This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Move initialization of PL_XPosix_ptrs[] to perl.c
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
a0ed51b3 35
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 42This file contains various utility functions for manipulating UTF8-encoded
72d33970 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
0cfa64bf 110 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 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 }
6588300d 193#ifdef UTF8_QUAD_MAX
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 }
6588300d 206#ifdef UTF8_QUAD_MAX
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 230of the string C<d>; C<d> should have at least C<UTF8_MAXBYTES+1> free
72d33970
FC
231bytes available. The return value is the pointer to the byte after the
232end of the new character. In other words,
07693fe6
KW
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
72d33970
FC
260bytes available. The return value is the pointer to the byte after the
261end of the new character. In other words,
de69f3af
KW
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
4c3cfd5d 282The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 283affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 284UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
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
492a624f
KW
369 if (IS_UTF8_CHAR_FAST(len))
370 return IS_UTF8_CHAR(buf, len) ? len : 0;
492a624f
KW
371 return is_utf8_char_slow(buf, len);
372}
373
374/*
87cea99e 375=for apidoc is_utf8_char
eebe1485 376
5da9da9e 377Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
378character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
379character is a valid UTF-8 character. The actual number of bytes in the UTF-8
380character will be returned if it is valid, otherwise 0.
9041c2e3 381
76848387 382This function is deprecated due to the possibility that malformed input could
a1433954 383cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 384instead.
e0328548 385
82686b01 386=cut */
76848387 387
067a85ef 388STRLEN
668b6d8d 389Perl_is_utf8_char(const U8 *s)
386d01d6 390{
7918f24d 391 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 392
76848387 393 /* Assumes we have enough space, which is why this is deprecated */
492a624f 394 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
395}
396
eaf7a4d2 397
6662521e 398/*
87cea99e 399=for apidoc is_utf8_string
6662521e 400
a1433954 401Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 402UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
e0328548
KW
403using C<strlen(s)> (which means if you use this option, that C<s> has to have a
404terminating NUL byte). Note that all characters being ASCII constitute 'a
405valid UTF-8 string'.
6662521e 406
a1433954 407See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 408
6662521e
GS
409=cut
410*/
411
8e84507e 412bool
668b6d8d 413Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 414{
35da51f7 415 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 416 const U8* x = s;
067a85ef 417
7918f24d 418 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 419
6662521e 420 while (x < send) {
1acdb0da 421 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
422 if (UTF8_IS_INVARIANT(*x)) {
423 x++;
424 }
1acdb0da
JH
425 else {
426 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
427 const STRLEN c = UTF8SKIP(x);
428 const U8* const next_char_ptr = x + c;
429
430 if (next_char_ptr > send) {
431 return FALSE;
432 }
433
768c67ee
JH
434 if (IS_UTF8_CHAR_FAST(c)) {
435 if (!IS_UTF8_CHAR(x, c))
e0328548 436 return FALSE;
3c614e38 437 }
e0328548
KW
438 else if (! is_utf8_char_slow(x, c)) {
439 return FALSE;
440 }
441 x = next_char_ptr;
1acdb0da 442 }
6662521e 443 }
768c67ee 444
067a85ef 445 return TRUE;
6662521e
GS
446}
447
67e989fb 448/*
814fafa7
NC
449Implemented as a macro in utf8.h
450
87cea99e 451=for apidoc is_utf8_string_loc
814fafa7 452
a1433954
KW
453Like L</is_utf8_string> but stores the location of the failure (in the
454case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
455"utf8ness success") in the C<ep>.
456
a1433954 457See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 458
87cea99e 459=for apidoc is_utf8_string_loclen
81cd54e3 460
a1433954
KW
461Like L</is_utf8_string>() but stores the location of the failure (in the
462case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
463"utf8ness success") in the C<ep>, and the number of UTF-8
464encoded characters in the C<el>.
465
a1433954 466See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
467
468=cut
469*/
470
471bool
668b6d8d 472Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 473{
35da51f7 474 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 475 const U8* x = s;
81cd54e3 476 STRLEN c;
3ebfea28 477 STRLEN outlen = 0;
7918f24d
NC
478
479 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 480
81cd54e3 481 while (x < send) {
e0328548
KW
482 const U8* next_char_ptr;
483
81cd54e3
JH
484 /* Inline the easy bits of is_utf8_char() here for speed... */
485 if (UTF8_IS_INVARIANT(*x))
e0328548 486 next_char_ptr = x + 1;
81cd54e3 487 else {
768c67ee 488 /* ... and call is_utf8_char() only if really needed. */
768c67ee 489 c = UTF8SKIP(x);
e0328548
KW
490 next_char_ptr = c + x;
491 if (next_char_ptr > send) {
492 goto out;
493 }
768c67ee
JH
494 if (IS_UTF8_CHAR_FAST(c)) {
495 if (!IS_UTF8_CHAR(x, c))
496 c = 0;
497 } else
498 c = is_utf8_char_slow(x, c);
768c67ee
JH
499 if (!c)
500 goto out;
81cd54e3 501 }
e0328548 502 x = next_char_ptr;
3ebfea28 503 outlen++;
81cd54e3 504 }
768c67ee
JH
505
506 out:
3ebfea28
AL
507 if (el)
508 *el = outlen;
509
768c67ee
JH
510 if (ep)
511 *ep = x;
3ebfea28 512 return (x == send);
81cd54e3
JH
513}
514
515/*
768c67ee 516
de69f3af 517=for apidoc utf8n_to_uvchr
378516de
KW
518
519THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 520Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 521
9041c2e3 522Bottom level UTF-8 decode routine.
de69f3af 523Returns the native code point value of the first character in the string C<s>,
746afd53
KW
524which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
525C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
526the length, in bytes, of that character.
949cf498
KW
527
528The value of C<flags> determines the behavior when C<s> does not point to a
529well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
530zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
531next possible position in C<s> that could begin a non-malformed character.
532Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
533
534Various ALLOW flags can be set in C<flags> to allow (and not warn on)
535individual types of malformations, such as the sequence being overlong (that
536is, when there is a shorter sequence that can express the same code point;
537overlong sequences are expressly forbidden in the UTF-8 standard due to
538potential security issues). Another malformation example is the first byte of
539a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
540flags. For allowed 0 length strings, this function returns 0; for allowed
541overlong sequences, the computed code point is returned; for all other allowed
542malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
543determinable reasonable value.
949cf498
KW
544
545The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
546flags) malformation is found. If this flag is set, the routine assumes that
547the caller will raise a warning, and this function will silently just set
d088425d
KW
548C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
549
550Note that this API requires disambiguation between successful decoding a NUL
551character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
552in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
553first byte of C<s> is 0 as well. If so, the input was a NUL; if not, the input
554had an error.
949cf498
KW
555
556Certain code points are considered problematic. These are Unicode surrogates,
746afd53 557Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 558By default these are considered regular code points, but certain situations
5eafe189 559warrant special handling for them. If C<flags> contains
949cf498
KW
560UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
561malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
562UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
563maximum) can be set to disallow these categories individually.
564
565The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
566UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
567for their respective categories, but otherwise the code points are considered
568valid (not malformations). To get a category to both be treated as a
569malformation and raise a warning, specify both the WARN and DISALLOW flags.
570(But note that warnings are not raised if lexically disabled nor if
571UTF8_CHECK_ONLY is also specified.)
572
573Very large code points (above 0x7FFF_FFFF) are considered more problematic than
574the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
575reasons: they requre at least 32 bits to represent them on ASCII platforms, are
576not representable at all on EBCDIC platforms, and the original UTF-8
577specification never went above this number (the current 0x10FFFF limit was
578imposed later). (The smaller ones, those that fit into 32 bits, are
579representable by a UV on ASCII platforms, but not by an IV, which means that
580the number of operations that can be performed on them is quite restricted.)
581The UTF-8 encoding on ASCII platforms for these large code points begins with a
582byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
583be treated as malformations, while allowing smaller above-Unicode code points.
584(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
585including these, as malformations.)
586Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 587the other WARN flags, but applies just to these code points.
949cf498
KW
588
589All other code points corresponding to Unicode characters, including private
590use and those yet to be assigned, are never considered malformed and never
591warn.
67e989fb 592
37607a96
PK
593=cut
594*/
67e989fb 595
a0ed51b3 596UV
de69f3af 597Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 598{
97aff369 599 dVAR;
d4c19fe8 600 const U8 * const s0 = s;
eb83ed87 601 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 602 U8 * send;
eb83ed87
KW
603 UV uv = *s;
604 STRLEN expectlen;
949cf498 605 SV* sv = NULL;
eb83ed87
KW
606 UV outlier_ret = 0; /* return value when input is in error or problematic
607 */
608 UV pack_warn = 0; /* Save result of packWARN() for later */
609 bool unexpected_non_continuation = FALSE;
610 bool overflowed = FALSE;
2f8f112e 611 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 612
eb83ed87 613 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 614
de69f3af 615 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 616
eb83ed87
KW
617 /* The order of malformation tests here is important. We should consume as
618 * few bytes as possible in order to not skip any valid character. This is
619 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
620 * http://unicode.org/reports/tr36 for more discussion as to why. For
621 * example, once we've done a UTF8SKIP, we can tell the expected number of
622 * bytes, and could fail right off the bat if the input parameters indicate
623 * that there are too few available. But it could be that just that first
624 * byte is garbled, and the intended character occupies fewer bytes. If we
625 * blindly assumed that the first byte is correct, and skipped based on
626 * that number, we could skip over a valid input character. So instead, we
627 * always examine the sequence byte-by-byte.
628 *
629 * We also should not consume too few bytes, otherwise someone could inject
630 * things. For example, an input could be deliberately designed to
631 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 632 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
633 * which is actually part of of the overflowing sequence), that could look
634 * legitimate to the caller, which could discard the initial partial
635 * sequence and process the rest, inappropriately */
636
637 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 638 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
639 if (retlen) {
640 *retlen = 0;
641 }
a0dbb045 642
eb83ed87
KW
643 if (flags & UTF8_ALLOW_EMPTY) {
644 return 0;
645 }
646 if (! (flags & UTF8_CHECK_ONLY)) {
647 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
648 }
0c443dc2
JH
649 goto malformed;
650 }
651
eb83ed87
KW
652 expectlen = UTF8SKIP(s);
653
654 /* A well-formed UTF-8 character, as the vast majority of calls to this
655 * function will be for, has this expected length. For efficiency, set
656 * things up here to return it. It will be overriden only in those rare
657 * cases where a malformation is found */
658 if (retlen) {
659 *retlen = expectlen;
660 }
661
662 /* An invariant is trivially well-formed */
1d72bdf6 663 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 664 return uv;
a0ed51b3 665 }
67e989fb 666
eb83ed87 667 /* A continuation character can't start a valid sequence */
b5b9af04 668 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
669 if (flags & UTF8_ALLOW_CONTINUATION) {
670 if (retlen) {
671 *retlen = 1;
672 }
673 return UNICODE_REPLACEMENT;
674 }
ba210ebe 675
eb83ed87
KW
676 if (! (flags & UTF8_CHECK_ONLY)) {
677 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
678 }
679 curlen = 1;
ba210ebe
JH
680 goto malformed;
681 }
9041c2e3 682
dcd27b3c
KW
683 /* Here is not a continuation byte, nor an invariant. The only thing left
684 * is a start byte (possibly for an overlong) */
685
1d72bdf6 686#ifdef EBCDIC
bc3632a8 687 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
688#endif
689
eb83ed87
KW
690 /* Remove the leading bits that indicate the number of bytes in the
691 * character's whole UTF-8 sequence, leaving just the bits that are part of
692 * the value */
693 uv &= UTF_START_MASK(expectlen);
ba210ebe 694
eb83ed87
KW
695 /* Now, loop through the remaining bytes in the character's sequence,
696 * accumulating each into the working value as we go. Be sure to not look
697 * past the end of the input string */
0b8d30e8
KW
698 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
699
eb83ed87 700 for (s = s0 + 1; s < send; s++) {
b5b9af04 701 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
702#ifndef EBCDIC /* Can't overflow in EBCDIC */
703 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
704
705 /* The original implementors viewed this malformation as more
706 * serious than the others (though I, khw, don't understand
707 * why, since other malformations also give very very wrong
708 * results), so there is no way to turn off checking for it.
709 * Set a flag, but keep going in the loop, so that we absorb
710 * the rest of the bytes that comprise the character. */
711 overflowed = TRUE;
712 overflow_byte = *s; /* Save for warning message's use */
713 }
714#endif
8850bf83 715 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
716 }
717 else {
718 /* Here, found a non-continuation before processing all expected
719 * bytes. This byte begins a new character, so quit, even if
720 * allowing this malformation. */
721 unexpected_non_continuation = TRUE;
722 break;
723 }
724 } /* End of loop through the character's bytes */
725
726 /* Save how many bytes were actually in the character */
727 curlen = s - s0;
728
729 /* The loop above finds two types of malformations: non-continuation and/or
730 * overflow. The non-continuation malformation is really a too-short
731 * malformation, as it means that the current character ended before it was
732 * expected to (being terminated prematurely by the beginning of the next
733 * character, whereas in the too-short malformation there just are too few
734 * bytes available to hold the character. In both cases, the check below
735 * that we have found the expected number of bytes would fail if executed.)
