This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: Reorder cases in switch to eliminate a goto
[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
27da23d5
JH
36static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 38
48ef279e 39/*
ccfc67b7 40=head1 Unicode Support
a0ed51b3 41
166f8a29
DM
42This file contains various utility functions for manipulating UTF8-encoded
43strings. For the uninitiated, this is a method of representing arbitrary
61296642 44Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
45characters in the ASCII range are unmodified, and a zero byte never appears
46within non-zero characters.
166f8a29 47
eaf7a4d2
CS
48=cut
49*/
50
51/*
52=for apidoc is_ascii_string
53
a1433954 54Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
55or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
56is, if they are invariant. On ASCII-ish machines, only ASCII characters
57fit this definition, hence the function's name.
eaf7a4d2 58
9f7e3d64
MH
59If C<len> is 0, it will be calculated using C<strlen(s)>.
60
a1433954 61See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
62
63=cut
64*/
65
66bool
668b6d8d 67Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
68{
69 const U8* const send = s + (len ? len : strlen((const char *)s));
70 const U8* x = s;
71
72 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
73
74 for (; x < send; ++x) {
75 if (!UTF8_IS_INVARIANT(*x))
76 break;
77 }
78
79 return x == send;
80}
81
82/*
378516de 83=for apidoc uvoffuni_to_utf8_flags
eebe1485 84
a27992cc 85THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
86Instead, B<Almost all code should use L</uvchr_to_utf8> or
87L</uvchr_to_utf8_flags>>.
a27992cc 88
de69f3af
KW
89This function is like them, but the input is a strict Unicode
90(as opposed to native) code point. Only in very rare circumstances should code
91not be using the native code point.
949cf498 92
de69f3af 93For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 94
eebe1485
SC
95=cut
96*/
97
dfe13c55 98U8 *
378516de 99Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 100{
378516de 101 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 102
d9432125
KW
103 if (UNI_IS_INVARIANT(uv)) {
104 *d++ = (U8) LATIN1_TO_NATIVE(uv);
105 return d;
106 }
107
979f77b6
KW
108 /* The first problematic code point is the first surrogate */
109 if (uv >= UNICODE_SURROGATE_FIRST
110 && ckWARN4_d(WARN_UTF8, WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
111 {
949cf498
KW
112 if (UNICODE_IS_SURROGATE(uv)) {
113 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 114 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
115 "UTF-16 surrogate U+%04"UVXf, uv);
116 }
117 if (flags & UNICODE_DISALLOW_SURROGATE) {
118 return NULL;
119 }
120 }
121 else if (UNICODE_IS_SUPER(uv)) {
122 if (flags & UNICODE_WARN_SUPER
123 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
124 {
8457b38f 125 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
126 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
127 }
128 if (flags & UNICODE_DISALLOW_SUPER
129 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
130 {
131 return NULL;
132 }
133 }
134 else if (UNICODE_IS_NONCHAR(uv)) {
135 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 136 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
137 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
138 uv);
139 }
140 if (flags & UNICODE_DISALLOW_NONCHAR) {
141 return NULL;
142 }
143 }
507b9800 144 }
d9432125 145
2d331972 146#if defined(EBCDIC)
d9432125 147 {
5aaebcb3 148 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
149 U8 *p = d+len-1;
150 while (p > d) {
bc3632a8 151 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
152 uv >>= UTF_ACCUMULATION_SHIFT;
153 }
bc3632a8 154 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
155 return d+len;
156 }
157#else /* Non loop style */
a0ed51b3 158 if (uv < 0x800) {
eb160463
GS
159 *d++ = (U8)(( uv >> 6) | 0xc0);
160 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
161 return d;
162 }
163 if (uv < 0x10000) {
eb160463
GS
164 *d++ = (U8)(( uv >> 12) | 0xe0);
165 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
166 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
167 return d;
168 }
169 if (uv < 0x200000) {
eb160463
GS
170 *d++ = (U8)(( uv >> 18) | 0xf0);
171 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
172 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
173 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
174 return d;
175 }
176 if (uv < 0x4000000) {
eb160463
GS
177 *d++ = (U8)(( uv >> 24) | 0xf8);
178 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
179 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
180 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
181 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
182 return d;
183 }
184 if (uv < 0x80000000) {
eb160463
GS
185 *d++ = (U8)(( uv >> 30) | 0xfc);
186 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
187 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
188 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
189 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
190 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
191 return d;
192 }
6b8eaf93 193#ifdef HAS_QUAD
d7578b48 194 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
195#endif
196 {
eb160463
GS
197 *d++ = 0xfe; /* Can't match U+FEFF! */
198 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
199 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
204 return d;
205 }
6b8eaf93 206#ifdef HAS_QUAD
a0ed51b3 207 {
eb160463
GS
208 *d++ = 0xff; /* Can't match U+FFFE! */
209 *d++ = 0x80; /* 6 Reserved bits */
210 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
211 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
212 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
220 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
221 return d;
222 }
223#endif
537124e4 224#endif /* Non loop style */
a0ed51b3 225}
646ca15d 226/*
07693fe6
KW
227=for apidoc uvchr_to_utf8
228
bcb1a2d4 229Adds the UTF-8 representation of the native code point C<uv> to the end
07693fe6
KW
230of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
231bytes available. The return value is the pointer to the byte after the
232end of the new character. In other words,
233
234 d = uvchr_to_utf8(d, uv);
235
236is the recommended wide native character-aware way of saying
237
238 *(d++) = uv;
239
de69f3af
KW
240This function accepts any UV as input. To forbid or warn on non-Unicode code
241points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
242
07693fe6
KW
243=cut
244*/
245
de69f3af
KW
246/* This is also a macro */
247PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
248
07693fe6
KW
249U8 *
250Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
251{
de69f3af 252 return uvchr_to_utf8(d, uv);
07693fe6
KW
253}
254
de69f3af
KW
255/*
256=for apidoc uvchr_to_utf8_flags
257
258Adds the UTF-8 representation of the native code point C<uv> to the end
259of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
260bytes available. The return value is the pointer to the byte after the
261end of the new character. In other words,
262
263 d = uvchr_to_utf8_flags(d, uv, flags);
264
265or, in most cases,
266
267 d = uvchr_to_utf8_flags(d, uv, 0);
268
269This is the Unicode-aware way of saying
270
271 *(d++) = uv;
272
273This function will convert to UTF-8 (and not warn) even code points that aren't
274legal Unicode or are problematic, unless C<flags> contains one or more of the
275following flags:
276
277If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
278the function will raise a warning, provided UTF8 warnings are enabled. If instead
279UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
280If both flags are set, the function will both warn and return NULL.
281
282The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags correspondingly
283affect how the function handles a Unicode non-character. And likewise, the
284UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags, affect the handling of
285code points that are
286above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
287even less portable) can be warned and/or disallowed even if other above-Unicode
288code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
289flags.
290
291And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
292above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
293DISALLOW flags.
294
295=cut
296*/
297
298/* This is also a macro */
299PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
300
07693fe6
KW
301U8 *
302Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
303{
de69f3af 304 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
305}
306
307/*
646ca15d 308
f7d739d1 309Tests if the first C<len> bytes of string C<s> form a valid UTF-8
bcb1a2d4
KW
310character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC) character is a
311valid UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
312will be returned if it is valid, otherwise 0.
313
314This is the "slow" version as opposed to the "fast" version which is
315the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
316difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
317or less you should use the IS_UTF8_CHAR(), for lengths of five or more
318you should use the _slow(). In practice this means that the _slow()
319will be used very rarely, since the maximum Unicode code point (as of
320Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
537124e4 321the "Perl extended UTF-8" (e.g, the infamous 'v-strings') will encode into
646ca15d
JH
322five bytes or more.
323
324=cut */
7af276bc 325PERL_STATIC_INLINE STRLEN
5f66b61c 326S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 327{
cd7e6c88 328 dTHX; /* The function called below requires thread context */
646ca15d 329
cd7e6c88 330 STRLEN actual_len;
646ca15d 331
cd7e6c88 332 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 333
18712bce 334 utf8n_to_uvchr(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 335
cd7e6c88 336 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 337}
9041c2e3
NIS
338
339/*
492a624f
KW
340=for apidoc is_utf8_char_buf
341
342Returns the number of bytes that comprise the first UTF-8 encoded character in
343buffer C<buf>. C<buf_end> should point to one position beyond the end of the
344buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
345encoded character.
346
347Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
348machines) is a valid UTF-8 character.
349
350=cut */
351
352STRLEN
353Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
354{
355
356 STRLEN len;
357
358 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
359
360 if (buf_end <= buf) {
361 return 0;
362 }
363
364 len = buf_end - buf;
365 if (len > UTF8SKIP(buf)) {
366 len = UTF8SKIP(buf);
367 }
368
369#ifdef IS_UTF8_CHAR
370 if (IS_UTF8_CHAR_FAST(len))
371 return IS_UTF8_CHAR(buf, len) ? len : 0;
372#endif /* #ifdef IS_UTF8_CHAR */
373 return is_utf8_char_slow(buf, len);
374}
375
376/*
87cea99e 377=for apidoc is_utf8_char
eebe1485 378
5da9da9e 379Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
380character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
381character is a valid UTF-8 character. The actual number of bytes in the UTF-8
382character will be returned if it is valid, otherwise 0.
9041c2e3 383
76848387 384This function is deprecated due to the possibility that malformed input could
a1433954 385cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 386instead.
e0328548 387
82686b01 388=cut */
76848387 389
067a85ef 390STRLEN
668b6d8d 391Perl_is_utf8_char(const U8 *s)
386d01d6 392{
7918f24d 393 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 394
76848387 395 /* Assumes we have enough space, which is why this is deprecated */
492a624f 396 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
397}
398
eaf7a4d2 399
6662521e 400/*
87cea99e 401=for apidoc is_utf8_string
6662521e 402
a1433954 403Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 404UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
405using C<strlen(s)> (which means if you use this option, that C<s> has to have a
406terminating NUL byte). Note that all characters being ASCII constitute 'a
407valid UTF-8 string'.
6662521e 408
a1433954 409See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 410
6662521e
GS
411=cut
412*/
413
8e84507e 414bool
668b6d8d 415Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 416{
35da51f7 417 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 418 const U8* x = s;
067a85ef 419
7918f24d 420 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 421
6662521e 422 while (x < send) {
1acdb0da 423 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
424 if (UTF8_IS_INVARIANT(*x)) {
425 x++;
426 }
1acdb0da
JH
427 else {
428 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
429 const STRLEN c = UTF8SKIP(x);
430 const U8* const next_char_ptr = x + c;
431
432 if (next_char_ptr > send) {
433 return FALSE;
434 }
435
768c67ee
JH
436 if (IS_UTF8_CHAR_FAST(c)) {
437 if (!IS_UTF8_CHAR(x, c))
e0328548 438 return FALSE;
3c614e38 439 }
e0328548
KW
440 else if (! is_utf8_char_slow(x, c)) {
441 return FALSE;
442 }
443 x = next_char_ptr;
1acdb0da 444 }
6662521e 445 }
768c67ee 446
067a85ef 447 return TRUE;
6662521e
GS
448}
449
67e989fb 450/*
814fafa7
NC
451Implemented as a macro in utf8.h
452
87cea99e 453=for apidoc is_utf8_string_loc
814fafa7 454
a1433954
KW
455Like L</is_utf8_string> but stores the location of the failure (in the
456case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
457"utf8ness success") in the C<ep>.
458
a1433954 459See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 460
87cea99e 461=for apidoc is_utf8_string_loclen
81cd54e3 462
a1433954
KW
463Like L</is_utf8_string>() but stores the location of the failure (in the
464case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
465"utf8ness success") in the C<ep>, and the number of UTF-8
466encoded characters in the C<el>.
467
a1433954 468See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
469
470=cut
471*/
472
473bool
668b6d8d 474Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 475{
35da51f7 476 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 477 const U8* x = s;
81cd54e3 478 STRLEN c;
3ebfea28 479 STRLEN outlen = 0;
7918f24d
NC
480
481 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 482
81cd54e3 483 while (x < send) {
e0328548
KW
484 const U8* next_char_ptr;
485
81cd54e3
JH
486 /* Inline the easy bits of is_utf8_char() here for speed... */
487 if (UTF8_IS_INVARIANT(*x))
e0328548 488 next_char_ptr = x + 1;
81cd54e3 489 else {
768c67ee 490 /* ... and call is_utf8_char() only if really needed. */
768c67ee 491 c = UTF8SKIP(x);
e0328548
KW
492 next_char_ptr = c + x;
493 if (next_char_ptr > send) {
494 goto out;
495 }
768c67ee
JH
496 if (IS_UTF8_CHAR_FAST(c)) {
497 if (!IS_UTF8_CHAR(x, c))
498 c = 0;
499 } else
500 c = is_utf8_char_slow(x, c);
768c67ee
JH
501 if (!c)
502 goto out;
81cd54e3 503 }
e0328548 504 x = next_char_ptr;
3ebfea28 505 outlen++;
81cd54e3 506 }
768c67ee
JH
507
508 out:
3ebfea28
AL
509 if (el)
510 *el = outlen;
511
768c67ee
JH
512 if (ep)
513 *ep = x;
3ebfea28 514 return (x == send);
81cd54e3
JH
515}
516
517/*
768c67ee 518
de69f3af 519=for apidoc utf8n_to_uvchr
378516de
KW
520
521THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 522Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 523
9041c2e3 524Bottom level UTF-8 decode routine.
de69f3af 525Returns the native code point value of the first character in the string C<s>,
746afd53
KW
526which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
527C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
528the length, in bytes, of that character.
949cf498
KW
529
530The value of C<flags> determines the behavior when C<s> does not point to a
531well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
532zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
533next possible position in C<s> that could begin a non-malformed character.
534Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
535
536Various ALLOW flags can be set in C<flags> to allow (and not warn on)
537individual types of malformations, such as the sequence being overlong (that
538is, when there is a shorter sequence that can express the same code point;
539overlong sequences are expressly forbidden in the UTF-8 standard due to
540potential security issues). Another malformation example is the first byte of
541a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
542flags. For allowed 0 length strings, this function returns 0; for allowed
543overlong sequences, the computed code point is returned; for all other allowed
544malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
545determinable reasonable value.
949cf498
KW
546
547The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
548flags) malformation is found. If this flag is set, the routine assumes that
549the caller will raise a warning, and this function will silently just set
d088425d
KW
550C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
551
552Note that this API requires disambiguation between successful decoding a NUL
553character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
554in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
555first byte of C<s> is 0 as well. If so, the input was a NUL; if not, the input
556had an error.
