the lowercased version is returned
2669(but note, as explained above, that there may be more.)
2670
a1433954
KW
2671The character at C<p> is assumed by this routine to be well-formed.
2672
d3e79532
JH
2673=cut */
2674
051a06d4
KW
2675/* Not currently externally documented, and subject to change:
2676 * <flags> is set iff locale semantics are to be used for code points < 256
2677 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2678 * were used in the calculation; otherwise unchanged. */
2679
2104c8d9 2680UV
051a06d4 2681Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2682{
051a06d4
KW
2683 UV result;
2684
97aff369 2685 dVAR;
7918f24d 2686
051a06d4 2687 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2688
968c5e6a 2689 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2690 if (flags) {
2691 result = toLOWER_LC(*p);
2692 }
2693 else {
81c6c7ce 2694 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2695 }
968c5e6a
KW
2696 }
2697 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2698 if (flags) {
2699 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2700 }
2701 else {
81c6c7ce
KW
2702 return to_lower_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
2703 ustrp, lenp);
051a06d4 2704 }
968c5e6a 2705 }
051a06d4
KW
2706 else { /* utf8, ord above 255 */
2707 result = CALL_LOWER_CASE(p, ustrp, lenp);
2708
2709 if (flags) {
2710 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2711 }
968c5e6a 2712
051a06d4
KW
2713 return result;
2714 }
2715
2716 /* Here, used locale rules. Convert back to utf8 */
2717 if (UTF8_IS_INVARIANT(result)) {
2718 *ustrp = (U8) result;
2719 *lenp = 1;
2720 }
2721 else {
2722 *ustrp = UTF8_EIGHT_BIT_HI(result);
2723 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2724 *lenp = 2;
2725 }
2726
2727 if (tainted_ptr) {
2728 *tainted_ptr = TRUE;
2729 }
2730 return result;
b4e400f9
JH
2731}
2732
d3e79532 2733/*
87cea99e 2734=for apidoc to_utf8_fold
d3e79532 2735
a1433954
KW
2736Convert the UTF-8 encoded character at C<p> to its foldcase version and
2737store that in UTF-8 in C<ustrp> and its length in bytes in C<lenp>. Note
2738that the C<ustrp> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since the
d3e79532
JH
2739foldcase version may be longer than the original character (up to
2740three characters).
2741
2742The first character of the foldcased version is returned
2743(but note, as explained above, that there may be more.)
2744
a1433954
KW
2745The character at C<p> is assumed by this routine to be well-formed.
2746
d3e79532
JH
2747=cut */
2748
051a06d4
KW
2749/* Not currently externally documented, and subject to change,
2750 * in <flags>
2751 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2752 * points < 256. Since foldcase is not defined in
2753 * POSIX, lowercase is used instead
2754 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2755 * otherwise simple folds
a0270393
KW
2756 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2757 * prohibited
051a06d4
KW
2758 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2759 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2760
b4e400f9 2761UV
051a06d4 2762Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2763{
97aff369 2764 dVAR;
7918f24d 2765
051a06d4
KW
2766 UV result;
2767
36bb2ab6 2768 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2769
a0270393
KW
2770 /* These are mutually exclusive */
2771 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2772
50ba90ff
KW
2773 assert(p != ustrp); /* Otherwise overwrites */
2774
a1dde8de 2775 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2776 if (flags & FOLD_FLAGS_LOCALE) {
2777 result = toLOWER_LC(*p);
2778 }
2779 else {
81c6c7ce
KW
2780 return _to_fold_latin1(*p, ustrp, lenp,
2781 cBOOL(flags & FOLD_FLAGS_FULL));
051a06d4 2782 }
a1dde8de
KW
2783 }
2784 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4
KW
2785 if (flags & FOLD_FLAGS_LOCALE) {
2786 result = toLOWER_LC(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)));
2787 }
2788 else {
81c6c7ce 2789 return _to_fold_latin1(TWO_BYTE_UTF8_TO_UNI(*p, *(p+1)),
a0270393
KW
2790 ustrp, lenp,
2791 cBOOL((flags & FOLD_FLAGS_FULL
2792 /* If ASCII safe, don't allow full
2793 * folding, as that could include SHARP
2794 * S => ss; otherwise there is no
2795 * crossing of ascii/non-ascii in the
2796 * latin1 range */
2797 && ! (flags & FOLD_FLAGS_NOMIX_ASCII))));
051a06d4 2798 }
a1dde8de 2799 }
051a06d4 2800 else { /* utf8, ord above 255 */
a0270393 2801 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2802
051a06d4 2803 if ((flags & FOLD_FLAGS_LOCALE)) {
a0270393 2804 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2805 }
a0270393
KW
2806 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2807 return result;
2808 }
2809 else {
2810 /* This is called when changing the case of a utf8-encoded
2811 * character above the Latin1 range, and the result should not