736 * Thus the non-continuation malformation is really unnecessary, being a
737 * subset of the too-short malformation. But there may be existing
738 * applications that are expecting the non-continuation type, so we retain
739 * it, and return it in preference to the too-short malformation. (If this
740 * code were being written from scratch, the two types might be collapsed
741 * into one.) I, khw, am also giving priority to returning the
742 * non-continuation and too-short malformations over overflow when multiple
743 * ones are present. I don't know of any real reason to prefer one over
744 * the other, except that it seems to me that multiple-byte errors trumps
745 * errors from a single byte */
b5b9af04 746 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
747 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
748 if (! (flags & UTF8_CHECK_ONLY)) {
749 if (curlen == 1) {
750 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
751 }
752 else {
753 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
754 }
755 }
eb83ed87
KW
756 goto malformed;
757 }
758 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
759
760 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
761 * as what the original expectations were. */
762 do_overlong_test = FALSE;
eb83ed87
KW
763 if (retlen) {
764 *retlen = curlen;
765 }
766 }
b5b9af04 767 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
768 if (! (flags & UTF8_ALLOW_SHORT)) {
769 if (! (flags & UTF8_CHECK_ONLY)) {
770 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 771 }
eb83ed87
KW
772 goto malformed;
773 }
774 uv = UNICODE_REPLACEMENT;
2f8f112e 775 do_overlong_test = FALSE;
eb83ed87
KW
776 if (retlen) {
777 *retlen = curlen;
778 }
779 }
780
ea5ced44 781#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 782 if (UNLIKELY(overflowed)) {
eb83ed87 783 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 784 goto malformed;
eb83ed87
KW
785 }
786#endif
787
2f8f112e 788 if (do_overlong_test
5aaebcb3 789 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
790 && ! (flags & UTF8_ALLOW_LONG))
791 {
eb83ed87
KW
792 /* The overlong malformation has lower precedence than the others.
793 * Note that if this malformation is allowed, we return the actual
794 * value, instead of the replacement character. This is because this
795 * value is actually well-defined. */
796 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 797 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
798 }
799 goto malformed;
800 }
801
1a89bb6c 802 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
803 * problematic code point that is not allowed by the input parameters. */
804 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
805 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
806 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
807 {
949cf498 808 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
809
810 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
811 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 812 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 813 && ckWARN_d(WARN_SURROGATE))
eb83ed87 814 {
111d382d 815 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 816 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
817 }
818 if (flags & UTF8_DISALLOW_SURROGATE) {
819 goto disallowed;
820 }
821 }
949cf498 822 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 823 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 824 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 825 {
111d382d 826 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 827 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 828 }
ea5ced44
KW
829#ifndef EBCDIC /* EBCDIC always allows FE, FF */
830
831 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
832 * points. We test for these after the regular SUPER ones, and
833 * before possibly bailing out, so that the more dire warning
834 * overrides the regular one, if applicable */
835 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
836 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
837 {
838 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
839 == UTF8_WARN_FE_FF
840 && ckWARN_d(WARN_UTF8))
841 {
842 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
843 pack_warn = packWARN(WARN_UTF8);
844 }
845 if (flags & UTF8_DISALLOW_FE_FF) {
846 goto disallowed;
847 }
848 }
849#endif
949cf498
KW
850 if (flags & UTF8_DISALLOW_SUPER) {
851 goto disallowed;
852 }
853 }
4190d317
KW
854 else if (UNICODE_IS_NONCHAR(uv)) {
855 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 856 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
857 {
858 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 859 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
860 }
861 if (flags & UTF8_DISALLOW_NONCHAR) {
862 goto disallowed;
863 }
864 }
949cf498 865
eb83ed87 866 if (sv) {
de69f3af
KW
867 outlier_ret = uv; /* Note we don't bother to convert to native,
868 as all the outlier code points are the same
869 in both ASCII and EBCDIC */
eb83ed87
KW
870 goto do_warn;
871 }
872
949cf498
KW
873 /* Here, this is not considered a malformed character, so drop through
874 * to return it */
a0ed51b3 875 }
ba210ebe 876
de69f3af 877 return UNI_TO_NATIVE(uv);
ba210ebe 878
eb83ed87
KW
879 /* There are three cases which get to beyond this point. In all 3 cases:
880 * <sv> if not null points to a string to print as a warning.
881 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
882 * set.
883 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
884 * This is done by initializing it to 0, and changing it only
885 * for case 1).
886 * The 3 cases are:
887 * 1) The input is valid but problematic, and to be warned about. The
888 * return value is the resultant code point; <*retlen> is set to
889 * <curlen>, the number of bytes that comprise the code point.
890 * <pack_warn> contains the result of packWARN() for the warning
891 * types. The entry point for this case is the label <do_warn>;
892 * 2) The input is a valid code point but disallowed by the parameters to
893 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
894 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
895 * comprise the code point. <pack_warn> contains the result of
896 * packWARN() for the warning types. The entry point for this case is
897 * the label <disallowed>.
898 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
899 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
900 * bytes that comprise the malformation. All such malformations are
901 * assumed to be warning type <utf8>. The entry point for this case
902 * is the label <malformed>.
903 */
949cf498 904
ba210ebe
JH
905malformed:
906
eb83ed87
KW
907 if (sv && ckWARN_d(WARN_UTF8)) {
908 pack_warn = packWARN(WARN_UTF8);
909 }
910
911disallowed:
912
fcc8fcf6 913 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 914 if (retlen)
10edeb5d 915 *retlen = ((STRLEN) -1);
ba210ebe
JH
916 return 0;
917 }
918
eb83ed87 919do_warn:
5b311467 920
eb83ed87
KW
921 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
922 if warnings are to be raised. */
f555bc63 923 const char * const string = SvPVX_const(sv);
a0dbb045 924
f555bc63
KW
925 if (PL_op)
926 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
927 else
928 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
929 }
930
eb83ed87
KW
931 if (retlen) {
932 *retlen = curlen;
933 }
ba210ebe 934
eb83ed87 935 return outlier_ret;
a0ed51b3
LW
936}
937
8e84507e 938/*
ec5f19d0
KW
939=for apidoc utf8_to_uvchr_buf
940
941Returns the native code point of the first character in the string C<s> which
942is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 943C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 944
524080c4
KW
945If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
946enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
947NULL) to -1. If those warnings are off, the computed value, if well-defined
948(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
949C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
950the next possible position in C<s> that could begin a non-malformed character.
de69f3af 951See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 952returned.
ec5f19d0
KW
953
954=cut
955*/
956
957
958UV
959Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
960{
ec5f19d0
KW
961 assert(s < send);
962
963 return utf8n_to_uvchr(s, send - s, retlen,
964 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
965}
966
27d6c58a 967/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 968 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 969 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
970
971UV
972Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
973{
010ab96b
KW
974 UV expectlen = UTF8SKIP(s);
975 const U8* send = s + expectlen;
9ff2f0f7 976 UV uv = *s;
3986bb7c 977
27d6c58a
KW
978 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
979
010ab96b
KW
980 if (retlen) {
981 *retlen = expectlen;
982 }
983
984 /* An invariant is trivially returned */
985 if (expectlen == 1) {
9ff2f0f7 986 return uv;
010ab96b
KW
987 }
988
9ff2f0f7
KW
989#ifdef EBCDIC
990 uv = NATIVE_UTF8_TO_I8(uv);
991#endif
992
010ab96b
KW
993 /* Remove the leading bits that indicate the number of bytes, leaving just
994 * the bits that are part of the value */
995 uv &= UTF_START_MASK(expectlen);
996
997 /* Now, loop through the remaining bytes, accumulating each into the
998 * working total as we go. (I khw tried unrolling the loop for up to 4
999 * bytes, but there was no performance improvement) */
1000 for (++s; s < send; s++) {
1001 uv = UTF8_ACCUMULATE(uv, *s);
1002 }
1003
3986bb7c 1004 return UNI_TO_NATIVE(uv);
010ab96b 1005
27d6c58a
KW
1006}
1007
ec5f19d0 1008/*
87cea99e 1009=for apidoc utf8_to_uvchr
9041c2e3 1010
6ee84de2 1011Returns the native code point of the first character in the string C<s>
1e54db1a 1012which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1013length, in bytes, of that character.
1014
4b88fb76 1015Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1016malformed input could cause reading beyond the end of the input buffer, which
1017is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 1018
524080c4
KW
1019If C<s> points to one of the detected malformations, and UTF8 warnings are
1020enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1021NULL) to -1. If those warnings are off, the computed value if well-defined (or
1022the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1023is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1024next possible position in C<s> that could begin a non-malformed character.
de69f3af 1025See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
1026
1027=cut
1028*/
1029
1030UV
7fc63493 1031Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 1032{
7918f24d
NC
1033 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
1034
2ff6c191 1035 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
1036}
1037
1038/*
ec5f19d0
KW
1039=for apidoc utf8_to_uvuni_buf
1040
de69f3af
KW
1041Only in very rare circumstances should code need to be dealing in Unicode
1042(as opposed to native) code points. In those few cases, use
1043C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1044
1045Returns the Unicode (not-native) code point of the first character in the
1046string C<s> which
ec5f19d0
KW
1047is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1048C<retlen> will be set to the length, in bytes, of that character.
1049
524080c4
KW
1050If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1051enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1052NULL) to -1. If those warnings are off, the computed value if well-defined (or
1053the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1054is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1055next possible position in C<s> that could begin a non-malformed character.
de69f3af 1056See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
1057
1058=cut
1059*/
1060
1061UV
1062Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1063{
1064 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1065
1066 assert(send > s);
1067
1068 /* Call the low level routine asking for checks */
de69f3af
KW
1069 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1070 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
1071}
1072
5495102a
KW
1073/* DEPRECATED!
1074 * Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1075 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1076 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1077
1078UV
1079Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1080{
1081 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1082
010ab96b 1083 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
27d6c58a
KW
1084}
1085
ec5f19d0 1086/*
87cea99e 1087=for apidoc utf8_to_uvuni
9041c2e3
NIS
1088
1089Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1090which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1091length, in bytes, of that character.
1092
4b88fb76 1093Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31 1094malformed input could cause reading beyond the end of the input buffer, which
4f83cdcd
KW
1095is one reason why this function is deprecated. The other is that only in
1096extremely limited circumstances should the Unicode versus native code point be
de69f3af 1097of any interest to you. See L</utf8_to_uvuni_buf> for alternatives.
9041c2e3 1098
524080c4
KW
1099If C<s> points to one of the detected malformations, and UTF8 warnings are
1100enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1101NULL) to -1. If those warnings are off, the computed value if well-defined (or
1102the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1103is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1104next possible position in C<s> that could begin a non-malformed character.
de69f3af 1105See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1106
1107=cut
1108*/
1109
1110UV
7fc63493 1111Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1112{
7918f24d
NC
1113 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1114
5495102a 1115 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
8e84507e
NIS
1116}
1117
b76347f2 1118/*
87cea99e 1119=for apidoc utf8_length
b76347f2
JH
1120
1121Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1122Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1123up past C<e>, croaks.
b76347f2
JH
1124
1125=cut
1126*/
1127
1128STRLEN
35a4481c 1129Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1130{
97aff369 1131 dVAR;
b76347f2
JH
1132 STRLEN len = 0;
1133
7918f24d
NC
1134 PERL_ARGS_ASSERT_UTF8_LENGTH;
1135
8850bf83
JH
1136 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1137 * the bitops (especially ~) can create illegal UTF-8.
1138 * In other words: in Perl UTF-8 is not just for Unicode. */
1139
a3b680e6
AL
1140 if (e < s)
1141 goto warn_and_return;
b76347f2 1142 while (s < e) {
4cbf4130 1143 s += UTF8SKIP(s);
8e91ec7f
AV
1144 len++;
1145 }
1146
1147 if (e != s) {
1148 len--;
1149 warn_and_return:
9b387841
NC
1150 if (PL_op)
1151 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1152 "%s in %s", unees, OP_DESC(PL_op));
1153 else
61a12c31 1154 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1155 }
1156
1157 return len;
1158}
1159
b06226ff 1160/*
87cea99e 1161=for apidoc utf8_distance
b06226ff 1162
1e54db1a 1163Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1164and C<b>.
1165
1166WARNING: use only if you *know* that the pointers point inside the
1167same UTF-8 buffer.
1168
37607a96
PK
1169=cut
1170*/
a0ed51b3 1171
02eb7b47 1172IV
35a4481c 1173Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1174{
7918f24d
NC
1175 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1176
bf1665bc 1177 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1178}
1179
b06226ff 1180/*
87cea99e 1181=for apidoc utf8_hop
b06226ff 1182
8850bf83
JH
1183Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1184forward or backward.
b06226ff
JH
1185
1186WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1187the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1188on the first byte of character or just after the last byte of a character.
b06226ff 1189
37607a96
PK
1190=cut
1191*/
a0ed51b3
LW
1192
1193U8 *
4373e329 1194Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1195{
7918f24d
NC
1196 PERL_ARGS_ASSERT_UTF8_HOP;
1197
96a5add6 1198 PERL_UNUSED_CONTEXT;
8850bf83
JH
1199 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1200 * the bitops (especially ~) can create illegal UTF-8.