949cf498
KW
557
558Certain code points are considered problematic. These are Unicode surrogates,
746afd53 559Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 560By default these are considered regular code points, but certain situations
5eafe189 561warrant special handling for them. If C<flags> contains
949cf498
KW
562UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
563malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
564UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
565maximum) can be set to disallow these categories individually.
566
567The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
568UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
569for their respective categories, but otherwise the code points are considered
570valid (not malformations). To get a category to both be treated as a
571malformation and raise a warning, specify both the WARN and DISALLOW flags.
572(But note that warnings are not raised if lexically disabled nor if
573UTF8_CHECK_ONLY is also specified.)
574
575Very large code points (above 0x7FFF_FFFF) are considered more problematic than
576the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
577reasons: they requre at least 32 bits to represent them on ASCII platforms, are
578not representable at all on EBCDIC platforms, and the original UTF-8
579specification never went above this number (the current 0x10FFFF limit was
580imposed later). (The smaller ones, those that fit into 32 bits, are
581representable by a UV on ASCII platforms, but not by an IV, which means that
582the number of operations that can be performed on them is quite restricted.)
583The UTF-8 encoding on ASCII platforms for these large code points begins with a
584byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
585be treated as malformations, while allowing smaller above-Unicode code points.
586(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
587including these, as malformations.) Similarly, UTF8_WARN_FE_FF acts just like
588the other WARN flags, but applies just to these code points.
949cf498
KW
589
590All other code points corresponding to Unicode characters, including private
591use and those yet to be assigned, are never considered malformed and never
592warn.
67e989fb 593
37607a96
PK
594=cut
595*/
67e989fb 596
a0ed51b3 597UV
de69f3af 598Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 599{
97aff369 600 dVAR;
d4c19fe8 601 const U8 * const s0 = s;
eb83ed87 602 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 603 U8 * send;
eb83ed87
KW
604 UV uv = *s;
605 STRLEN expectlen;
949cf498 606 SV* sv = NULL;
eb83ed87
KW
607 UV outlier_ret = 0; /* return value when input is in error or problematic
608 */
609 UV pack_warn = 0; /* Save result of packWARN() for later */
610 bool unexpected_non_continuation = FALSE;
611 bool overflowed = FALSE;
2f8f112e 612 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 613
eb83ed87 614 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 615
de69f3af 616 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 617
eb83ed87
KW
618 /* The order of malformation tests here is important. We should consume as
619 * few bytes as possible in order to not skip any valid character. This is
620 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
621 * http://unicode.org/reports/tr36 for more discussion as to why. For
622 * example, once we've done a UTF8SKIP, we can tell the expected number of
623 * bytes, and could fail right off the bat if the input parameters indicate
624 * that there are too few available. But it could be that just that first
625 * byte is garbled, and the intended character occupies fewer bytes. If we
626 * blindly assumed that the first byte is correct, and skipped based on
627 * that number, we could skip over a valid input character. So instead, we
628 * always examine the sequence byte-by-byte.
629 *
630 * We also should not consume too few bytes, otherwise someone could inject
631 * things. For example, an input could be deliberately designed to
632 * overflow, and if this code bailed out immediately upon discovering that,
633 * returning to the caller *retlen pointing to the very next byte (one
634 * which is actually part of of the overflowing sequence), that could look
635 * legitimate to the caller, which could discard the initial partial
636 * sequence and process the rest, inappropriately */
637
638 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 639 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
640 if (retlen) {
641 *retlen = 0;
642 }
a0dbb045 643
eb83ed87
KW
644 if (flags & UTF8_ALLOW_EMPTY) {
645 return 0;
646 }
647 if (! (flags & UTF8_CHECK_ONLY)) {
648 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
649 }
0c443dc2
JH
650 goto malformed;
651 }
652
eb83ed87
KW
653 expectlen = UTF8SKIP(s);
654
655 /* A well-formed UTF-8 character, as the vast majority of calls to this
656 * function will be for, has this expected length. For efficiency, set
657 * things up here to return it. It will be overriden only in those rare
658 * cases where a malformation is found */
659 if (retlen) {
660 *retlen = expectlen;
661 }
662
663 /* An invariant is trivially well-formed */
1d72bdf6 664 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 665 return uv;
a0ed51b3 666 }
67e989fb 667
eb83ed87 668 /* A continuation character can't start a valid sequence */
b5b9af04 669 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
670 if (flags & UTF8_ALLOW_CONTINUATION) {
671 if (retlen) {
672 *retlen = 1;
673 }
674 return UNICODE_REPLACEMENT;
675 }
ba210ebe 676
eb83ed87
KW
677 if (! (flags & UTF8_CHECK_ONLY)) {
678 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
679 }
680 curlen = 1;
ba210ebe
JH
681 goto malformed;
682 }
9041c2e3 683
dcd27b3c
KW
684 /* Here is not a continuation byte, nor an invariant. The only thing left
685 * is a start byte (possibly for an overlong) */
686
1d72bdf6 687#ifdef EBCDIC
bc3632a8 688 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
689#endif
690
eb83ed87
KW
691 /* Remove the leading bits that indicate the number of bytes in the
692 * character's whole UTF-8 sequence, leaving just the bits that are part of
693 * the value */
694 uv &= UTF_START_MASK(expectlen);
ba210ebe 695
eb83ed87
KW
696 /* Now, loop through the remaining bytes in the character's sequence,
697 * accumulating each into the working value as we go. Be sure to not look
698 * past the end of the input string */
0b8d30e8
KW
699 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
700
eb83ed87 701 for (s = s0 + 1; s < send; s++) {
b5b9af04 702 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
703#ifndef EBCDIC /* Can't overflow in EBCDIC */
704 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
705
706 /* The original implementors viewed this malformation as more
707 * serious than the others (though I, khw, don't understand
708 * why, since other malformations also give very very wrong
709 * results), so there is no way to turn off checking for it.
710 * Set a flag, but keep going in the loop, so that we absorb
711 * the rest of the bytes that comprise the character. */
712 overflowed = TRUE;
713 overflow_byte = *s; /* Save for warning message's use */
714 }
715#endif
8850bf83 716 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
717 }
718 else {
719 /* Here, found a non-continuation before processing all expected
720 * bytes. This byte begins a new character, so quit, even if
721 * allowing this malformation. */
722 unexpected_non_continuation = TRUE;
723 break;
724 }
725 } /* End of loop through the character's bytes */
726
727 /* Save how many bytes were actually in the character */
728 curlen = s - s0;
729
730 /* The loop above finds two types of malformations: non-continuation and/or
731 * overflow. The non-continuation malformation is really a too-short
732 * malformation, as it means that the current character ended before it was
733 * expected to (being terminated prematurely by the beginning of the next
734 * character, whereas in the too-short malformation there just are too few
735 * bytes available to hold the character. In both cases, the check below
736 * that we have found the expected number of bytes would fail if executed.)
737 * Thus the non-continuation malformation is really unnecessary, being a
738 * subset of the too-short malformation. But there may be existing
739 * applications that are expecting the non-continuation type, so we retain
740 * it, and return it in preference to the too-short malformation. (If this
741 * code were being written from scratch, the two types might be collapsed
742 * into one.) I, khw, am also giving priority to returning the
743 * non-continuation and too-short malformations over overflow when multiple
744 * ones are present. I don't know of any real reason to prefer one over
745 * the other, except that it seems to me that multiple-byte errors trumps
746 * errors from a single byte */
b5b9af04 747 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
748 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
749 if (! (flags & UTF8_CHECK_ONLY)) {
750 if (curlen == 1) {
751 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
752 }
753 else {
754 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
755 }
756 }
eb83ed87
KW
757 goto malformed;
758 }
759 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
760
761 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
762 * as what the original expectations were. */
763 do_overlong_test = FALSE;
eb83ed87
KW
764 if (retlen) {
765 *retlen = curlen;
766 }
767 }
b5b9af04 768 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
769 if (! (flags & UTF8_ALLOW_SHORT)) {
770 if (! (flags & UTF8_CHECK_ONLY)) {
771 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 772 }
eb83ed87
KW
773 goto malformed;
774 }
775 uv = UNICODE_REPLACEMENT;
2f8f112e 776 do_overlong_test = FALSE;
eb83ed87
KW
777 if (retlen) {
778 *retlen = curlen;
779 }
780 }
781
782#ifndef EBCDIC /* EBCDIC allows FE, FF, can't overflow */
2f8f112e 783 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
eb83ed87
KW
784 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
785 {
786 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
787 * generation of the sv, since no warnings are raised under CHECK */
788 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY)) == UTF8_WARN_FE_FF
789 && ckWARN_d(WARN_UTF8))
790 {
42303544
KW
791 /* This message is deliberately not of the same syntax as the other
792 * messages for malformations, for backwards compatibility in the
793 * unlikely event that code is relying on its precise earlier text
794 */
eb83ed87
KW
795 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s Code point beginning with byte 0x%02X is not Unicode, and not portable", malformed_text, *s0));
796 pack_warn = packWARN(WARN_UTF8);
797 }
798 if (flags & UTF8_DISALLOW_FE_FF) {
799 goto malformed;
ba210ebe 800 }
ba210ebe 801 }
b5b9af04 802 if (UNLIKELY(overflowed)) {
ba210ebe 803
eb83ed87
KW
804 /* If the first byte is FF, it will overflow a 32-bit word. If the
805 * first byte is FE, it will overflow a signed 32-bit word. The
806 * above preserves backward compatibility, since its message was used
807 * in earlier versions of this code in preference to overflow */
808 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 809 goto malformed;
eb83ed87
KW
810 }
811#endif
812
2f8f112e 813 if (do_overlong_test
5aaebcb3 814 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
815 && ! (flags & UTF8_ALLOW_LONG))
816 {
eb83ed87
KW
817 /* The overlong malformation has lower precedence than the others.
818 * Note that if this malformation is allowed, we return the actual
819 * value, instead of the replacement character. This is because this
820 * value is actually well-defined. */
821 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 822 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (%d byte%s, need %d, after start byte 0x%02x)", malformed_text, (int)expectlen, expectlen == 1 ? "": "s", OFFUNISKIP(uv), *s0));
eb83ed87
KW
823 }
824 goto malformed;
825 }
826
827 /* Here, the input is considered to be well-formed , but could be a
828 * problematic code point that is not allowed by the input parameters. */
829 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
830 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
831 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
832 {
949cf498 833 if (UNICODE_IS_SURROGATE(uv)) {
eb83ed87
KW
834 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
835 && ckWARN2_d(WARN_UTF8, WARN_SURROGATE))
836 {
111d382d 837 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
eb83ed87 838 pack_warn = packWARN2(WARN_UTF8, WARN_SURROGATE);
949cf498
KW
839 }
840 if (flags & UTF8_DISALLOW_SURROGATE) {
841 goto disallowed;
842 }
843 }
949cf498 844 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87
KW
845 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
846 && ckWARN2_d(WARN_UTF8, WARN_NON_UNICODE))
847 {
111d382d 848 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
eb83ed87 849 pack_warn = packWARN2(WARN_UTF8, WARN_NON_UNICODE);
949cf498
KW
850 }
851 if (flags & UTF8_DISALLOW_SUPER) {
852 goto disallowed;
853 }
854 }
4190d317
KW
855 else if (UNICODE_IS_NONCHAR(uv)) {
856 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
857 && ckWARN2_d(WARN_UTF8, WARN_NONCHAR))
858 {
859 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
860 pack_warn = packWARN2(WARN_UTF8, WARN_NONCHAR);
861 }
862 if (flags & UTF8_DISALLOW_NONCHAR) {
863 goto disallowed;
864 }
865 }
949cf498 866
eb83ed87 867 if (sv) {
de69f3af
KW
868 outlier_ret = uv; /* Note we don't bother to convert to native,
869 as all the outlier code points are the same
870 in both ASCII and EBCDIC */
eb83ed87
KW
871 goto do_warn;
872 }
873
949cf498
KW
874 /* Here, this is not considered a malformed character, so drop through
875 * to return it */
a0ed51b3 876 }
ba210ebe 877
de69f3af 878 return UNI_TO_NATIVE(uv);
ba210ebe 879
eb83ed87
KW
880 /* There are three cases which get to beyond this point. In all 3 cases:
881 * <sv> if not null points to a string to print as a warning.
882 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
883 * set.
884 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
885 * This is done by initializing it to 0, and changing it only
886 * for case 1).
887 * The 3 cases are:
888 * 1) The input is valid but problematic, and to be warned about. The
889 * return value is the resultant code point; <*retlen> is set to
890 * <curlen>, the number of bytes that comprise the code point.
891 * <pack_warn> contains the result of packWARN() for the warning
892 * types. The entry point for this case is the label <do_warn>;
893 * 2) The input is a valid code point but disallowed by the parameters to
894 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
895 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
896 * comprise the code point. <pack_warn> contains the result of
897 * packWARN() for the warning types. The entry point for this case is
898 * the label <disallowed>.
899 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
900 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
901 * bytes that comprise the malformation. All such malformations are
902 * assumed to be warning type <utf8>. The entry point for this case
903 * is the label <malformed>.