2812 * contain an ASCII character. */
2813
2814 UV original; /* To store the first code point of <p> */
2815
2816 /* Look at every character in the result; if any cross the
2817 * boundary, the whole thing is disallowed */
2818 U8* s = ustrp;
2819 U8* e = ustrp + *lenp;
2820 while (s < e) {
2821 if (isASCII(*s)) {
2822 /* Crossed, have to return the original */
2823 original = valid_utf8_to_uvchr(p, lenp);
2824 Copy(p, ustrp, *lenp, char);
2825 return original;
2826 }
2827 s += UTF8SKIP(s);
2828 }
051a06d4 2829
a0270393
KW
2830 /* Here, no characters crossed, result is ok as-is */
2831 return result;
2832 }
051a06d4
KW
2833 }
2834
2835 /* Here, used locale rules. Convert back to utf8 */
2836 if (UTF8_IS_INVARIANT(result)) {
2837 *ustrp = (U8) result;
2838 *lenp = 1;
2839 }
2840 else {
2841 *ustrp = UTF8_EIGHT_BIT_HI(result);
2842 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2843 *lenp = 2;
2844 }
2845
2846 if (tainted_ptr) {
2847 *tainted_ptr = TRUE;
2848 }
2849 return result;
a0ed51b3
LW
2850}
2851
711a919c 2852/* Note:
f90a9a02 2853 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2854 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2855 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2856 */
c4a5db0c 2857
a0ed51b3 2858SV*
7fc63493 2859Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2860{
c4a5db0c
KW
2861 PERL_ARGS_ASSERT_SWASH_INIT;
2862
2863 /* Returns a copy of a swash initiated by the called function. This is the
2864 * public interface, and returning a copy prevents others from doing
2865 * mischief on the original */
2866
9a53f6cf 2867 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, FALSE, NULL, FALSE));
c4a5db0c
KW
2868}
2869
2870SV*
9a53f6cf 2871Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, bool return_if_undef, SV* invlist, bool passed_in_invlist_has_user_defined_property)
c4a5db0c
KW
2872{
2873 /* Initialize and return a swash, creating it if necessary. It does this
9a53f6cf 2874 * by calling utf8_heavy.pl in the general case.
c4a5db0c
KW
2875 *
2876 * This interface should only be used by functions that won't destroy or
2877 * adversely change the swash, as doing so affects all other uses of the
2878 * swash in the program; the general public should use 'Perl_swash_init'
2879 * instead.
2880 *
2881 * pkg is the name of the package that <name> should be in.
2882 * name is the name of the swash to find. Typically it is a Unicode
2883 * property name, including user-defined ones
2884 * listsv is a string to initialize the swash with. It must be of the form
2885 * documented as the subroutine return value in
2886 * L<perlunicode/User-Defined Character Properties>
2887 * minbits is the number of bits required to represent each data element.
2888 * It is '1' for binary properties.
2889 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf
KW
2890 * return_if_undef is TRUE if the routine shouldn't croak if it can't find
2891 * the requested property
2892 * invlist is an inversion list to initialize the swash with (or NULL)
2893 * has_user_defined_property is TRUE if <invlist> has some component that
2894 * came from a user-defined property
2895 *
2896 * Thus there are three possible inputs to find the swash: <name>,
2897 * <listsv>, and <invlist>. At least one must be specified. The result
2898 * will be the union of the specified ones, although <listsv>'s various
2899 * actions can intersect, etc. what <name> gives.
2900 *
2901 * <invlist> is only valid for binary properties */
c4a5db0c 2902
27da23d5 2903 dVAR;
c4a5db0c 2904 SV* retval = &PL_sv_undef;
9a53f6cf
KW
2905
2906 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2907 assert(! invlist || minbits == 1);
2908
2909 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2910 * so */
2911 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2912 dSP;
2913 const size_t pkg_len = strlen(pkg);
2914 const size_t name_len = strlen(name);
2915 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2916 SV* errsv_save;
2917 GV *method;
2918
2919 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2920
2921 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2922 ENTER;
69794297
KW
2923 SAVEHINTS();
2924 save_re_context();
2925 if (PL_parser && PL_parser->error_count)
2926 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2927 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2928 if (!method) { /* demand load utf8 */
2929 ENTER;
2930 errsv_save = newSVsv(ERRSV);
2931 /* It is assumed that callers of this routine are not passing in
2932 * any user derived data. */
2933 /* Need to do this after save_re_context() as it will set
2934 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2935 * in Perl_magic_get). Even line to create errsv_save can turn on
2936 * PL_tainted. */
2937 SAVEBOOL(PL_tainted);
2938 PL_tainted = 0;
2939 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2940 NULL);