1201 * In other words: in Perl UTF-8 is not just for Unicode. */
1202
a0ed51b3
LW
1203 if (off >= 0) {
1204 while (off--)
1205 s += UTF8SKIP(s);
1206 }
1207 else {
1208 while (off++) {
1209 s--;
8850bf83
JH
1210 while (UTF8_IS_CONTINUATION(*s))
1211 s--;
a0ed51b3
LW
1212 }
1213 }
4373e329 1214 return (U8 *)s;
a0ed51b3
LW
1215}
1216
6940069f 1217/*
fed3ba5d
NC
1218=for apidoc bytes_cmp_utf8
1219
a1433954 1220Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1221sequence of characters (stored as UTF-8)
1222in 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
72d33970
FC
1227longer string. -2 or +2 is returned if
1228there was a difference between characters
fed3ba5d
NC
1229within the strings.
1230
1231=cut
1232*/
1233
1234int
1235Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1236{
1237 const U8 *const bend = b + blen;
1238 const U8 *const uend = u + ulen;
1239
1240 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1241
1242 PERL_UNUSED_CONTEXT;
1243
1244 while (b < bend && u < uend) {
1245 U8 c = *u++;
1246 if (!UTF8_IS_INVARIANT(c)) {
1247 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1248 if (u < uend) {
1249 U8 c1 = *u++;
1250 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1251 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1252 } else {
1253 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1254 "Malformed UTF-8 character "
1255 "(unexpected non-continuation byte 0x%02x"
1256 ", immediately after start byte 0x%02x)"
1257 /* Dear diag.t, it's in the pod. */
1258 "%s%s", c1, c,
1259 PL_op ? " in " : "",
1260 PL_op ? OP_DESC(PL_op) : "");
1261 return -2;
1262 }
1263 } else {
1264 if (PL_op)
1265 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1266 "%s in %s", unees, OP_DESC(PL_op));
1267 else
61a12c31 1268 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1269 return -2; /* Really want to return undef :-) */
1270 }
1271 } else {
1272 return -2;
1273 }
1274 }
1275 if (*b != c) {
1276 return *b < c ? -2 : +2;
1277 }
1278 ++b;
1279 }
1280
1281 if (b == bend && u == uend)
1282 return 0;
1283
1284 return b < bend ? +1 : -1;
1285}
1286
1287/*
87cea99e 1288=for apidoc utf8_to_bytes
6940069f 1289
2bbc8d55 1290Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1291Unlike L</bytes_to_utf8>, this over-writes the original string, and
1292updates C<len> to contain the new length.
67e989fb 1293Returns zero on failure, setting C<len> to -1.
6940069f 1294
a1433954 1295If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1296
6940069f
GS
1297=cut
1298*/
1299
1300U8 *
37607a96 1301Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1302{
d4c19fe8
AL
1303 U8 * const save = s;
1304 U8 * const send = s + *len;
6940069f 1305 U8 *d;
246fae53 1306
7918f24d
NC
1307 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1308
1e54db1a 1309 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1310 while (s < send) {
d59937ca
KW
1311 if (! UTF8_IS_INVARIANT(*s)) {
1312 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1313 *len = ((STRLEN) -1);
1314 return 0;
1315 }
1316 s++;
dcad2880 1317 }
d59937ca 1318 s++;
246fae53 1319 }
dcad2880
JH
1320
1321 d = s = save;
6940069f 1322 while (s < send) {
80e0b38f
KW
1323 U8 c = *s++;
1324 if (! UTF8_IS_INVARIANT(c)) {
1325 /* Then it is two-byte encoded */
1326 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1327 s++;
1328 }
1329 *d++ = c;
6940069f
GS
1330 }
1331 *d = '\0';
246fae53 1332 *len = d - save;
6940069f
GS
1333 return save;
1334}
1335
1336/*
87cea99e 1337=for apidoc bytes_from_utf8
f9a63242 1338
2bbc8d55 1339Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1340Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1341the newly-created string, and updates C<len> to contain the new
1342length. Returns the original string if no conversion occurs, C<len>
72d33970 1343is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
13440 if C<s> is converted or consisted entirely of characters that are invariant
1345in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1346
37607a96
PK
1347=cut
1348*/
f9a63242
JH
1349
1350U8 *
e1ec3a88 1351Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1352{
f9a63242 1353 U8 *d;
e1ec3a88
AL
1354 const U8 *start = s;
1355 const U8 *send;
f9a63242
JH
1356 I32 count = 0;
1357
7918f24d
NC
1358 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1359
96a5add6 1360 PERL_UNUSED_CONTEXT;
f9a63242 1361 if (!*is_utf8)
73d840c0 1362 return (U8 *)start;
f9a63242 1363
1e54db1a 1364 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1365 for (send = s + *len; s < send;) {
d59937ca
KW
1366 if (! UTF8_IS_INVARIANT(*s)) {
1367 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1368 return (U8 *)start;
d59937ca
KW
1369 }
1370 count++;
1371 s++;
db42d148 1372 }
d59937ca 1373 s++;
f9a63242
JH
1374 }
1375
35da51f7 1376 *is_utf8 = FALSE;
f9a63242 1377
212542aa 1378 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1379 s = start; start = d;
f9a63242
JH
1380 while (s < send) {
1381 U8 c = *s++;
1a91c45d 1382 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1383 /* Then it is two-byte encoded */
1a91c45d
KW
1384 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1385 s++;
c4d5f83a
NIS
1386 }
1387 *d++ = c;
f9a63242
JH
1388 }
1389 *d = '\0';
1390 *len = d - start;
73d840c0 1391 return (U8 *)start;
f9a63242
JH
1392}
1393
1394/*
87cea99e 1395=for apidoc bytes_to_utf8
6940069f 1396
ff97e5cf
KW
1397Converts a string C<s> of length C<len> bytes from the native encoding into
1398UTF-8.
6662521e 1399Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1400reflect the new length in bytes.
6940069f 1401
2bbc8d55
SP
1402A NUL character will be written after the end of the string.
1403
1404If you want to convert to UTF-8 from encodings other than
1405the native (Latin1 or EBCDIC),
a1433954 1406see L</sv_recode_to_utf8>().
c9ada85f 1407
497711e7 1408=cut
6940069f
GS
1409*/
1410
c682ebef
FC
1411/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1412 likewise need duplication. */
1413
6940069f 1414U8*
35a4481c 1415Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1416{
35a4481c 1417 const U8 * const send = s + (*len);
6940069f
GS
1418 U8 *d;
1419 U8 *dst;
7918f24d
NC
1420
1421 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1422 PERL_UNUSED_CONTEXT;
6940069f 1423
212542aa 1424 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1425 dst = d;
1426
1427 while (s < send) {
55d09dc8
KW
1428 append_utf8_from_native_byte(*s, &d);
1429 s++;
6940069f
GS
1430 }
1431 *d = '\0';
6662521e 1432 *len = d-dst;
6940069f
GS
1433 return dst;
1434}
1435
a0ed51b3 1436/*
dea0fc0b 1437 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1438 *
1439 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1440 * We optimize for native, for obvious reasons. */
1441
1442U8*
dea0fc0b 1443Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1444{
dea0fc0b
JH
1445 U8* pend;
1446 U8* dstart = d;
1447
7918f24d
NC
1448 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1449
dea0fc0b 1450 if (bytelen & 1)
f5992bc4 1451 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1452
1453 pend = p + bytelen;
1454
a0ed51b3 1455 while (p < pend) {
dea0fc0b
JH
1456 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1457 p += 2;
56d37426
KW
1458 if (UNI_IS_INVARIANT(uv)) {
1459 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1460 continue;
1461 }
56d37426
KW
1462 if (uv <= MAX_UTF8_TWO_BYTE) {
1463 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1464 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1465 continue;
1466 }
46956fad
KW
1467#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1468#define LAST_HIGH_SURROGATE 0xDBFF
1469#define FIRST_LOW_SURROGATE 0xDC00
1470#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1471 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1472 if (p >= pend) {
dea0fc0b 1473 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1474 } else {
1475 UV low = (p[0] << 8) + p[1];
1476 p += 2;
46956fad 1477 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1478 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1479 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1480 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1481 }
46956fad 1482 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1483 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1484 }
56d37426
KW
1485#ifdef EBCDIC
1486 d = uvoffuni_to_utf8_flags(d, uv, 0);
1487#else
a0ed51b3 1488 if (uv < 0x10000) {
eb160463
GS
1489 *d++ = (U8)(( uv >> 12) | 0xe0);
1490 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1491 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1492 continue;
1493 }
1494 else {
eb160463
GS
1495 *d++ = (U8)(( uv >> 18) | 0xf0);
1496 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1497 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1498 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1499 continue;
1500 }
56d37426 1501#endif
a0ed51b3 1502 }
dea0fc0b 1503 *newlen = d - dstart;
a0ed51b3
LW
1504 return d;
1505}
1506
1507/* Note: this one is slightly destructive of the source. */
1508
1509U8*
dea0fc0b 1510Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1511{
1512 U8* s = (U8*)p;
d4c19fe8 1513 U8* const send = s + bytelen;
7918f24d
NC
1514
1515 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1516
e0ea5e2d
NC
1517 if (bytelen & 1)
1518 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1519 (UV)bytelen);
1520
a0ed51b3 1521 while (s < send) {
d4c19fe8 1522 const U8 tmp = s[0];
a0ed51b3
LW
1523 s[0] = s[1];
1524 s[1] = tmp;
1525 s += 2;
1526 }
dea0fc0b 1527 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1528}
1529
922e8cb4
KW
1530bool
1531Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1532{
1533 U8 tmpbuf[UTF8_MAXBYTES+1];
1534 uvchr_to_utf8(tmpbuf, c);
1535 return _is_utf8_FOO(classnum, tmpbuf);
1536}
1537
f9ae8fb6
JD
1538/* Internal function so we can deprecate the external one, and call
1539 this one from other deprecated functions in this file */
1540
1541PERL_STATIC_INLINE bool
61b19385
KW
1542S_is_utf8_idfirst(pTHX_ const U8 *p)
1543{
1544 dVAR;
1545
1546 if (*p == '_')
1547 return TRUE;
1548 /* is_utf8_idstart would be more logical. */
1549 return is_utf8_common(p, &PL_utf8_idstart, "IdStart");
1550}
1551
5092f92a 1552bool
84afefe6 1553Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1554{
89ebb4a3 1555 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1556 uvchr_to_utf8(tmpbuf, c);
61b19385 1557 return S_is_utf8_idfirst(aTHX_ tmpbuf);
a0ed51b3
LW
1558}
1559
1560bool
eba68aa0
KW
1561Perl__is_uni_perl_idcont(pTHX_ UV c)
1562{
1563 U8 tmpbuf[UTF8_MAXBYTES+1];
1564 uvchr_to_utf8(tmpbuf, c);
1565 return _is_utf8_perl_idcont(tmpbuf);
1566}
1567
1568bool
f91dcd13
KW
1569Perl__is_uni_perl_idstart(pTHX_ UV c)
1570{
1571 U8 tmpbuf[UTF8_MAXBYTES+1];
1572 uvchr_to_utf8(tmpbuf, c);
1573 return _is_utf8_perl_idstart(tmpbuf);
1574}
1575
3a4c58c9
KW
1576UV
1577Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1578{
1579 /* We have the latin1-range values compiled into the core, so just use
1580 * those, converting the result to utf8. The only difference between upper
1581 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1582 * either "SS" or "Ss". Which one to use is passed into the routine in
1583 * 'S_or_s' to avoid a test */
1584
1585 UV converted = toUPPER_LATIN1_MOD(c);
1586
1587 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1588
1589 assert(S_or_s == 'S' || S_or_s == 's');
1590
6f2d5cbc 1591 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1592 characters in this range */
3a4c58c9
KW
1593 *p = (U8) converted;
1594 *lenp = 1;
1595 return converted;
1596 }
1597
1598 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1599 * which it maps to one of them, so as to only have to have one check for
1600 * it in the main case */
1601 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1602 switch (c) {
1603 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1604 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1605 break;
1606 case MICRO_SIGN:
1607 converted = GREEK_CAPITAL_LETTER_MU;
1608 break;
1609 case LATIN_SMALL_LETTER_SHARP_S:
1610 *(p)++ = 'S';
1611 *p = S_or_s;
1612 *lenp = 2;
1613 return 'S';
1614 default:
1615 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1616 assert(0); /* NOTREACHED */
3a4c58c9
KW
1617 }
1618 }
1619
1620 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1621 *p = UTF8_TWO_BYTE_LO(converted);
1622 *lenp = 2;
1623
1624 return converted;
1625}
1626
50bda2c3
KW
1627/* Call the function to convert a UTF-8 encoded character to the specified case.
1628 * Note that there may be more than one character in the result.