904 */
949cf498 905
ba210ebe
JH
906malformed:
907
eb83ed87
KW
908 if (sv && ckWARN_d(WARN_UTF8)) {
909 pack_warn = packWARN(WARN_UTF8);
910 }
911
912disallowed:
913
fcc8fcf6 914 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 915 if (retlen)
10edeb5d 916 *retlen = ((STRLEN) -1);
ba210ebe
JH
917 return 0;
918 }
919
eb83ed87 920do_warn:
5b311467 921
eb83ed87
KW
922 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
923 if warnings are to be raised. */
f555bc63 924 const char * const string = SvPVX_const(sv);
a0dbb045 925
f555bc63
KW
926 if (PL_op)
927 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
928 else
929 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
930 }
931
eb83ed87
KW
932 if (retlen) {
933 *retlen = curlen;
934 }
ba210ebe 935
eb83ed87 936 return outlier_ret;
a0ed51b3
LW
937}
938
8e84507e 939/*
ec5f19d0
KW
940=for apidoc utf8_to_uvchr_buf
941
942Returns the native code point of the first character in the string C<s> which
943is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 944C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 945
524080c4
KW
946If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
947enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
948NULL) to -1. If those warnings are off, the computed value, if well-defined
949(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
950C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
951the next possible position in C<s> that could begin a non-malformed character.
de69f3af 952See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 953returned.
ec5f19d0
KW
954
955=cut
956*/
957
958
959UV
960Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
961{
ec5f19d0
KW
962 assert(s < send);
963
964 return utf8n_to_uvchr(s, send - s, retlen,
965 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
966}
967
27d6c58a 968/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 969 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 970 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
971
972UV
973Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
974{
010ab96b
KW
975 UV expectlen = UTF8SKIP(s);
976 const U8* send = s + expectlen;
9ff2f0f7 977 UV uv = *s;
3986bb7c 978
27d6c58a
KW
979 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
980
010ab96b
KW
981 if (retlen) {
982 *retlen = expectlen;
983 }
984
985 /* An invariant is trivially returned */
986 if (expectlen == 1) {
9ff2f0f7 987 return uv;
010ab96b
KW
988 }
989
9ff2f0f7
KW
990#ifdef EBCDIC
991 uv = NATIVE_UTF8_TO_I8(uv);
992#endif
993
010ab96b
KW
994 /* Remove the leading bits that indicate the number of bytes, leaving just
995 * the bits that are part of the value */
996 uv &= UTF_START_MASK(expectlen);
997
998 /* Now, loop through the remaining bytes, accumulating each into the
999 * working total as we go. (I khw tried unrolling the loop for up to 4
1000 * bytes, but there was no performance improvement) */
1001 for (++s; s < send; s++) {
1002 uv = UTF8_ACCUMULATE(uv, *s);
1003 }
1004
3986bb7c 1005 return UNI_TO_NATIVE(uv);
010ab96b 1006
27d6c58a
KW
1007}
1008
ec5f19d0 1009/*
87cea99e 1010=for apidoc utf8_to_uvchr
9041c2e3 1011
6ee84de2 1012Returns the native code point of the first character in the string C<s>
1e54db1a 1013which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1014length, in bytes, of that character.
1015
4b88fb76 1016Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1017malformed input could cause reading beyond the end of the input buffer, which
1018is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 1019
524080c4
KW
1020If C<s> points to one of the detected malformations, and UTF8 warnings are
1021enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1022NULL) to -1. If those warnings are off, the computed value if well-defined (or
1023the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1024is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1025next possible position in C<s> that could begin a non-malformed character.
de69f3af 1026See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
1027
1028=cut
1029*/
1030
1031UV
7fc63493 1032Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 1033{
7918f24d
NC
1034 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
1035
2ff6c191 1036 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
1037}
1038
1039/*
ec5f19d0
KW
1040=for apidoc utf8_to_uvuni_buf
1041
de69f3af
KW
1042Only in very rare circumstances should code need to be dealing in Unicode
1043(as opposed to native) code points. In those few cases, use
1044C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1045
1046Returns the Unicode (not-native) code point of the first character in the
1047string C<s> which
ec5f19d0
KW
1048is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1049C<retlen> will be set to the length, in bytes, of that character.
1050
524080c4
KW
1051If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1052enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1053NULL) to -1. If those warnings are off, the computed value if well-defined (or
1054the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1055is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1056next possible position in C<s> that could begin a non-malformed character.
de69f3af 1057See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
1058
1059=cut
1060*/
1061
1062UV
1063Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1064{
1065 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1066
1067 assert(send > s);
1068
1069 /* Call the low level routine asking for checks */
de69f3af
KW
1070 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1071 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
1072}
1073
5495102a
KW
1074/* DEPRECATED!
1075 * Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1076 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1077 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1078
1079UV
1080Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1081{
1082 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1083
010ab96b 1084 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
27d6c58a
KW
1085}
1086
ec5f19d0 1087/*
87cea99e 1088=for apidoc utf8_to_uvuni
9041c2e3
NIS
1089
1090Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1091which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1092length, in bytes, of that character.
1093
4b88fb76 1094Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31 1095malformed input could cause reading beyond the end of the input buffer, which
4f83cdcd
KW
1096is one reason why this function is deprecated. The other is that only in
1097extremely limited circumstances should the Unicode versus native code point be
de69f3af 1098of any interest to you. See L</utf8_to_uvuni_buf> for alternatives.
9041c2e3 1099
524080c4
KW
1100If C<s> points to one of the detected malformations, and UTF8 warnings are
1101enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1102NULL) to -1. If those warnings are off, the computed value if well-defined (or
1103the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1104is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1105next possible position in C<s> that could begin a non-malformed character.
de69f3af 1106See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1107
1108=cut
1109*/
1110
1111UV
7fc63493 1112Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1113{
7918f24d
NC
1114 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1115
5495102a 1116 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
8e84507e
NIS
1117}
1118
b76347f2 1119/*
87cea99e 1120=for apidoc utf8_length
b76347f2
JH
1121
1122Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1123Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1124up past C<e>, croaks.
b76347f2
JH
1125
1126=cut
1127*/
1128
1129STRLEN
35a4481c 1130Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1131{
97aff369 1132 dVAR;
b76347f2
JH
1133 STRLEN len = 0;
1134
7918f24d
NC
1135 PERL_ARGS_ASSERT_UTF8_LENGTH;
1136
8850bf83
JH
1137 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1138 * the bitops (especially ~) can create illegal UTF-8.
1139 * In other words: in Perl UTF-8 is not just for Unicode. */
1140
a3b680e6
AL
1141 if (e < s)
1142 goto warn_and_return;
b76347f2 1143 while (s < e) {
4cbf4130 1144 s += UTF8SKIP(s);
8e91ec7f
AV
1145 len++;
1146 }
1147
1148 if (e != s) {
1149 len--;
1150 warn_and_return:
9b387841
NC
1151 if (PL_op)
1152 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1153 "%s in %s", unees, OP_DESC(PL_op));
1154 else
61a12c31 1155 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1156 }
1157
1158 return len;
1159}
1160
b06226ff 1161/*
87cea99e 1162=for apidoc utf8_distance
b06226ff 1163
1e54db1a 1164Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1165and C<b>.
1166
1167WARNING: use only if you *know* that the pointers point inside the
1168same UTF-8 buffer.
1169
37607a96
PK
1170=cut
1171*/
a0ed51b3 1172
02eb7b47 1173IV
35a4481c 1174Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1175{
7918f24d
NC
1176 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1177
bf1665bc 1178 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1179}
1180
b06226ff 1181/*
87cea99e 1182=for apidoc utf8_hop
b06226ff 1183
8850bf83
JH
1184Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1185forward or backward.
b06226ff
JH
1186
1187WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1188the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1189on the first byte of character or just after the last byte of a character.
b06226ff 1190
37607a96
PK
1191=cut
1192*/
a0ed51b3
LW
1193
1194U8 *
4373e329 1195Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1196{
7918f24d
NC
1197 PERL_ARGS_ASSERT_UTF8_HOP;
1198
96a5add6 1199 PERL_UNUSED_CONTEXT;
8850bf83
JH
1200 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1201 * the bitops (especially ~) can create illegal UTF-8.
1202 * In other words: in Perl UTF-8 is not just for Unicode. */
1203
a0ed51b3
LW
1204 if (off >= 0) {
1205 while (off--)
1206 s += UTF8SKIP(s);
1207 }
1208 else {
1209 while (off++) {
1210 s--;
8850bf83
JH
1211 while (UTF8_IS_CONTINUATION(*s))
1212 s--;
a0ed51b3
LW
1213 }
1214 }
4373e329 1215 return (U8 *)s;
a0ed51b3
LW
1216}
1217
6940069f 1218/*
fed3ba5d
NC
1219=for apidoc bytes_cmp_utf8
1220
a1433954
KW
1221Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1222sequence of characters (stored as UTF-8) in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1223equal, -1 or -2 if the first string is less than the second string, +1 or +2
1224if the first string is greater than the second string.
1225
1226-1 or +1 is returned if the shorter string was identical to the start of the
1227longer string. -2 or +2 is returned if the was a difference between characters
1228within the strings.
1229
1230=cut
1231*/
1232
1233int
1234Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1235{
1236 const U8 *const bend = b + blen;
1237 const U8 *const uend = u + ulen;
1238
1239 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1240
1241 PERL_UNUSED_CONTEXT;
1242
1243 while (b < bend && u < uend) {
1244 U8 c = *u++;
1245 if (!UTF8_IS_INVARIANT(c)) {
1246 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1247 if (u < uend) {
1248 U8 c1 = *u++;
1249 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1250 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1251 } else {
1252 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1253 "Malformed UTF-8 character "
1254 "(unexpected non-continuation byte 0x%02x"
1255 ", immediately after start byte 0x%02x)"
1256 /* Dear diag.t, it's in the pod. */
1257 "%s%s", c1, c,
1258 PL_op ? " in " : "",
1259 PL_op ? OP_DESC(PL_op) : "");
1260 return -2;
1261 }
1262 } else {
1263 if (PL_op)
1264 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1265 "%s in %s", unees, OP_DESC(PL_op));
1266 else
61a12c31 1267 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1268 return -2; /* Really want to return undef :-) */
1269 }
1270 } else {
1271 return -2;
1272 }
1273 }
1274 if (*b != c) {
1275 return *b < c ? -2 : +2;
1276 }
1277 ++b;
1278 }
1279
1280 if (b == bend && u == uend)
1281 return 0;
1282
1283 return b < bend ? +1 : -1;
1284}
1285
1286/*
87cea99e 1287=for apidoc utf8_to_bytes
6940069f 1288
2bbc8d55 1289Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1290Unlike L</bytes_to_utf8>, this over-writes the original string, and
1291updates C<len> to contain the new length.
67e989fb 1292Returns zero on failure, setting C<len> to -1.
6940069f 1293
a1433954 1294If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1295
6940069f
GS
1296=cut
1297*/
1298
1299U8 *
37607a96 1300Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1301{
d4c19fe8
AL
1302 U8 * const save = s;
1303 U8 * const send = s + *len;
6940069f 1304 U8 *d;
246fae53 1305
7918f24d
NC
1306 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1307
1e54db1a 1308 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1309 while (s < send) {
d59937ca
KW
1310 if (! UTF8_IS_INVARIANT(*s)) {
1311 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1312 *len = ((STRLEN) -1);
1313 return 0;
1314 }
1315 s++;
dcad2880 1316 }
d59937ca 1317 s++;
246fae53 1318 }
dcad2880
JH
1319
1320 d = s = save;
6940069f 1321 while (s < send) {
80e0b38f
KW
1322 U8 c = *s++;
1323 if (! UTF8_IS_INVARIANT(c)) {
1324 /* Then it is two-byte encoded */
1325 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1326 s++;
1327 }
1328 *d++ = c;
6940069f
GS
1329 }
1330 *d = '\0';
246fae53 1331 *len = d - save;
6940069f
GS
1332 return save;
1333}
1334
1335/*
87cea99e 1336=for apidoc bytes_from_utf8
f9a63242 1337
2bbc8d55 1338Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1339Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1340the newly-created string, and updates C<len> to contain the new
1341length. Returns the original string if no conversion occurs, C<len>
1342is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
13430 if C<s> is converted or consisted entirely of characters that are invariant
1344in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1345
37607a96
PK
1346=cut
1347*/
f9a63242
JH
1348
1349U8 *
e1ec3a88 1350Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1351{
f9a63242 1352 U8 *d;
e1ec3a88
AL
1353 const U8 *start = s;
1354 const U8 *send;
f9a63242
JH
1355 I32 count = 0;
1356
7918f24d
NC
1357 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1358
96a5add6 1359 PERL_UNUSED_CONTEXT;
f9a63242 1360 if (!*is_utf8)
73d840c0 1361 return (U8 *)start;
f9a63242 1362
1e54db1a 1363 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1364 for (send = s + *len; s < send;) {
d59937ca
KW
1365 if (! UTF8_IS_INVARIANT(*s)) {
1366 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1367 return (U8 *)start;
d59937ca
KW
1368 }
1369 count++;
1370 s++;
db42d148 1371 }
d59937ca 1372 s++;
f9a63242
JH
1373 }
1374
35da51f7 1375 *is_utf8 = FALSE;
f9a63242 1376
212542aa 1377 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1378 s = start; start = d;
f9a63242
JH
1379 while (s < send) {
1380 U8 c = *s++;
1a91c45d 1381 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1382 /* Then it is two-byte encoded */
1a91c45d
KW
1383 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1384 s++;
c4d5f83a
NIS
1385 }
1386 *d++ = c;
f9a63242
JH
1387 }
1388 *d = '\0';
1389 *len = d - start;
73d840c0 1390 return (U8 *)start;
f9a63242
JH
1391}
1392
1393/*
87cea99e 1394=for apidoc bytes_to_utf8
6940069f 1395
ff97e5cf
KW
1396Converts a string C<s> of length C<len> bytes from the native encoding into
1397UTF-8.
6662521e 1398Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1399reflect the new length in bytes.
6940069f 1400
2bbc8d55
SP
1401A NUL character will be written after the end of the string.