2941 if (!SvTRUE(ERRSV))
2942 sv_setsv(ERRSV, errsv_save);
2943 SvREFCNT_dec(errsv_save);
2944 LEAVE;
2945 }
2946 SPAGAIN;
2947 PUSHMARK(SP);
2948 EXTEND(SP,5);
2949 mPUSHp(pkg, pkg_len);
2950 mPUSHp(name, name_len);
2951 PUSHs(listsv);
2952 mPUSHi(minbits);
2953 mPUSHi(none);
2954 PUTBACK;
f8be5cf0 2955 errsv_save = newSVsv(ERRSV);
69794297
KW
2956 /* If we already have a pointer to the method, no need to use
2957 * call_method() to repeat the lookup. */
2958 if (method ? call_sv(MUTABLE_SV(method), G_SCALAR)
2959 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2960 {
2961 retval = *PL_stack_sp--;
2962 SvREFCNT_inc(retval);
2963 }
f8be5cf0
JH
2964 if (!SvTRUE(ERRSV))
2965 sv_setsv(ERRSV, errsv_save);
2966 SvREFCNT_dec(errsv_save);
ce3b816e 2967 LEAVE;
69794297
KW
2968 POPSTACK;
2969 if (IN_PERL_COMPILETIME) {
2970 CopHINTS_set(PL_curcop, PL_hints);
2971 }
2972 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2973 if (SvPOK(retval))
2974
2975 /* If caller wants to handle missing properties, let them */
2976 if (return_if_undef) {
2977 return NULL;
2978 }
2979 Perl_croak(aTHX_
2980 "Can't find Unicode property definition \"%"SVf"\"",
2981 SVfARG(retval));
2982 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2983 }
9a53f6cf 2984 } /* End of calling the module to find the swash */
36eb48b4
KW
2985
2986 /* Make sure there is an inversion list for binary properties */
2987 if (minbits == 1) {
2988 SV** swash_invlistsvp = NULL;
2989 SV* swash_invlist = NULL;
9a53f6cf 2990 bool invlist_in_swash_is_valid = FALSE;
88d45d28 2991 HV* swash_hv = NULL;
36eb48b4 2992
9a53f6cf
KW
2993 /* If this operation fetched a swash, get its already existing
2994 * inversion list or create one for it */
2995 if (retval != &PL_sv_undef) {
36eb48b4
KW
2996 swash_hv = MUTABLE_HV(SvRV(retval));
2997
2998 swash_invlistsvp = hv_fetchs(swash_hv, "INVLIST", FALSE);
9a53f6cf
KW
2999 if (swash_invlistsvp) {
3000 swash_invlist = *swash_invlistsvp;
3001 invlist_in_swash_is_valid = TRUE;
3002 }
3003 else {
36eb48b4 3004 swash_invlist = _swash_to_invlist(retval);
9a53f6cf
KW
3005 }
3006 }
3007
3008 /* If an inversion list was passed in, have to include it */
3009 if (invlist) {
3010
3011 /* Any fetched swash will by now have an inversion list in it;
3012 * otherwise <swash_invlist> will be NULL, indicating that we
3013 * didn't fetch a swash */
3014 if (swash_invlist) {
3015
3016 /* Add the passed-in inversion list, which invalidates the one
3017 * already stored in the swash */
3018 invlist_in_swash_is_valid = FALSE;
3019 _invlist_union(invlist, swash_invlist, &swash_invlist);
3020 }
3021 else {
3022
3023 /* Here, there is no swash already. Set up a minimal one */
3024 swash_hv = newHV();
3025 retval = newRV_inc(MUTABLE_SV(swash_hv));
3026 swash_invlist = invlist;
3027 }
3028
3029 if (passed_in_invlist_has_user_defined_property) {
3030 if (! hv_stores(swash_hv, "USER_DEFINED", newSVuv(1))) {
3031 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3032 }
3033 }
3034 }
3035
3036 /* Here, we have computed the union of all the passed-in data. It may
3037 * be that there was an inversion list in the swash which didn't get
3038 * touched; otherwise save the one computed one */
3039 if (! invlist_in_swash_is_valid) {
69794297
KW
3040 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "INVLIST", swash_invlist))
3041 {
3042 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3043 }
9a53f6cf 3044 }
36eb48b4
KW
3045 }
3046
a0ed51b3
LW
3047 return retval;
3048}
3049
035d37be
JH
3050
3051/* This API is wrong for special case conversions since we may need to
3052 * return several Unicode characters for a single Unicode character
3053 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3054 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
3055 * multiple values. --jhi
3056 * For those, you should use to_utf8_case() instead */
b0e3252e 3057/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3058
3059/* Note:
3060 * Returns the value of property/mapping C<swash> for the first character
3061 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3062 * assumed to be in utf8. If C<do_utf8> is false, the string C<ptr> is
3063 * assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3064 *
3065 * A "swash" is a hash which contains initially the keys/values set up by
3066 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3067 * property for all possible code points. Things are stored in a compact form
3068 * (see utf8_heavy.pl) so that calculation is required to find the actual
3069 * property value for a given code point. As code points are looked up, new
3070 * key/value pairs are added to the hash, so that the calculation doesn't have
3071 * to ever be re-done. Further, each calculation is done, not just for the
3072 * desired one, but for a whole block of code points adjacent to that one.