1629 * INP is a pointer to the first byte of the input character
1630 * OUTP will be set to the first byte of the string of changed characters. It
1631 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1632 * LENP will be set to the length in bytes of the string of changed characters
1633 *
1634 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1635#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1636#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1637#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1638
1639/* This additionally has the input parameter SPECIALS, which if non-zero will
1640 * cause this to use the SPECIALS hash for folding (meaning get full case
1641 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1642#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1643
84afefe6
JH
1644UV
1645Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1646{
3a4c58c9
KW
1647 dVAR;
1648
a1433954
KW
1649 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1650 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1651 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1652 * the changed version may be longer than the original character.
1653 *
1654 * The ordinal of the first character of the changed version is returned
1655 * (but note, as explained above, that there may be more.) */
1656
7918f24d
NC
1657 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1658
3a4c58c9
KW
1659 if (c < 256) {
1660 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1661 }
1662
0ebc6274 1663 uvchr_to_utf8(p, c);
3a4c58c9 1664 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1665}
1666
84afefe6
JH
1667UV
1668Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1669{
3a4c58c9
KW
1670 dVAR;
1671
7918f24d
NC
1672 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1673
3a4c58c9
KW
1674 if (c < 256) {
1675 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1676 }
1677
0ebc6274 1678 uvchr_to_utf8(p, c);
3a4c58c9 1679 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1680}
1681
afc16117
KW
1682STATIC U8
1683S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1684{
1685 /* We have the latin1-range values compiled into the core, so just use
1686 * those, converting the result to utf8. Since the result is always just
a1433954 1687 * one character, we allow <p> to be NULL */
afc16117
KW
1688
1689 U8 converted = toLOWER_LATIN1(c);
1690
1691 if (p != NULL) {
6f2d5cbc 1692 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1693 *p = converted;
1694 *lenp = 1;
1695 }
1696 else {
1697 *p = UTF8_TWO_BYTE_HI(converted);
1698 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1699 *lenp = 2;
1700 }
1701 }
1702 return converted;
1703}
1704
84afefe6
JH
1705UV
1706Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1707{
968c5e6a
KW
1708 dVAR;
1709
7918f24d
NC
1710 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1711
afc16117
KW
1712 if (c < 256) {
1713 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1714 }
1715
afc16117 1716 uvchr_to_utf8(p, c);
968c5e6a 1717 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1718}
1719
84afefe6 1720UV
51910141 1721Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1722{
51910141 1723 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1724 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1725 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1726 *
1727 * Not to be used for locale folds
51910141 1728 */
f673fad4 1729
a1dde8de
KW
1730 UV converted;
1731
1732 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1733
1ca267a5
KW
1734 assert (! (flags & FOLD_FLAGS_LOCALE));
1735
a1dde8de
KW
1736 if (c == MICRO_SIGN) {
1737 converted = GREEK_SMALL_LETTER_MU;
1738 }
51910141 1739 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1740
1741 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1742 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1743 * under those circumstances. */
1744 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1745 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1746 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1747 p, *lenp, U8);
1748 return LATIN_SMALL_LETTER_LONG_S;
1749 }
1750 else {
4f489194
KW
1751 *(p)++ = 's';
1752 *p = 's';
1753 *lenp = 2;
1754 return 's';
1ca267a5 1755 }
a1dde8de
KW
1756 }
1757 else { /* In this range the fold of all other characters is their lower
1758 case */
1759 converted = toLOWER_LATIN1(c);
1760 }
1761
6f2d5cbc 1762 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1763 *p = (U8) converted;
1764 *lenp = 1;
1765 }
1766 else {
1767 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1768 *p = UTF8_TWO_BYTE_LO(converted);
1769 *lenp = 2;
1770 }
1771
1772 return converted;
1773}
1774
1775UV
a0270393 1776Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, const U8 flags)
84afefe6 1777{
4b593389 1778
a0270393
KW
1779 /* Not currently externally documented, and subject to change
1780 * <flags> bits meanings:
1781 * FOLD_FLAGS_FULL iff full folding is to be used;
1782 * FOLD_FLAGS_LOCALE iff in locale
1783 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1784 */
4b593389 1785
36bb2ab6 1786 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1787
a1dde8de 1788 if (c < 256) {
a0270393 1789 UV result = _to_fold_latin1((U8) c, p, lenp,
1ca267a5 1790 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1791 /* It is illegal for the fold to cross the 255/256 boundary under
1792 * locale; in this case return the original */
1793 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1794 ? c
1795 : result;
a1dde8de
KW
1796 }
1797
a0270393
KW
1798 /* If no special needs, just use the macro */
1799 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1800 uvchr_to_utf8(p, c);
1801 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1802 }
1803 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1804 the special flags. */
1805 U8 utf8_c[UTF8_MAXBYTES + 1];
1806 uvchr_to_utf8(utf8_c, c);
1807 return _to_utf8_fold_flags(utf8_c, p, lenp, flags, NULL);
1808 }
84afefe6
JH
1809}
1810
26483009 1811PERL_STATIC_INLINE bool
5141f98e 1812S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
bde6a22d
NC
1813 const char *const swashname)
1814{
ea317ccb
KW
1815 /* returns a boolean giving whether or not the UTF8-encoded character that
1816 * starts at <p> is in the swash indicated by <swashname>. <swash>
1817 * contains a pointer to where the swash indicated by <swashname>
1818 * is to be stored; which this routine will do, so that future calls will
1819 * look at <*swash> and only generate a swash if it is not null
1820 *
1821 * Note that it is assumed that the buffer length of <p> is enough to
1822 * contain all the bytes that comprise the character. Thus, <*p> should
1823 * have been checked before this call for mal-formedness enough to assure
1824 * that. */
1825
97aff369 1826 dVAR;
7918f24d
NC
1827
1828 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1829
492a624f 1830 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1831 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1832 * as far as there being enough bytes available in it to accommodate the
1833 * character without reading beyond the end, and pass that number on to the
1834 * validating routine */
28123549
KW
1835 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
1836 if (ckWARN_d(WARN_UTF8)) {
1837 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1838 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1839 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1840 what the malformation is */
1841 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1842 }
1843 }
1844 return FALSE;
1845 }
87367d5f
KW
1846 if (!*swash) {
1847 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1848 *swash = _core_swash_init("utf8", swashname, &PL_sv_undef, 1, 0, NULL, &flags);
1849 }
28123549 1850
bde6a22d
NC
1851 return swash_fetch(*swash, p, TRUE) != 0;
1852}
1853
1854bool
922e8cb4
KW
1855Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1856{
1857 dVAR;
1858
1859 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1860
1861 assert(classnum < _FIRST_NON_SWASH_CC);
1862
1863 return is_utf8_common(p, &PL_utf8_swash_ptrs[classnum], swash_property_names[classnum]);
1864}
1865
1866bool
7fc63493 1867Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1868{
97aff369 1869 dVAR;
7918f24d
NC
1870
1871 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1872
61b19385 1873 return S_is_utf8_idfirst(aTHX_ p);
82686b01
JH
1874}
1875
1876bool
c11ff943
KW
1877Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1878{
1879 dVAR;
1880
1881 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1882
1883 if (*p == '_')
1884 return TRUE;
1885 /* is_utf8_idstart would be more logical. */
1886 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart");
1887}
1888
1889bool
d65654cb 1890Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
b6912c02
KW
1891{
1892 dVAR;
1893
d65654cb 1894 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
b6912c02
KW
1895
1896 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart");
1897}
1898
1899bool
eba68aa0
KW
1900Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1901{
1902 dVAR;
1903
1904 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1905
1906 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont");
1907}
1908
1909
1910bool
7fc63493 1911Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1912{
97aff369 1913 dVAR;
7918f24d
NC
1914
1915 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
1916
d4c19fe8 1917 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue");
a0ed51b3
LW
1918}
1919
1920bool
c11ff943
KW
1921Perl_is_utf8_xidcont(pTHX_ const U8 *p)
1922{
1923 dVAR;
1924
1925 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
1926
c11ff943
KW
1927 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue");
1928}
1929
1930bool
7dbf68d2
KW
1931Perl__is_utf8_mark(pTHX_ const U8 *p)
1932{
1933 dVAR;
1934
1935 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1936
1937 return is_utf8_common(p, &PL_utf8_mark, "IsM");
1938}
1939
6b5c0936 1940/*
87cea99e 1941=for apidoc to_utf8_case
6b5c0936 1942
6fae5207 1943C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1944the character that is being converted. This routine assumes that the character
1945at C<p> is well-formed.
6b5c0936 1946
6fae5207
KW
1947C<ustrp> is a pointer to the character buffer to put the
1948conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1949of the result.
1950
6fae5207 1951C<swashp> is a pointer to the swash to use.
6b5c0936 1952
a1433954 1953Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1954and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1955but not always, a multicharacter mapping), is tried first.
6b5c0936 1956
4a8240a3
KW
1957C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1958any special mappings; C<""> means to use the special mappings. Values other
1959than these two are treated as the name of the hash containing the special
1960mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1961
6fae5207 1962C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1963%utf8::ToLower.
1964
1965=cut */
6b5c0936 1966
2104c8d9 1967UV
9a957fbc
AL
1968Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1969 SV **swashp, const char *normal, const char *special)
a0ed51b3 1970{
97aff369 1971 dVAR;
0134edef 1972 STRLEN len = 0;
f4cd282c 1973 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1974
1975 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1976
9ae3ac1a
KW
1977 /* Note that swash_fetch() doesn't output warnings for these because it
1978 * assumes we will */
8457b38f 1979 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1980 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1981 if (ckWARN_d(WARN_SURROGATE)) {
1982 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1983 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1984 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1985 }
9ae3ac1a
KW
1986 }
1987 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1988 if (ckWARN_d(WARN_NON_UNICODE)) {
1989 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1990 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1991 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1992 }
9ae3ac1a
KW
1993 }
1994
1995 /* Note that non-characters are perfectly legal, so no warning should
1996 * be given */
1997 }
1998
0134edef 1999 if (!*swashp) /* load on-demand */
5ab9d2ef 2000 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2001
a6f87d8c 2002 if (special) {
0134edef 2003 /* It might be "special" (sometimes, but not always,
2a37f04d 2004 * a multicharacter mapping) */
4a8240a3 2005 HV *hv = NULL;
b08cf34e
JH
2006 SV **svp;
2007
4a8240a3
KW
2008 /* If passed in the specials name, use that; otherwise use any
2009 * given in the swash */
2010 if (*special != '\0') {
2011 hv = get_hv(special, 0);
2012 }
2013 else {
2014 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2015 if (svp) {
2016 hv = MUTABLE_HV(SvRV(*svp));
2017 }
2018 }
2019
176fe009
KW
2020 if (hv
2021 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
2022 && (*svp))
2023 {
cfd0369c 2024 const char *s;
47654450 2025
cfd0369c 2026 s = SvPV_const(*svp, len);
47654450 2027 if (len == 1)
f4cd282c 2028 /* EIGHTBIT */
c80e42f3 2029 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2030 else {
d2dcd0fb 2031 Copy(s, ustrp, len, U8);
29e98929 2032 }
983ffd37 2033 }
0134edef
JH
2034 }
2035
2036 if (!len && *swashp) {
f4cd282c 2037 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 2038
0134edef
JH
2039 if (uv2) {
2040 /* It was "normal" (a single character mapping). */
f4cd282c 2041 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2042 }
2043 }
1feea2c7 2044
cbe07460
KW
2045 if (len) {
2046 if (lenp) {
2047 *lenp = len;
2048 }
2049 return valid_utf8_to_uvchr(ustrp, 0);
2050 }
2051
2052 /* Here, there was no mapping defined, which means that the code point maps
2053 * to itself. Return the inputs */
bfdf22ec 2054 len = UTF8SKIP(p);
ca9fab46
KW
2055 if (p != ustrp) { /* Don't copy onto itself */
2056 Copy(p, ustrp, len, U8);
2057 }
0134edef 2058
2a37f04d
JH
2059 if (lenp)
2060 *lenp = len;
2061
f4cd282c 2062 return uv1;
cbe07460 2063
a0ed51b3
LW
2064}
2065
051a06d4
KW
2066STATIC UV
2067S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2068{
2069 /* This is called when changing the case of a utf8-encoded character above
2070 * the Latin1 range, and the operation is in locale. If the result
2071 * contains a character that crosses the 255/256 boundary, disallow the
2072 * change, and return the original code point. See L<perlfunc/lc> for why;
2073 *
a1433954
KW
2074 * p points to the original string whose case was changed; assumed
2075 * by this routine to be well-formed
051a06d4
KW
2076 * result the code point of the first character in the changed-case string
2077 * ustrp points to the changed-case string (<result> represents its first char)
2078 * lenp points to the length of <ustrp> */
2079
2080 UV original; /* To store the first code point of <p> */
2081
2082 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2083
a4f12ed7 2084 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2085
2086 /* We know immediately if the first character in the string crosses the
2087 * boundary, so can skip */
2088 if (result > 255) {
2089
2090 /* Look at every character in the result; if any cross the
2091 * boundary, the whole thing is disallowed */
2092 U8* s = ustrp + UTF8SKIP(ustrp);
2093 U8* e = ustrp + *lenp;
2094 while (s < e) {
a4f12ed7 2095 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2096 goto bad_crossing;
2097 }
2098 s += UTF8SKIP(s);
2099 }
2100
2101 /* Here, no characters crossed, result is ok as-is */
2102 return result;
2103 }
2104
2105bad_crossing:
2106
2107 /* Failed, have to return the original */
4b88fb76 2108 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2109 Copy(p, ustrp, *lenp, char);
2110 return original;
2111}
2112
d3e79532 2113/*
87cea99e 2114=for apidoc to_utf8_upper
d3e79532 2115
1f607577 2116Instead use L</toUPPER_utf8>.