1402
1403If you want to convert to UTF-8 from encodings other than
1404the native (Latin1 or EBCDIC),
a1433954 1405see L</sv_recode_to_utf8>().
c9ada85f 1406
497711e7 1407=cut
6940069f
GS
1408*/
1409
c682ebef
FC
1410/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1411 likewise need duplication. */
1412
6940069f 1413U8*
35a4481c 1414Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1415{
35a4481c 1416 const U8 * const send = s + (*len);
6940069f
GS
1417 U8 *d;
1418 U8 *dst;
7918f24d
NC
1419
1420 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1421 PERL_UNUSED_CONTEXT;
6940069f 1422
212542aa 1423 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1424 dst = d;
1425
1426 while (s < send) {
55d09dc8
KW
1427 append_utf8_from_native_byte(*s, &d);
1428 s++;
6940069f
GS
1429 }
1430 *d = '\0';
6662521e 1431 *len = d-dst;
6940069f
GS
1432 return dst;
1433}
1434
a0ed51b3 1435/*
dea0fc0b 1436 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1437 *
1438 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1439 * We optimize for native, for obvious reasons. */
1440
1441U8*
dea0fc0b 1442Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1443{
dea0fc0b
JH
1444 U8* pend;
1445 U8* dstart = d;
1446
7918f24d
NC
1447 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1448
dea0fc0b 1449 if (bytelen & 1)
f5992bc4 1450 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1451
1452 pend = p + bytelen;
1453
a0ed51b3 1454 while (p < pend) {
dea0fc0b
JH
1455 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1456 p += 2;
56d37426
KW
1457 if (UNI_IS_INVARIANT(uv)) {
1458 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1459 continue;
1460 }
56d37426
KW
1461 if (uv <= MAX_UTF8_TWO_BYTE) {
1462 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1463 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1464 continue;
1465 }
46956fad
KW
1466#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1467#define LAST_HIGH_SURROGATE 0xDBFF
1468#define FIRST_LOW_SURROGATE 0xDC00
1469#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1470 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1471 if (p >= pend) {
dea0fc0b 1472 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1473 } else {
1474 UV low = (p[0] << 8) + p[1];
1475 p += 2;
46956fad 1476 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1477 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1478 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1479 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1480 }
46956fad 1481 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1482 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1483 }
56d37426
KW
1484#ifdef EBCDIC
1485 d = uvoffuni_to_utf8_flags(d, uv, 0);
1486#else
a0ed51b3 1487 if (uv < 0x10000) {
eb160463
GS
1488 *d++ = (U8)(( uv >> 12) | 0xe0);
1489 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1490 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1491 continue;
1492 }
1493 else {
eb160463
GS
1494 *d++ = (U8)(( uv >> 18) | 0xf0);
1495 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1496 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1497 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1498 continue;
1499 }
56d37426 1500#endif
a0ed51b3 1501 }
dea0fc0b 1502 *newlen = d - dstart;
a0ed51b3
LW
1503 return d;
1504}
1505
1506/* Note: this one is slightly destructive of the source. */
1507
1508U8*
dea0fc0b 1509Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1510{
1511 U8* s = (U8*)p;
d4c19fe8 1512 U8* const send = s + bytelen;
7918f24d
NC
1513
1514 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1515
e0ea5e2d
NC
1516 if (bytelen & 1)
1517 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1518 (UV)bytelen);
1519
a0ed51b3 1520 while (s < send) {
d4c19fe8 1521 const U8 tmp = s[0];
a0ed51b3
LW
1522 s[0] = s[1];
1523 s[1] = tmp;
1524 s += 2;
1525 }
dea0fc0b 1526 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1527}
1528
922e8cb4
KW
1529bool
1530Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1531{
1532 U8 tmpbuf[UTF8_MAXBYTES+1];
1533 uvchr_to_utf8(tmpbuf, c);
1534 return _is_utf8_FOO(classnum, tmpbuf);
1535}
1536
c3fd2246
KW
1537/* for now these are all defined (inefficiently) in terms of the utf8 versions.
1538 * Note that the macros in handy.h that call these short-circuit calling them
1539 * for Latin-1 range inputs */
a0ed51b3
LW
1540
1541bool
84afefe6 1542Perl_is_uni_alnum(pTHX_ UV c)
a0ed51b3 1543{
89ebb4a3 1544 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1545 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1546 return _is_utf8_FOO(_CC_WORDCHAR, tmpbuf);
a0ed51b3
LW
1547}
1548
1549bool
5092f92a
KW
1550Perl_is_uni_alnumc(pTHX_ UV c)
1551{
1552 U8 tmpbuf[UTF8_MAXBYTES+1];
1553 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1554 return _is_utf8_FOO(_CC_ALPHANUMERIC, tmpbuf);
5092f92a
KW
1555}
1556
f9ae8fb6
JD
1557/* Internal function so we can deprecate the external one, and call
1558 this one from other deprecated functions in this file */
1559
1560PERL_STATIC_INLINE bool
61b19385
KW
1561S_is_utf8_idfirst(pTHX_ const U8 *p)
1562{
1563 dVAR;
1564
1565 if (*p == '_')
1566 return TRUE;
1567 /* is_utf8_idstart would be more logical. */
1568 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1569}
1570
5092f92a 1571bool
84afefe6 1572Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1573{
89ebb4a3 1574 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1575 uvchr_to_utf8(tmpbuf, c);
61b19385 1576 return S_is_utf8_idfirst(aTHX_ tmpbuf);
a0ed51b3
LW
1577}
1578
1579bool
eba68aa0
KW
1580Perl__is_uni_perl_idcont(pTHX_ UV c)
1581{
1582 U8 tmpbuf[UTF8_MAXBYTES+1];
1583 uvchr_to_utf8(tmpbuf, c);
1584 return _is_utf8_perl_idcont(tmpbuf);
1585}
1586
1587bool
f91dcd13
KW
1588Perl__is_uni_perl_idstart(pTHX_ UV c)
1589{
1590 U8 tmpbuf[UTF8_MAXBYTES+1];
1591 uvchr_to_utf8(tmpbuf, c);
1592 return _is_utf8_perl_idstart(tmpbuf);
1593}
1594
1595bool
84afefe6 1596Perl_is_uni_alpha(pTHX_ UV c)
a0ed51b3 1597{
89ebb4a3 1598 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1599 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1600 return _is_utf8_FOO(_CC_ALPHA, tmpbuf);
a0ed51b3
LW
1601}
1602
1603bool
84afefe6 1604Perl_is_uni_ascii(pTHX_ UV c)
4d61ec05 1605{
bc39fe24 1606 return isASCII(c);
4d61ec05
GS
1607}
1608
1609bool
bdd8600f
KW
1610Perl_is_uni_blank(pTHX_ UV c)
1611{
2cafb56b 1612 return isBLANK_uni(c);
bdd8600f
KW
1613}
1614
1615bool
84afefe6 1616Perl_is_uni_space(pTHX_ UV c)
a0ed51b3 1617{
add4123a 1618 return isSPACE_uni(c);
a0ed51b3
LW
1619}
1620
1621bool
84afefe6 1622Perl_is_uni_digit(pTHX_ UV c)
a0ed51b3 1623{
89ebb4a3 1624 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1625 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1626 return _is_utf8_FOO(_CC_DIGIT, tmpbuf);
a0ed51b3
LW
1627}
1628
1629bool
84afefe6 1630Perl_is_uni_upper(pTHX_ UV c)
a0ed51b3 1631{
89ebb4a3 1632 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1633 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1634 return _is_utf8_FOO(_CC_UPPER, tmpbuf);
a0ed51b3
LW
1635}
1636
1637bool
84afefe6 1638Perl_is_uni_lower(pTHX_ UV c)
a0ed51b3 1639{
89ebb4a3 1640 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1641 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1642 return _is_utf8_FOO(_CC_LOWER, tmpbuf);
a0ed51b3
LW
1643}
1644
1645bool
84afefe6 1646Perl_is_uni_cntrl(pTHX_ UV c)
b8c5462f 1647{
7b952154 1648 return isCNTRL_L1(c);
b8c5462f
JH
1649}
1650
1651bool
84afefe6 1652Perl_is_uni_graph(pTHX_ UV c)
b8c5462f 1653{
89ebb4a3 1654 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1655 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1656 return _is_utf8_FOO(_CC_GRAPH, tmpbuf);
b8c5462f
JH
1657}
1658
1659bool
84afefe6 1660Perl_is_uni_print(pTHX_ UV c)
a0ed51b3 1661{
89ebb4a3 1662 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1663 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1664 return _is_utf8_FOO(_CC_PRINT, tmpbuf);
a0ed51b3
LW
1665}
1666
b8c5462f 1667bool
84afefe6 1668Perl_is_uni_punct(pTHX_ UV c)
b8c5462f 1669{
89ebb4a3 1670 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1671 uvchr_to_utf8(tmpbuf, c);
922e8cb4 1672 return _is_utf8_FOO(_CC_PUNCT, tmpbuf);
b8c5462f
JH
1673}
1674
4d61ec05 1675bool
84afefe6 1676Perl_is_uni_xdigit(pTHX_ UV c)
4d61ec05 1677{
4ac6419d 1678 return isXDIGIT_uni(c);
4d61ec05
GS
1679}
1680
3a4c58c9
KW
1681UV
1682Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1683{
1684 /* We have the latin1-range values compiled into the core, so just use
1685 * those, converting the result to utf8. The only difference between upper
1686 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1687 * either "SS" or "Ss". Which one to use is passed into the routine in
1688 * 'S_or_s' to avoid a test */
1689
1690 UV converted = toUPPER_LATIN1_MOD(c);
1691
1692 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1693
1694 assert(S_or_s == 'S' || S_or_s == 's');
1695
6f2d5cbc 1696 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1697 characters in this range */
3a4c58c9
KW
1698 *p = (U8) converted;
1699 *lenp = 1;
1700 return converted;
1701 }
1702
1703 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1704 * which it maps to one of them, so as to only have to have one check for
1705 * it in the main case */
1706 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1707 switch (c) {
1708 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1709 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1710 break;
1711 case MICRO_SIGN:
1712 converted = GREEK_CAPITAL_LETTER_MU;
1713 break;
1714 case LATIN_SMALL_LETTER_SHARP_S:
1715 *(p)++ = 'S';
1716 *p = S_or_s;
1717 *lenp = 2;
1718 return 'S';
1719 default:
1720 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1721 assert(0); /* NOTREACHED */
3a4c58c9
KW
1722 }
1723 }
1724
1725 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1726 *p = UTF8_TWO_BYTE_LO(converted);
1727 *lenp = 2;
1728
1729 return converted;
1730}
1731
50bda2c3
KW
1732/* Call the function to convert a UTF-8 encoded character to the specified case.
1733 * Note that there may be more than one character in the result.
1734 * INP is a pointer to the first byte of the input character
1735 * OUTP will be set to the first byte of the string of changed characters. It
1736 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1737 * LENP will be set to the length in bytes of the string of changed characters
1738 *
1739 * The functions return the ordinal of the first character in the string of OUTP */
f90a9a02
KW
1740#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "utf8::ToSpecUc")
1741#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "utf8::ToSpecTc")
1742#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "utf8::ToSpecLc")
50bda2c3
KW
1743
1744/* This additionally has the input parameter SPECIALS, which if non-zero will
1745 * cause this to use the SPECIALS hash for folding (meaning get full case
1746 * folding); otherwise, when zero, this implies a simple case fold */
f90a9a02 1747#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 1748
84afefe6
JH
1749UV
1750Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1751{
3a4c58c9
KW
1752 dVAR;
1753
a1433954
KW
1754 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1755 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1756 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1757 * the changed version may be longer than the original character.