3073 * For binary properties on ASCII machines, the block is usually for 64 code
3074 * points, starting with a code point evenly divisible by 64. Thus if the
3075 * property value for code point 257 is requested, the code goes out and
3076 * calculates the property values for all 64 code points between 256 and 319,
3077 * and stores these as a single 64-bit long bit vector, called a "swatch",
3078 * under the key for code point 256. The key is the UTF-8 encoding for code
3079 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3080 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3081 * for code point 258 is then requested, this code realizes that it would be
3082 * stored under the key for 256, and would find that value and extract the
3083 * relevant bit, offset from 256.
3084 *
3085 * Non-binary properties are stored in as many bits as necessary to represent
3086 * their values (32 currently, though the code is more general than that), not
3087 * as single bits, but the principal is the same: the value for each key is a
3088 * vector that encompasses the property values for all code points whose UTF-8
3089 * representations are represented by the key. That is, for all code points
3090 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3091 * bytes of that.
680c470c 3092 */
a0ed51b3 3093UV
680c470c 3094Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3095{
27da23d5 3096 dVAR;
ef8f7699 3097 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3098 U32 klen;
3099 U32 off;
a0ed51b3 3100 STRLEN slen;
7d85a32c 3101 STRLEN needents;
cfd0369c 3102 const U8 *tmps = NULL;
a0ed51b3 3103 U32 bit;
979f2922 3104 SV *swatch;
3568d838 3105 U8 tmputf8[2];
35da51f7 3106 const UV c = NATIVE_TO_ASCII(*ptr);
3568d838 3107
7918f24d
NC
3108 PERL_ARGS_ASSERT_SWASH_FETCH;
3109
dbe7a391 3110 /* Convert to utf8 if not already */
3568d838 3111 if (!do_utf8 && !UNI_IS_INVARIANT(c)) {
979f2922
ST
3112 tmputf8[0] = (U8)UTF8_EIGHT_BIT_HI(c);
3113 tmputf8[1] = (U8)UTF8_EIGHT_BIT_LO(c);
3114 ptr = tmputf8;
3568d838
JH
3115 }
3116 /* Given a UTF-X encoded char 0xAA..0xYY,0xZZ
37e2e78e 3117 * then the "swatch" is a vec() for all the chars which start
3568d838
JH
3118 * with 0xAA..0xYY
3119 * So the key in the hash (klen) is length of encoded char -1
3120 */
3121 klen = UTF8SKIP(ptr) - 1;
3122 off = ptr[klen];
a0ed51b3 3123
979f2922 3124 if (klen == 0) {
37e2e78e 3125 /* If char is invariant then swatch is for all the invariant chars
1e54db1a 3126 * In both UTF-8 and UTF-8-MOD that happens to be UTF_CONTINUATION_MARK
7d85a32c 3127 */
979f2922
ST
3128 needents = UTF_CONTINUATION_MARK;
3129 off = NATIVE_TO_UTF(ptr[klen]);
3130 }
3131 else {
7d85a32c 3132 /* If char is encoded then swatch is for the prefix */
979f2922
ST
3133 needents = (1 << UTF_ACCUMULATION_SHIFT);
3134 off = NATIVE_TO_UTF(ptr[klen]) & UTF_CONTINUATION_MASK;
3135 }
7d85a32c 3136
a0ed51b3
LW
3137 /*
3138 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3139 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3140 * it's nothing to sniff at.) Pity we usually come through at least
3141 * two function calls to get here...
3142 *
3143 * NB: this code assumes that swatches are never modified, once generated!
3144 */
3145
3568d838 3146 if (hv == PL_last_swash_hv &&
a0ed51b3 3147 klen == PL_last_swash_klen &&
27da23d5 3148 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3149 {
3150 tmps = PL_last_swash_tmps;
3151 slen = PL_last_swash_slen;
3152 }
3153 else {
3154 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3155 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3156
b0e3252e 3157 /* If not cached, generate it via swatch_get */
979f2922
ST
3158 if (!svp || !SvPOK(*svp)
3159 || !(tmps = (const U8*)SvPV_const(*svp, slen))) {
2b9d42f0
NIS
3160 /* We use utf8n_to_uvuni() as we want an index into
3161 Unicode tables, not a native character number.
3162 */
aec46f14 3163 const UV code_point = utf8n_to_uvuni(ptr, UTF8_MAXBYTES, 0,
872c91ae
JH
3164 ckWARN(WARN_UTF8) ?