a1433954 2117
d3e79532
JH
2118=cut */
2119
051a06d4
KW
2120/* Not currently externally documented, and subject to change:
2121 * <flags> is set iff locale semantics are to be used for code points < 256
2122 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2123 * were used in the calculation; otherwise unchanged. */
2124
2104c8d9 2125UV
051a06d4 2126Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2127{
97aff369 2128 dVAR;
7918f24d 2129
051a06d4
KW
2130 UV result;
2131
2132 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2133
3a4c58c9 2134 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2135 if (flags) {
2136 result = toUPPER_LC(*p);
2137 }
2138 else {
81c6c7ce 2139 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2140 }
3a4c58c9
KW
2141 }
2142 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2143 if (flags) {
a6d8b88b 2144 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2145 result = toUPPER_LC(c);
051a06d4
KW
2146 }
2147 else {
94bb8c36 2148 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2149 ustrp, lenp, 'S');
051a06d4
KW
2150 }
2151 }
2152 else { /* utf8, ord above 255 */
2153 result = CALL_UPPER_CASE(p, ustrp, lenp);
2154
2155 if (flags) {
2156 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2157 }
2158 return result;
2159 }
2160
2161 /* Here, used locale rules. Convert back to utf8 */
2162 if (UTF8_IS_INVARIANT(result)) {
2163 *ustrp = (U8) result;
2164 *lenp = 1;
2165 }
2166 else {
62cb07ea
KW
2167 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2168 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2169 *lenp = 2;
3a4c58c9
KW
2170 }
2171
051a06d4
KW
2172 if (tainted_ptr) {
2173 *tainted_ptr = TRUE;
2174 }
2175 return result;
983ffd37 2176}
a0ed51b3 2177
d3e79532 2178/*
87cea99e 2179=for apidoc to_utf8_title
d3e79532 2180
1f607577 2181Instead use L</toTITLE_utf8>.
a1433954 2182
d3e79532
JH
2183=cut */
2184
051a06d4
KW
2185/* Not currently externally documented, and subject to change:
2186 * <flags> is set iff locale semantics are to be used for code points < 256
2187 * Since titlecase is not defined in POSIX, uppercase is used instead
2188 * for these/
2189 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2190 * were used in the calculation; otherwise unchanged. */
2191
983ffd37 2192UV
051a06d4 2193Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
983ffd37 2194{
97aff369 2195 dVAR;
7918f24d 2196
051a06d4
KW
2197 UV result;
2198
2199 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2200
3a4c58c9 2201 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2202 if (flags) {
2203 result = toUPPER_LC(*p);
2204 }
2205 else {
81c6c7ce 2206 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2207 }
3a4c58c9
KW
2208 }
2209 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2210 if (flags) {
a6d8b88b 2211 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2212 result = toUPPER_LC(c);
051a06d4
KW
2213 }
2214 else {
94bb8c36 2215 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2216 ustrp, lenp, 's');
051a06d4
KW
2217 }
2218 }
2219 else { /* utf8, ord above 255 */
2220 result = CALL_TITLE_CASE(p, ustrp, lenp);
2221
2222 if (flags) {
2223 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2224 }
2225 return result;
2226 }
2227
2228 /* Here, used locale rules. Convert back to utf8 */
2229 if (UTF8_IS_INVARIANT(result)) {
2230 *ustrp = (U8) result;
2231 *lenp = 1;
2232 }
2233 else {
62cb07ea
KW
2234 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2235 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2236 *lenp = 2;
3a4c58c9
KW
2237 }
2238
051a06d4
KW
2239 if (tainted_ptr) {
2240 *tainted_ptr = TRUE;
2241 }
2242 return result;
a0ed51b3
LW
2243}
2244
d3e79532 2245/*
87cea99e 2246=for apidoc to_utf8_lower
d3e79532 2247
1f607577 2248Instead use L</toLOWER_utf8>.
a1433954 2249
d3e79532
JH
2250=cut */
2251
051a06d4
KW
2252/* Not currently externally documented, and subject to change:
2253 * <flags> is set iff locale semantics are to be used for code points < 256
2254 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2255 * were used in the calculation; otherwise unchanged. */
2256
2104c8d9 2257UV
051a06d4 2258Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, const bool flags, bool* tainted_ptr)
a0ed51b3 2259{
051a06d4
KW
2260 UV result;
2261
97aff369 2262 dVAR;
7918f24d 2263
051a06d4 2264 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2265
968c5e6a 2266 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2267 if (flags) {
2268 result = toLOWER_LC(*p);
2269 }
2270 else {
81c6c7ce 2271 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2272 }
968c5e6a
KW
2273 }
2274 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2275 if (flags) {
a6d8b88b 2276 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2277 result = toLOWER_LC(c);
051a06d4
KW
2278 }
2279 else {
94bb8c36 2280 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2281 ustrp, lenp);
051a06d4 2282 }
968c5e6a 2283 }
051a06d4
KW
2284 else { /* utf8, ord above 255 */
2285 result = CALL_LOWER_CASE(p, ustrp, lenp);
2286
2287 if (flags) {
2288 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2289 }
968c5e6a 2290
051a06d4
KW
2291 return result;
2292 }
2293
2294 /* Here, used locale rules. Convert back to utf8 */
2295 if (UTF8_IS_INVARIANT(result)) {
2296 *ustrp = (U8) result;
2297 *lenp = 1;
2298 }
2299 else {
62cb07ea
KW
2300 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2301 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2302 *lenp = 2;
2303 }
2304
2305 if (tainted_ptr) {
2306 *tainted_ptr = TRUE;
2307 }
2308 return result;
b4e400f9
JH
2309}
2310
d3e79532 2311/*
87cea99e 2312=for apidoc to_utf8_fold
d3e79532 2313
1f607577 2314Instead use L</toFOLD_utf8>.
a1433954 2315
d3e79532
JH
2316=cut */
2317
051a06d4
KW
2318/* Not currently externally documented, and subject to change,
2319 * in <flags>
2320 * bit FOLD_FLAGS_LOCALE is set iff locale semantics are to be used for code
2321 * points < 256. Since foldcase is not defined in
2322 * POSIX, lowercase is used instead
2323 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2324 * otherwise simple folds
a0270393
KW
2325 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2326 * prohibited
051a06d4
KW
2327 * <tainted_ptr> if non-null, *tainted_ptr will be set TRUE iff locale rules
2328 * were used in the calculation; otherwise unchanged. */
36bb2ab6 2329
b4e400f9 2330UV
051a06d4 2331Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags, bool* tainted_ptr)
b4e400f9 2332{
97aff369 2333 dVAR;
7918f24d 2334
051a06d4
KW
2335 UV result;
2336
36bb2ab6 2337 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2338
a0270393
KW
2339 /* These are mutually exclusive */
2340 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2341
50ba90ff
KW
2342 assert(p != ustrp); /* Otherwise overwrites */
2343
a1dde8de 2344 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2345 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2346 result = toFOLD_LC(*p);
051a06d4
KW
2347 }
2348 else {
81c6c7ce 2349 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2350 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2351 }
a1dde8de
KW
2352 }
2353 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2354 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2355 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2356 result = toFOLD_LC(c);
051a06d4
KW
2357 }
2358 else {
94bb8c36 2359 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2360 ustrp, lenp,
1ca267a5 2361 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2362 }
a1dde8de 2363 }
051a06d4 2364 else { /* utf8, ord above 255 */
a0270393 2365 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2366
1ca267a5
KW
2367 if (flags & FOLD_FLAGS_LOCALE) {
2368
9fc2026f 2369 /* Special case these characters, as what normally gets returned
1ca267a5
KW
2370 * under locale doesn't work */
2371 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2372 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2373 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2374 {
2375 goto return_long_s;
2376 }
9fc2026f
KW
2377 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2378 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2379 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2380 {
2381 goto return_ligature_st;
2382 }
a0270393 2383 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2384 }
a0270393
KW
2385 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2386 return result;
2387 }
2388 else {
2389 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2390 * character above the ASCII range, and the result should not
2391 * contain an ASCII character. */
a0270393
KW
2392
2393 UV original; /* To store the first code point of <p> */
2394
2395 /* Look at every character in the result; if any cross the
2396 * boundary, the whole thing is disallowed */
2397 U8* s = ustrp;
2398 U8* e = ustrp + *lenp;
2399 while (s < e) {
2400 if (isASCII(*s)) {
2401 /* Crossed, have to return the original */
2402 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2403
9fc2026f 2404 /* But in these instances, there is an alternative we can
1ca267a5 2405 * return that is valid */
9fc2026f
KW
2406 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2407 || original == LATIN_SMALL_LETTER_SHARP_S)
2408 {
1ca267a5
KW
2409 goto return_long_s;
2410 }
9fc2026f
KW
2411 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2412 goto return_ligature_st;
2413 }
a0270393
KW
2414 Copy(p, ustrp, *lenp, char);
2415 return original;
2416 }
2417 s += UTF8SKIP(s);
2418 }
051a06d4 2419
a0270393
KW
2420 /* Here, no characters crossed, result is ok as-is */
2421 return result;
2422 }
051a06d4
KW
2423 }
2424
2425 /* Here, used locale rules. Convert back to utf8 */
2426 if (UTF8_IS_INVARIANT(result)) {
2427 *ustrp = (U8) result;
2428 *lenp = 1;
2429 }
2430 else {
62cb07ea
KW
2431 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2432 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2433 *lenp = 2;
2434 }
2435
2436 if (tainted_ptr) {
2437 *tainted_ptr = TRUE;
2438 }
2439 return result;
1ca267a5
KW
2440
2441 return_long_s:
2442 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2443 * folds to a string of two of these characters. By returning this
2444 * instead, then, e.g.,
2445 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2446 * works. */
2447
2448 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2449 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2450 ustrp, *lenp, U8);
2451 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2452
2453 return_ligature_st:
2454 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2455 * have the other one fold to it */
2456
2457 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2458 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2459 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2460}
2461
711a919c 2462/* Note:
f90a9a02 2463 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2464 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2465 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2466 */
c4a5db0c 2467
a0ed51b3 2468SV*
7fc63493 2469Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2470{
c4a5db0c
KW
2471 PERL_ARGS_ASSERT_SWASH_INIT;
2472
2473 /* Returns a copy of a swash initiated by the called function. This is the
2474 * public interface, and returning a copy prevents others from doing
2475 * mischief on the original */
2476
5d3d13d1 2477 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2478}
2479
2480SV*
5d3d13d1 2481Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2482{
2483 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2484 * by calling utf8_heavy.pl in the general case. The returned value may be
2485 * the swash's inversion list instead if the input parameters allow it.
2486 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2487 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2488 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2489 *
2490 * This interface should only be used by functions that won't destroy or
2491 * adversely change the swash, as doing so affects all other uses of the
2492 * swash in the program; the general public should use 'Perl_swash_init'
2493 * instead.
2494 *
2495 * pkg is the name of the package that <name> should be in.
2496 * name is the name of the swash to find. Typically it is a Unicode
2497 * property name, including user-defined ones
2498 * listsv is a string to initialize the swash with. It must be of the form
2499 * documented as the subroutine return value in
2500 * L<perlunicode/User-Defined Character Properties>
2501 * minbits is the number of bits required to represent each data element.
2502 * It is '1' for binary properties.
2503 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2504 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2505 * flags_p if non-NULL is the address of various input and output flag bits
2506 * to the routine, as follows: ('I' means is input to the routine;
2507 * 'O' means output from the routine. Only flags marked O are
2508 * meaningful on return.)