1758 *
1759 * The ordinal of the first character of the changed version is returned
1760 * (but note, as explained above, that there may be more.) */
1761
7918f24d
NC
1762 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1763
3a4c58c9
KW
1764 if (c < 256) {
1765 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1766 }
1767
0ebc6274 1768 uvchr_to_utf8(p, c);
3a4c58c9 1769 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1770}
1771
84afefe6
JH
1772UV
1773Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1774{
3a4c58c9
KW
1775 dVAR;
1776
7918f24d
NC
1777 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1778
3a4c58c9
KW
1779 if (c < 256) {
1780 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1781 }
1782
0ebc6274 1783 uvchr_to_utf8(p, c);
3a4c58c9 1784 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1785}
1786
afc16117
KW
1787STATIC U8
1788S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1789{
1790 /* We have the latin1-range values compiled into the core, so just use
1791 * those, converting the result to utf8. Since the result is always just
a1433954 1792 * one character, we allow <p> to be NULL */
afc16117
KW
1793
1794 U8 converted = toLOWER_LATIN1(c);
1795
1796 if (p != NULL) {
6f2d5cbc 1797 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1798 *p = converted;
1799 *lenp = 1;
1800 }
1801 else {
1802 *p = UTF8_TWO_BYTE_HI(converted);
1803 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1804 *lenp = 2;
1805 }
1806 }
1807 return converted;
1808}
1809
84afefe6
JH
1810UV
1811Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1812{
968c5e6a
KW
1813 dVAR;
1814
7918f24d
NC
1815 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1816
afc16117
KW
1817 if (c < 256) {
1818 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1819 }
1820
afc16117 1821 uvchr_to_utf8(p, c);
968c5e6a 1822 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1823}
1824
84afefe6 1825UV
51910141 1826Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1827{
51910141 1828 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1829 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1830 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1831 *
1832 * Not to be used for locale folds
51910141 1833 */
f673fad4 1834
a1dde8de
KW
1835 UV converted;
1836
1837 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1838
1ca267a5
KW
1839 assert (! (flags & FOLD_FLAGS_LOCALE));
1840
a1dde8de
KW
1841 if (c == MICRO_SIGN) {
1842 converted = GREEK_SMALL_LETTER_MU;
1843 }
51910141 1844 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1845
1846 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1847 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1848 * under those circumstances. */
1849 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1850 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1851 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1852 p, *lenp, U8);
1853 return LATIN_SMALL_LETTER_LONG_S;
1854 }
1855 else {
4f489194
KW
1856 *(p)++ = 's';
1857 *p = 's';
1858 *lenp = 2;
1859 return 's';
1ca267a5 1860 }
a1dde8de
KW
1861 }
1862 else { /* In this range the fold of all other characters is their lower
1863 case */
1864 converted = toLOWER_LATIN1(c);
1865 }
1866
6f2d5cbc 1867 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1868 *p = (U8) converted;
1869 *lenp = 1;
1870 }
1871 else {
1872 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1873 *p = UTF8_TWO_BYTE_LO(converted);
1874 *lenp = 2;
1875 }
1876
1877 return converted;
1878}
1879
1880UV
a0270393 1881Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
84afefe6 1882{
4b593389 1883
a0270393
KW
1884 /* Not currently externally documented, and subject to change
1885 * <flags> bits meanings:
1886 * FOLD_FLAGS_FULL iff full folding is to be used;
1887 * FOLD_FLAGS_LOCALE iff in locale
1888 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1889 */
4b593389 1890
36bb2ab6 1891 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1892
a1dde8de 1893 if (c < 256) {
a0270393 1894 UV result = _to_fold_latin1((U8) c, p, lenp,
1ca267a5 1895 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1896 /* It is illegal for the fold to cross the 255/256 boundary under
1897 * locale; in this case return the original */
1898 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1899 ? c
1900 : result;
a1dde8de
KW
1901 }
1902
a0270393
KW
1903 /* If no special needs, just use the macro */
1904 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1905 uvchr_to_utf8(p, c);
1906 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1907 }
1908 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1909 the special flags. */
1910 U8 utf8_c[UTF8_MAXBYTES + 1];
1911 uvchr_to_utf8(utf8_c, c);
1912 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1913 }
84afefe6
JH
1914}
1915
a0ed51b3 1916bool
84afefe6 1917Perl_is_uni_alnum_lc(pTHX_ UV c)
a0ed51b3 1918{
edfb3318 1919 if (c < 256) {
f4cd282c 1920 return isALNUM_LC(c);
edfb3318 1921 }
922e8cb4 1922 return _is_uni_FOO(_CC_WORDCHAR, c);
a0ed51b3
LW
1923}
1924
1925bool
5092f92a
KW
1926Perl_is_uni_alnumc_lc(pTHX_ UV c)
1927{
1928 if (c < 256) {
f4cd282c 1929 return isALPHANUMERIC_LC(c);
5092f92a 1930 }
922e8cb4 1931 return _is_uni_FOO(_CC_ALPHANUMERIC, c);
5092f92a
KW
1932}
1933
1934bool
84afefe6 1935Perl_is_uni_idfirst_lc(pTHX_ UV c)
a0ed51b3 1936{
edfb3318 1937 if (c < 256) {
f4cd282c 1938 return isIDFIRST_LC(c);
edfb3318
KW
1939 }
1940 return _is_uni_perl_idstart(c);
a0ed51b3
LW
1941}
1942
1943bool
84afefe6 1944Perl_is_uni_alpha_lc(pTHX_ UV c)
a0ed51b3 1945{
edfb3318 1946 if (c < 256) {
f4cd282c 1947 return isALPHA_LC(c);
edfb3318 1948 }
922e8cb4 1949 return _is_uni_FOO(_CC_ALPHA, c);
a0ed51b3
LW
1950}
1951
1952bool
84afefe6 1953Perl_is_uni_ascii_lc(pTHX_ UV c)
4d61ec05 1954{
edfb3318 1955 if (c < 256) {
f4cd282c 1956 return isASCII_LC(c);
edfb3318
KW
1957 }
1958 return 0;
4d61ec05
GS
1959}
1960
1961bool
bdd8600f
KW
1962Perl_is_uni_blank_lc(pTHX_ UV c)
1963{
edfb3318 1964 if (c < 256) {
f4cd282c 1965 return isBLANK_LC(c);
edfb3318 1966 }
61b19385 1967 return isBLANK_uni(c);
bdd8600f
KW
1968}
1969
1970bool
84afefe6 1971Perl_is_uni_space_lc(pTHX_ UV c)
a0ed51b3 1972{
edfb3318 1973 if (c < 256) {
f4cd282c 1974 return isSPACE_LC(c);
edfb3318 1975 }
61b19385 1976 return isSPACE_uni(c);
a0ed51b3
LW
1977}
1978
1979bool
84afefe6 1980Perl_is_uni_digit_lc(pTHX_ UV c)
a0ed51b3 1981{
edfb3318 1982 if (c < 256) {
f4cd282c 1983 return isDIGIT_LC(c);
edfb3318 1984 }
922e8cb4 1985 return _is_uni_FOO(_CC_DIGIT, c);
a0ed51b3
LW
1986}
1987
1988bool
84afefe6 1989Perl_is_uni_upper_lc(pTHX_ UV c)
a0ed51b3 1990{
edfb3318 1991 if (c < 256) {
f4cd282c 1992 return isUPPER_LC(c);
edfb3318 1993 }
922e8cb4 1994 return _is_uni_FOO(_CC_UPPER, c);
a0ed51b3
LW
1995}
1996
1997bool
84afefe6 1998Perl_is_uni_lower_lc(pTHX_ UV c)
a0ed51b3 1999{
edfb3318 2000 if (c < 256) {
f4cd282c 2001 return isLOWER_LC(c);
edfb3318 2002 }
922e8cb4 2003 return _is_uni_FOO(_CC_LOWER, c);
a0ed51b3
LW
2004}
2005
2006bool
84afefe6 2007Perl_is_uni_cntrl_lc(pTHX_ UV c)
b8c5462f 2008{
edfb3318 2009 if (c < 256) {
f4cd282c 2010 return isCNTRL_LC(c);
edfb3318 2011 }
61b19385 2012 return 0;
b8c5462f
JH
2013}
2014
2015bool
84afefe6 2016Perl_is_uni_graph_lc(pTHX_ UV c)
b8c5462f 2017{
edfb3318 2018 if (c < 256) {
f4cd282c 2019 return isGRAPH_LC(c);
edfb3318 2020 }
922e8cb4 2021 return _is_uni_FOO(_CC_GRAPH, c);
b8c5462f
JH
2022}
2023
2024bool
84afefe6 2025Perl_is_uni_print_lc(pTHX_ UV c)
a0ed51b3 2026{
edfb3318 2027 if (c < 256) {
f4cd282c 2028 return isPRINT_LC(c);
edfb3318 2029 }
922e8cb4 2030 return _is_uni_FOO(_CC_PRINT, c);
a0ed51b3
LW
2031}
2032
b8c5462f 2033bool
84afefe6 2034Perl_is_uni_punct_lc(pTHX_ UV c)
b8c5462f 2035{
edfb3318 2036 if (c < 256) {
f4cd282c 2037 return isPUNCT_LC(c);
edfb3318 2038 }
922e8cb4 2039 return _is_uni_FOO(_CC_PUNCT, c);
b8c5462f
JH
2040}
2041
4d61ec05 2042bool
84afefe6 2043Perl_is_uni_xdigit_lc(pTHX_ UV c)
4d61ec05 2044{
edfb3318 2045 if (c < 256) {
f4cd282c 2046 return isXDIGIT_LC(c);
edfb3318 2047 }
61b19385 2048 return isXDIGIT_uni(c);
4d61ec05
GS
2049}
2050
b7ac61fa
JH
2051U32
2052Perl_to_uni_upper_lc(pTHX_ U32 c)
2053{
ee099d14
JH
2054 /* XXX returns only the first character -- do not use XXX */
2055 /* XXX no locale support yet */
2056 STRLEN len;
89ebb4a3 2057 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 2058 return (U32)to_uni_upper(c, tmpbuf, &len);
b7ac61fa
JH
2059}
2060
2061U32
2062Perl_to_uni_title_lc(pTHX_ U32 c)
2063{
ee099d14
JH
2064 /* XXX returns only the first character XXX -- do not use XXX */
2065 /* XXX no locale support yet */
2066 STRLEN len;
89ebb4a3 2067 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 2068 return (U32)to_uni_title(c, tmpbuf, &len);
b7ac61fa
JH
2069}
2070
2071U32
2072Perl_to_uni_lower_lc(pTHX_ U32 c)
2073{
ee099d14
JH
2074 /* XXX returns only the first character -- do not use XXX */
2075 /* XXX no locale support yet */
2076 STRLEN len;
89ebb4a3 2077 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
ee099d14 2078 return (U32)to_uni_lower(c, tmpbuf, &len);
b7ac61fa
JH
2079}
2080
26483009 2081PERL_STATIC_INLINE bool
5141f98e 2082S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
2083 const char *const swashname)
2084{
ea317ccb
KW
2085 /* returns a boolean giving whether or not the UTF8-encoded character that
2086 * starts at <p> is in the swash indicated by <swashname>. <swash>
2087 * contains a pointer to where the swash indicated by <swashname>
2088 * is to be stored; which this routine will do, so that future calls will
2089 * look at <*swash> and only generate a swash if it is not null
2090 *
2091 * Note that it is assumed that the buffer length of <p> is enough to
2092 * contain all the bytes that comprise the character. Thus, <*p> should
2093 * have been checked before this call for mal-formedness enough to assure
2094 * that. */
2095
97aff369 2096 dVAR;
7918f24d
NC
2097
2098 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2099
492a624f 2100 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2101 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2102 * as far as there being enough bytes available in it to accommodate the
2103 * character without reading beyond the end, and pass that number on to the
2104 * validating routine */
28123549
KW
2105 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
2106 if (ckWARN_d(WARN_UTF8)) {
2107 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 2108 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
2109 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
2110 what the malformation is */
2111 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
2112 }
2113 }
2114 return FALSE;
2115 }
87367d5f
KW
2116 if (!*swash) {
2117 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2118 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
2119 }
28123549 2120
bde6a22d
NC
2121 return swash_fetch(*swash, p, TRUE) != 0;
2122}
2123
2124bool
922e8cb4
KW
2125Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2126{
2127 dVAR;
2128
2129 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2130
2131 assert(classnum < _FIRST_NON_SWASH_CC);
2132
2133 return is_utf8_common(p, &PL_utf8_swash_ptrs[classnum], swash_property_names[classnum]);
2134}
2135
2136bool
7fc63493 2137Perl_is_utf8_alnum(pTHX_ const U8 *p)
a0ed51b3 2138{
97aff369 2139 dVAR;
7918f24d
NC
2140
2141 PERL_ARGS_ASSERT_IS_UTF8_ALNUM;
2142
671c33bf
NC
2143 /* NOTE: "IsWord", not "IsAlnum", since Alnum is a true
2144 * descendant of isalnum(3), in other words, it doesn't
2145 * contain the '_'. --jhi */
03940dc2 2146 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_WORDCHAR], "IsWord");
a0ed51b3
LW
2147}
2148
2149bool
5092f92a
KW
2150Perl_is_utf8_alnumc(pTHX_ const U8 *p)
2151{
2152 dVAR;
2153
2154 PERL_ARGS_ASSERT_IS_UTF8_ALNUMC;
2155
03940dc2 2156 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHANUMERIC], "IsAlnum");
5092f92a
KW
2157}
2158
2159bool
7fc63493 2160Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 2161{
97aff369 2162 dVAR;
7918f24d
NC
2163
2164 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
2165
61b19385 2166 return S_is_utf8_idfirst(aTHX_ p);
82686b01
JH
2167}
2168
2169bool
c11ff943
KW
2170Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
2171{
2172 dVAR;
2173
2174 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
2175
2176 if (*p == '_')
2177 return TRUE;
2178 /* is_utf8_idstart would be more logical. */
2179 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
2180}
2181
2182bool
d65654cb 2183Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
b6912c02
KW
2184{
2185 dVAR;
2186
d65654cb 2187 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
b6912c02
KW
2188
2189 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
2190}
2191
2192bool
eba68aa0
KW
2193Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2194{
2195 dVAR;
2196
2197 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2198
2199 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont");
2200}
2201
2202
2203bool
7fc63493 2204Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2205{
97aff369 2206 dVAR;
7918f24d
NC
2207
2208 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
2209
d4c19fe8 2210 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
2211}
2212
2213bool
c11ff943
KW
2214Perl_is_utf8_xidcont(pTHX_ const U8 *p)
2215{
2216 dVAR;
2217
2218 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
2219
c11ff943
KW
2220 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
2221}
2222
2223bool
7fc63493 2224Perl_is_utf8_alpha(pTHX_ const U8 *p)
a0ed51b3 2225{
97aff369 2226 dVAR;
7918f24d
NC
2227
2228 PERL_ARGS_ASSERT_IS_UTF8_ALPHA;
2229
03940dc2 2230 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_ALPHA], "IsAlpha");
a0ed51b3
LW
2231}
2232
2233bool
7fc63493 2234Perl_is_utf8_ascii(pTHX_ const U8 *p)
b8c5462f 2235{
97aff369 2236 dVAR;
7918f24d
NC
2237
2238 PERL_ARGS_ASSERT_IS_UTF8_ASCII;
2239
bc39fe24
KW
2240 /* ASCII characters are the same whether in utf8 or not. So the macro
2241 * works on both utf8 and non-utf8 representations. */
2242 return isASCII(*p);
b8c5462f
JH
2243}
2244
2245bool
bdd8600f
KW
2246Perl_is_utf8_blank(pTHX_ const U8 *p)
2247{
2248 dVAR;
2249
2250 PERL_ARGS_ASSERT_IS_UTF8_BLANK;
2251
2cafb56b 2252 return isBLANK_utf8(p);
bdd8600f
KW
2253}
2254
2255bool
7fc63493 2256Perl_is_utf8_space(pTHX_ const U8 *p)
a0ed51b3 2257{
97aff369 2258 dVAR;
7918f24d
NC
2259
2260 PERL_ARGS_ASSERT_IS_UTF8_SPACE;
2261
add4123a 2262 return isSPACE_utf8(p);
a0ed51b3
LW
2263}
2264
2265bool
d1eb3177
YO
2266Perl_is_utf8_perl_space(pTHX_ const U8 *p)
2267{
2268 dVAR;
2269
2270 PERL_ARGS_ASSERT_IS_UTF8_PERL_SPACE;
2271
c4428693
KW
2272 /* Only true if is an ASCII space-like character, and ASCII is invariant
2273 * under utf8, so can just use the macro */
2274 return isSPACE_A(*p);
d1eb3177
YO
2275}
2276
2277bool
2278Perl_is_utf8_perl_word(pTHX_ const U8 *p)
2279{
2280 dVAR;
2281
2282 PERL_ARGS_ASSERT_IS_UTF8_PERL_WORD;
2283
c4428693
KW
2284 /* Only true if is an ASCII word character, and ASCII is invariant
2285 * under utf8, so can just use the macro */
2286 return isWORDCHAR_A(*p);
d1eb3177
YO
2287}
2288
2289bool
7fc63493 2290Perl_is_utf8_digit(pTHX_ const U8 *p)
a0ed51b3 2291{
97aff369 2292 dVAR;
7918f24d
NC
2293
2294 PERL_ARGS_ASSERT_IS_UTF8_DIGIT;
2295
03940dc2 2296 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_DIGIT], "IsDigit");
a0ed51b3
LW
2297}
2298
2299bool
d1eb3177
YO
2300Perl_is_utf8_posix_digit(pTHX_ const U8 *p)
2301{
2302 dVAR;
2303
2304 PERL_ARGS_ASSERT_IS_UTF8_POSIX_DIGIT;
2305
c4428693
KW
2306 /* Only true if is an ASCII digit character, and ASCII is invariant
2307 * under utf8, so can just use the macro */
2308 return isDIGIT_A(*p);
d1eb3177
YO
2309}
2310
2311bool
7fc63493 2312Perl_is_utf8_upper(pTHX_ const U8 *p)
a0ed51b3 2313{
97aff369 2314 dVAR;
7918f24d
NC
2315
2316 PERL_ARGS_ASSERT_IS_UTF8_UPPER;
2317
03940dc2 2318 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_UPPER], "IsUppercase");
a0ed51b3
LW
2319}
2320
2321bool
7fc63493 2322Perl_is_utf8_lower(pTHX_ const U8 *p)
a0ed51b3 2323{
97aff369 2324 dVAR;
7918f24d
NC
2325
2326 PERL_ARGS_ASSERT_IS_UTF8_LOWER;
2327
03940dc2 2328 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_LOWER], "IsLowercase");
a0ed51b3
LW
2329}
2330
2331bool
7fc63493 2332Perl_is_utf8_cntrl(pTHX_ const U8 *p)
b8c5462f 2333{
97aff369 2334 dVAR;
7918f24d
NC
2335
2336 PERL_ARGS_ASSERT_IS_UTF8_CNTRL;
2337
a35d759a 2338 return isCNTRL_utf8(p);
b8c5462f
JH
2339}
2340
2341bool
7fc63493 2342Perl_is_utf8_graph(pTHX_ const U8 *p)
b8c5462f 2343{
97aff369 2344 dVAR;
7918f24d
NC
2345
2346 PERL_ARGS_ASSERT_IS_UTF8_GRAPH;
2347
03940dc2 2348 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_GRAPH], "IsGraph");
b8c5462f
JH
2349}
2350
2351bool
7fc63493 2352Perl_is_utf8_print(pTHX_ const U8 *p)
a0ed51b3 2353{
97aff369 2354 dVAR;
7918f24d
NC
2355
2356 PERL_ARGS_ASSERT_IS_UTF8_PRINT;
2357
03940dc2 2358 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PRINT], "IsPrint");
a0ed51b3
LW
2359}
2360
2361bool
7fc63493 2362Perl_is_utf8_punct(pTHX_ const U8 *p)
b8c5462f 2363{
97aff369 2364 dVAR;
7918f24d
NC
2365
2366 PERL_ARGS_ASSERT_IS_UTF8_PUNCT;
2367
03940dc2 2368 return is_utf8_common(p, &PL_utf8_swash_ptrs[_CC_PUNCT], "IsPunct");
b8c5462f
JH
2369}
2370
2371bool
7fc63493 2372Perl_is_utf8_xdigit(pTHX_ const U8 *p)
b8c5462f 2373{
97aff369 2374 dVAR;
7918f24d
NC
2375
2376 PERL_ARGS_ASSERT_IS_UTF8_XDIGIT;
2377
4ac6419d 2378 return is_XDIGIT_utf8(p);
b8c5462f
JH
2379}
2380
2381bool
7dbf68d2
KW
2382Perl__is_utf8_mark(pTHX_ const U8 *p)
2383{
2384 dVAR;
2385
2386 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2387
2388 return is_utf8_common(p, &PL_utf8_mark, "IsM");
2389}
2390
2391
2392bool
7fc63493 2393Perl_is_utf8_mark(pTHX_ const U8 *p)
a0ed51b3 2394{
97aff369 2395 dVAR;
7918f24d
NC
2396
2397 PERL_ARGS_ASSERT_IS_UTF8_MARK;
2398
7dbf68d2 2399 return _is_utf8_mark(p);
a0ed51b3
LW
2400}
2401
6b5c0936 2402/*
87cea99e 2403=for apidoc to_utf8_case
6b5c0936 2404
a1433954
KW
2405The C<p> contains the pointer to the UTF-8 string encoding
2406the character that is being converted. This routine assumes that the character
2407at C<p> is well-formed.