3165 0 : UTF8_ALLOW_ANY);
b0e3252e 3166 swatch = swatch_get(swash,
979f2922 3167 /* On EBCDIC & ~(0xA0-1) isn't a useful thing to do */
361ee0fe 3168 (klen) ? (code_point & ~((UV)needents - 1)) : 0,
979f2922
ST
3169 needents);
3170
923e4eb5 3171 if (IN_PERL_COMPILETIME)
623e6609 3172 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3173
979f2922 3174 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3175
979f2922
ST
3176 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3177 || (slen << 3) < needents)
5637ef5b
NC
3178 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3179 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3180 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3181 }
3182
3183 PL_last_swash_hv = hv;
16d8f38a 3184 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3185 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3186 /* FIXME change interpvar.h? */
3187 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3188 PL_last_swash_slen = slen;
3189 if (klen)
3190 Copy(ptr, PL_last_swash_key, klen, U8);
3191 }
3192
9faf8d75 3193 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3194 case 1:
3195 bit = 1 << (off & 7);
3196 off >>= 3;
3197 return (tmps[off] & bit) != 0;
3198 case 8:
3199 return tmps[off];
3200 case 16:
3201 off <<= 1;
3202 return (tmps[off] << 8) + tmps[off + 1] ;
3203 case 32:
3204 off <<= 2;
3205 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3206 }
5637ef5b
NC
3207 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3208 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3209 NORETURN_FUNCTION_END;
a0ed51b3 3210}
2b9d42f0 3211
319009ee
KW
3212/* Read a single line of the main body of the swash input text. These are of
3213 * the form:
3214 * 0053 0056 0073
3215 * where each number is hex. The first two numbers form the minimum and
3216 * maximum of a range, and the third is the value associated with the range.
3217 * Not all swashes should have a third number
3218 *
3219 * On input: l points to the beginning of the line to be examined; it points
3220 * to somewhere in the string of the whole input text, and is
3221 * terminated by a \n or the null string terminator.
3222 * lend points to the null terminator of that string
3223 * wants_value is non-zero if the swash expects a third number
3224 * typestr is the name of the swash's mapping, like 'ToLower'
3225 * On output: *min, *max, and *val are set to the values read from the line.
3226 * returns a pointer just beyond the line examined. If there was no
3227 * valid min number on the line, returns lend+1
3228 */
3229
3230STATIC U8*
3231S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3232 const bool wants_value, const U8* const typestr)
3233{
3234 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3235 STRLEN numlen; /* Length of the number */
02470786
KW
3236 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3237 | PERL_SCAN_DISALLOW_PREFIX
3238 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3239
3240 /* nl points to the next \n in the scan */
3241 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3242
3243 /* Get the first number on the line: the range minimum */
3244 numlen = lend - l;
3245 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3246 if (numlen) /* If found a hex number, position past it */
3247 l += numlen;
3248 else if (nl) { /* Else, go handle next line, if any */
3249 return nl + 1; /* 1 is length of "\n" */
3250 }
3251 else { /* Else, no next line */
3252 return lend + 1; /* to LIST's end at which \n is not found */
3253 }
3254
3255 /* The max range value follows, separated by a BLANK */
3256 if (isBLANK(*l)) {
3257 ++l;
02470786
KW
3258 flags = PERL_SCAN_SILENT_ILLDIGIT
3259 | PERL_SCAN_DISALLOW_PREFIX
3260 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3261 numlen = lend - l;
3262 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3263 if (numlen)
3264 l += numlen;
3265 else /* If no value here, it is a single element range */
3266 *max = *min;
3267
3268 /* Non-binary tables have a third entry: what the first element of the
3269 * range maps to */
3270 if (wants_value) {
3271 if (isBLANK(*l)) {
3272 ++l;
f90a9a02
KW
3273
3274 /* The ToLc, etc table mappings are not in hex, and must be
3275 * corrected by adding the code point to them */
3276 if (typeto) {
3277 char *after_strtol = (char *) lend;
3278 *val = Strtol((char *)l, &after_strtol, 10);
3279 l = (U8 *) after_strtol;
f90a9a02
KW
3280 }
3281 else { /* Other tables are in hex, and are the correct result
3282 without tweaking */
a9d188b3
KW
3283 flags = PERL_SCAN_SILENT_ILLDIGIT
3284 | PERL_SCAN_DISALLOW_PREFIX
3285 | PERL_SCAN_SILENT_NON_PORTABLE;
3286 numlen = lend - l;
3287 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3288 if (numlen)
3289 l += numlen;
3290 else
3291 *val = 0;
f90a9a02 3292 }
319009ee
KW
3293 }
3294 else {
3295 *val = 0;
3296 if (typeto) {
dcbac5bb 3297 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3298 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3299 typestr, l);
3300 }
3301 }
3302 }
3303 else
3304 *val = 0; /* bits == 1, then any val should be ignored */
3305 }
3306 else { /* Nothing following range min, should be single element with no
3307 mapping expected */
3308 *max = *min;
3309 if (wants_value) {
3310 *val = 0;
3311 if (typeto) {
dcbac5bb 3312 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3313 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3314 }
3315 }
3316 else
3317 *val = 0; /* bits == 1, then val should be ignored */
3318 }
3319
3320 /* Position to next line if any, or EOF */
3321 if (nl)
3322 l = nl + 1;
3323 else
3324 l = lend;
3325
3326 return l;
3327}
3328
979f2922
ST
3329/* Note:
3330 * Returns a swatch (a bit vector string) for a code point sequence
3331 * that starts from the value C<start> and comprises the number C<span>.