2509 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2510 * came from a user-defined property. (I O)
5d3d13d1
KW
2511 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2512 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2513 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2514 * return of an inversion list instead of a swash hash if this routine
2515 * thinks that would result in faster execution of swash_fetch() later
2516 * on. (I)
9a53f6cf
KW
2517 *
2518 * Thus there are three possible inputs to find the swash: <name>,
2519 * <listsv>, and <invlist>. At least one must be specified. The result
2520 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2521 * actions can intersect, etc. what <name> gives. To avoid going out to
2522 * disk at all, <invlist> should specify completely what the swash should
2523 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2524 *
2525 * <invlist> is only valid for binary properties */
c4a5db0c 2526
27da23d5 2527 dVAR;
c4a5db0c 2528 SV* retval = &PL_sv_undef;
83199d38 2529 HV* swash_hv = NULL;
87367d5f
KW
2530 const int invlist_swash_boundary =
2531 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2532 ? 512 /* Based on some benchmarking, but not extensive, see commit
2533 message */
2534 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2535
2536 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2537 assert(! invlist || minbits == 1);
2538
2539 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2540 * so */
2541 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2542 dSP;
2543 const size_t pkg_len = strlen(pkg);
2544 const size_t name_len = strlen(name);
2545 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2546 SV* errsv_save;
2547 GV *method;
2548
2549 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2550
2551 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2552 ENTER;
69794297
KW
2553 SAVEHINTS();
2554 save_re_context();
650f067c
JL
2555 /* We might get here via a subroutine signature which uses a utf8
2556 * parameter name, at which point PL_subname will have been set
2557 * but not yet used. */
2558 save_item(PL_subname);
69794297
KW
2559 if (PL_parser && PL_parser->error_count)
2560 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2561 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2562 if (!method) { /* demand load utf8 */
2563 ENTER;
db2c6cb3
FC
2564 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2565 GvSV(PL_errgv) = NULL;
69794297
KW
2566 /* It is assumed that callers of this routine are not passing in
2567 * any user derived data. */
2568 /* Need to do this after save_re_context() as it will set
2569 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2570 * in Perl_magic_get). Even line to create errsv_save can turn on
2571 * PL_tainted. */
284167a5
SM
2572#ifndef NO_TAINT_SUPPORT
2573 SAVEBOOL(TAINT_get);
2574 TAINT_NOT;
2575#endif
69794297
KW
2576 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2577 NULL);
eed484f9 2578 {
db2c6cb3
FC
2579 /* Not ERRSV, as there is no need to vivify a scalar we are
2580 about to discard. */
2581 SV * const errsv = GvSV(PL_errgv);
2582 if (!SvTRUE(errsv)) {
2583 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2584 SvREFCNT_dec(errsv);
2585 }
eed484f9 2586 }
69794297
KW
2587 LEAVE;
2588 }
2589 SPAGAIN;
2590 PUSHMARK(SP);
2591 EXTEND(SP,5);
2592 mPUSHp(pkg, pkg_len);
2593 mPUSHp(name, name_len);
2594 PUSHs(listsv);
2595 mPUSHi(minbits);
2596 mPUSHi(none);
2597 PUTBACK;
db2c6cb3
FC
2598 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2599 GvSV(PL_errgv) = NULL;
69794297
KW
2600 /* If we already have a pointer to the method, no need to use
2601 * call_method() to repeat the lookup. */
c41800a8
KW
2602 if (method
2603 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2604 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2605 {
2606 retval = *PL_stack_sp--;
2607 SvREFCNT_inc(retval);
2608 }
eed484f9 2609 {
db2c6cb3
FC
2610 /* Not ERRSV. See above. */
2611 SV * const errsv = GvSV(PL_errgv);
2612 if (!SvTRUE(errsv)) {
2613 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2614 SvREFCNT_dec(errsv);
2615 }
eed484f9 2616 }
ce3b816e 2617 LEAVE;
69794297
KW
2618 POPSTACK;
2619 if (IN_PERL_COMPILETIME) {
2620 CopHINTS_set(PL_curcop, PL_hints);
2621 }
2622 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2623 if (SvPOK(retval))
2624
2625 /* If caller wants to handle missing properties, let them */
5d3d13d1 2626 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
2627 return NULL;
2628 }
2629 Perl_croak(aTHX_
2630 "Can't find Unicode property definition \"%"SVf"\"",
2631 SVfARG(retval));
2632 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2633 }
9a53f6cf 2634 } /* End of calling the module to find the swash */
36eb48b4 2635
83199d38
KW
2636 /* If this operation fetched a swash, and we will need it later, get it */
2637 if (retval != &PL_sv_undef
2638 && (minbits == 1 || (flags_p
2639 && ! (*flags_p
2640 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2641 {
2642 swash_hv = MUTABLE_HV(SvRV(retval));
2643
2644 /* If we don't already know that there is a user-defined component to
2645 * this swash, and the user has indicated they wish to know if there is
2646 * one (by passing <flags_p>), find out */
2647 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2648 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2649 if (user_defined && SvUV(*user_defined)) {
2650 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2651 }
2652 }
2653 }
2654
36eb48b4
KW
2655 /* Make sure there is an inversion list for binary properties */
2656 if (minbits == 1) {
2657 SV** swash_invlistsvp = NULL;
2658 SV* swash_invlist = NULL;
9a53f6cf 2659 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2660 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2661 an unclaimed reference count */
36eb48b4 2662
9a53f6cf 2663 /* If this operation fetched a swash, get its already existing
83199d38 2664 * inversion list, or create one for it */
36eb48b4 2665
83199d38 2666 if (swash_hv) {
5c9f4bd2 2667 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2668 if (swash_invlistsvp) {
2669 swash_invlist = *swash_invlistsvp;
2670 invlist_in_swash_is_valid = TRUE;
2671 }
2672 else {
36eb48b4 2673 swash_invlist = _swash_to_invlist(retval);
02c85471 2674 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2675 }
2676 }
2677
2678 /* If an inversion list was passed in, have to include it */
2679 if (invlist) {
2680
2681 /* Any fetched swash will by now have an inversion list in it;
2682 * otherwise <swash_invlist> will be NULL, indicating that we
2683 * didn't fetch a swash */
2684 if (swash_invlist) {
2685
2686 /* Add the passed-in inversion list, which invalidates the one
2687 * already stored in the swash */
2688 invlist_in_swash_is_valid = FALSE;
2689 _invlist_union(invlist, swash_invlist, &swash_invlist);
2690 }
2691 else {
2692
87367d5f
KW
2693 /* Here, there is no swash already. Set up a minimal one, if
2694 * we are going to return a swash */
2695 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2696 swash_hv = newHV();
4aca0fe6 2697 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2698 }
9a53f6cf
KW
2699 swash_invlist = invlist;
2700 }
9a53f6cf
KW
2701 }
2702
2703 /* Here, we have computed the union of all the passed-in data. It may
2704 * be that there was an inversion list in the swash which didn't get
2705 * touched; otherwise save the one computed one */
87367d5f
KW
2706 if (! invlist_in_swash_is_valid
2707 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2708 {
5c9f4bd2 2709 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2710 {
2711 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2712 }
cc34d8c5
FC
2713 /* We just stole a reference count. */
2714 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2715 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2716 }
87367d5f 2717
c41800a8 2718 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2719 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2720 SvREFCNT_dec(retval);
02c85471
FC
2721 if (!swash_invlist_unclaimed)
2722 SvREFCNT_inc_simple_void_NN(swash_invlist);
2723 retval = newRV_noinc(swash_invlist);
87367d5f 2724 }
36eb48b4
KW
2725 }
2726
a0ed51b3
LW
2727 return retval;
2728}
2729
035d37be
JH
2730
2731/* This API is wrong for special case conversions since we may need to
2732 * return several Unicode characters for a single Unicode character
2733 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2734 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2735 * multiple values. --jhi
2736 * For those, you should use to_utf8_case() instead */
b0e3252e 2737/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2738
2739/* Note:
2740 * Returns the value of property/mapping C<swash> for the first character
2741 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2742 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2743 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2744 *
2745 * A "swash" is a hash which contains initially the keys/values set up by
2746 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2747 * property for all possible code points. Things are stored in a compact form
2748 * (see utf8_heavy.pl) so that calculation is required to find the actual
2749 * property value for a given code point. As code points are looked up, new
2750 * key/value pairs are added to the hash, so that the calculation doesn't have
2751 * to ever be re-done. Further, each calculation is done, not just for the
2752 * desired one, but for a whole block of code points adjacent to that one.
2753 * For binary properties on ASCII machines, the block is usually for 64 code
2754 * points, starting with a code point evenly divisible by 64. Thus if the
2755 * property value for code point 257 is requested, the code goes out and
2756 * calculates the property values for all 64 code points between 256 and 319,
2757 * and stores these as a single 64-bit long bit vector, called a "swatch",
2758 * under the key for code point 256. The key is the UTF-8 encoding for code
2759 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2760 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2761 * for code point 258 is then requested, this code realizes that it would be
2762 * stored under the key for 256, and would find that value and extract the
2763 * relevant bit, offset from 256.
2764 *
2765 * Non-binary properties are stored in as many bits as necessary to represent
2766 * their values (32 currently, though the code is more general than that), not
2767 * as single bits, but the principal is the same: the value for each key is a
2768 * vector that encompasses the property values for all code points whose UTF-8
2769 * representations are represented by the key. That is, for all code points
2770 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2771 * bytes of that.
680c470c 2772 */
a0ed51b3 2773UV
680c470c 2774Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2775{
27da23d5 2776 dVAR;
ef8f7699 2777 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2778 U32 klen;
2779 U32 off;
a0ed51b3 2780 STRLEN slen;
7d85a32c 2781 STRLEN needents;
cfd0369c 2782 const U8 *tmps = NULL;
a0ed51b3 2783 U32 bit;
979f2922 2784 SV *swatch;
08fb1ac5 2785 const U8 c = *ptr;
3568d838 2786
7918f24d
NC
2787 PERL_ARGS_ASSERT_SWASH_FETCH;
2788
87367d5f
KW
2789 /* If it really isn't a hash, it isn't really swash; must be an inversion
2790 * list */
2791 if (SvTYPE(hv) != SVt_PVHV) {
2792 return _invlist_contains_cp((SV*)hv,
2793 (do_utf8)
2794 ? valid_utf8_to_uvchr(ptr, NULL)
2795 : c);
2796 }
2797
08fb1ac5
KW
2798 /* We store the values in a "swatch" which is a vec() value in a swash
2799 * hash. Code points 0-255 are a single vec() stored with key length
2800 * (klen) 0. All other code points have a UTF-8 representation
2801 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2802 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2803 * length for them is the length of the encoded char - 1. ptr[klen] is the
2804 * final byte in the sequence representing the character */
2805 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2806 klen = 0;
2807 needents = 256;
2808 off = c;
3568d838 2809 }
08fb1ac5
KW
2810 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2811 klen = 0;
2812 needents = 256;
2813 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2814 }
2815 else {
08fb1ac5
KW
2816 klen = UTF8SKIP(ptr) - 1;
2817
2818 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2819 * the vec is the final byte in the sequence. (In EBCDIC this is
2820 * converted to I8 to get consecutive values.) To help you visualize
2821 * all this:
2822 * Straight 1047 After final byte
2823 * UTF-8 UTF-EBCDIC I8 transform
2824 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2825 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2826 * ...
2827 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2828 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2829 * ...
2830 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2831 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2832 * ...
2833 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2834 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2835 * ...
2836 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2837 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2838 *
2839 * (There are no discontinuities in the elided (...) entries.)
2840 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2841 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2842 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2843 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2844 * index into the vec() swatch (after subtracting 0x80, which we
2845 * actually do with an '&').
2846 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2847 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2848 * dicontinuities which go away by transforming it into I8, and we
2849 * effectively subtract 0xA0 to get the index. */
979f2922 2850 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2851 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2852 }
7d85a32c 2853
a0ed51b3
LW
2854 /*
2855 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2856 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2857 * it's nothing to sniff at.) Pity we usually come through at least
2858 * two function calls to get here...
2859 *
2860 * NB: this code assumes that swatches are never modified, once generated!
2861 */
2862
3568d838 2863 if (hv == PL_last_swash_hv &&
a0ed51b3 2864 klen == PL_last_swash_klen &&
27da23d5 2865 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2866 {
2867 tmps = PL_last_swash_tmps;
2868 slen = PL_last_swash_slen;
2869 }
2870 else {
2871 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2872 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2873
b0e3252e 2874 /* If not cached, generate it via swatch_get */
979f2922 2875 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2876 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2877 {
2878 if (klen) {
2879 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2880 swatch = swatch_get(swash,
2881 code_point & ~((UV)needents - 1),
2882 needents);
2883 }
2884 else { /* For the first 256 code points, the swatch has a key of
2885 length 0 */
2886 swatch = swatch_get(swash, 0, needents);
2887 }
979f2922 2888
923e4eb5 2889 if (IN_PERL_COMPILETIME)
623e6609 2890 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2891
979f2922 2892 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2893
979f2922
ST
2894 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2895 || (slen << 3) < needents)
5637ef5b
NC
2896 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2897 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2898 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2899 }
2900
2901 PL_last_swash_hv = hv;
16d8f38a 2902 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2903 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2904 /* FIXME change interpvar.h? */
2905 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2906 PL_last_swash_slen = slen;
2907 if (klen)
2908 Copy(ptr, PL_last_swash_key, klen, U8);
2909 }
2910
9faf8d75 2911 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
2912 case 1:
2913 bit = 1 << (off & 7);
2914 off >>= 3;
2915 return (tmps[off] & bit) != 0;
2916 case 8:
2917 return tmps[off];
2918 case 16:
2919 off <<= 1;
2920 return (tmps[off] << 8) + tmps[off + 1] ;
2921 case 32:
2922 off <<= 2;
2923 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
2924 }
5637ef5b
NC
2925 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2926 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2927 NORETURN_FUNCTION_END;
a0ed51b3 2928}
2b9d42f0 2929
319009ee
KW
2930/* Read a single line of the main body of the swash input text. These are of
2931 * the form:
2932 * 0053 0056 0073
2933 * where each number is hex. The first two numbers form the minimum and
2934 * maximum of a range, and the third is the value associated with the range.