6b5c0936 2408
a1433954
KW
2409The C<ustrp> is a pointer to the character buffer to put the
2410conversion result to. The C<lenp> is a pointer to the length
6b5c0936
JH
2411of the result.
2412
a1433954 2413The C<swashp> is a pointer to the swash to use.
6b5c0936 2414
a1433954
KW
2415Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2416and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. The C<special> (usually,
0134edef 2417but not always, a multicharacter mapping), is tried first.
6b5c0936 2418
a1433954 2419The C<special> is a string like "utf8::ToSpecLower", which means the
0134edef
JH
2420hash %utf8::ToSpecLower. The access to the hash is through
2421Perl_to_utf8_case().
6b5c0936 2422
a1433954 2423The C<normal> is a string like "ToLower" which means the swash
0134edef
JH
2424%utf8::ToLower.
2425
2426=cut */
6b5c0936 2427
2104c8d9 2428UV
9a957fbc
AL
2429Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2430 SV **swashp, const char *normal, const char *special)
a0ed51b3 2431{
97aff369 2432 dVAR;
0134edef 2433 STRLEN len = 0;
f4cd282c 2434 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
2435
2436 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2437
9ae3ac1a
KW
2438 /* Note that swash_fetch() doesn't output warnings for these because it
2439 * assumes we will */
8457b38f 2440 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2441 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2442 if (ckWARN_d(WARN_SURROGATE)) {
2443 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2444 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2445 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2446 }
9ae3ac1a
KW
2447 }
2448 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2449 if (ckWARN_d(WARN_NON_UNICODE)) {
2450 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2451 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2452 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2453 }
9ae3ac1a
KW
2454 }
2455
2456 /* Note that non-characters are perfectly legal, so no warning should
2457 * be given */
2458 }
2459
0134edef 2460 if (!*swashp) /* load on-demand */
5ab9d2ef 2461 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2462
a6f87d8c 2463 if (special) {
0134edef 2464 /* It might be "special" (sometimes, but not always,
2a37f04d 2465 * a multicharacter mapping) */
6673a63c 2466 HV * const hv = get_hv(special, 0);
b08cf34e
JH
2467 SV **svp;
2468
35da51f7 2469 if (hv &&
f4cd282c 2470 (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE)) &&
b08cf34e 2471 (*svp)) {
cfd0369c 2472 const char *s;
47654450 2473
cfd0369c 2474 s = SvPV_const(*svp, len);
47654450 2475 if (len == 1)
f4cd282c 2476 /* EIGHTBIT */
c80e42f3 2477 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2478 else {
d2dcd0fb 2479 Copy(s, ustrp, len, U8);
29e98929 2480 }
983ffd37 2481 }
0134edef
JH
2482 }
2483
2484 if (!len && *swashp) {
f4cd282c 2485 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 2486
0134edef
JH
2487 if (uv2) {
2488 /* It was "normal" (a single character mapping). */
f4cd282c 2489 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2490 }
2491 }
1feea2c7 2492
cbe07460
KW
2493 if (len) {
2494 if (lenp) {
2495 *lenp = len;
2496 }
2497 return valid_utf8_to_uvchr(ustrp, 0);
2498 }
2499
2500 /* Here, there was no mapping defined, which means that the code point maps
2501 * to itself. Return the inputs */
bfdf22ec 2502 len = UTF8SKIP(p);
ca9fab46
KW
2503 if (p != ustrp) { /* Don't copy onto itself */
2504 Copy(p, ustrp, len, U8);
2505 }
0134edef 2506
2a37f04d
JH
2507 if (lenp)
2508 *lenp = len;
2509
f4cd282c 2510 return uv1;
cbe07460 2511
a0ed51b3
LW
2512}
2513
051a06d4
KW
2514STATIC UV
2515S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2516{
2517 /* This is called when changing the case of a utf8-encoded character above
2518 * the Latin1 range, and the operation is in locale. If the result
2519 * contains a character that crosses the 255/256 boundary, disallow the
2520 * change, and return the original code point. See L<perlfunc/lc> for why;
2521 *
a1433954
KW
2522 * p points to the original string whose case was changed; assumed
2523 * by this routine to be well-formed
051a06d4
KW
2524 * result the code point of the first character in the changed-case string
2525 * ustrp points to the changed-case string (<result> represents its first char)
2526 * lenp points to the length of <ustrp> */
2527
2528 UV original; /* To store the first code point of <p> */
2529
2530 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2531
a4f12ed7 2532 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2533
2534 /* We know immediately if the first character in the string crosses the
2535 * boundary, so can skip */
2536 if (result > 255) {
2537
2538 /* Look at every character in the result; if any cross the
2539 * boundary, the whole thing is disallowed */
2540 U8* s = ustrp + UTF8SKIP(ustrp);
2541 U8* e = ustrp + *lenp;
2542 while (s < e) {
a4f12ed7 2543 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2544 goto bad_crossing;
2545 }
2546 s += UTF8SKIP(s);
2547 }
2548
2549 /* Here, no characters crossed, result is ok as-is */
2550 return result;
2551 }
2552
2553bad_crossing:
2554
2555 /* Failed, have to return the original */
4b88fb76 2556 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2557 Copy(p, ustrp, *lenp, char);
2558 return original;
2559}
2560
d3e79532 2561/*
87cea99e 2562=for apidoc to_utf8_upper
d3e79532 2563
1f607577 2564Instead use L</toUPPER_utf8>.
a1433954 2565
d3e79532
JH
2566=cut */
2567
051a06d4
KW
2568/* Not currently externally documented, and subject to change:
2569 * <flags> is set iff locale semantics are to be used for code points < 256
2570 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2571 * were used in the calculation; otherwise unchanged. */
2572
2104c8d9 2573UV
051a06d4 2574Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2575{
97aff369 2576 dVAR;
7918f24d 2577
051a06d4
KW
2578 UV result;
2579
2580 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2581
3a4c58c9 2582 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2583 if (flags) {
2584 result = toUPPER_LC(*p);
2585 }
2586 else {
81c6c7ce 2587 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2588 }
3a4c58c9
KW
2589 }
2590 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2591 if (flags) {
94bb8c36 2592 result = toUPPER_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
051a06d4
KW
2593 }
2594 else {
94bb8c36 2595 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2596 ustrp, lenp, 'S');
051a06d4
KW
2597 }
2598 }
2599 else { /* utf8, ord above 255 */
2600 result = CALL_UPPER_CASE(p, ustrp, lenp);
2601
2602 if (flags) {
2603 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2604 }
2605 return result;
2606 }
2607
2608 /* Here, used locale rules. Convert back to utf8 */
2609 if (UTF8_IS_INVARIANT(result)) {
2610 *ustrp = (U8) result;
2611 *lenp = 1;
2612 }
2613 else {
2614 *ustrp = UTF8_EIGHT_BIT_HI(result);
2615 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2616 *lenp = 2;
3a4c58c9
KW
2617 }
2618
051a06d4
KW
2619 if (tainted_ptr) {
2620 *tainted_ptr = TRUE;
2621 }
2622 return result;
983ffd37 2623}
a0ed51b3 2624
d3e79532 2625/*
87cea99e 2626=for apidoc to_utf8_title
d3e79532 2627
1f607577 2628Instead use L</toTITLE_utf8>.
a1433954 2629
d3e79532
JH
2630=cut */
2631
051a06d4
KW
2632/* Not currently externally documented, and subject to change:
2633 * <flags> is set iff locale semantics are to be used for code points < 256
2634 * Since titlecase is not defined in POSIX, uppercase is used instead
2635 * for these/
2636 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2637 * were used in the calculation; otherwise unchanged. */
2638
983ffd37 2639UV
051a06d4 2640Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2641{
97aff369 2642 dVAR;
7918f24d 2643
051a06d4
KW
2644 UV result;
2645
2646 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2647
3a4c58c9 2648 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2649 if (flags) {
2650 result = toUPPER_LC(*p);
2651 }
2652 else {
81c6c7ce 2653 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2654 }
3a4c58c9
KW
2655 }
2656 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2657 if (flags) {
94bb8c36 2658 result = toUPPER_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
051a06d4
KW
2659 }
2660 else {
94bb8c36 2661 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2662 ustrp, lenp, 's');
051a06d4
KW
2663 }
2664 }
2665 else { /* utf8, ord above 255 */
2666 result = CALL_TITLE_CASE(p, ustrp, lenp);
2667
2668 if (flags) {
2669 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2670 }
2671 return result;
2672 }
2673
2674 /* Here, used locale rules. Convert back to utf8 */
2675 if (UTF8_IS_INVARIANT(result)) {
2676 *ustrp = (U8) result;
2677 *lenp = 1;
2678 }
2679 else {
2680 *ustrp = UTF8_EIGHT_BIT_HI(result);
2681 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2682 *lenp = 2;
3a4c58c9
KW
2683 }
2684
051a06d4
KW
2685 if (tainted_ptr) {
2686 *tainted_ptr = TRUE;
2687 }
2688 return result;
a0ed51b3
LW
2689}
2690
d3e79532 2691/*
87cea99e 2692=for apidoc to_utf8_lower
d3e79532 2693
1f607577 2694Instead use L</toLOWER_utf8>.
a1433954 2695
d3e79532
JH
2696=cut */
2697
051a06d4
KW
2698/* Not currently externally documented, and subject to change:
2699 * <flags> is set iff locale semantics are to be used for code points < 256
2700 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2701 * were used in the calculation; otherwise unchanged. */
2702
2104c8d9 2703UV
051a06d4 2704Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2705{
051a06d4
KW
2706 UV result;
2707
97aff369 2708 dVAR;
7918f24d 2709
051a06d4 2710 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2711
968c5e6a 2712 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2713 if (flags) {
2714 result = toLOWER_LC(*p);
2715 }
2716 else {
81c6c7ce 2717 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2718 }
968c5e6a
KW
2719 }
2720 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2721 if (flags) {
94bb8c36 2722 result = toLOWER_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
051a06d4
KW
2723 }
2724 else {
94bb8c36 2725 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2726 ustrp, lenp);
051a06d4 2727 }
968c5e6a 2728 }
051a06d4
KW
2729 else { /* utf8, ord above 255 */
2730 result = CALL_LOWER_CASE(p, ustrp, lenp);
2731
2732 if (flags) {
2733 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2734 }
968c5e6a 2735
051a06d4
KW
2736 return result;
2737 }
2738
2739 /* Here, used locale rules. Convert back to utf8 */
2740 if (UTF8_IS_INVARIANT(result)) {
2741 *ustrp = (U8) result;
2742 *lenp = 1;
2743 }
2744 else {
2745 *ustrp = UTF8_EIGHT_BIT_HI(result);
2746 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2747 *lenp = 2;
2748 }
2749
2750 if (tainted_ptr) {
2751 *tainted_ptr = TRUE;
2752 }
2753 return result;
b4e400f9
JH
2754}
2755
d3e79532 2756/*
87cea99e 2757=for apidoc to_utf8_fold
d3e79532 2758
1f607577 2759Instead use L</toFOLD_utf8>.