3332 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3333 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3334 */
3335STATIC SV*
b0e3252e 3336S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3337{
3338 SV *swatch;
77f9f126 3339 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3340 STRLEN lcur, xcur, scur;
ef8f7699 3341 HV *const hv = MUTABLE_HV(SvRV(swash));
36eb48b4
KW
3342 SV** const invlistsvp = hv_fetchs(hv, "INVLIST", FALSE);
3343
88d45d28
KW
3344 SV** listsvp = NULL; /* The string containing the main body of the table */
3345 SV** extssvp = NULL;
3346 SV** invert_it_svp = NULL;
3347 U8* typestr = NULL;
786861f5
KW
3348 STRLEN bits;
3349 STRLEN octets; /* if bits == 1, then octets == 0 */
3350 UV none;
3351 UV end = start + span;
972dd592 3352
36eb48b4 3353 if (invlistsvp == NULL) {
786861f5
KW
3354 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3355 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3356 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3357 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3358 listsvp = hv_fetchs(hv, "LIST", FALSE);
3359 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3360
3361 bits = SvUV(*bitssvp);
3362 none = SvUV(*nonesvp);
3363 typestr = (U8*)SvPV_nolen(*typesvp);
3364 }
36eb48b4
KW
3365 else {
3366 bits = 1;
3367 none = 0;
3368 }
786861f5 3369 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3370
b0e3252e 3371 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3372
979f2922 3373 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3374 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3375 (UV)bits);
979f2922
ST
3376 }
3377
84ea5ef6
KW
3378 /* If overflowed, use the max possible */
3379 if (end < start) {
3380 end = UV_MAX;
3381 span = end - start;
3382 }
3383
979f2922 3384 /* create and initialize $swatch */
979f2922 3385 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3386 swatch = newSV(scur);
3387 SvPOK_on(swatch);
979f2922
ST
3388 s = (U8*)SvPVX(swatch);
3389 if (octets && none) {
0bd48802 3390 const U8* const e = s + scur;
979f2922
ST
3391 while (s < e) {
3392 if (bits == 8)
3393 *s++ = (U8)(none & 0xff);
3394 else if (bits == 16) {
3395 *s++ = (U8)((none >> 8) & 0xff);
3396 *s++ = (U8)( none & 0xff);
3397 }
3398 else if (bits == 32) {
3399 *s++ = (U8)((none >> 24) & 0xff);
3400 *s++ = (U8)((none >> 16) & 0xff);
3401 *s++ = (U8)((none >> 8) & 0xff);
3402 *s++ = (U8)( none & 0xff);
3403 }
3404 }
3405 *s = '\0';
3406 }
3407 else {
3408 (void)memzero((U8*)s, scur + 1);
3409 }
3410 SvCUR_set(swatch, scur);
3411 s = (U8*)SvPVX(swatch);
3412
36eb48b4
KW
3413 if (invlistsvp) { /* If has an inversion list set up use that */
3414 _invlist_populate_swatch(*invlistsvp, start, end, s);
3415 return swatch;
3416 }
3417
3418 /* read $swash->{LIST} */
979f2922
ST
3419 l = (U8*)SvPV(*listsvp, lcur);
3420 lend = l + lcur;
3421 while (l < lend) {
8ed25d53 3422 UV min, max, val, upper;
319009ee
KW
3423 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3424 cBOOL(octets), typestr);
3425 if (l > lend) {
979f2922
ST
3426 break;
3427 }
3428
972dd592 3429 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3430 if (max < start)
3431 continue;
3432
8ed25d53
KW
3433 /* <end> is generally 1 beyond where we want to set things, but at the
3434 * platform's infinity, where we can't go any higher, we want to
3435 * include the code point at <end> */
3436 upper = (max < end)
3437 ? max
3438 : (max != UV_MAX || end != UV_MAX)
3439 ? end - 1
3440 : end;
3441
979f2922 3442 if (octets) {
35da51f7 3443 UV key;
979f2922
ST
3444 if (min < start) {
3445 if (!none || val < none) {
3446 val += start - min;
3447 }
3448 min = start;
3449 }
8ed25d53 3450 for (key = min; key <= upper; key++) {
979f2922 3451 STRLEN offset;
979f2922
ST
3452 /* offset must be non-negative (start <= min <= key < end) */
3453 offset = octets * (key - start);
3454 if (bits == 8)
3455 s[offset] = (U8)(val & 0xff);
3456 else if (bits == 16) {
3457 s[offset ] = (U8)((val >> 8) & 0xff);
3458 s[offset + 1] = (U8)( val & 0xff);
3459 }
3460 else if (bits == 32) {
3461 s[offset ] = (U8)((val >> 24) & 0xff);
3462 s[offset + 1] = (U8)((val >> 16) & 0xff);
3463 s[offset + 2] = (U8)((val >> 8) & 0xff);
3464 s[offset + 3] = (U8)( val & 0xff);
3465 }
3466
3467 if (!none || val < none)
3468 ++val;
3469 }
3470 }
711a919c 3471 else { /* bits == 1, then val should be ignored */
35da51f7 3472 UV key;
979f2922
ST
3473 if (min < start)
3474 min = start;
6cb05c12 3475
8ed25d53 3476 for (key = min; key <= upper; key++) {
0bd48802 3477 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3478 s[offset >> 3] |= 1 << (offset & 7);