2935 * Not all swashes should have a third number
2936 *
2937 * On input: l points to the beginning of the line to be examined; it points
2938 * to somewhere in the string of the whole input text, and is
2939 * terminated by a \n or the null string terminator.
2940 * lend points to the null terminator of that string
2941 * wants_value is non-zero if the swash expects a third number
2942 * typestr is the name of the swash's mapping, like 'ToLower'
2943 * On output: *min, *max, and *val are set to the values read from the line.
2944 * returns a pointer just beyond the line examined. If there was no
2945 * valid min number on the line, returns lend+1
2946 */
2947
2948STATIC U8*
2949S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2950 const bool wants_value, const U8* const typestr)
2951{
2952 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2953 STRLEN numlen; /* Length of the number */
02470786
KW
2954 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2955 | PERL_SCAN_DISALLOW_PREFIX
2956 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2957
2958 /* nl points to the next \n in the scan */
2959 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2960
2961 /* Get the first number on the line: the range minimum */
2962 numlen = lend - l;
2963 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2964 if (numlen) /* If found a hex number, position past it */
2965 l += numlen;
2966 else if (nl) { /* Else, go handle next line, if any */
2967 return nl + 1; /* 1 is length of "\n" */
2968 }
2969 else { /* Else, no next line */
2970 return lend + 1; /* to LIST's end at which \n is not found */
2971 }
2972
2973 /* The max range value follows, separated by a BLANK */
2974 if (isBLANK(*l)) {
2975 ++l;
02470786
KW
2976 flags = PERL_SCAN_SILENT_ILLDIGIT
2977 | PERL_SCAN_DISALLOW_PREFIX
2978 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2979 numlen = lend - l;
2980 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2981 if (numlen)
2982 l += numlen;
2983 else /* If no value here, it is a single element range */
2984 *max = *min;
2985
2986 /* Non-binary tables have a third entry: what the first element of the
24303724 2987 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2988 if (wants_value) {
2989 if (isBLANK(*l)) {
2990 ++l;
f2a7d0fc
KW
2991 flags = PERL_SCAN_SILENT_ILLDIGIT
2992 | PERL_SCAN_DISALLOW_PREFIX
2993 | PERL_SCAN_SILENT_NON_PORTABLE;
2994 numlen = lend - l;
2995 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2996 if (numlen)
2997 l += numlen;
2998 else
2999 *val = 0;
319009ee
KW
3000 }
3001 else {
3002 *val = 0;
3003 if (typeto) {
dcbac5bb 3004 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3005 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3006 typestr, l);
3007 }
3008 }
3009 }
3010 else
3011 *val = 0; /* bits == 1, then any val should be ignored */
3012 }
3013 else { /* Nothing following range min, should be single element with no
3014 mapping expected */
3015 *max = *min;
3016 if (wants_value) {
3017 *val = 0;
3018 if (typeto) {
dcbac5bb 3019 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3020 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3021 }
3022 }
3023 else
3024 *val = 0; /* bits == 1, then val should be ignored */
3025 }
3026
3027 /* Position to next line if any, or EOF */
3028 if (nl)
3029 l = nl + 1;
3030 else
3031 l = lend;
3032
3033 return l;
3034}
3035
979f2922
ST
3036/* Note:
3037 * Returns a swatch (a bit vector string) for a code point sequence
3038 * that starts from the value C<start> and comprises the number C<span>.
3039 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3040 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3041 */
3042STATIC SV*
b0e3252e 3043S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3044{
3045 SV *swatch;
77f9f126 3046 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3047 STRLEN lcur, xcur, scur;
ef8f7699 3048 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3049 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3050
88d45d28
KW
3051 SV** listsvp = NULL; /* The string containing the main body of the table */
3052 SV** extssvp = NULL;
3053 SV** invert_it_svp = NULL;
3054 U8* typestr = NULL;
786861f5
KW
3055 STRLEN bits;
3056 STRLEN octets; /* if bits == 1, then octets == 0 */
3057 UV none;
3058 UV end = start + span;
972dd592 3059
36eb48b4 3060 if (invlistsvp == NULL) {
786861f5
KW
3061 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3062 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3063 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3064 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3065 listsvp = hv_fetchs(hv, "LIST", FALSE);
3066 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3067
3068 bits = SvUV(*bitssvp);
3069 none = SvUV(*nonesvp);
3070 typestr = (U8*)SvPV_nolen(*typesvp);
3071 }
36eb48b4
KW
3072 else {
3073 bits = 1;
3074 none = 0;
3075 }
786861f5 3076 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3077
b0e3252e 3078 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3079
979f2922 3080 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3081 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3082 (UV)bits);
979f2922
ST
3083 }
3084
84ea5ef6
KW
3085 /* If overflowed, use the max possible */
3086 if (end < start) {
3087 end = UV_MAX;
3088 span = end - start;
3089 }
3090
979f2922 3091 /* create and initialize $swatch */
979f2922 3092 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3093 swatch = newSV(scur);
3094 SvPOK_on(swatch);
979f2922
ST
3095 s = (U8*)SvPVX(swatch);
3096 if (octets && none) {
0bd48802 3097 const U8* const e = s + scur;
979f2922
ST
3098 while (s < e) {
3099 if (bits == 8)
3100 *s++ = (U8)(none & 0xff);
3101 else if (bits == 16) {
3102 *s++ = (U8)((none >> 8) & 0xff);
3103 *s++ = (U8)( none & 0xff);
3104 }
3105 else if (bits == 32) {
3106 *s++ = (U8)((none >> 24) & 0xff);
3107 *s++ = (U8)((none >> 16) & 0xff);
3108 *s++ = (U8)((none >> 8) & 0xff);
3109 *s++ = (U8)( none & 0xff);
3110 }
3111 }
3112 *s = '\0';
3113 }
3114 else {
3115 (void)memzero((U8*)s, scur + 1);
3116 }
3117 SvCUR_set(swatch, scur);
3118 s = (U8*)SvPVX(swatch);
3119
36eb48b4
KW
3120 if (invlistsvp) { /* If has an inversion list set up use that */
3121 _invlist_populate_swatch(*invlistsvp, start, end, s);
3122 return swatch;
3123 }
3124
3125 /* read $swash->{LIST} */
979f2922
ST
3126 l = (U8*)SvPV(*listsvp, lcur);
3127 lend = l + lcur;
3128 while (l < lend) {
8ed25d53 3129 UV min, max, val, upper;
319009ee
KW
3130 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3131 cBOOL(octets), typestr);
3132 if (l > lend) {
979f2922
ST
3133 break;
3134 }
3135
972dd592 3136 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3137 if (max < start)
3138 continue;
3139
8ed25d53
KW
3140 /* <end> is generally 1 beyond where we want to set things, but at the
3141 * platform's infinity, where we can't go any higher, we want to
3142 * include the code point at <end> */
3143 upper = (max < end)
3144 ? max
3145 : (max != UV_MAX || end != UV_MAX)
3146 ? end - 1
3147 : end;
3148
979f2922 3149 if (octets) {
35da51f7 3150 UV key;
979f2922
ST
3151 if (min < start) {
3152 if (!none || val < none) {
3153 val += start - min;
3154 }
3155 min = start;
3156 }
8ed25d53 3157 for (key = min; key <= upper; key++) {
979f2922 3158 STRLEN offset;
979f2922
ST
3159 /* offset must be non-negative (start <= min <= key < end) */
3160 offset = octets * (key - start);
3161 if (bits == 8)
3162 s[offset] = (U8)(val & 0xff);
3163 else if (bits == 16) {
3164 s[offset ] = (U8)((val >> 8) & 0xff);
3165 s[offset + 1] = (U8)( val & 0xff);
3166 }
3167 else if (bits == 32) {
3168 s[offset ] = (U8)((val >> 24) & 0xff);
3169 s[offset + 1] = (U8)((val >> 16) & 0xff);
3170 s[offset + 2] = (U8)((val >> 8) & 0xff);
3171 s[offset + 3] = (U8)( val & 0xff);
3172 }
3173
3174 if (!none || val < none)
3175 ++val;
3176 }
3177 }
711a919c 3178 else { /* bits == 1, then val should be ignored */
35da51f7 3179 UV key;
979f2922
ST
3180 if (min < start)
3181 min = start;
6cb05c12 3182
8ed25d53 3183 for (key = min; key <= upper; key++) {
0bd48802 3184 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3185 s[offset >> 3] |= 1 << (offset & 7);
3186 }
3187 }
3188 } /* while */
979f2922 3189
9479a769 3190 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3191 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3192
3193 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3194 * be 0, and their inversion should also be 0, as we don't succeed any
3195 * Unicode property matches for non-Unicode code points */
3196 if (start <= PERL_UNICODE_MAX) {
3197
3198 /* The code below assumes that we never cross the
3199 * Unicode/above-Unicode boundary in a range, as otherwise we would
3200 * have to figure out where to stop flipping the bits. Since this
3201 * boundary is divisible by a large power of 2, and swatches comes
3202 * in small powers of 2, this should be a valid assumption */
3203 assert(start + span - 1 <= PERL_UNICODE_MAX);
3204
507a8485
KW
3205 send = s + scur;
3206 while (s < send) {
3207 *s = ~(*s);
3208 s++;
3209 }
0bda3001 3210 }
77f9f126
KW
3211 }
3212
d73c39c5
KW
3213 /* read $swash->{EXTRAS}
3214 * This code also copied to swash_to_invlist() below */
979f2922
ST
3215 x = (U8*)SvPV(*extssvp, xcur);
3216 xend = x + xcur;
3217 while (x < xend) {
3218 STRLEN namelen;
3219 U8 *namestr;
3220 SV** othersvp;
3221 HV* otherhv;
3222 STRLEN otherbits;
3223 SV **otherbitssvp, *other;
711a919c 3224 U8 *s, *o, *nl;
979f2922
ST
3225 STRLEN slen, olen;
3226
35da51f7 3227 const U8 opc = *x++;
979f2922
ST
3228 if (opc == '\n')
3229 continue;
3230
3231 nl = (U8*)memchr(x, '\n', xend - x);
3232
3233 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3234 if (nl) {
3235 x = nl + 1; /* 1 is length of "\n" */
3236 continue;
3237 }
3238 else {
3239 x = xend; /* to EXTRAS' end at which \n is not found */
3240 break;
3241 }
3242 }
3243
3244 namestr = x;
3245 if (nl) {
3246 namelen = nl - namestr;
3247 x = nl + 1;
3248 }
3249 else {
3250 namelen = xend - namestr;
3251 x = xend;
3252 }
3253
3254 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3255 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3256 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3257 otherbits = (STRLEN)SvUV(*otherbitssvp);
3258 if (bits < otherbits)
5637ef5b
NC
3259 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3260 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3261
3262 /* The "other" swatch must be destroyed after. */
b0e3252e 3263 other = swatch_get(*othersvp, start, span);
979f2922
ST
3264 o = (U8*)SvPV(other, olen);
3265
3266 if (!olen)
b0e3252e 3267 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3268
3269 s = (U8*)SvPV(swatch, slen);
3270 if (bits == 1 && otherbits == 1) {
3271 if (slen != olen)
5637ef5b
NC
3272 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3273 "mismatch, slen=%"UVuf", olen=%"UVuf,
3274 (UV)slen, (UV)olen);
979f2922
ST
3275
3276 switch (opc) {
3277 case '+':
3278 while (slen--)
3279 *s++ |= *o++;
3280 break;
3281 case '!':
3282 while (slen--)
3283 *s++ |= ~*o++;
3284 break;
3285 case '-':
3286 while (slen--)
3287 *s++ &= ~*o++;
3288 break;
3289 case '&':
3290 while (slen--)
3291 *s++ &= *o++;
3292 break;
3293 default:
3294 break;
3295 }
3296 }
711a919c 3297 else {
979f2922
ST
3298 STRLEN otheroctets = otherbits >> 3;
3299 STRLEN offset = 0;
35da51f7 3300 U8* const send = s + slen;
979f2922
ST
3301
3302 while (s < send) {
3303 UV otherval = 0;
3304
3305 if (otherbits == 1) {
3306 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3307 ++offset;
3308 }
3309 else {
3310 STRLEN vlen = otheroctets;
3311 otherval = *o++;
3312 while (--vlen) {
3313 otherval <<= 8;
3314 otherval |= *o++;
3315 }
3316 }
3317
711a919c 3318 if (opc == '+' && otherval)
6f207bd3 3319 NOOP; /* replace with otherval */
979f2922
ST
3320 else if (opc == '!' && !otherval)
3321 otherval = 1;
3322 else if (opc == '-' && otherval)
3323 otherval = 0;
3324 else if (opc == '&' && !otherval)
3325 otherval = 0;
3326 else {
711a919c 3327 s += octets; /* no replacement */
979f2922
ST
3328 continue;
3329 }
3330
3331 if (bits == 8)
3332 *s++ = (U8)( otherval & 0xff);
3333 else if (bits == 16) {
3334 *s++ = (U8)((otherval >> 8) & 0xff);
3335 *s++ = (U8)( otherval & 0xff);
3336 }
3337 else if (bits == 32) {
3338 *s++ = (U8)((otherval >> 24) & 0xff);
3339 *s++ = (U8)((otherval >> 16) & 0xff);
3340 *s++ = (U8)((otherval >> 8) & 0xff);
3341 *s++ = (U8)( otherval & 0xff);
3342 }
3343 }
3344 }
3345 sv_free(other); /* through with it! */
3346 } /* while */
3347 return swatch;
3348}
3349
064c021d 3350HV*
4c2e1131 3351Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3352{
3353
79a2a0e8 3354 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3355 * Can't be used on a property that is subject to user override, as it
3356 * relies on the value of SPECIALS in the swash which would be set by
3357 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3358 * for overridden properties
064c021d
KW
3359 *
3360 * Returns a hash which is the inversion and closure of a swash mapping.