a1433954 2760
d3e79532
JH
2761=cut */
2762
051a06d4
KW
2763/* Not currently externally documented, and subject to change,
2764 * in <flags>
2765 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2766 * points < 256. Since foldcase is not defined in
2767 * POSIX, lowercase is used instead
2768 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2769 * otherwise simple folds
a0270393
KW
2770 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2771 * prohibited
051a06d4
KW
2772 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2773 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2774
b4e400f9 2775UV
051a06d4 2776Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2777{
97aff369 2778 dVAR;
7918f24d 2779
051a06d4
KW
2780 UV result;
2781
36bb2ab6 2782 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2783
a0270393
KW
2784 /* These are mutually exclusive */
2785 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2786
50ba90ff
KW
2787 assert(p != ustrp); /* Otherwise overwrites */
2788
a1dde8de 2789 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2790 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2791 result = toFOLD_LC(*p);
051a06d4
KW
2792 }
2793 else {
81c6c7ce 2794 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2795 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2796 }
a1dde8de
KW
2797 }
2798 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2799 if (flags & FOLD_FLAGS_LOCALE) {
94bb8c36 2800 result = toFOLD_LC(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)));
051a06d4
KW
2801 }
2802 else {
94bb8c36 2803 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2804 ustrp, lenp,
1ca267a5 2805 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2806 }
a1dde8de 2807 }
051a06d4 2808 else { /* utf8, ord above 255 */
a0270393 2809 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2810
1ca267a5
KW
2811 if (flags & FOLD_FLAGS_LOCALE) {
2812
9fc2026f 2813 /* Special case these characters, as what normally gets returned
1ca267a5
KW
2814 * under locale doesn't work */
2815 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2816 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2817 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2818 {
2819 goto return_long_s;
2820 }
9fc2026f
KW
2821 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2822 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2823 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2824 {
2825 goto return_ligature_st;
2826 }
a0270393 2827 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2828 }
a0270393
KW
2829 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2830 return result;
2831 }
2832 else {
2833 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2834 * character above the ASCII range, and the result should not
2835 * contain an ASCII character. */
a0270393
KW
2836
2837 UV original; /* To store the first code point of <p> */
2838
2839 /* Look at every character in the result; if any cross the
2840 * boundary, the whole thing is disallowed */
2841 U8* s = ustrp;
2842 U8* e = ustrp + *lenp;
2843 while (s < e) {
2844 if (isASCII(*s)) {
2845 /* Crossed, have to return the original */
2846 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2847
9fc2026f 2848 /* But in these instances, there is an alternative we can
1ca267a5 2849 * return that is valid */
9fc2026f
KW
2850 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2851 || original == LATIN_SMALL_LETTER_SHARP_S)
2852 {
1ca267a5
KW
2853 goto return_long_s;
2854 }
9fc2026f
KW
2855 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2856 goto return_ligature_st;
2857 }
a0270393
KW
2858 Copy(p, ustrp, *lenp, char);
2859 return original;
2860 }
2861 s += UTF8SKIP(s);
2862 }
051a06d4 2863
a0270393
KW
2864 /* Here, no characters crossed, result is ok as-is */
2865 return result;
2866 }
051a06d4
KW
2867 }
2868
2869 /* Here, used locale rules. Convert back to utf8 */
2870 if (UTF8_IS_INVARIANT(result)) {
2871 *ustrp = (U8) result;
2872 *lenp = 1;
2873 }
2874 else {
2875 *ustrp = UTF8_EIGHT_BIT_HI(result);
2876 *(ustrp + 1) = UTF8_EIGHT_BIT_LO(result);
2877 *lenp = 2;
2878 }
2879
2880 if (tainted_ptr) {
2881 *tainted_ptr = TRUE;
2882 }
2883 return result;
1ca267a5
KW
2884
2885 return_long_s:
2886 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2887 * folds to a string of two of these characters. By returning this
2888 * instead, then, e.g.,
2889 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2890 * works. */
2891
2892 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2893 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2894 ustrp, *lenp, U8);
2895 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2896
2897 return_ligature_st:
2898 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2899 * have the other one fold to it */
2900
2901 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2902 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2903 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2904}
2905
711a919c 2906/* Note:
f90a9a02 2907 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2908 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2909 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2910 */
c4a5db0c 2911
a0ed51b3 2912SV*
7fc63493 2913Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2914{
c4a5db0c
KW
2915 PERL_ARGS_ASSERT_SWASH_INIT;
2916
2917 /* Returns a copy of a swash initiated by the called function. This is the
2918 * public interface, and returning a copy prevents others from doing
2919 * mischief on the original */
2920
5d3d13d1 2921 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2922}
2923
2924SV*
5d3d13d1 2925Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2926{
2927 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2928 * by calling utf8_heavy.pl in the general case. The returned value may be
2929 * the swash's inversion list instead if the input parameters allow it.
2930 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2931 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2932 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2933 *
2934 * This interface should only be used by functions that won't destroy or
2935 * adversely change the swash, as doing so affects all other uses of the
2936 * swash in the program; the general public should use 'Perl_swash_init'
2937 * instead.
2938 *
2939 * pkg is the name of the package that <name> should be in.
2940 * name is the name of the swash to find. Typically it is a Unicode
2941 * property name, including user-defined ones
2942 * listsv is a string to initialize the swash with. It must be of the form
2943 * documented as the subroutine return value in
2944 * L<perlunicode/User-Defined Character Properties>
2945 * minbits is the number of bits required to represent each data element.
2946 * It is '1' for binary properties.
2947 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2948 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2949 * flags_p if non-NULL is the address of various input and output flag bits
2950 * to the routine, as follows: ('I' means is input to the routine;
2951 * 'O' means output from the routine. Only flags marked O are
2952 * meaningful on return.)
2953 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2954 * came from a user-defined property. (I O)
5d3d13d1
KW
2955 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2956 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2957 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2958 * return of an inversion list instead of a swash hash if this routine
2959 * thinks that would result in faster execution of swash_fetch() later
2960 * on. (I)
9a53f6cf
KW
2961 *
2962 * Thus there are three possible inputs to find the swash: <name>,
2963 * <listsv>, and <invlist>. At least one must be specified. The result
2964 * will be the union of the specified ones, although <listsv>'s various
2965 * actions can intersect, etc. what <name> gives.
2966 *
2967 * <invlist> is only valid for binary properties */
c4a5db0c 2968
27da23d5 2969 dVAR;
c4a5db0c 2970 SV* retval = &PL_sv_undef;
83199d38 2971 HV* swash_hv = NULL;
87367d5f
KW
2972 const int invlist_swash_boundary =
2973 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2974 ? 512 /* Based on some benchmarking, but not extensive, see commit
2975 message */
2976 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2977
2978 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2979 assert(! invlist || minbits == 1);
2980
2981 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2982 * so */
2983 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2984 dSP;
2985 const size_t pkg_len = strlen(pkg);
2986 const size_t name_len = strlen(name);
2987 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2988 SV* errsv_save;
2989 GV *method;
2990
2991 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2992
2993 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2994 ENTER;
69794297
KW
2995 SAVEHINTS();
2996 save_re_context();
650f067c
JL
2997 /* We might get here via a subroutine signature which uses a utf8
2998 * parameter name, at which point PL_subname will have been set
2999 * but not yet used. */
3000 save_item(PL_subname);
69794297
KW
3001 if (PL_parser && PL_parser->error_count)
3002 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3003 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3004 if (!method) { /* demand load utf8 */
3005 ENTER;
db2c6cb3
FC
3006 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3007 GvSV(PL_errgv) = NULL;
69794297
KW
3008 /* It is assumed that callers of this routine are not passing in
3009 * any user derived data. */
3010 /* Need to do this after save_re_context() as it will set
3011 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3012 * in Perl_magic_get). Even line to create errsv_save can turn on
3013 * PL_tainted. */
284167a5
SM
3014#ifndef NO_TAINT_SUPPORT
3015 SAVEBOOL(TAINT_get);
3016 TAINT_NOT;
3017#endif
69794297
KW
3018 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3019 NULL);
eed484f9 3020 {
db2c6cb3
FC
3021 /* Not ERRSV, as there is no need to vivify a scalar we are
3022 about to discard. */
3023 SV * const errsv = GvSV(PL_errgv);
3024 if (!SvTRUE(errsv)) {
3025 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3026 SvREFCNT_dec(errsv);
3027 }
eed484f9 3028 }
69794297
KW
3029 LEAVE;
3030 }
3031 SPAGAIN;
3032 PUSHMARK(SP);
3033 EXTEND(SP,5);
3034 mPUSHp(pkg, pkg_len);
3035 mPUSHp(name, name_len);
3036 PUSHs(listsv);
3037 mPUSHi(minbits);
3038 mPUSHi(none);
3039 PUTBACK;
db2c6cb3
FC
3040 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3041 GvSV(PL_errgv) = NULL;
69794297
KW
3042 /* If we already have a pointer to the method, no need to use
3043 * call_method() to repeat the lookup. */
c41800a8
KW
3044 if (method
3045 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3046 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3047 {
3048 retval = *PL_stack_sp--;
3049 SvREFCNT_inc(retval);
3050 }
eed484f9 3051 {
db2c6cb3
FC
3052 /* Not ERRSV. See above. */
3053 SV * const errsv = GvSV(PL_errgv);
3054 if (!SvTRUE(errsv)) {
3055 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3056 SvREFCNT_dec(errsv);
3057 }
eed484f9 3058 }
ce3b816e 3059 LEAVE;
69794297
KW
3060 POPSTACK;
3061 if (IN_PERL_COMPILETIME) {
3062 CopHINTS_set(PL_curcop, PL_hints);
3063 }
3064 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3065 if (SvPOK(retval))
3066
3067 /* If caller wants to handle missing properties, let them */
5d3d13d1 3068 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
3069 return NULL;
3070 }
3071 Perl_croak(aTHX_
3072 "Can't find Unicode property definition \"%"SVf"\"",
3073 SVfARG(retval));
3074 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
3075 }
9a53f6cf 3076 } /* End of calling the module to find the swash */
36eb48b4 3077
83199d38
KW
3078 /* If this operation fetched a swash, and we will need it later, get it */
3079 if (retval != &PL_sv_undef
3080 && (minbits == 1 || (flags_p
3081 && ! (*flags_p
3082 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3083 {
3084 swash_hv = MUTABLE_HV(SvRV(retval));
3085
3086 /* If we don't already know that there is a user-defined component to
3087 * this swash, and the user has indicated they wish to know if there is
3088 * one (by passing <flags_p>), find out */
3089 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3090 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3091 if (user_defined && SvUV(*user_defined)) {
3092 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3093 }
3094 }
3095 }
3096
36eb48b4
KW
3097 /* Make sure there is an inversion list for binary properties */
3098 if (minbits == 1) {
3099 SV** swash_invlistsvp = NULL;
3100 SV* swash_invlist = NULL;
9a53f6cf 3101 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3102 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3103 an unclaimed reference count */
36eb48b4 3104
9a53f6cf 3105 /* If this operation fetched a swash, get its already existing
83199d38 3106 * inversion list, or create one for it */
36eb48b4 3107
83199d38 3108 if (swash_hv) {
5c9f4bd2 3109 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3110 if (swash_invlistsvp) {
3111 swash_invlist = *swash_invlistsvp;
3112 invlist_in_swash_is_valid = TRUE;
3113 }
3114 else {
36eb48b4 3115 swash_invlist = _swash_to_invlist(retval);
02c85471 3116 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3117 }
3118 }
3119
3120 /* If an inversion list was passed in, have to include it */
3121 if (invlist) {
3122
3123 /* Any fetched swash will by now have an inversion list in it;
3124 * otherwise <swash_invlist> will be NULL, indicating that we
3125 * didn't fetch a swash */
3126 if (swash_invlist) {
3127
3128 /* Add the passed-in inversion list, which invalidates the one
3129 * already stored in the swash */
3130 invlist_in_swash_is_valid = FALSE;
3131 _invlist_union(invlist, swash_invlist, &swash_invlist);
3132 }
3133 else {
3134
87367d5f
KW
3135 /* Here, there is no swash already. Set up a minimal one, if
3136 * we are going to return a swash */
3137 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3138 swash_hv = newHV();
4aca0fe6 3139 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3140 }
9a53f6cf
KW
3141 swash_invlist = invlist;
3142 }
9a53f6cf
KW
3143 }
3144
3145 /* Here, we have computed the union of all the passed-in data. It may
3146 * be that there was an inversion list in the swash which didn't get
3147 * touched; otherwise save the one computed one */
87367d5f
KW
3148 if (! invlist_in_swash_is_valid
3149 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3150 {
5c9f4bd2 3151 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3152 {
3153 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3154 }
cc34d8c5
FC
3155 /* We just stole a reference count. */
3156 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3157 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3158 }
87367d5f 3159
c41800a8 3160 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3161 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3162 SvREFCNT_dec(retval);
02c85471
FC
3163 if (!swash_invlist_unclaimed)
3164 SvREFCNT_inc_simple_void_NN(swash_invlist);
3165 retval = newRV_noinc(swash_invlist);
87367d5f 3166 }
36eb48b4
KW
3167 }
3168
a0ed51b3
LW
3169 return retval;
3170}
3171
035d37be
JH
3172
3173/* This API is wrong for special case conversions since we may need to
3174 * return several Unicode characters for a single Unicode character
3175 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3176 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
3177 * multiple values. --jhi
3178 * For those, you should use to_utf8_case() instead */
b0e3252e 3179/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3180
3181/* Note:
3182 * Returns the value of property/mapping C<swash> for the first character
3183 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
3184 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
3185 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3186 *
3187 * A "swash" is a hash which contains initially the keys/values set up by
3188 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3189 * property for all possible code points. Things are stored in a compact form
3190 * (see utf8_heavy.pl) so that calculation is required to find the actual
3191 * property value for a given code point. As code points are looked up, new
3192 * key/value pairs are added to the hash, so that the calculation doesn't have
3193 * to ever be re-done. Further, each calculation is done, not just for the
3194 * desired one, but for a whole block of code points adjacent to that one.
3195 * For binary properties on ASCII machines, the block is usually for 64 code
3196 * points, starting with a code point evenly divisible by 64. Thus if the
3197 * property value for code point 257 is requested, the code goes out and
3198 * calculates the property values for all 64 code points between 256 and 319,
3199 * and stores these as a single 64-bit long bit vector, called a "swatch",
3200 * under the key for code point 256. The key is the UTF-8 encoding for code
3201 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3202 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3203 * for code point 258 is then requested, this code realizes that it would be
3204 * stored under the key for 256, and would find that value and extract the
3205 * relevant bit, offset from 256.