3479 }
3480 }
3481 } /* while */
979f2922 3482
9479a769 3483 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3484 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3485
3486 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3487 * be 0, and their inversion should also be 0, as we don't succeed any
3488 * Unicode property matches for non-Unicode code points */
3489 if (start <= PERL_UNICODE_MAX) {
3490
3491 /* The code below assumes that we never cross the
3492 * Unicode/above-Unicode boundary in a range, as otherwise we would
3493 * have to figure out where to stop flipping the bits. Since this
3494 * boundary is divisible by a large power of 2, and swatches comes
3495 * in small powers of 2, this should be a valid assumption */
3496 assert(start + span - 1 <= PERL_UNICODE_MAX);
3497
507a8485
KW
3498 send = s + scur;
3499 while (s < send) {
3500 *s = ~(*s);
3501 s++;
3502 }
0bda3001 3503 }
77f9f126
KW
3504 }
3505
d73c39c5
KW
3506 /* read $swash->{EXTRAS}
3507 * This code also copied to swash_to_invlist() below */
979f2922
ST
3508 x = (U8*)SvPV(*extssvp, xcur);
3509 xend = x + xcur;
3510 while (x < xend) {
3511 STRLEN namelen;
3512 U8 *namestr;
3513 SV** othersvp;
3514 HV* otherhv;
3515 STRLEN otherbits;
3516 SV **otherbitssvp, *other;
711a919c 3517 U8 *s, *o, *nl;
979f2922
ST
3518 STRLEN slen, olen;
3519
35da51f7 3520 const U8 opc = *x++;
979f2922
ST
3521 if (opc == '\n')
3522 continue;
3523
3524 nl = (U8*)memchr(x, '\n', xend - x);
3525
3526 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3527 if (nl) {
3528 x = nl + 1; /* 1 is length of "\n" */
3529 continue;
3530 }
3531 else {
3532 x = xend; /* to EXTRAS' end at which \n is not found */
3533 break;
3534 }
3535 }
3536
3537 namestr = x;
3538 if (nl) {
3539 namelen = nl - namestr;
3540 x = nl + 1;
3541 }
3542 else {
3543 namelen = xend - namestr;
3544 x = xend;
3545 }
3546
3547 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3548 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3549 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3550 otherbits = (STRLEN)SvUV(*otherbitssvp);
3551 if (bits < otherbits)
5637ef5b
NC
3552 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3553 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3554
3555 /* The "other" swatch must be destroyed after. */
b0e3252e 3556 other = swatch_get(*othersvp, start, span);
979f2922
ST
3557 o = (U8*)SvPV(other, olen);
3558
3559 if (!olen)
b0e3252e 3560 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3561
3562 s = (U8*)SvPV(swatch, slen);
3563 if (bits == 1 && otherbits == 1) {
3564 if (slen != olen)
5637ef5b
NC
3565 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3566 "mismatch, slen=%"UVuf", olen=%"UVuf,
3567 (UV)slen, (UV)olen);
979f2922
ST
3568
3569 switch (opc) {
3570 case '+':
3571 while (slen--)
3572 *s++ |= *o++;
3573 break;
3574 case '!':
3575 while (slen--)
3576 *s++ |= ~*o++;
3577 break;
3578 case '-':
3579 while (slen--)
3580 *s++ &= ~*o++;
3581 break;
3582 case '&':
3583 while (slen--)
3584 *s++ &= *o++;
3585 break;
3586 default:
3587 break;
3588 }
3589 }
711a919c 3590 else {
979f2922
ST
3591 STRLEN otheroctets = otherbits >> 3;
3592 STRLEN offset = 0;
35da51f7 3593 U8* const send = s + slen;
979f2922
ST
3594
3595 while (s < send) {
3596 UV otherval = 0;
3597
3598 if (otherbits == 1) {
3599 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3600 ++offset;
3601 }
3602 else {
3603 STRLEN vlen = otheroctets;
3604 otherval = *o++;
3605 while (--vlen) {
3606 otherval <<= 8;
3607 otherval |= *o++;
3608 }
3609 }
3610
711a919c 3611 if (opc == '+' && otherval)
6f207bd3 3612 NOOP; /* replace with otherval */
979f2922
ST
3613 else if (opc == '!' && !otherval)
3614 otherval = 1;
3615 else if (opc == '-' && otherval)
3616 otherval = 0;
3617 else if (opc == '&' && !otherval)
3618 otherval = 0;
3619 else {
711a919c 3620 s += octets; /* no replacement */
979f2922
ST
3621 continue;
3622 }
3623
3624 if (bits == 8)
3625 *s++ = (U8)( otherval & 0xff);
3626 else if (bits == 16) {
3627 *s++ = (U8)((otherval >> 8) & 0xff);
3628 *s++ = (U8)( otherval & 0xff);
3629 }
3630 else if (bits == 32) {
3631 *s++ = (U8)((otherval >> 24) & 0xff);
3632 *s++ = (U8)((otherval >> 16) & 0xff);
3633 *s++ = (U8)((otherval >> 8) & 0xff);
3634 *s++ = (U8)( otherval & 0xff);
3635 }
3636 }
3637 }
3638 sv_free(other); /* through with it! */
3639 } /* while */
3640 return swatch;
3641}
3642
064c021d 3643HV*
4c2e1131 3644Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3645{
3646
5662e334
KW
3647 /* Subject to change or removal. For use only in one place in regcomp.c.