3361 * For example, consider the input lines:
3362 * 004B 006B
3363 * 004C 006C
3364 * 212A 006B
3365 *
3366 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3367 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3368 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3369 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3370 *
3371 * Essentially, for any code point, it gives all the code points that map to
3372 * it, or the list of 'froms' for that point.
3373 *
5662e334
KW
3374 * Currently it ignores any additions or deletions from other swashes,
3375 * looking at just the main body of the swash, and if there are SPECIALS
3376 * in the swash, at that hash
3377 *
3378 * The specials hash can be extra code points, and most likely consists of
3379 * maps from single code points to multiple ones (each expressed as a string
3380 * of utf8 characters). This function currently returns only 1-1 mappings.
3381 * However consider this possible input in the specials hash:
3382 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3383 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3384 *
3385 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3386 * currently handle. But it also means that FB05 and FB06 are equivalent in
3387 * a 1-1 mapping which we should handle, and this relationship may not be in
3388 * the main table. Therefore this function examines all the multi-char
3389 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3390
3391 U8 *l, *lend;
3392 STRLEN lcur;
3393 HV *const hv = MUTABLE_HV(SvRV(swash));
3394
923b6d4e
KW
3395 /* The string containing the main body of the table. This will have its
3396 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3397 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3398
3399 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3400 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3401 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3402 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3403 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3404 const STRLEN bits = SvUV(*bitssvp);
3405 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3406 const UV none = SvUV(*nonesvp);
5662e334 3407 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3408
3409 HV* ret = newHV();
3410
3411 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3412
3413 /* Must have at least 8 bits to get the mappings */
3414 if (bits != 8 && bits != 16 && bits != 32) {
3415 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3416 (UV)bits);
3417 }
3418
5662e334
KW
3419 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3420 mapping to more than one character */
3421
3422 /* Construct an inverse mapping hash for the specials */
3423 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3424 HV * specials_inverse = newHV();
3425 char *char_from; /* the lhs of the map */
3426 I32 from_len; /* its byte length */
3427 char *char_to; /* the rhs of the map */
3428 I32 to_len; /* its byte length */
3429 SV *sv_to; /* and in a sv */
3430 AV* from_list; /* list of things that map to each 'to' */
3431
3432 hv_iterinit(specials_hv);
3433
3434 /* The keys are the characters (in utf8) that map to the corresponding
3435 * utf8 string value. Iterate through the list creating the inverse
3436 * list. */
3437 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3438 SV** listp;
3439 if (! SvPOK(sv_to)) {
5637ef5b
NC
3440 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3441 "unexpectedly is not a string, flags=%lu",
3442 (unsigned long)SvFLAGS(sv_to));
5662e334 3443 }
4b88fb76 3444 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %"UVXf", First char of to is %"UVXf"\n", valid_utf8_to_uvchr((U8*) char_from, 0), valid_utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
5662e334
KW
3445
3446 /* Each key in the inverse list is a mapped-to value, and the key's
3447 * hash value is a list of the strings (each in utf8) that map to
3448 * it. Those strings are all one character long */
3449 if ((listp = hv_fetch(specials_inverse,
3450 SvPVX(sv_to),
3451 SvCUR(sv_to), 0)))
3452 {
3453 from_list = (AV*) *listp;
3454 }
3455 else { /* No entry yet for it: create one */
3456 from_list = newAV();
3457 if (! hv_store(specials_inverse,
3458 SvPVX(sv_to),
3459 SvCUR(sv_to),
3460 (SV*) from_list, 0))
3461 {
3462 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3463 }
3464 }
3465
3466 /* Here have the list associated with this 'to' (perhaps newly
3467 * created and empty). Just add to it. Note that we ASSUME that
3468 * the input is guaranteed to not have duplications, so we don't
3469 * check for that. Duplications just slow down execution time. */
3470 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3471 }
3472
3473 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3474 * it looking for cases like the FB05/FB06 examples above. There would
3475 * be an entry in the hash like
3476 * 'st' => [ FB05, FB06 ]
3477 * In this example we will create two lists that get stored in the
3478 * returned hash, 'ret':
3479 * FB05 => [ FB05, FB06 ]
3480 * FB06 => [ FB05, FB06 ]
3481 *
3482 * Note that there is nothing to do if the array only has one element.
3483 * (In the normal 1-1 case handled below, we don't have to worry about
3484 * two lists, as everything gets tied to the single list that is
3485 * generated for the single character 'to'. But here, we are omitting
3486 * that list, ('st' in the example), so must have multiple lists.) */
3487 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3488 &char_to, &to_len)))
3489 {
3490 if (av_len(from_list) > 0) {
c70927a6 3491 SSize_t i;
5662e334
KW
3492
3493 /* We iterate over all combinations of i,j to place each code
3494 * point on each list */
3495 for (i = 0; i <= av_len(from_list); i++) {
c70927a6 3496 SSize_t j;
5662e334
KW
3497 AV* i_list = newAV();
3498 SV** entryp = av_fetch(from_list, i, FALSE);
3499 if (entryp == NULL) {
3500 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3501 }
3502 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3503 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3504 }
3505 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3506 (SV*) i_list, FALSE))
3507 {
3508 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3509 }
3510
4b88fb76 3511 /* For debugging: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
5662e334
KW
3512 for (j = 0; j <= av_len(from_list); j++) {
3513 entryp = av_fetch(from_list, j, FALSE);
3514 if (entryp == NULL) {
3515 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3516 }
3517
3518 /* When i==j this adds itself to the list */
4b88fb76
KW
3519 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3520 (U8*) SvPVX(*entryp),
3521 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3522 0)));
4637d003 3523 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
5662e334
KW
3524 }
3525 }
3526 }
3527 }
3528 SvREFCNT_dec(specials_inverse); /* done with it */
3529 } /* End of specials */
3530
064c021d
KW
3531 /* read $swash->{LIST} */
3532 l = (U8*)SvPV(*listsvp, lcur);
3533 lend = l + lcur;
3534
3535 /* Go through each input line */
3536 while (l < lend) {
3537 UV min, max, val;
3538 UV inverse;
3539 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3540 cBOOL(octets), typestr);
3541 if (l > lend) {
3542 break;
3543 }
3544
3545 /* Each element in the range is to be inverted */
3546 for (inverse = min; inverse <= max; inverse++) {
3547 AV* list;
064c021d
KW
3548 SV** listp;
3549 IV i;
3550 bool found_key = FALSE;
5662e334 3551 bool found_inverse = FALSE;
064c021d
KW
3552
3553 /* The key is the inverse mapping */
3554 char key[UTF8_MAXBYTES+1];
c80e42f3 3555 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3556 STRLEN key_len = key_end - key;
3557
064c021d
KW
3558 /* Get the list for the map */
3559 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3560 list = (AV*) *listp;
3561 }
3562 else { /* No entry yet for it: create one */
3563 list = newAV();
3564 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3565 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3566 }
3567 }
3568
5662e334
KW
3569 /* Look through list to see if this inverse mapping already is
3570 * listed, or if there is a mapping to itself already */
508f7cfa 3571 for (i = 0; i <= av_len(list); i++) {
064c021d
KW
3572 SV** entryp = av_fetch(list, i, FALSE);
3573 SV* entry;
3574 if (entryp == NULL) {
3575 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3576 }
3577 entry = *entryp;
5662e334 3578 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
56ca34ca 3579 if (SvUV(entry) == val) {
064c021d 3580 found_key = TRUE;
5662e334
KW
3581 }
3582 if (SvUV(entry) == inverse) {
3583 found_inverse = TRUE;
3584 }
3585
3586 /* No need to continue searching if found everything we are
3587 * looking for */
3588 if (found_key && found_inverse) {
064c021d
KW
3589 break;
3590 }
3591 }
56ca34ca
KW
3592
3593 /* Make sure there is a mapping to itself on the list */
064c021d 3594 if (! found_key) {
d397ff6a 3595 av_push(list, newSVuv(val));
4637d003 3596 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3597 }
3598
3599
3600 /* Simply add the value to the list */
5662e334
KW
3601 if (! found_inverse) {
3602 av_push(list, newSVuv(inverse));
4637d003 3603 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3604 }
064c021d 3605
b0e3252e 3606 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3607 * range. That makes more compact tables possible. You can
3608 * express the capitalization, for example, of all consecutive
3609 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3610 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3611 * and it's not documented; it appears to be used only in
b0e3252e 3612 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3613 * in case */
064c021d
KW
3614 if (!none || val < none) {
3615 ++val;
3616 }
3617 }
3618 }
3619
3620 return ret;
3621}
3622
a25abddc 3623SV*
d764b54e
KW
3624Perl__swash_to_invlist(pTHX_ SV* const swash)
3625{
3626
ed92f1b3
KW
3627 /* Subject to change or removal. For use only in one place in regcomp.c.
3628 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3629
3630 U8 *l, *lend;
3631 char *loc;
3632 STRLEN lcur;
3633 HV *const hv = MUTABLE_HV(SvRV(swash));
3634 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3635 U8 empty[] = "";
923b6d4e
KW
3636 SV** listsvp;
3637 SV** typesvp;
3638 SV** bitssvp;
3639 SV** extssvp;
3640 SV** invert_it_svp;
d764b54e 3641
923b6d4e
KW
3642 U8* typestr;
3643 STRLEN bits;
3644 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3645 U8 *x, *xend;
3646 STRLEN xcur;
d764b54e 3647
a25abddc 3648 SV* invlist;
d764b54e 3649
b81740c0
KW
3650 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3651
923b6d4e
KW
3652 /* If not a hash, it must be the swash's inversion list instead */
3653 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3654 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3655 }
3656
3657 /* The string containing the main body of the table */
3658 listsvp = hv_fetchs(hv, "LIST", FALSE);
3659 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3660 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3661 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3662 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3663
3664 typestr = (U8*)SvPV_nolen(*typesvp);
3665 bits = SvUV(*bitssvp);
3666 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3667
d764b54e 3668 /* read $swash->{LIST} */
b443038a
KW
3669 if (SvPOK(*listsvp)) {
3670 l = (U8*)SvPV(*listsvp, lcur);
3671 }
3672 else {
3673 /* LIST legitimately doesn't contain a string during compilation phases
3674 * of Perl itself, before the Unicode tables are generated. In this
3675 * case, just fake things up by creating an empty list */
3676 l = empty;
3677 lcur = 0;
3678 }
d764b54e
KW
3679 loc = (char *) l;
3680 lend = l + lcur;
3681
31aa6e0b
KW
3682 if (*l == 'V') { /* Inversion list format */
3683 char *after_strtol = (char *) lend;
3684 UV element0;
3685 UV* other_elements_ptr;
3686
3687 /* The first number is a count of the rest */
3688 l++;
3689 elements = Strtoul((char *)l, &after_strtol, 10);
3690 l = (U8 *) after_strtol;
3691
3692 /* Get the 0th element, which is needed to setup the inversion list */
3693 element0 = (UV) Strtoul((char *)l, &after_strtol, 10);
3694 l = (U8 *) after_strtol;
3695 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3696 elements--;
3697
3698 /* Then just populate the rest of the input */
3699 while (elements-- > 0) {
3700 if (l > lend) {
3701 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3702 }
3703 *other_elements_ptr++ = (UV) Strtoul((char *)l, &after_strtol, 10);
3704 l = (U8 *) after_strtol;
3705 }
3706 }
3707 else {
3708
1784d2f9
KW
3709 /* Scan the input to count the number of lines to preallocate array
3710 * size based on worst possible case, which is each line in the input
3711 * creates 2 elements in the inversion list: 1) the beginning of a
3712 * range in the list; 2) the beginning of a range not in the list. */
3713 while ((loc = (strchr(loc, '\n'))) != NULL) {
3714 elements += 2;
3715 loc++;
3716 }
d764b54e 3717
1784d2f9
KW
3718 /* If the ending is somehow corrupt and isn't a new line, add another
3719 *