3206 *
3207 * Non-binary properties are stored in as many bits as necessary to represent
3208 * their values (32 currently, though the code is more general than that), not
3209 * as single bits, but the principal is the same: the value for each key is a
3210 * vector that encompasses the property values for all code points whose UTF-8
3211 * representations are represented by the key. That is, for all code points
3212 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3213 * bytes of that.
680c470c 3214 */
a0ed51b3 3215UV
680c470c 3216Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3217{
27da23d5 3218 dVAR;
ef8f7699 3219 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3220 U32 klen;
3221 U32 off;
a0ed51b3 3222 STRLEN slen;
7d85a32c 3223 STRLEN needents;
cfd0369c 3224 const U8 *tmps = NULL;
a0ed51b3 3225 U32 bit;
979f2922 3226 SV *swatch;
08fb1ac5 3227 const U8 c = *ptr;
3568d838 3228
7918f24d
NC
3229 PERL_ARGS_ASSERT_SWASH_FETCH;
3230
87367d5f
KW
3231 /* If it really isn't a hash, it isn't really swash; must be an inversion
3232 * list */
3233 if (SvTYPE(hv) != SVt_PVHV) {
3234 return _invlist_contains_cp((SV*)hv,
3235 (do_utf8)
3236 ? valid_utf8_to_uvchr(ptr, NULL)
3237 : c);
3238 }
3239
08fb1ac5
KW
3240 /* We store the values in a "swatch" which is a vec() value in a swash
3241 * hash. Code points 0-255 are a single vec() stored with key length
3242 * (klen) 0. All other code points have a UTF-8 representation
3243 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3244 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3245 * length for them is the length of the encoded char - 1. ptr[klen] is the
3246 * final byte in the sequence representing the character */
3247 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3248 klen = 0;
3249 needents = 256;
3250 off = c;
3568d838 3251 }
08fb1ac5
KW
3252 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3253 klen = 0;
3254 needents = 256;
3255 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
3256 }
3257 else {
08fb1ac5
KW
3258 klen = UTF8SKIP(ptr) - 1;
3259
3260 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3261 * the vec is the final byte in the sequence. (In EBCDIC this is
3262 * converted to I8 to get consecutive values.) To help you visualize
3263 * all this:
3264 * Straight 1047 After final byte
3265 * UTF-8 UTF-EBCDIC I8 transform
3266 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3267 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3268 * ...
3269 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3270 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3271 * ...
3272 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3273 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3274 * ...
3275 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3276 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3277 * ...
3278 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3279 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3280 *
3281 * (There are no discontinuities in the elided (...) entries.)
3282 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3283 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3284 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3285 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3286 * index into the vec() swatch (after subtracting 0x80, which we
3287 * actually do with an '&').
3288 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3289 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3290 * dicontinuities which go away by transforming it into I8, and we
3291 * effectively subtract 0xA0 to get the index. */
979f2922 3292 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 3293 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 3294 }
7d85a32c 3295
a0ed51b3
LW
3296 /*
3297 * This single-entry cache saves about 1/3 of the utf8 overhead in test
3298 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3299 * it's nothing to sniff at.) Pity we usually come through at least
3300 * two function calls to get here...
3301 *
3302 * NB: this code assumes that swatches are never modified, once generated!
3303 */
3304
3568d838 3305 if (hv == PL_last_swash_hv &&
a0ed51b3 3306 klen == PL_last_swash_klen &&
27da23d5 3307 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3308 {
3309 tmps = PL_last_swash_tmps;
3310 slen = PL_last_swash_slen;
3311 }
3312 else {
3313 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3314 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3315
b0e3252e 3316 /* If not cached, generate it via swatch_get */
979f2922 3317 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
3318 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3319 {
3320 if (klen) {
3321 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3322 swatch = swatch_get(swash,
3323 code_point & ~((UV)needents - 1),
3324 needents);
3325 }
3326 else { /* For the first 256 code points, the swatch has a key of
3327 length 0 */
3328 swatch = swatch_get(swash, 0, needents);
3329 }
979f2922 3330
923e4eb5 3331 if (IN_PERL_COMPILETIME)
623e6609 3332 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3333
979f2922 3334 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3335
979f2922
ST
3336 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3337 || (slen << 3) < needents)
5637ef5b
NC
3338 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3339 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3340 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3341 }
3342
3343 PL_last_swash_hv = hv;
16d8f38a 3344 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3345 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3346 /* FIXME change interpvar.h? */
3347 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3348 PL_last_swash_slen = slen;
3349 if (klen)
3350 Copy(ptr, PL_last_swash_key, klen, U8);
3351 }
3352
9faf8d75 3353 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
3354 case 1:
3355 bit = 1 << (off & 7);
3356 off >>= 3;
3357 return (tmps[off] & bit) != 0;
3358 case 8:
3359 return tmps[off];
3360 case 16:
3361 off <<= 1;
3362 return (tmps[off] << 8) + tmps[off + 1] ;
3363 case 32:
3364 off <<= 2;
3365 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
3366 }
5637ef5b
NC
3367 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3368 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3369 NORETURN_FUNCTION_END;
a0ed51b3 3370}
2b9d42f0 3371
319009ee
KW
3372/* Read a single line of the main body of the swash input text. These are of
3373 * the form:
3374 * 0053 0056 0073
3375 * where each number is hex. The first two numbers form the minimum and
3376 * maximum of a range, and the third is the value associated with the range.
3377 * Not all swashes should have a third number
3378 *
3379 * On input: l points to the beginning of the line to be examined; it points
3380 * to somewhere in the string of the whole input text, and is
3381 * terminated by a \n or the null string terminator.
3382 * lend points to the null terminator of that string
3383 * wants_value is non-zero if the swash expects a third number
3384 * typestr is the name of the swash's mapping, like 'ToLower'
3385 * On output: *min, *max, and *val are set to the values read from the line.
3386 * returns a pointer just beyond the line examined. If there was no
3387 * valid min number on the line, returns lend+1
3388 */
3389
3390STATIC U8*
3391S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3392 const bool wants_value, const U8* const typestr)
3393{
3394 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3395 STRLEN numlen; /* Length of the number */
02470786
KW
3396 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3397 | PERL_SCAN_DISALLOW_PREFIX
3398 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3399
3400 /* nl points to the next \n in the scan */
3401 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3402
3403 /* Get the first number on the line: the range minimum */
3404 numlen = lend - l;
3405 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3406 if (numlen) /* If found a hex number, position past it */
3407 l += numlen;
3408 else if (nl) { /* Else, go handle next line, if any */
3409 return nl + 1; /* 1 is length of "\n" */
3410 }
3411 else { /* Else, no next line */
3412 return lend + 1; /* to LIST's end at which \n is not found */
3413 }
3414
3415 /* The max range value follows, separated by a BLANK */
3416 if (isBLANK(*l)) {
3417 ++l;
02470786
KW
3418 flags = PERL_SCAN_SILENT_ILLDIGIT
3419 | PERL_SCAN_DISALLOW_PREFIX
3420 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3421 numlen = lend - l;
3422 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3423 if (numlen)
3424 l += numlen;
3425 else /* If no value here, it is a single element range */
3426 *max = *min;
3427
3428 /* Non-binary tables have a third entry: what the first element of the
3429 * range maps to */
3430 if (wants_value) {
3431 if (isBLANK(*l)) {
3432 ++l;
f90a9a02
KW
3433
3434 /* The ToLc, etc table mappings are not in hex, and must be
3435 * corrected by adding the code point to them */
3436 if (typeto) {
3437 char *after_strtol = (char *) lend;
3438 *val = Strtol((char *)l, &after_strtol, 10);
3439 l = (U8 *) after_strtol;
f90a9a02
KW
3440 }
3441 else { /* Other tables are in hex, and are the correct result
3442 without tweaking */
a9d188b3
KW
3443 flags = PERL_SCAN_SILENT_ILLDIGIT
3444 | PERL_SCAN_DISALLOW_PREFIX
3445 | PERL_SCAN_SILENT_NON_PORTABLE;
3446 numlen = lend - l;
3447 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3448 if (numlen)
3449 l += numlen;
3450 else
3451 *val = 0;
f90a9a02 3452 }
319009ee
KW
3453 }
3454 else {
3455 *val = 0;
3456 if (typeto) {
dcbac5bb 3457 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3458 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3459 typestr, l);
3460 }
3461 }
3462 }
3463 else
3464 *val = 0; /* bits == 1, then any val should be ignored */
3465 }
3466 else { /* Nothing following range min, should be single element with no
3467 mapping expected */
3468 *max = *min;
3469 if (wants_value) {
3470 *val = 0;
3471 if (typeto) {
dcbac5bb 3472 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3473 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3474 }
3475 }
3476 else
3477 *val = 0; /* bits == 1, then val should be ignored */
3478 }
3479
3480 /* Position to next line if any, or EOF */
3481 if (nl)
3482 l = nl + 1;
3483 else
3484 l = lend;
3485
3486 return l;
3487}
3488
979f2922
ST
3489/* Note:
3490 * Returns a swatch (a bit vector string) for a code point sequence
3491 * that starts from the value C<start> and comprises the number C<span>.
3492 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3493 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3494 */
3495STATIC SV*
b0e3252e 3496S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3497{
3498 SV *swatch;
77f9f126 3499 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3500 STRLEN lcur, xcur, scur;
ef8f7699 3501 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3502 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3503
88d45d28
KW
3504 SV** listsvp = NULL; /* The string containing the main body of the table */
3505 SV** extssvp = NULL;
3506 SV** invert_it_svp = NULL;
3507 U8* typestr = NULL;
786861f5
KW
3508 STRLEN bits;
3509 STRLEN octets; /* if bits == 1, then octets == 0 */
3510 UV none;
3511 UV end = start + span;
972dd592 3512
36eb48b4 3513 if (invlistsvp == NULL) {
786861f5
KW
3514 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3515 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3516 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3517 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3518 listsvp = hv_fetchs(hv, "LIST", FALSE);
3519 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3520
3521 bits = SvUV(*bitssvp);
3522 none = SvUV(*nonesvp);
3523 typestr = (U8*)SvPV_nolen(*typesvp);
3524 }
36eb48b4
KW
3525 else {
3526 bits = 1;
3527 none = 0;
3528 }
786861f5 3529 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3530
b0e3252e 3531 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3532
979f2922 3533 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3534 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3535 (UV)bits);
979f2922
ST
3536 }
3537
84ea5ef6
KW
3538 /* If overflowed, use the max possible */
3539 if (end < start) {
3540 end = UV_MAX;
3541 span = end - start;
3542 }
3543
979f2922 3544 /* create and initialize $swatch */
979f2922 3545 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3546 swatch = newSV(scur);
3547 SvPOK_on(swatch);
979f2922
ST
3548 s = (U8*)SvPVX(swatch);
3549 if (octets && none) {
0bd48802 3550 const U8* const e = s + scur;
979f2922
ST
3551 while (s < e) {
3552 if (bits == 8)
3553 *s++ = (U8)(none & 0xff);
3554 else if (bits == 16) {
3555 *s++ = (U8)((none >> 8) & 0xff);
3556 *s++ = (U8)( none & 0xff);
3557 }
3558 else if (bits == 32) {
3559 *s++ = (U8)((none >> 24) & 0xff);
3560 *s++ = (U8)((none >> 16) & 0xff);
3561 *s++ = (U8)((none >> 8) & 0xff);
3562 *s++ = (U8)( none & 0xff);
3563 }
3564 }
3565 *s = '\0';
3566 }
3567 else {
3568 (void)memzero((U8*)s, scur + 1);
3569 }
3570 SvCUR_set(swatch, scur);
3571 s = (U8*)SvPVX(swatch);
3572
36eb48b4
KW
3573 if (invlistsvp) { /* If has an inversion list set up use that */
3574 _invlist_populate_swatch(*invlistsvp, start, end, s);
3575 return swatch;
3576 }
3577
3578 /* read $swash->{LIST} */
979f2922
ST
3579 l = (U8*)SvPV(*listsvp, lcur);
3580 lend = l + lcur;
3581 while (l < lend) {
8ed25d53 3582 UV min, max, val, upper;
319009ee
KW
3583 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3584 cBOOL(octets), typestr);
3585 if (l > lend) {
979f2922
ST
3586 break;
3587 }
3588
972dd592 3589 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3590 if (max < start)
3591 continue;
3592
8ed25d53
KW
3593 /* <end> is generally 1 beyond where we want to set things, but at the
3594 * platform's infinity, where we can't go any higher, we want to
3595 * include the code point at <end> */
3596 upper = (max < end)
3597 ? max
3598 : (max != UV_MAX || end != UV_MAX)
3599 ? end - 1
3600 : end;
3601
979f2922 3602 if (octets) {
35da51f7 3603 UV key;
979f2922
ST
3604 if (min < start) {
3605 if (!none || val < none) {
3606 val += start - min;
3607 }
3608 min = start;
3609 }
8ed25d53 3610 for (key = min; key <= upper; key++) {
979f2922 3611 STRLEN offset;
979f2922
ST
3612 /* offset must be non-negative (start <= min <= key < end) */
3613 offset = octets * (key - start);
3614 if (bits == 8)
3615 s[offset] = (U8)(val & 0xff);
3616 else if (bits == 16) {
3617 s[offset ] = (U8)((val >> 8) & 0xff);
3618 s[offset + 1] = (U8)( val & 0xff);
3619 }
3620 else if (bits == 32) {
3621 s[offset ] = (U8)((val >> 24) & 0xff);
3622 s[offset + 1] = (U8)((val >> 16) & 0xff);
3623 s[offset + 2] = (U8)((val >> 8) & 0xff);
3624 s[offset + 3] = (U8)( val & 0xff);
3625 }
3626
3627 if (!none || val < none)
3628 ++val;
3629 }
3630 }
711a919c 3631 else { /* bits == 1, then val should be ignored */
35da51f7 3632 UV key;
979f2922
ST
3633 if (min < start)
3634 min = start;
6cb05c12 3635
8ed25d53 3636 for (key = min; key <= upper; key++) {
0bd48802 3637 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3638 s[offset >> 3] |= 1 << (offset & 7);
3639 }
3640 }
3641 } /* while */
979f2922 3642
9479a769 3643 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3644 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3645
3646 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3647 * be 0, and their inversion should also be 0, as we don't succeed any
3648 * Unicode property matches for non-Unicode code points */
3649 if (start <= PERL_UNICODE_MAX) {
3650
3651 /* The code below assumes that we never cross the
3652 * Unicode/above-Unicode boundary in a range, as otherwise we would
3653 * have to figure out where