3648 * Can't be used on a property that is subject to user override, as it
3649 * relies on the value of SPECIALS in the swash which would be set by
3650 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3651 * for overridden properties
064c021d
KW
3652 *
3653 * Returns a hash which is the inversion and closure of a swash mapping.
3654 * For example, consider the input lines:
3655 * 004B 006B
3656 * 004C 006C
3657 * 212A 006B
3658 *
3659 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3660 * 006C. The value for each key is an array. For 006C, the array would
3661 * have a two elements, the utf8 for itself, and for 004C. For 006B, there
3662 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3663 *
3664 * Essentially, for any code point, it gives all the code points that map to
3665 * it, or the list of 'froms' for that point.
3666 *
5662e334
KW
3667 * Currently it ignores any additions or deletions from other swashes,
3668 * looking at just the main body of the swash, and if there are SPECIALS
3669 * in the swash, at that hash
3670 *
3671 * The specials hash can be extra code points, and most likely consists of
3672 * maps from single code points to multiple ones (each expressed as a string
3673 * of utf8 characters). This function currently returns only 1-1 mappings.
3674 * However consider this possible input in the specials hash:
3675 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3676 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3677 *
3678 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3679 * currently handle. But it also means that FB05 and FB06 are equivalent in
3680 * a 1-1 mapping which we should handle, and this relationship may not be in
3681 * the main table. Therefore this function examines all the multi-char
3682 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3683
3684 U8 *l, *lend;
3685 STRLEN lcur;
3686 HV *const hv = MUTABLE_HV(SvRV(swash));
3687
3688 /* The string containing the main body of the table */
3689 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3690
3691 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3692 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3693 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3694 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3695 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3696 const STRLEN bits = SvUV(*bitssvp);
3697 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3698 const UV none = SvUV(*nonesvp);
5662e334 3699 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3700
3701 HV* ret = newHV();
3702
3703 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3704
3705 /* Must have at least 8 bits to get the mappings */
3706 if (bits != 8 && bits != 16 && bits != 32) {
3707 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3708 (UV)bits);
3709 }
3710
5662e334
KW
3711 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3712 mapping to more than one character */
3713
3714 /* Construct an inverse mapping hash for the specials */
3715 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3716 HV * specials_inverse = newHV();
3717 char *char_from; /* the lhs of the map */
3718 I32 from_len; /* its byte length */
3719 char *char_to; /* the rhs of the map */
3720 I32 to_len; /* its byte length */
3721 SV *sv_to; /* and in a sv */
3722 AV* from_list; /* list of things that map to each 'to' */
3723
3724 hv_iterinit(specials_hv);
3725
3726 /* The keys are the characters (in utf8) that map to the corresponding
3727 * utf8 string value. Iterate through the list creating the inverse
3728 * list. */
3729 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3730 SV** listp;
3731 if (! SvPOK(sv_to)) {
5637ef5b
NC
3732 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3733 "unexpectedly is not a string, flags=%lu",
3734 (unsigned long)SvFLAGS(sv_to));
5662e334 3735 }
4b88fb76 3736 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %"UVXf", First char of to is %"UVXf"\n", valid_utf8_to_uvchr((U8*) char_from, 0), valid_utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
5662e334
KW
3737
3738 /* Each key in the inverse list is a mapped-to value, and the key's
3739 * hash value is a list of the strings (each in utf8) that map to
3740 * it. Those strings are all one character long */
3741 if ((listp = hv_fetch(specials_inverse,
3742 SvPVX(sv_to),
3743 SvCUR(sv_to), 0)))
3744 {
3745 from_list = (AV*) *listp;
3746 }
3747 else { /* No entry yet for it: create one */
3748 from_list = newAV();
3749 if (! hv_store(specials_inverse,
3750 SvPVX(sv_to),
3751 SvCUR(sv_to),
3752 (SV*) from_list, 0))
3753 {
3754 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3755 }
3756 }
3757
3758 /* Here have the list associated with this 'to' (perhaps newly
3759 * created and empty). Just add to it. Note that we ASSUME that
3760 * the input is guaranteed to not have duplications, so we don't
3761 * check for that. Duplications just slow down execution time. */
3762 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3763 }
3764
3765 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3766 * it looking for cases like the FB05/FB06 examples above. There would
3767 * be an entry in the hash like
3768 * 'st' => [ FB05, FB06 ]